diff --git "a/openai_whisper-base.en/AudioEncoder.mlmodelc/model.mil" "b/openai_whisper-base.en/AudioEncoder.mlmodelc/model.mil" new file mode 100644--- /dev/null +++ "b/openai_whisper-base.en/AudioEncoder.mlmodelc/model.mil" @@ -0,0 +1,3351 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3401.3.1"}, {"coremlc-version", "3401.4.1"}, {"coremltools-component-torch", "2.5.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] +{ + func main(tensor melspectrogram_features) { + string var_66_pad_type_0 = const()[name = string("op_66_pad_type_0"), val = string("custom")]; + tensor var_66_pad_0 = const()[name = string("op_66_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_66_strides_0 = const()[name = string("op_66_strides_0"), val = tensor([1, 1])]; + tensor var_66_dilations_0 = const()[name = string("op_66_dilations_0"), val = tensor([1, 1])]; + int32 var_66_groups_0 = const()[name = string("op_66_groups_0"), val = int32(1)]; + tensor var_41_to_fp16 = const()[name = string("op_41_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; + tensor var_47_to_fp16 = const()[name = string("op_47_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(245888)))]; + tensor var_66_cast_fp16 = conv(bias = var_47_to_fp16, dilations = var_66_dilations_0, groups = var_66_groups_0, pad = var_66_pad_0, pad_type = var_66_pad_type_0, strides = var_66_strides_0, weight = var_41_to_fp16, x = melspectrogram_features)[name = string("op_66_cast_fp16")]; + string hidden_states_1_mode_0 = const()[name = string("hidden_states_1_mode_0"), val = string("EXACT")]; + tensor hidden_states_1_cast_fp16 = gelu(mode = hidden_states_1_mode_0, x = var_66_cast_fp16)[name = string("hidden_states_1_cast_fp16")]; + string var_106_pad_type_0 = const()[name = string("op_106_pad_type_0"), val = string("custom")]; + tensor var_106_pad_0 = const()[name = string("op_106_pad_0"), val = tensor([0, 0, 1, 1])]; + tensor var_106_strides_0 = const()[name = string("op_106_strides_0"), val = tensor([2, 2])]; + tensor var_106_dilations_0 = const()[name = string("op_106_dilations_0"), val = tensor([1, 1])]; + int32 var_106_groups_0 = const()[name = string("op_106_groups_0"), val = int32(1)]; + tensor var_81_to_fp16 = const()[name = string("op_81_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(246976)))]; + tensor var_87_to_fp16 = const()[name = string("op_87_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1819904)))]; + tensor var_106_cast_fp16 = conv(bias = var_87_to_fp16, dilations = var_106_dilations_0, groups = var_106_groups_0, pad = var_106_pad_0, pad_type = var_106_pad_type_0, strides = var_106_strides_0, weight = var_81_to_fp16, x = hidden_states_1_cast_fp16)[name = string("op_106_cast_fp16")]; + string hidden_states_3_mode_0 = const()[name = string("hidden_states_3_mode_0"), val = string("EXACT")]; + tensor hidden_states_3_cast_fp16 = gelu(mode = hidden_states_3_mode_0, x = var_106_cast_fp16)[name = string("hidden_states_3_cast_fp16")]; + tensor var_124_to_fp16 = const()[name = string("op_124_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1820992)))]; + tensor inputs_1_cast_fp16 = add(x = hidden_states_3_cast_fp16, y = var_124_to_fp16)[name = string("inputs_1_cast_fp16")]; + int32 var_134 = const()[name = string("op_134"), val = int32(3)]; + int32 var_147 = const()[name = string("op_147"), val = int32(1)]; + tensor out_1_axes_0 = const()[name = string("out_1_axes_0"), val = tensor([1])]; + fp16 var_164_to_fp16 = const()[name = string("op_164_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_164_to_fp16, x = inputs_1_cast_fp16)[name = string("out_1_cast_fp16")]; + tensor obj_1_mean_0_to_fp16 = const()[name = string("obj_1_mean_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3357056)))]; + tensor obj_1_variance_0_to_fp16 = const()[name = string("obj_1_variance_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3358144)))]; + tensor obj_1_gamma_0_to_fp16 = const()[name = string("obj_1_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3359232)))]; + tensor obj_1_beta_0_to_fp16 = const()[name = string("obj_1_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3360320)))]; + fp16 obj_1_epsilon_0_to_fp16 = const()[name = string("obj_1_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_1_cast_fp16 = batch_norm(beta = obj_1_beta_0_to_fp16, epsilon = obj_1_epsilon_0_to_fp16, gamma = obj_1_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_1_cast_fp16)[name = string("obj_1_cast_fp16")]; + string query_1_pad_type_0 = const()[name = string("query_1_pad_type_0"), val = string("valid")]; + tensor query_1_strides_0 = const()[name = string("query_1_strides_0"), val = tensor([1, 1])]; + tensor query_1_pad_0 = const()[name = string("query_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_1_dilations_0 = const()[name = string("query_1_dilations_0"), val = tensor([1, 1])]; + int32 query_1_groups_0 = const()[name = string("query_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3361408)))]; + tensor layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3885760)))]; + tensor query_1_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_bias_to_fp16, dilations = query_1_dilations_0, groups = query_1_groups_0, pad = query_1_pad_0, pad_type = query_1_pad_type_0, strides = query_1_strides_0, weight = layers_0_self_attn_q_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("query_1_cast_fp16")]; + string key_1_pad_type_0 = const()[name = string("key_1_pad_type_0"), val = string("valid")]; + tensor key_1_strides_0 = const()[name = string("key_1_strides_0"), val = tensor([1, 1])]; + tensor key_1_pad_0 = const()[name = string("key_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_1_dilations_0 = const()[name = string("key_1_dilations_0"), val = tensor([1, 1])]; + int32 key_1_groups_0 = const()[name = string("key_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3886848)))]; + tensor key_1_cast_fp16 = conv(dilations = key_1_dilations_0, groups = key_1_groups_0, pad = key_1_pad_0, pad_type = key_1_pad_type_0, strides = key_1_strides_0, weight = layers_0_self_attn_k_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("key_1_cast_fp16")]; + string value_1_pad_type_0 = const()[name = string("value_1_pad_type_0"), val = string("valid")]; + tensor value_1_strides_0 = const()[name = string("value_1_strides_0"), val = tensor([1, 1])]; + tensor value_1_pad_0 = const()[name = string("value_1_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_1_dilations_0 = const()[name = string("value_1_dilations_0"), val = tensor([1, 1])]; + int32 value_1_groups_0 = const()[name = string("value_1_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4411200)))]; + tensor layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4935552)))]; + tensor value_1_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_bias_to_fp16, dilations = value_1_dilations_0, groups = value_1_groups_0, pad = value_1_pad_0, pad_type = value_1_pad_type_0, strides = value_1_strides_0, weight = layers_0_self_attn_v_proj_weight_to_fp16, x = obj_1_cast_fp16)[name = string("value_1_cast_fp16")]; + tensor var_202_begin_0 = const()[name = string("op_202_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_202_end_0 = const()[name = string("op_202_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_202_end_mask_0 = const()[name = string("op_202_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_202_cast_fp16 = slice_by_index(begin = var_202_begin_0, end = var_202_end_0, end_mask = var_202_end_mask_0, x = query_1_cast_fp16)[name = string("op_202_cast_fp16")]; + tensor var_206_begin_0 = const()[name = string("op_206_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_206_end_0 = const()[name = string("op_206_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_206_end_mask_0 = const()[name = string("op_206_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_206_cast_fp16 = slice_by_index(begin = var_206_begin_0, end = var_206_end_0, end_mask = var_206_end_mask_0, x = query_1_cast_fp16)[name = string("op_206_cast_fp16")]; + tensor var_210_begin_0 = const()[name = string("op_210_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_210_end_0 = const()[name = string("op_210_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_210_end_mask_0 = const()[name = string("op_210_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_210_cast_fp16 = slice_by_index(begin = var_210_begin_0, end = var_210_end_0, end_mask = var_210_end_mask_0, x = query_1_cast_fp16)[name = string("op_210_cast_fp16")]; + tensor var_214_begin_0 = const()[name = string("op_214_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_214_end_0 = const()[name = string("op_214_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_214_end_mask_0 = const()[name = string("op_214_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_214_cast_fp16 = slice_by_index(begin = var_214_begin_0, end = var_214_end_0, end_mask = var_214_end_mask_0, x = query_1_cast_fp16)[name = string("op_214_cast_fp16")]; + tensor var_218_begin_0 = const()[name = string("op_218_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_218_end_0 = const()[name = string("op_218_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_218_end_mask_0 = const()[name = string("op_218_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_218_cast_fp16 = slice_by_index(begin = var_218_begin_0, end = var_218_end_0, end_mask = var_218_end_mask_0, x = query_1_cast_fp16)[name = string("op_218_cast_fp16")]; + tensor var_222_begin_0 = const()[name = string("op_222_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_222_end_0 = const()[name = string("op_222_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_222_end_mask_0 = const()[name = string("op_222_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_222_cast_fp16 = slice_by_index(begin = var_222_begin_0, end = var_222_end_0, end_mask = var_222_end_mask_0, x = query_1_cast_fp16)[name = string("op_222_cast_fp16")]; + tensor var_226_begin_0 = const()[name = string("op_226_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_226_end_0 = const()[name = string("op_226_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_226_end_mask_0 = const()[name = string("op_226_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_226_cast_fp16 = slice_by_index(begin = var_226_begin_0, end = var_226_end_0, end_mask = var_226_end_mask_0, x = query_1_cast_fp16)[name = string("op_226_cast_fp16")]; + tensor var_230_begin_0 = const()[name = string("op_230_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_230_end_0 = const()[name = string("op_230_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_230_end_mask_0 = const()[name = string("op_230_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_230_cast_fp16 = slice_by_index(begin = var_230_begin_0, end = var_230_end_0, end_mask = var_230_end_mask_0, x = query_1_cast_fp16)[name = string("op_230_cast_fp16")]; + tensor var_239_begin_0 = const()[name = string("op_239_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_239_end_0 = const()[name = string("op_239_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_239_end_mask_0 = const()[name = string("op_239_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_239_cast_fp16 = slice_by_index(begin = var_239_begin_0, end = var_239_end_0, end_mask = var_239_end_mask_0, x = var_202_cast_fp16)[name = string("op_239_cast_fp16")]; + tensor var_246_begin_0 = const()[name = string("op_246_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_246_end_0 = const()[name = string("op_246_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_246_end_mask_0 = const()[name = string("op_246_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_246_cast_fp16 = slice_by_index(begin = var_246_begin_0, end = var_246_end_0, end_mask = var_246_end_mask_0, x = var_202_cast_fp16)[name = string("op_246_cast_fp16")]; + tensor var_253_begin_0 = const()[name = string("op_253_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_253_end_0 = const()[name = string("op_253_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_253_end_mask_0 = const()[name = string("op_253_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_253_cast_fp16 = slice_by_index(begin = var_253_begin_0, end = var_253_end_0, end_mask = var_253_end_mask_0, x = var_202_cast_fp16)[name = string("op_253_cast_fp16")]; + tensor var_260_begin_0 = const()[name = string("op_260_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_260_end_0 = const()[name = string("op_260_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_260_end_mask_0 = const()[name = string("op_260_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_260_cast_fp16 = slice_by_index(begin = var_260_begin_0, end = var_260_end_0, end_mask = var_260_end_mask_0, x = var_202_cast_fp16)[name = string("op_260_cast_fp16")]; + tensor var_267_begin_0 = const()[name = string("op_267_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_267_end_0 = const()[name = string("op_267_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_267_end_mask_0 = const()[name = string("op_267_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_267_cast_fp16 = slice_by_index(begin = var_267_begin_0, end = var_267_end_0, end_mask = var_267_end_mask_0, x = var_206_cast_fp16)[name = string("op_267_cast_fp16")]; + tensor var_274_begin_0 = const()[name = string("op_274_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_274_end_0 = const()[name = string("op_274_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_274_end_mask_0 = const()[name = string("op_274_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_274_cast_fp16 = slice_by_index(begin = var_274_begin_0, end = var_274_end_0, end_mask = var_274_end_mask_0, x = var_206_cast_fp16)[name = string("op_274_cast_fp16")]; + tensor var_281_begin_0 = const()[name = string("op_281_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_281_end_0 = const()[name = string("op_281_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_281_end_mask_0 = const()[name = string("op_281_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_281_cast_fp16 = slice_by_index(begin = var_281_begin_0, end = var_281_end_0, end_mask = var_281_end_mask_0, x = var_206_cast_fp16)[name = string("op_281_cast_fp16")]; + tensor var_288_begin_0 = const()[name = string("op_288_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_288_end_0 = const()[name = string("op_288_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_288_end_mask_0 = const()[name = string("op_288_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_288_cast_fp16 = slice_by_index(begin = var_288_begin_0, end = var_288_end_0, end_mask = var_288_end_mask_0, x = var_206_cast_fp16)[name = string("op_288_cast_fp16")]; + tensor var_295_begin_0 = const()[name = string("op_295_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_295_end_0 = const()[name = string("op_295_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_295_end_mask_0 = const()[name = string("op_295_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_295_cast_fp16 = slice_by_index(begin = var_295_begin_0, end = var_295_end_0, end_mask = var_295_end_mask_0, x = var_210_cast_fp16)[name = string("op_295_cast_fp16")]; + tensor var_302_begin_0 = const()[name = string("op_302_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_302_end_0 = const()[name = string("op_302_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_302_end_mask_0 = const()[name = string("op_302_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_302_cast_fp16 = slice_by_index(begin = var_302_begin_0, end = var_302_end_0, end_mask = var_302_end_mask_0, x = var_210_cast_fp16)[name = string("op_302_cast_fp16")]; + tensor var_309_begin_0 = const()[name = string("op_309_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_309_end_0 = const()[name = string("op_309_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_309_end_mask_0 = const()[name = string("op_309_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_309_cast_fp16 = slice_by_index(begin = var_309_begin_0, end = var_309_end_0, end_mask = var_309_end_mask_0, x = var_210_cast_fp16)[name = string("op_309_cast_fp16")]; + tensor var_316_begin_0 = const()[name = string("op_316_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_316_end_0 = const()[name = string("op_316_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_316_end_mask_0 = const()[name = string("op_316_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_316_cast_fp16 = slice_by_index(begin = var_316_begin_0, end = var_316_end_0, end_mask = var_316_end_mask_0, x = var_210_cast_fp16)[name = string("op_316_cast_fp16")]; + tensor var_323_begin_0 = const()[name = string("op_323_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_323_end_0 = const()[name = string("op_323_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_323_end_mask_0 = const()[name = string("op_323_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_323_cast_fp16 = slice_by_index(begin = var_323_begin_0, end = var_323_end_0, end_mask = var_323_end_mask_0, x = var_214_cast_fp16)[name = string("op_323_cast_fp16")]; + tensor var_330_begin_0 = const()[name = string("op_330_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_330_end_0 = const()[name = string("op_330_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_330_end_mask_0 = const()[name = string("op_330_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_330_cast_fp16 = slice_by_index(begin = var_330_begin_0, end = var_330_end_0, end_mask = var_330_end_mask_0, x = var_214_cast_fp16)[name = string("op_330_cast_fp16")]; + tensor var_337_begin_0 = const()[name = string("op_337_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_337_end_0 = const()[name = string("op_337_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_337_end_mask_0 = const()[name = string("op_337_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_337_cast_fp16 = slice_by_index(begin = var_337_begin_0, end = var_337_end_0, end_mask = var_337_end_mask_0, x = var_214_cast_fp16)[name = string("op_337_cast_fp16")]; + tensor var_344_begin_0 = const()[name = string("op_344_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_344_end_0 = const()[name = string("op_344_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_344_end_mask_0 = const()[name = string("op_344_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_344_cast_fp16 = slice_by_index(begin = var_344_begin_0, end = var_344_end_0, end_mask = var_344_end_mask_0, x = var_214_cast_fp16)[name = string("op_344_cast_fp16")]; + tensor var_351_begin_0 = const()[name = string("op_351_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_351_end_0 = const()[name = string("op_351_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_351_end_mask_0 = const()[name = string("op_351_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_351_cast_fp16 = slice_by_index(begin = var_351_begin_0, end = var_351_end_0, end_mask = var_351_end_mask_0, x = var_218_cast_fp16)[name = string("op_351_cast_fp16")]; + tensor var_358_begin_0 = const()[name = string("op_358_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_358_end_0 = const()[name = string("op_358_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_358_end_mask_0 = const()[name = string("op_358_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_358_cast_fp16 = slice_by_index(begin = var_358_begin_0, end = var_358_end_0, end_mask = var_358_end_mask_0, x = var_218_cast_fp16)[name = string("op_358_cast_fp16")]; + tensor var_365_begin_0 = const()[name = string("op_365_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_365_end_0 = const()[name = string("op_365_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_365_end_mask_0 = const()[name = string("op_365_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_365_cast_fp16 = slice_by_index(begin = var_365_begin_0, end = var_365_end_0, end_mask = var_365_end_mask_0, x = var_218_cast_fp16)[name = string("op_365_cast_fp16")]; + tensor var_372_begin_0 = const()[name = string("op_372_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_372_end_0 = const()[name = string("op_372_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_372_end_mask_0 = const()[name = string("op_372_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_372_cast_fp16 = slice_by_index(begin = var_372_begin_0, end = var_372_end_0, end_mask = var_372_end_mask_0, x = var_218_cast_fp16)[name = string("op_372_cast_fp16")]; + tensor var_379_begin_0 = const()[name = string("op_379_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_379_end_0 = const()[name = string("op_379_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_379_end_mask_0 = const()[name = string("op_379_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_379_cast_fp16 = slice_by_index(begin = var_379_begin_0, end = var_379_end_0, end_mask = var_379_end_mask_0, x = var_222_cast_fp16)[name = string("op_379_cast_fp16")]; + tensor var_386_begin_0 = const()[name = string("op_386_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_386_end_0 = const()[name = string("op_386_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_386_end_mask_0 = const()[name = string("op_386_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_386_cast_fp16 = slice_by_index(begin = var_386_begin_0, end = var_386_end_0, end_mask = var_386_end_mask_0, x = var_222_cast_fp16)[name = string("op_386_cast_fp16")]; + tensor var_393_begin_0 = const()[name = string("op_393_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_393_end_0 = const()[name = string("op_393_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_393_end_mask_0 = const()[name = string("op_393_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_393_cast_fp16 = slice_by_index(begin = var_393_begin_0, end = var_393_end_0, end_mask = var_393_end_mask_0, x = var_222_cast_fp16)[name = string("op_393_cast_fp16")]; + tensor var_400_begin_0 = const()[name = string("op_400_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_400_end_0 = const()[name = string("op_400_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_400_end_mask_0 = const()[name = string("op_400_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_400_cast_fp16 = slice_by_index(begin = var_400_begin_0, end = var_400_end_0, end_mask = var_400_end_mask_0, x = var_222_cast_fp16)[name = string("op_400_cast_fp16")]; + tensor var_407_begin_0 = const()[name = string("op_407_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_407_end_0 = const()[name = string("op_407_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_407_end_mask_0 = const()[name = string("op_407_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_407_cast_fp16 = slice_by_index(begin = var_407_begin_0, end = var_407_end_0, end_mask = var_407_end_mask_0, x = var_226_cast_fp16)[name = string("op_407_cast_fp16")]; + tensor var_414_begin_0 = const()[name = string("op_414_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_414_end_0 = const()[name = string("op_414_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_414_end_mask_0 = const()[name = string("op_414_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_414_cast_fp16 = slice_by_index(begin = var_414_begin_0, end = var_414_end_0, end_mask = var_414_end_mask_0, x = var_226_cast_fp16)[name = string("op_414_cast_fp16")]; + tensor var_421_begin_0 = const()[name = string("op_421_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_421_end_0 = const()[name = string("op_421_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_421_end_mask_0 = const()[name = string("op_421_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_421_cast_fp16 = slice_by_index(begin = var_421_begin_0, end = var_421_end_0, end_mask = var_421_end_mask_0, x = var_226_cast_fp16)[name = string("op_421_cast_fp16")]; + tensor var_428_begin_0 = const()[name = string("op_428_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_428_end_0 = const()[name = string("op_428_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_428_end_mask_0 = const()[name = string("op_428_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_428_cast_fp16 = slice_by_index(begin = var_428_begin_0, end = var_428_end_0, end_mask = var_428_end_mask_0, x = var_226_cast_fp16)[name = string("op_428_cast_fp16")]; + tensor var_435_begin_0 = const()[name = string("op_435_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_435_end_0 = const()[name = string("op_435_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_435_end_mask_0 = const()[name = string("op_435_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_435_cast_fp16 = slice_by_index(begin = var_435_begin_0, end = var_435_end_0, end_mask = var_435_end_mask_0, x = var_230_cast_fp16)[name = string("op_435_cast_fp16")]; + tensor var_442_begin_0 = const()[name = string("op_442_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_442_end_0 = const()[name = string("op_442_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_442_end_mask_0 = const()[name = string("op_442_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_442_cast_fp16 = slice_by_index(begin = var_442_begin_0, end = var_442_end_0, end_mask = var_442_end_mask_0, x = var_230_cast_fp16)[name = string("op_442_cast_fp16")]; + tensor var_449_begin_0 = const()[name = string("op_449_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_449_end_0 = const()[name = string("op_449_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_449_end_mask_0 = const()[name = string("op_449_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_449_cast_fp16 = slice_by_index(begin = var_449_begin_0, end = var_449_end_0, end_mask = var_449_end_mask_0, x = var_230_cast_fp16)[name = string("op_449_cast_fp16")]; + tensor var_456_begin_0 = const()[name = string("op_456_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_456_end_0 = const()[name = string("op_456_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_456_end_mask_0 = const()[name = string("op_456_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_456_cast_fp16 = slice_by_index(begin = var_456_begin_0, end = var_456_end_0, end_mask = var_456_end_mask_0, x = var_230_cast_fp16)[name = string("op_456_cast_fp16")]; + tensor k_1_perm_0 = const()[name = string("k_1_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_461_begin_0 = const()[name = string("op_461_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_461_end_0 = const()[name = string("op_461_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_461_end_mask_0 = const()[name = string("op_461_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_1_cast_fp16 = transpose(perm = k_1_perm_0, x = key_1_cast_fp16)[name = string("transpose_5")]; + tensor var_461_cast_fp16 = slice_by_index(begin = var_461_begin_0, end = var_461_end_0, end_mask = var_461_end_mask_0, x = k_1_cast_fp16)[name = string("op_461_cast_fp16")]; + tensor var_465_begin_0 = const()[name = string("op_465_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_465_end_0 = const()[name = string("op_465_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_465_end_mask_0 = const()[name = string("op_465_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_465_cast_fp16 = slice_by_index(begin = var_465_begin_0, end = var_465_end_0, end_mask = var_465_end_mask_0, x = k_1_cast_fp16)[name = string("op_465_cast_fp16")]; + tensor var_469_begin_0 = const()[name = string("op_469_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_469_end_0 = const()[name = string("op_469_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_469_end_mask_0 = const()[name = string("op_469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_469_cast_fp16 = slice_by_index(begin = var_469_begin_0, end = var_469_end_0, end_mask = var_469_end_mask_0, x = k_1_cast_fp16)[name = string("op_469_cast_fp16")]; + tensor var_473_begin_0 = const()[name = string("op_473_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_473_end_0 = const()[name = string("op_473_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_473_end_mask_0 = const()[name = string("op_473_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_473_cast_fp16 = slice_by_index(begin = var_473_begin_0, end = var_473_end_0, end_mask = var_473_end_mask_0, x = k_1_cast_fp16)[name = string("op_473_cast_fp16")]; + tensor var_477_begin_0 = const()[name = string("op_477_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_477_end_0 = const()[name = string("op_477_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_477_end_mask_0 = const()[name = string("op_477_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_477_cast_fp16 = slice_by_index(begin = var_477_begin_0, end = var_477_end_0, end_mask = var_477_end_mask_0, x = k_1_cast_fp16)[name = string("op_477_cast_fp16")]; + tensor var_481_begin_0 = const()[name = string("op_481_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_481_end_0 = const()[name = string("op_481_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_481_end_mask_0 = const()[name = string("op_481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_481_cast_fp16 = slice_by_index(begin = var_481_begin_0, end = var_481_end_0, end_mask = var_481_end_mask_0, x = k_1_cast_fp16)[name = string("op_481_cast_fp16")]; + tensor var_485_begin_0 = const()[name = string("op_485_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_485_end_0 = const()[name = string("op_485_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_485_end_mask_0 = const()[name = string("op_485_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_485_cast_fp16 = slice_by_index(begin = var_485_begin_0, end = var_485_end_0, end_mask = var_485_end_mask_0, x = k_1_cast_fp16)[name = string("op_485_cast_fp16")]; + tensor var_489_begin_0 = const()[name = string("op_489_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_489_end_0 = const()[name = string("op_489_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_489_end_mask_0 = const()[name = string("op_489_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_489_cast_fp16 = slice_by_index(begin = var_489_begin_0, end = var_489_end_0, end_mask = var_489_end_mask_0, x = k_1_cast_fp16)[name = string("op_489_cast_fp16")]; + tensor var_491_begin_0 = const()[name = string("op_491_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_491_end_0 = const()[name = string("op_491_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_491_end_mask_0 = const()[name = string("op_491_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_491_cast_fp16 = slice_by_index(begin = var_491_begin_0, end = var_491_end_0, end_mask = var_491_end_mask_0, x = value_1_cast_fp16)[name = string("op_491_cast_fp16")]; + tensor var_495_begin_0 = const()[name = string("op_495_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_495_end_0 = const()[name = string("op_495_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_495_end_mask_0 = const()[name = string("op_495_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_495_cast_fp16 = slice_by_index(begin = var_495_begin_0, end = var_495_end_0, end_mask = var_495_end_mask_0, x = value_1_cast_fp16)[name = string("op_495_cast_fp16")]; + tensor var_499_begin_0 = const()[name = string("op_499_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_499_end_0 = const()[name = string("op_499_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_499_end_mask_0 = const()[name = string("op_499_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_499_cast_fp16 = slice_by_index(begin = var_499_begin_0, end = var_499_end_0, end_mask = var_499_end_mask_0, x = value_1_cast_fp16)[name = string("op_499_cast_fp16")]; + tensor var_503_begin_0 = const()[name = string("op_503_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_503_end_0 = const()[name = string("op_503_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_503_end_mask_0 = const()[name = string("op_503_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_503_cast_fp16 = slice_by_index(begin = var_503_begin_0, end = var_503_end_0, end_mask = var_503_end_mask_0, x = value_1_cast_fp16)[name = string("op_503_cast_fp16")]; + tensor var_507_begin_0 = const()[name = string("op_507_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_507_end_0 = const()[name = string("op_507_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_507_end_mask_0 = const()[name = string("op_507_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_507_cast_fp16 = slice_by_index(begin = var_507_begin_0, end = var_507_end_0, end_mask = var_507_end_mask_0, x = value_1_cast_fp16)[name = string("op_507_cast_fp16")]; + tensor var_511_begin_0 = const()[name = string("op_511_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_511_end_0 = const()[name = string("op_511_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_511_end_mask_0 = const()[name = string("op_511_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_511_cast_fp16 = slice_by_index(begin = var_511_begin_0, end = var_511_end_0, end_mask = var_511_end_mask_0, x = value_1_cast_fp16)[name = string("op_511_cast_fp16")]; + tensor var_515_begin_0 = const()[name = string("op_515_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_515_end_0 = const()[name = string("op_515_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_515_end_mask_0 = const()[name = string("op_515_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_515_cast_fp16 = slice_by_index(begin = var_515_begin_0, end = var_515_end_0, end_mask = var_515_end_mask_0, x = value_1_cast_fp16)[name = string("op_515_cast_fp16")]; + tensor var_519_begin_0 = const()[name = string("op_519_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_519_end_0 = const()[name = string("op_519_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_519_end_mask_0 = const()[name = string("op_519_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_519_cast_fp16 = slice_by_index(begin = var_519_begin_0, end = var_519_end_0, end_mask = var_519_end_mask_0, x = value_1_cast_fp16)[name = string("op_519_cast_fp16")]; + string _SplitHeadsQ__mh_w_1_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_1_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_1_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_1_equation_0, values = (var_461_cast_fp16, var_239_cast_fp16))[name = string("_SplitHeadsQ__mh_w_1_cast_fp16")]; + string _SplitHeadsQ__mh_w_3_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_3_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_3_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_3_equation_0, values = (var_461_cast_fp16, var_246_cast_fp16))[name = string("_SplitHeadsQ__mh_w_3_cast_fp16")]; + string _SplitHeadsQ__mh_w_5_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_5_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_5_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_5_equation_0, values = (var_461_cast_fp16, var_253_cast_fp16))[name = string("_SplitHeadsQ__mh_w_5_cast_fp16")]; + string _SplitHeadsQ__mh_w_7_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_7_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_7_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_7_equation_0, values = (var_461_cast_fp16, var_260_cast_fp16))[name = string("_SplitHeadsQ__mh_w_7_cast_fp16")]; + string _SplitHeadsQ__mh_w_9_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_9_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_9_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_9_equation_0, values = (var_465_cast_fp16, var_267_cast_fp16))[name = string("_SplitHeadsQ__mh_w_9_cast_fp16")]; + string _SplitHeadsQ__mh_w_11_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_11_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_11_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_11_equation_0, values = (var_465_cast_fp16, var_274_cast_fp16))[name = string("_SplitHeadsQ__mh_w_11_cast_fp16")]; + string _SplitHeadsQ__mh_w_13_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_13_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_13_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_13_equation_0, values = (var_465_cast_fp16, var_281_cast_fp16))[name = string("_SplitHeadsQ__mh_w_13_cast_fp16")]; + string _SplitHeadsQ__mh_w_15_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_15_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_15_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_15_equation_0, values = (var_465_cast_fp16, var_288_cast_fp16))[name = string("_SplitHeadsQ__mh_w_15_cast_fp16")]; + string _SplitHeadsQ__mh_w_17_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_17_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_17_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_17_equation_0, values = (var_469_cast_fp16, var_295_cast_fp16))[name = string("_SplitHeadsQ__mh_w_17_cast_fp16")]; + string _SplitHeadsQ__mh_w_19_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_19_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_19_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_19_equation_0, values = (var_469_cast_fp16, var_302_cast_fp16))[name = string("_SplitHeadsQ__mh_w_19_cast_fp16")]; + string _SplitHeadsQ__mh_w_21_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_21_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_21_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_21_equation_0, values = (var_469_cast_fp16, var_309_cast_fp16))[name = string("_SplitHeadsQ__mh_w_21_cast_fp16")]; + string _SplitHeadsQ__mh_w_23_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_23_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_23_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_23_equation_0, values = (var_469_cast_fp16, var_316_cast_fp16))[name = string("_SplitHeadsQ__mh_w_23_cast_fp16")]; + string _SplitHeadsQ__mh_w_25_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_25_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_25_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_25_equation_0, values = (var_473_cast_fp16, var_323_cast_fp16))[name = string("_SplitHeadsQ__mh_w_25_cast_fp16")]; + string _SplitHeadsQ__mh_w_27_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_27_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_27_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_27_equation_0, values = (var_473_cast_fp16, var_330_cast_fp16))[name = string("_SplitHeadsQ__mh_w_27_cast_fp16")]; + string _SplitHeadsQ__mh_w_29_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_29_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_29_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_29_equation_0, values = (var_473_cast_fp16, var_337_cast_fp16))[name = string("_SplitHeadsQ__mh_w_29_cast_fp16")]; + string _SplitHeadsQ__mh_w_31_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_31_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_31_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_31_equation_0, values = (var_473_cast_fp16, var_344_cast_fp16))[name = string("_SplitHeadsQ__mh_w_31_cast_fp16")]; + string _SplitHeadsQ__mh_w_33_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_33_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_33_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_33_equation_0, values = (var_477_cast_fp16, var_351_cast_fp16))[name = string("_SplitHeadsQ__mh_w_33_cast_fp16")]; + string _SplitHeadsQ__mh_w_35_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_35_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_35_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_35_equation_0, values = (var_477_cast_fp16, var_358_cast_fp16))[name = string("_SplitHeadsQ__mh_w_35_cast_fp16")]; + string _SplitHeadsQ__mh_w_37_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_37_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_37_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_37_equation_0, values = (var_477_cast_fp16, var_365_cast_fp16))[name = string("_SplitHeadsQ__mh_w_37_cast_fp16")]; + string _SplitHeadsQ__mh_w_39_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_39_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_39_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_39_equation_0, values = (var_477_cast_fp16, var_372_cast_fp16))[name = string("_SplitHeadsQ__mh_w_39_cast_fp16")]; + string _SplitHeadsQ__mh_w_41_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_41_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_41_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_41_equation_0, values = (var_481_cast_fp16, var_379_cast_fp16))[name = string("_SplitHeadsQ__mh_w_41_cast_fp16")]; + string _SplitHeadsQ__mh_w_43_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_43_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_43_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_43_equation_0, values = (var_481_cast_fp16, var_386_cast_fp16))[name = string("_SplitHeadsQ__mh_w_43_cast_fp16")]; + string _SplitHeadsQ__mh_w_45_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_45_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_45_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_45_equation_0, values = (var_481_cast_fp16, var_393_cast_fp16))[name = string("_SplitHeadsQ__mh_w_45_cast_fp16")]; + string _SplitHeadsQ__mh_w_47_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_47_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_47_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_47_equation_0, values = (var_481_cast_fp16, var_400_cast_fp16))[name = string("_SplitHeadsQ__mh_w_47_cast_fp16")]; + string _SplitHeadsQ__mh_w_49_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_49_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_49_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_49_equation_0, values = (var_485_cast_fp16, var_407_cast_fp16))[name = string("_SplitHeadsQ__mh_w_49_cast_fp16")]; + string _SplitHeadsQ__mh_w_51_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_51_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_51_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_51_equation_0, values = (var_485_cast_fp16, var_414_cast_fp16))[name = string("_SplitHeadsQ__mh_w_51_cast_fp16")]; + string _SplitHeadsQ__mh_w_53_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_53_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_53_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_53_equation_0, values = (var_485_cast_fp16, var_421_cast_fp16))[name = string("_SplitHeadsQ__mh_w_53_cast_fp16")]; + string _SplitHeadsQ__mh_w_55_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_55_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_55_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_55_equation_0, values = (var_485_cast_fp16, var_428_cast_fp16))[name = string("_SplitHeadsQ__mh_w_55_cast_fp16")]; + string _SplitHeadsQ__mh_w_57_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_57_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_57_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_57_equation_0, values = (var_489_cast_fp16, var_435_cast_fp16))[name = string("_SplitHeadsQ__mh_w_57_cast_fp16")]; + string _SplitHeadsQ__mh_w_59_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_59_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_59_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_59_equation_0, values = (var_489_cast_fp16, var_442_cast_fp16))[name = string("_SplitHeadsQ__mh_w_59_cast_fp16")]; + string _SplitHeadsQ__mh_w_61_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_61_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_61_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_61_equation_0, values = (var_489_cast_fp16, var_449_cast_fp16))[name = string("_SplitHeadsQ__mh_w_61_cast_fp16")]; + string _SplitHeadsQ__mh_w_63_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_63_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_63_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_63_equation_0, values = (var_489_cast_fp16, var_456_cast_fp16))[name = string("_SplitHeadsQ__mh_w_63_cast_fp16")]; + fp16 var_586_to_fp16 = const()[name = string("op_586_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_1_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_1_cast_fp16, y = var_586_to_fp16)[name = string("aw_chunk_1_cast_fp16")]; + fp16 var_588_to_fp16 = const()[name = string("op_588_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_3_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_3_cast_fp16, y = var_588_to_fp16)[name = string("aw_chunk_3_cast_fp16")]; + fp16 var_590_to_fp16 = const()[name = string("op_590_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_5_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_5_cast_fp16, y = var_590_to_fp16)[name = string("aw_chunk_5_cast_fp16")]; + fp16 var_592_to_fp16 = const()[name = string("op_592_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_7_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_7_cast_fp16, y = var_592_to_fp16)[name = string("aw_chunk_7_cast_fp16")]; + fp16 var_594_to_fp16 = const()[name = string("op_594_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_9_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_9_cast_fp16, y = var_594_to_fp16)[name = string("aw_chunk_9_cast_fp16")]; + fp16 var_596_to_fp16 = const()[name = string("op_596_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_11_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_11_cast_fp16, y = var_596_to_fp16)[name = string("aw_chunk_11_cast_fp16")]; + fp16 var_598_to_fp16 = const()[name = string("op_598_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_13_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_13_cast_fp16, y = var_598_to_fp16)[name = string("aw_chunk_13_cast_fp16")]; + fp16 var_600_to_fp16 = const()[name = string("op_600_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_15_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_15_cast_fp16, y = var_600_to_fp16)[name = string("aw_chunk_15_cast_fp16")]; + fp16 var_602_to_fp16 = const()[name = string("op_602_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_17_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_17_cast_fp16, y = var_602_to_fp16)[name = string("aw_chunk_17_cast_fp16")]; + fp16 var_604_to_fp16 = const()[name = string("op_604_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_19_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_19_cast_fp16, y = var_604_to_fp16)[name = string("aw_chunk_19_cast_fp16")]; + fp16 var_606_to_fp16 = const()[name = string("op_606_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_21_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_21_cast_fp16, y = var_606_to_fp16)[name = string("aw_chunk_21_cast_fp16")]; + fp16 var_608_to_fp16 = const()[name = string("op_608_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_23_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_23_cast_fp16, y = var_608_to_fp16)[name = string("aw_chunk_23_cast_fp16")]; + fp16 var_610_to_fp16 = const()[name = string("op_610_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_25_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_25_cast_fp16, y = var_610_to_fp16)[name = string("aw_chunk_25_cast_fp16")]; + fp16 var_612_to_fp16 = const()[name = string("op_612_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_27_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_27_cast_fp16, y = var_612_to_fp16)[name = string("aw_chunk_27_cast_fp16")]; + fp16 var_614_to_fp16 = const()[name = string("op_614_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_29_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_29_cast_fp16, y = var_614_to_fp16)[name = string("aw_chunk_29_cast_fp16")]; + fp16 var_616_to_fp16 = const()[name = string("op_616_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_31_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_31_cast_fp16, y = var_616_to_fp16)[name = string("aw_chunk_31_cast_fp16")]; + fp16 var_618_to_fp16 = const()[name = string("op_618_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_33_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_33_cast_fp16, y = var_618_to_fp16)[name = string("aw_chunk_33_cast_fp16")]; + fp16 var_620_to_fp16 = const()[name = string("op_620_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_35_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_35_cast_fp16, y = var_620_to_fp16)[name = string("aw_chunk_35_cast_fp16")]; + fp16 var_622_to_fp16 = const()[name = string("op_622_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_37_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_37_cast_fp16, y = var_622_to_fp16)[name = string("aw_chunk_37_cast_fp16")]; + fp16 var_624_to_fp16 = const()[name = string("op_624_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_39_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_39_cast_fp16, y = var_624_to_fp16)[name = string("aw_chunk_39_cast_fp16")]; + fp16 var_626_to_fp16 = const()[name = string("op_626_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_41_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_41_cast_fp16, y = var_626_to_fp16)[name = string("aw_chunk_41_cast_fp16")]; + fp16 var_628_to_fp16 = const()[name = string("op_628_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_43_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_43_cast_fp16, y = var_628_to_fp16)[name = string("aw_chunk_43_cast_fp16")]; + fp16 var_630_to_fp16 = const()[name = string("op_630_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_45_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_45_cast_fp16, y = var_630_to_fp16)[name = string("aw_chunk_45_cast_fp16")]; + fp16 var_632_to_fp16 = const()[name = string("op_632_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_47_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_47_cast_fp16, y = var_632_to_fp16)[name = string("aw_chunk_47_cast_fp16")]; + fp16 var_634_to_fp16 = const()[name = string("op_634_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_49_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_49_cast_fp16, y = var_634_to_fp16)[name = string("aw_chunk_49_cast_fp16")]; + fp16 var_636_to_fp16 = const()[name = string("op_636_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_51_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_51_cast_fp16, y = var_636_to_fp16)[name = string("aw_chunk_51_cast_fp16")]; + fp16 var_638_to_fp16 = const()[name = string("op_638_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_53_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_53_cast_fp16, y = var_638_to_fp16)[name = string("aw_chunk_53_cast_fp16")]; + fp16 var_640_to_fp16 = const()[name = string("op_640_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_55_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_55_cast_fp16, y = var_640_to_fp16)[name = string("aw_chunk_55_cast_fp16")]; + fp16 var_642_to_fp16 = const()[name = string("op_642_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_57_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_57_cast_fp16, y = var_642_to_fp16)[name = string("aw_chunk_57_cast_fp16")]; + fp16 var_644_to_fp16 = const()[name = string("op_644_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_59_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_59_cast_fp16, y = var_644_to_fp16)[name = string("aw_chunk_59_cast_fp16")]; + fp16 var_646_to_fp16 = const()[name = string("op_646_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_61_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_61_cast_fp16, y = var_646_to_fp16)[name = string("aw_chunk_61_cast_fp16")]; + fp16 var_648_to_fp16 = const()[name = string("op_648_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_63_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_63_cast_fp16, y = var_648_to_fp16)[name = string("aw_chunk_63_cast_fp16")]; + tensor var_650_cast_fp16 = softmax(axis = var_147, x = aw_chunk_1_cast_fp16)[name = string("op_650_cast_fp16")]; + tensor var_651_cast_fp16 = softmax(axis = var_147, x = aw_chunk_3_cast_fp16)[name = string("op_651_cast_fp16")]; + tensor var_652_cast_fp16 = softmax(axis = var_147, x = aw_chunk_5_cast_fp16)[name = string("op_652_cast_fp16")]; + tensor var_653_cast_fp16 = softmax(axis = var_147, x = aw_chunk_7_cast_fp16)[name = string("op_653_cast_fp16")]; + tensor var_654_cast_fp16 = softmax(axis = var_147, x = aw_chunk_9_cast_fp16)[name = string("op_654_cast_fp16")]; + tensor var_655_cast_fp16 = softmax(axis = var_147, x = aw_chunk_11_cast_fp16)[name = string("op_655_cast_fp16")]; + tensor var_656_cast_fp16 = softmax(axis = var_147, x = aw_chunk_13_cast_fp16)[name = string("op_656_cast_fp16")]; + tensor var_657_cast_fp16 = softmax(axis = var_147, x = aw_chunk_15_cast_fp16)[name = string("op_657_cast_fp16")]; + tensor var_658_cast_fp16 = softmax(axis = var_147, x = aw_chunk_17_cast_fp16)[name = string("op_658_cast_fp16")]; + tensor var_659_cast_fp16 = softmax(axis = var_147, x = aw_chunk_19_cast_fp16)[name = string("op_659_cast_fp16")]; + tensor var_660_cast_fp16 = softmax(axis = var_147, x = aw_chunk_21_cast_fp16)[name = string("op_660_cast_fp16")]; + tensor var_661_cast_fp16 = softmax(axis = var_147, x = aw_chunk_23_cast_fp16)[name = string("op_661_cast_fp16")]; + tensor var_662_cast_fp16 = softmax(axis = var_147, x = aw_chunk_25_cast_fp16)[name = string("op_662_cast_fp16")]; + tensor var_663_cast_fp16 = softmax(axis = var_147, x = aw_chunk_27_cast_fp16)[name = string("op_663_cast_fp16")]; + tensor var_664_cast_fp16 = softmax(axis = var_147, x = aw_chunk_29_cast_fp16)[name = string("op_664_cast_fp16")]; + tensor var_665_cast_fp16 = softmax(axis = var_147, x = aw_chunk_31_cast_fp16)[name = string("op_665_cast_fp16")]; + tensor var_666_cast_fp16 = softmax(axis = var_147, x = aw_chunk_33_cast_fp16)[name = string("op_666_cast_fp16")]; + tensor var_667_cast_fp16 = softmax(axis = var_147, x = aw_chunk_35_cast_fp16)[name = string("op_667_cast_fp16")]; + tensor var_668_cast_fp16 = softmax(axis = var_147, x = aw_chunk_37_cast_fp16)[name = string("op_668_cast_fp16")]; + tensor var_669_cast_fp16 = softmax(axis = var_147, x = aw_chunk_39_cast_fp16)[name = string("op_669_cast_fp16")]; + tensor var_670_cast_fp16 = softmax(axis = var_147, x = aw_chunk_41_cast_fp16)[name = string("op_670_cast_fp16")]; + tensor var_671_cast_fp16 = softmax(axis = var_147, x = aw_chunk_43_cast_fp16)[name = string("op_671_cast_fp16")]; + tensor var_672_cast_fp16 = softmax(axis = var_147, x = aw_chunk_45_cast_fp16)[name = string("op_672_cast_fp16")]; + tensor var_673_cast_fp16 = softmax(axis = var_147, x = aw_chunk_47_cast_fp16)[name = string("op_673_cast_fp16")]; + tensor var_674_cast_fp16 = softmax(axis = var_147, x = aw_chunk_49_cast_fp16)[name = string("op_674_cast_fp16")]; + tensor var_675_cast_fp16 = softmax(axis = var_147, x = aw_chunk_51_cast_fp16)[name = string("op_675_cast_fp16")]; + tensor var_676_cast_fp16 = softmax(axis = var_147, x = aw_chunk_53_cast_fp16)[name = string("op_676_cast_fp16")]; + tensor var_677_cast_fp16 = softmax(axis = var_147, x = aw_chunk_55_cast_fp16)[name = string("op_677_cast_fp16")]; + tensor var_678_cast_fp16 = softmax(axis = var_147, x = aw_chunk_57_cast_fp16)[name = string("op_678_cast_fp16")]; + tensor var_679_cast_fp16 = softmax(axis = var_147, x = aw_chunk_59_cast_fp16)[name = string("op_679_cast_fp16")]; + tensor var_680_cast_fp16 = softmax(axis = var_147, x = aw_chunk_61_cast_fp16)[name = string("op_680_cast_fp16")]; + tensor var_681_cast_fp16 = softmax(axis = var_147, x = aw_chunk_63_cast_fp16)[name = string("op_681_cast_fp16")]; + string var_683_equation_0 = const()[name = string("op_683_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_683_cast_fp16 = einsum(equation = var_683_equation_0, values = (var_491_cast_fp16, var_650_cast_fp16))[name = string("op_683_cast_fp16")]; + string var_685_equation_0 = const()[name = string("op_685_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_685_cast_fp16 = einsum(equation = var_685_equation_0, values = (var_491_cast_fp16, var_651_cast_fp16))[name = string("op_685_cast_fp16")]; + string var_687_equation_0 = const()[name = string("op_687_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_687_cast_fp16 = einsum(equation = var_687_equation_0, values = (var_491_cast_fp16, var_652_cast_fp16))[name = string("op_687_cast_fp16")]; + string var_689_equation_0 = const()[name = string("op_689_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_689_cast_fp16 = einsum(equation = var_689_equation_0, values = (var_491_cast_fp16, var_653_cast_fp16))[name = string("op_689_cast_fp16")]; + string var_691_equation_0 = const()[name = string("op_691_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_691_cast_fp16 = einsum(equation = var_691_equation_0, values = (var_495_cast_fp16, var_654_cast_fp16))[name = string("op_691_cast_fp16")]; + string var_693_equation_0 = const()[name = string("op_693_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_693_cast_fp16 = einsum(equation = var_693_equation_0, values = (var_495_cast_fp16, var_655_cast_fp16))[name = string("op_693_cast_fp16")]; + string var_695_equation_0 = const()[name = string("op_695_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_695_cast_fp16 = einsum(equation = var_695_equation_0, values = (var_495_cast_fp16, var_656_cast_fp16))[name = string("op_695_cast_fp16")]; + string var_697_equation_0 = const()[name = string("op_697_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_697_cast_fp16 = einsum(equation = var_697_equation_0, values = (var_495_cast_fp16, var_657_cast_fp16))[name = string("op_697_cast_fp16")]; + string var_699_equation_0 = const()[name = string("op_699_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_699_cast_fp16 = einsum(equation = var_699_equation_0, values = (var_499_cast_fp16, var_658_cast_fp16))[name = string("op_699_cast_fp16")]; + string var_701_equation_0 = const()[name = string("op_701_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_701_cast_fp16 = einsum(equation = var_701_equation_0, values = (var_499_cast_fp16, var_659_cast_fp16))[name = string("op_701_cast_fp16")]; + string var_703_equation_0 = const()[name = string("op_703_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_703_cast_fp16 = einsum(equation = var_703_equation_0, values = (var_499_cast_fp16, var_660_cast_fp16))[name = string("op_703_cast_fp16")]; + string var_705_equation_0 = const()[name = string("op_705_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_705_cast_fp16 = einsum(equation = var_705_equation_0, values = (var_499_cast_fp16, var_661_cast_fp16))[name = string("op_705_cast_fp16")]; + string var_707_equation_0 = const()[name = string("op_707_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_707_cast_fp16 = einsum(equation = var_707_equation_0, values = (var_503_cast_fp16, var_662_cast_fp16))[name = string("op_707_cast_fp16")]; + string var_709_equation_0 = const()[name = string("op_709_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_709_cast_fp16 = einsum(equation = var_709_equation_0, values = (var_503_cast_fp16, var_663_cast_fp16))[name = string("op_709_cast_fp16")]; + string var_711_equation_0 = const()[name = string("op_711_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_711_cast_fp16 = einsum(equation = var_711_equation_0, values = (var_503_cast_fp16, var_664_cast_fp16))[name = string("op_711_cast_fp16")]; + string var_713_equation_0 = const()[name = string("op_713_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_713_cast_fp16 = einsum(equation = var_713_equation_0, values = (var_503_cast_fp16, var_665_cast_fp16))[name = string("op_713_cast_fp16")]; + string var_715_equation_0 = const()[name = string("op_715_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_715_cast_fp16 = einsum(equation = var_715_equation_0, values = (var_507_cast_fp16, var_666_cast_fp16))[name = string("op_715_cast_fp16")]; + string var_717_equation_0 = const()[name = string("op_717_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_717_cast_fp16 = einsum(equation = var_717_equation_0, values = (var_507_cast_fp16, var_667_cast_fp16))[name = string("op_717_cast_fp16")]; + string var_719_equation_0 = const()[name = string("op_719_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_719_cast_fp16 = einsum(equation = var_719_equation_0, values = (var_507_cast_fp16, var_668_cast_fp16))[name = string("op_719_cast_fp16")]; + string var_721_equation_0 = const()[name = string("op_721_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_721_cast_fp16 = einsum(equation = var_721_equation_0, values = (var_507_cast_fp16, var_669_cast_fp16))[name = string("op_721_cast_fp16")]; + string var_723_equation_0 = const()[name = string("op_723_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_723_cast_fp16 = einsum(equation = var_723_equation_0, values = (var_511_cast_fp16, var_670_cast_fp16))[name = string("op_723_cast_fp16")]; + string var_725_equation_0 = const()[name = string("op_725_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_725_cast_fp16 = einsum(equation = var_725_equation_0, values = (var_511_cast_fp16, var_671_cast_fp16))[name = string("op_725_cast_fp16")]; + string var_727_equation_0 = const()[name = string("op_727_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_727_cast_fp16 = einsum(equation = var_727_equation_0, values = (var_511_cast_fp16, var_672_cast_fp16))[name = string("op_727_cast_fp16")]; + string var_729_equation_0 = const()[name = string("op_729_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_729_cast_fp16 = einsum(equation = var_729_equation_0, values = (var_511_cast_fp16, var_673_cast_fp16))[name = string("op_729_cast_fp16")]; + string var_731_equation_0 = const()[name = string("op_731_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_731_cast_fp16 = einsum(equation = var_731_equation_0, values = (var_515_cast_fp16, var_674_cast_fp16))[name = string("op_731_cast_fp16")]; + string var_733_equation_0 = const()[name = string("op_733_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_733_cast_fp16 = einsum(equation = var_733_equation_0, values = (var_515_cast_fp16, var_675_cast_fp16))[name = string("op_733_cast_fp16")]; + string var_735_equation_0 = const()[name = string("op_735_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_735_cast_fp16 = einsum(equation = var_735_equation_0, values = (var_515_cast_fp16, var_676_cast_fp16))[name = string("op_735_cast_fp16")]; + string var_737_equation_0 = const()[name = string("op_737_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_737_cast_fp16 = einsum(equation = var_737_equation_0, values = (var_515_cast_fp16, var_677_cast_fp16))[name = string("op_737_cast_fp16")]; + string var_739_equation_0 = const()[name = string("op_739_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_739_cast_fp16 = einsum(equation = var_739_equation_0, values = (var_519_cast_fp16, var_678_cast_fp16))[name = string("op_739_cast_fp16")]; + string var_741_equation_0 = const()[name = string("op_741_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_741_cast_fp16 = einsum(equation = var_741_equation_0, values = (var_519_cast_fp16, var_679_cast_fp16))[name = string("op_741_cast_fp16")]; + string var_743_equation_0 = const()[name = string("op_743_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_743_cast_fp16 = einsum(equation = var_743_equation_0, values = (var_519_cast_fp16, var_680_cast_fp16))[name = string("op_743_cast_fp16")]; + string var_745_equation_0 = const()[name = string("op_745_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_745_cast_fp16 = einsum(equation = var_745_equation_0, values = (var_519_cast_fp16, var_681_cast_fp16))[name = string("op_745_cast_fp16")]; + bool var_747_interleave_0 = const()[name = string("op_747_interleave_0"), val = bool(false)]; + tensor var_747_cast_fp16 = concat(axis = var_134, interleave = var_747_interleave_0, values = (var_683_cast_fp16, var_685_cast_fp16, var_687_cast_fp16, var_689_cast_fp16))[name = string("op_747_cast_fp16")]; + bool var_749_interleave_0 = const()[name = string("op_749_interleave_0"), val = bool(false)]; + tensor var_749_cast_fp16 = concat(axis = var_134, interleave = var_749_interleave_0, values = (var_691_cast_fp16, var_693_cast_fp16, var_695_cast_fp16, var_697_cast_fp16))[name = string("op_749_cast_fp16")]; + bool var_751_interleave_0 = const()[name = string("op_751_interleave_0"), val = bool(false)]; + tensor var_751_cast_fp16 = concat(axis = var_134, interleave = var_751_interleave_0, values = (var_699_cast_fp16, var_701_cast_fp16, var_703_cast_fp16, var_705_cast_fp16))[name = string("op_751_cast_fp16")]; + bool var_753_interleave_0 = const()[name = string("op_753_interleave_0"), val = bool(false)]; + tensor var_753_cast_fp16 = concat(axis = var_134, interleave = var_753_interleave_0, values = (var_707_cast_fp16, var_709_cast_fp16, var_711_cast_fp16, var_713_cast_fp16))[name = string("op_753_cast_fp16")]; + bool var_755_interleave_0 = const()[name = string("op_755_interleave_0"), val = bool(false)]; + tensor var_755_cast_fp16 = concat(axis = var_134, interleave = var_755_interleave_0, values = (var_715_cast_fp16, var_717_cast_fp16, var_719_cast_fp16, var_721_cast_fp16))[name = string("op_755_cast_fp16")]; + bool var_757_interleave_0 = const()[name = string("op_757_interleave_0"), val = bool(false)]; + tensor var_757_cast_fp16 = concat(axis = var_134, interleave = var_757_interleave_0, values = (var_723_cast_fp16, var_725_cast_fp16, var_727_cast_fp16, var_729_cast_fp16))[name = string("op_757_cast_fp16")]; + bool var_759_interleave_0 = const()[name = string("op_759_interleave_0"), val = bool(false)]; + tensor var_759_cast_fp16 = concat(axis = var_134, interleave = var_759_interleave_0, values = (var_731_cast_fp16, var_733_cast_fp16, var_735_cast_fp16, var_737_cast_fp16))[name = string("op_759_cast_fp16")]; + bool var_761_interleave_0 = const()[name = string("op_761_interleave_0"), val = bool(false)]; + tensor var_761_cast_fp16 = concat(axis = var_134, interleave = var_761_interleave_0, values = (var_739_cast_fp16, var_741_cast_fp16, var_743_cast_fp16, var_745_cast_fp16))[name = string("op_761_cast_fp16")]; + bool input_1_interleave_0 = const()[name = string("input_1_interleave_0"), val = bool(false)]; + tensor input_1_cast_fp16 = concat(axis = var_147, interleave = input_1_interleave_0, values = (var_747_cast_fp16, var_749_cast_fp16, var_751_cast_fp16, var_753_cast_fp16, var_755_cast_fp16, var_757_cast_fp16, var_759_cast_fp16, var_761_cast_fp16))[name = string("input_1_cast_fp16")]; + string obj_3_pad_type_0 = const()[name = string("obj_3_pad_type_0"), val = string("valid")]; + tensor obj_3_strides_0 = const()[name = string("obj_3_strides_0"), val = tensor([1, 1])]; + tensor obj_3_pad_0 = const()[name = string("obj_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_3_dilations_0 = const()[name = string("obj_3_dilations_0"), val = tensor([1, 1])]; + int32 obj_3_groups_0 = const()[name = string("obj_3_groups_0"), val = int32(1)]; + tensor layers_0_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4936640)))]; + tensor layers_0_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_0_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5460992)))]; + tensor obj_3_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_bias_to_fp16, dilations = obj_3_dilations_0, groups = obj_3_groups_0, pad = obj_3_pad_0, pad_type = obj_3_pad_type_0, strides = obj_3_strides_0, weight = layers_0_self_attn_o_proj_weight_to_fp16, x = input_1_cast_fp16)[name = string("obj_3_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = obj_3_cast_fp16)[name = string("inputs_3_cast_fp16")]; + tensor out_3_axes_0 = const()[name = string("out_3_axes_0"), val = tensor([1])]; + fp16 var_780_to_fp16 = const()[name = string("op_780_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_780_to_fp16, x = inputs_3_cast_fp16)[name = string("out_3_cast_fp16")]; + tensor input_3_gamma_0_to_fp16 = const()[name = string("input_3_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5462080)))]; + tensor input_3_beta_0_to_fp16 = const()[name = string("input_3_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5463168)))]; + fp16 input_3_epsilon_0_to_fp16 = const()[name = string("input_3_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_3_cast_fp16 = batch_norm(beta = input_3_beta_0_to_fp16, epsilon = input_3_epsilon_0_to_fp16, gamma = input_3_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_3_cast_fp16)[name = string("input_3_cast_fp16")]; + string input_5_pad_type_0 = const()[name = string("input_5_pad_type_0"), val = string("valid")]; + tensor input_5_strides_0 = const()[name = string("input_5_strides_0"), val = tensor([1, 1])]; + tensor input_5_pad_0 = const()[name = string("input_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_5_dilations_0 = const()[name = string("input_5_dilations_0"), val = tensor([1, 1])]; + int32 input_5_groups_0 = const()[name = string("input_5_groups_0"), val = int32(1)]; + tensor layers_0_fc1_weight_to_fp16 = const()[name = string("layers_0_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(5464256)))]; + tensor layers_0_fc1_bias_to_fp16 = const()[name = string("layers_0_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7561472)))]; + tensor input_5_cast_fp16 = conv(bias = layers_0_fc1_bias_to_fp16, dilations = input_5_dilations_0, groups = input_5_groups_0, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = input_5_strides_0, weight = layers_0_fc1_weight_to_fp16, x = input_3_cast_fp16)[name = string("input_5_cast_fp16")]; + string input_7_mode_0 = const()[name = string("input_7_mode_0"), val = string("EXACT")]; + tensor input_7_cast_fp16 = gelu(mode = input_7_mode_0, x = input_5_cast_fp16)[name = string("input_7_cast_fp16")]; + string hidden_states_5_pad_type_0 = const()[name = string("hidden_states_5_pad_type_0"), val = string("valid")]; + tensor hidden_states_5_strides_0 = const()[name = string("hidden_states_5_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_5_pad_0 = const()[name = string("hidden_states_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_5_dilations_0 = const()[name = string("hidden_states_5_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_5_groups_0 = const()[name = string("hidden_states_5_groups_0"), val = int32(1)]; + tensor layers_0_fc2_weight_to_fp16 = const()[name = string("layers_0_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7565632)))]; + tensor layers_0_fc2_bias_to_fp16 = const()[name = string("layers_0_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9662848)))]; + tensor hidden_states_5_cast_fp16 = conv(bias = layers_0_fc2_bias_to_fp16, dilations = hidden_states_5_dilations_0, groups = hidden_states_5_groups_0, pad = hidden_states_5_pad_0, pad_type = hidden_states_5_pad_type_0, strides = hidden_states_5_strides_0, weight = layers_0_fc2_weight_to_fp16, x = input_7_cast_fp16)[name = string("hidden_states_5_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = hidden_states_5_cast_fp16)[name = string("inputs_5_cast_fp16")]; + int32 var_809 = const()[name = string("op_809"), val = int32(3)]; + int32 var_822 = const()[name = string("op_822"), val = int32(1)]; + tensor out_5_axes_0 = const()[name = string("out_5_axes_0"), val = tensor([1])]; + fp16 var_839_to_fp16 = const()[name = string("op_839_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_839_to_fp16, x = inputs_5_cast_fp16)[name = string("out_5_cast_fp16")]; + tensor obj_5_gamma_0_to_fp16 = const()[name = string("obj_5_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9663936)))]; + tensor obj_5_beta_0_to_fp16 = const()[name = string("obj_5_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9665024)))]; + fp16 obj_5_epsilon_0_to_fp16 = const()[name = string("obj_5_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_5_cast_fp16 = batch_norm(beta = obj_5_beta_0_to_fp16, epsilon = obj_5_epsilon_0_to_fp16, gamma = obj_5_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_5_cast_fp16)[name = string("obj_5_cast_fp16")]; + string query_3_pad_type_0 = const()[name = string("query_3_pad_type_0"), val = string("valid")]; + tensor query_3_strides_0 = const()[name = string("query_3_strides_0"), val = tensor([1, 1])]; + tensor query_3_pad_0 = const()[name = string("query_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_3_dilations_0 = const()[name = string("query_3_dilations_0"), val = tensor([1, 1])]; + int32 query_3_groups_0 = const()[name = string("query_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9666112)))]; + tensor layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10190464)))]; + tensor query_3_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_bias_to_fp16, dilations = query_3_dilations_0, groups = query_3_groups_0, pad = query_3_pad_0, pad_type = query_3_pad_type_0, strides = query_3_strides_0, weight = layers_1_self_attn_q_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("query_3_cast_fp16")]; + string key_3_pad_type_0 = const()[name = string("key_3_pad_type_0"), val = string("valid")]; + tensor key_3_strides_0 = const()[name = string("key_3_strides_0"), val = tensor([1, 1])]; + tensor key_3_pad_0 = const()[name = string("key_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_3_dilations_0 = const()[name = string("key_3_dilations_0"), val = tensor([1, 1])]; + int32 key_3_groups_0 = const()[name = string("key_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10191552)))]; + tensor key_3_cast_fp16 = conv(dilations = key_3_dilations_0, groups = key_3_groups_0, pad = key_3_pad_0, pad_type = key_3_pad_type_0, strides = key_3_strides_0, weight = layers_1_self_attn_k_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("key_3_cast_fp16")]; + string value_3_pad_type_0 = const()[name = string("value_3_pad_type_0"), val = string("valid")]; + tensor value_3_strides_0 = const()[name = string("value_3_strides_0"), val = tensor([1, 1])]; + tensor value_3_pad_0 = const()[name = string("value_3_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_3_dilations_0 = const()[name = string("value_3_dilations_0"), val = tensor([1, 1])]; + int32 value_3_groups_0 = const()[name = string("value_3_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10715904)))]; + tensor layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11240256)))]; + tensor value_3_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_bias_to_fp16, dilations = value_3_dilations_0, groups = value_3_groups_0, pad = value_3_pad_0, pad_type = value_3_pad_type_0, strides = value_3_strides_0, weight = layers_1_self_attn_v_proj_weight_to_fp16, x = obj_5_cast_fp16)[name = string("value_3_cast_fp16")]; + tensor var_877_begin_0 = const()[name = string("op_877_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_877_end_0 = const()[name = string("op_877_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_877_end_mask_0 = const()[name = string("op_877_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_877_cast_fp16 = slice_by_index(begin = var_877_begin_0, end = var_877_end_0, end_mask = var_877_end_mask_0, x = query_3_cast_fp16)[name = string("op_877_cast_fp16")]; + tensor var_881_begin_0 = const()[name = string("op_881_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_881_end_0 = const()[name = string("op_881_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_881_end_mask_0 = const()[name = string("op_881_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_881_cast_fp16 = slice_by_index(begin = var_881_begin_0, end = var_881_end_0, end_mask = var_881_end_mask_0, x = query_3_cast_fp16)[name = string("op_881_cast_fp16")]; + tensor var_885_begin_0 = const()[name = string("op_885_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_885_end_0 = const()[name = string("op_885_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_885_end_mask_0 = const()[name = string("op_885_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_885_cast_fp16 = slice_by_index(begin = var_885_begin_0, end = var_885_end_0, end_mask = var_885_end_mask_0, x = query_3_cast_fp16)[name = string("op_885_cast_fp16")]; + tensor var_889_begin_0 = const()[name = string("op_889_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_889_end_0 = const()[name = string("op_889_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_889_end_mask_0 = const()[name = string("op_889_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_889_cast_fp16 = slice_by_index(begin = var_889_begin_0, end = var_889_end_0, end_mask = var_889_end_mask_0, x = query_3_cast_fp16)[name = string("op_889_cast_fp16")]; + tensor var_893_begin_0 = const()[name = string("op_893_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_893_end_0 = const()[name = string("op_893_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_893_end_mask_0 = const()[name = string("op_893_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_893_cast_fp16 = slice_by_index(begin = var_893_begin_0, end = var_893_end_0, end_mask = var_893_end_mask_0, x = query_3_cast_fp16)[name = string("op_893_cast_fp16")]; + tensor var_897_begin_0 = const()[name = string("op_897_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_897_end_0 = const()[name = string("op_897_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_897_end_mask_0 = const()[name = string("op_897_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_897_cast_fp16 = slice_by_index(begin = var_897_begin_0, end = var_897_end_0, end_mask = var_897_end_mask_0, x = query_3_cast_fp16)[name = string("op_897_cast_fp16")]; + tensor var_901_begin_0 = const()[name = string("op_901_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_901_end_0 = const()[name = string("op_901_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_901_end_mask_0 = const()[name = string("op_901_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_901_cast_fp16 = slice_by_index(begin = var_901_begin_0, end = var_901_end_0, end_mask = var_901_end_mask_0, x = query_3_cast_fp16)[name = string("op_901_cast_fp16")]; + tensor var_905_begin_0 = const()[name = string("op_905_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_905_end_0 = const()[name = string("op_905_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_905_end_mask_0 = const()[name = string("op_905_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_905_cast_fp16 = slice_by_index(begin = var_905_begin_0, end = var_905_end_0, end_mask = var_905_end_mask_0, x = query_3_cast_fp16)[name = string("op_905_cast_fp16")]; + tensor var_914_begin_0 = const()[name = string("op_914_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_914_end_0 = const()[name = string("op_914_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_914_end_mask_0 = const()[name = string("op_914_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_914_cast_fp16 = slice_by_index(begin = var_914_begin_0, end = var_914_end_0, end_mask = var_914_end_mask_0, x = var_877_cast_fp16)[name = string("op_914_cast_fp16")]; + tensor var_921_begin_0 = const()[name = string("op_921_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_921_end_0 = const()[name = string("op_921_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_921_end_mask_0 = const()[name = string("op_921_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_921_cast_fp16 = slice_by_index(begin = var_921_begin_0, end = var_921_end_0, end_mask = var_921_end_mask_0, x = var_877_cast_fp16)[name = string("op_921_cast_fp16")]; + tensor var_928_begin_0 = const()[name = string("op_928_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_928_end_0 = const()[name = string("op_928_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_928_end_mask_0 = const()[name = string("op_928_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_928_cast_fp16 = slice_by_index(begin = var_928_begin_0, end = var_928_end_0, end_mask = var_928_end_mask_0, x = var_877_cast_fp16)[name = string("op_928_cast_fp16")]; + tensor var_935_begin_0 = const()[name = string("op_935_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_935_end_0 = const()[name = string("op_935_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_935_end_mask_0 = const()[name = string("op_935_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_935_cast_fp16 = slice_by_index(begin = var_935_begin_0, end = var_935_end_0, end_mask = var_935_end_mask_0, x = var_877_cast_fp16)[name = string("op_935_cast_fp16")]; + tensor var_942_begin_0 = const()[name = string("op_942_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_942_end_0 = const()[name = string("op_942_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_942_end_mask_0 = const()[name = string("op_942_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_942_cast_fp16 = slice_by_index(begin = var_942_begin_0, end = var_942_end_0, end_mask = var_942_end_mask_0, x = var_881_cast_fp16)[name = string("op_942_cast_fp16")]; + tensor var_949_begin_0 = const()[name = string("op_949_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_949_end_0 = const()[name = string("op_949_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_949_end_mask_0 = const()[name = string("op_949_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_949_cast_fp16 = slice_by_index(begin = var_949_begin_0, end = var_949_end_0, end_mask = var_949_end_mask_0, x = var_881_cast_fp16)[name = string("op_949_cast_fp16")]; + tensor var_956_begin_0 = const()[name = string("op_956_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_956_end_0 = const()[name = string("op_956_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_956_end_mask_0 = const()[name = string("op_956_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_956_cast_fp16 = slice_by_index(begin = var_956_begin_0, end = var_956_end_0, end_mask = var_956_end_mask_0, x = var_881_cast_fp16)[name = string("op_956_cast_fp16")]; + tensor var_963_begin_0 = const()[name = string("op_963_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_963_end_0 = const()[name = string("op_963_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_963_end_mask_0 = const()[name = string("op_963_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_963_cast_fp16 = slice_by_index(begin = var_963_begin_0, end = var_963_end_0, end_mask = var_963_end_mask_0, x = var_881_cast_fp16)[name = string("op_963_cast_fp16")]; + tensor var_970_begin_0 = const()[name = string("op_970_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_970_end_0 = const()[name = string("op_970_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_970_end_mask_0 = const()[name = string("op_970_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_970_cast_fp16 = slice_by_index(begin = var_970_begin_0, end = var_970_end_0, end_mask = var_970_end_mask_0, x = var_885_cast_fp16)[name = string("op_970_cast_fp16")]; + tensor var_977_begin_0 = const()[name = string("op_977_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_977_end_0 = const()[name = string("op_977_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_977_end_mask_0 = const()[name = string("op_977_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_977_cast_fp16 = slice_by_index(begin = var_977_begin_0, end = var_977_end_0, end_mask = var_977_end_mask_0, x = var_885_cast_fp16)[name = string("op_977_cast_fp16")]; + tensor var_984_begin_0 = const()[name = string("op_984_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_984_end_0 = const()[name = string("op_984_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_984_end_mask_0 = const()[name = string("op_984_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_984_cast_fp16 = slice_by_index(begin = var_984_begin_0, end = var_984_end_0, end_mask = var_984_end_mask_0, x = var_885_cast_fp16)[name = string("op_984_cast_fp16")]; + tensor var_991_begin_0 = const()[name = string("op_991_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_991_end_0 = const()[name = string("op_991_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_991_end_mask_0 = const()[name = string("op_991_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_991_cast_fp16 = slice_by_index(begin = var_991_begin_0, end = var_991_end_0, end_mask = var_991_end_mask_0, x = var_885_cast_fp16)[name = string("op_991_cast_fp16")]; + tensor var_998_begin_0 = const()[name = string("op_998_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_998_end_0 = const()[name = string("op_998_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_998_end_mask_0 = const()[name = string("op_998_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_998_cast_fp16 = slice_by_index(begin = var_998_begin_0, end = var_998_end_0, end_mask = var_998_end_mask_0, x = var_889_cast_fp16)[name = string("op_998_cast_fp16")]; + tensor var_1005_begin_0 = const()[name = string("op_1005_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1005_end_0 = const()[name = string("op_1005_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1005_end_mask_0 = const()[name = string("op_1005_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1005_cast_fp16 = slice_by_index(begin = var_1005_begin_0, end = var_1005_end_0, end_mask = var_1005_end_mask_0, x = var_889_cast_fp16)[name = string("op_1005_cast_fp16")]; + tensor var_1012_begin_0 = const()[name = string("op_1012_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1012_end_0 = const()[name = string("op_1012_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1012_end_mask_0 = const()[name = string("op_1012_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1012_cast_fp16 = slice_by_index(begin = var_1012_begin_0, end = var_1012_end_0, end_mask = var_1012_end_mask_0, x = var_889_cast_fp16)[name = string("op_1012_cast_fp16")]; + tensor var_1019_begin_0 = const()[name = string("op_1019_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1019_end_0 = const()[name = string("op_1019_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1019_end_mask_0 = const()[name = string("op_1019_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1019_cast_fp16 = slice_by_index(begin = var_1019_begin_0, end = var_1019_end_0, end_mask = var_1019_end_mask_0, x = var_889_cast_fp16)[name = string("op_1019_cast_fp16")]; + tensor var_1026_begin_0 = const()[name = string("op_1026_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1026_end_0 = const()[name = string("op_1026_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1026_end_mask_0 = const()[name = string("op_1026_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1026_cast_fp16 = slice_by_index(begin = var_1026_begin_0, end = var_1026_end_0, end_mask = var_1026_end_mask_0, x = var_893_cast_fp16)[name = string("op_1026_cast_fp16")]; + tensor var_1033_begin_0 = const()[name = string("op_1033_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1033_end_0 = const()[name = string("op_1033_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1033_end_mask_0 = const()[name = string("op_1033_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1033_cast_fp16 = slice_by_index(begin = var_1033_begin_0, end = var_1033_end_0, end_mask = var_1033_end_mask_0, x = var_893_cast_fp16)[name = string("op_1033_cast_fp16")]; + tensor var_1040_begin_0 = const()[name = string("op_1040_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1040_end_0 = const()[name = string("op_1040_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1040_end_mask_0 = const()[name = string("op_1040_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1040_cast_fp16 = slice_by_index(begin = var_1040_begin_0, end = var_1040_end_0, end_mask = var_1040_end_mask_0, x = var_893_cast_fp16)[name = string("op_1040_cast_fp16")]; + tensor var_1047_begin_0 = const()[name = string("op_1047_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1047_end_0 = const()[name = string("op_1047_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1047_end_mask_0 = const()[name = string("op_1047_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1047_cast_fp16 = slice_by_index(begin = var_1047_begin_0, end = var_1047_end_0, end_mask = var_1047_end_mask_0, x = var_893_cast_fp16)[name = string("op_1047_cast_fp16")]; + tensor var_1054_begin_0 = const()[name = string("op_1054_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1054_end_0 = const()[name = string("op_1054_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1054_end_mask_0 = const()[name = string("op_1054_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1054_cast_fp16 = slice_by_index(begin = var_1054_begin_0, end = var_1054_end_0, end_mask = var_1054_end_mask_0, x = var_897_cast_fp16)[name = string("op_1054_cast_fp16")]; + tensor var_1061_begin_0 = const()[name = string("op_1061_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1061_end_0 = const()[name = string("op_1061_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1061_end_mask_0 = const()[name = string("op_1061_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1061_cast_fp16 = slice_by_index(begin = var_1061_begin_0, end = var_1061_end_0, end_mask = var_1061_end_mask_0, x = var_897_cast_fp16)[name = string("op_1061_cast_fp16")]; + tensor var_1068_begin_0 = const()[name = string("op_1068_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1068_end_0 = const()[name = string("op_1068_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1068_end_mask_0 = const()[name = string("op_1068_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1068_cast_fp16 = slice_by_index(begin = var_1068_begin_0, end = var_1068_end_0, end_mask = var_1068_end_mask_0, x = var_897_cast_fp16)[name = string("op_1068_cast_fp16")]; + tensor var_1075_begin_0 = const()[name = string("op_1075_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1075_end_0 = const()[name = string("op_1075_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1075_end_mask_0 = const()[name = string("op_1075_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1075_cast_fp16 = slice_by_index(begin = var_1075_begin_0, end = var_1075_end_0, end_mask = var_1075_end_mask_0, x = var_897_cast_fp16)[name = string("op_1075_cast_fp16")]; + tensor var_1082_begin_0 = const()[name = string("op_1082_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1082_end_0 = const()[name = string("op_1082_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1082_end_mask_0 = const()[name = string("op_1082_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1082_cast_fp16 = slice_by_index(begin = var_1082_begin_0, end = var_1082_end_0, end_mask = var_1082_end_mask_0, x = var_901_cast_fp16)[name = string("op_1082_cast_fp16")]; + tensor var_1089_begin_0 = const()[name = string("op_1089_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1089_end_0 = const()[name = string("op_1089_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1089_end_mask_0 = const()[name = string("op_1089_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1089_cast_fp16 = slice_by_index(begin = var_1089_begin_0, end = var_1089_end_0, end_mask = var_1089_end_mask_0, x = var_901_cast_fp16)[name = string("op_1089_cast_fp16")]; + tensor var_1096_begin_0 = const()[name = string("op_1096_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1096_end_0 = const()[name = string("op_1096_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1096_end_mask_0 = const()[name = string("op_1096_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1096_cast_fp16 = slice_by_index(begin = var_1096_begin_0, end = var_1096_end_0, end_mask = var_1096_end_mask_0, x = var_901_cast_fp16)[name = string("op_1096_cast_fp16")]; + tensor var_1103_begin_0 = const()[name = string("op_1103_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1103_end_0 = const()[name = string("op_1103_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1103_end_mask_0 = const()[name = string("op_1103_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1103_cast_fp16 = slice_by_index(begin = var_1103_begin_0, end = var_1103_end_0, end_mask = var_1103_end_mask_0, x = var_901_cast_fp16)[name = string("op_1103_cast_fp16")]; + tensor var_1110_begin_0 = const()[name = string("op_1110_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1110_end_0 = const()[name = string("op_1110_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1110_end_mask_0 = const()[name = string("op_1110_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1110_cast_fp16 = slice_by_index(begin = var_1110_begin_0, end = var_1110_end_0, end_mask = var_1110_end_mask_0, x = var_905_cast_fp16)[name = string("op_1110_cast_fp16")]; + tensor var_1117_begin_0 = const()[name = string("op_1117_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1117_end_0 = const()[name = string("op_1117_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1117_end_mask_0 = const()[name = string("op_1117_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1117_cast_fp16 = slice_by_index(begin = var_1117_begin_0, end = var_1117_end_0, end_mask = var_1117_end_mask_0, x = var_905_cast_fp16)[name = string("op_1117_cast_fp16")]; + tensor var_1124_begin_0 = const()[name = string("op_1124_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1124_end_0 = const()[name = string("op_1124_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1124_end_mask_0 = const()[name = string("op_1124_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1124_cast_fp16 = slice_by_index(begin = var_1124_begin_0, end = var_1124_end_0, end_mask = var_1124_end_mask_0, x = var_905_cast_fp16)[name = string("op_1124_cast_fp16")]; + tensor var_1131_begin_0 = const()[name = string("op_1131_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1131_end_0 = const()[name = string("op_1131_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1131_end_mask_0 = const()[name = string("op_1131_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1131_cast_fp16 = slice_by_index(begin = var_1131_begin_0, end = var_1131_end_0, end_mask = var_1131_end_mask_0, x = var_905_cast_fp16)[name = string("op_1131_cast_fp16")]; + tensor k_3_perm_0 = const()[name = string("k_3_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_1136_begin_0 = const()[name = string("op_1136_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1136_end_0 = const()[name = string("op_1136_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_1136_end_mask_0 = const()[name = string("op_1136_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_3_cast_fp16 = transpose(perm = k_3_perm_0, x = key_3_cast_fp16)[name = string("transpose_4")]; + tensor var_1136_cast_fp16 = slice_by_index(begin = var_1136_begin_0, end = var_1136_end_0, end_mask = var_1136_end_mask_0, x = k_3_cast_fp16)[name = string("op_1136_cast_fp16")]; + tensor var_1140_begin_0 = const()[name = string("op_1140_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_1140_end_0 = const()[name = string("op_1140_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_1140_end_mask_0 = const()[name = string("op_1140_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1140_cast_fp16 = slice_by_index(begin = var_1140_begin_0, end = var_1140_end_0, end_mask = var_1140_end_mask_0, x = k_3_cast_fp16)[name = string("op_1140_cast_fp16")]; + tensor var_1144_begin_0 = const()[name = string("op_1144_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_1144_end_0 = const()[name = string("op_1144_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_1144_end_mask_0 = const()[name = string("op_1144_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1144_cast_fp16 = slice_by_index(begin = var_1144_begin_0, end = var_1144_end_0, end_mask = var_1144_end_mask_0, x = k_3_cast_fp16)[name = string("op_1144_cast_fp16")]; + tensor var_1148_begin_0 = const()[name = string("op_1148_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_1148_end_0 = const()[name = string("op_1148_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_1148_end_mask_0 = const()[name = string("op_1148_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1148_cast_fp16 = slice_by_index(begin = var_1148_begin_0, end = var_1148_end_0, end_mask = var_1148_end_mask_0, x = k_3_cast_fp16)[name = string("op_1148_cast_fp16")]; + tensor var_1152_begin_0 = const()[name = string("op_1152_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_1152_end_0 = const()[name = string("op_1152_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_1152_end_mask_0 = const()[name = string("op_1152_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1152_cast_fp16 = slice_by_index(begin = var_1152_begin_0, end = var_1152_end_0, end_mask = var_1152_end_mask_0, x = k_3_cast_fp16)[name = string("op_1152_cast_fp16")]; + tensor var_1156_begin_0 = const()[name = string("op_1156_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_1156_end_0 = const()[name = string("op_1156_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_1156_end_mask_0 = const()[name = string("op_1156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1156_cast_fp16 = slice_by_index(begin = var_1156_begin_0, end = var_1156_end_0, end_mask = var_1156_end_mask_0, x = k_3_cast_fp16)[name = string("op_1156_cast_fp16")]; + tensor var_1160_begin_0 = const()[name = string("op_1160_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_1160_end_0 = const()[name = string("op_1160_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_1160_end_mask_0 = const()[name = string("op_1160_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1160_cast_fp16 = slice_by_index(begin = var_1160_begin_0, end = var_1160_end_0, end_mask = var_1160_end_mask_0, x = k_3_cast_fp16)[name = string("op_1160_cast_fp16")]; + tensor var_1164_begin_0 = const()[name = string("op_1164_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_1164_end_0 = const()[name = string("op_1164_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_1164_end_mask_0 = const()[name = string("op_1164_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1164_cast_fp16 = slice_by_index(begin = var_1164_begin_0, end = var_1164_end_0, end_mask = var_1164_end_mask_0, x = k_3_cast_fp16)[name = string("op_1164_cast_fp16")]; + tensor var_1166_begin_0 = const()[name = string("op_1166_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1166_end_0 = const()[name = string("op_1166_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1166_end_mask_0 = const()[name = string("op_1166_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1166_cast_fp16 = slice_by_index(begin = var_1166_begin_0, end = var_1166_end_0, end_mask = var_1166_end_mask_0, x = value_3_cast_fp16)[name = string("op_1166_cast_fp16")]; + tensor var_1170_begin_0 = const()[name = string("op_1170_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1170_end_0 = const()[name = string("op_1170_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_1170_end_mask_0 = const()[name = string("op_1170_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1170_cast_fp16 = slice_by_index(begin = var_1170_begin_0, end = var_1170_end_0, end_mask = var_1170_end_mask_0, x = value_3_cast_fp16)[name = string("op_1170_cast_fp16")]; + tensor var_1174_begin_0 = const()[name = string("op_1174_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1174_end_0 = const()[name = string("op_1174_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_1174_end_mask_0 = const()[name = string("op_1174_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1174_cast_fp16 = slice_by_index(begin = var_1174_begin_0, end = var_1174_end_0, end_mask = var_1174_end_mask_0, x = value_3_cast_fp16)[name = string("op_1174_cast_fp16")]; + tensor var_1178_begin_0 = const()[name = string("op_1178_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1178_end_0 = const()[name = string("op_1178_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_1178_end_mask_0 = const()[name = string("op_1178_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1178_cast_fp16 = slice_by_index(begin = var_1178_begin_0, end = var_1178_end_0, end_mask = var_1178_end_mask_0, x = value_3_cast_fp16)[name = string("op_1178_cast_fp16")]; + tensor var_1182_begin_0 = const()[name = string("op_1182_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1182_end_0 = const()[name = string("op_1182_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_1182_end_mask_0 = const()[name = string("op_1182_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1182_cast_fp16 = slice_by_index(begin = var_1182_begin_0, end = var_1182_end_0, end_mask = var_1182_end_mask_0, x = value_3_cast_fp16)[name = string("op_1182_cast_fp16")]; + tensor var_1186_begin_0 = const()[name = string("op_1186_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1186_end_0 = const()[name = string("op_1186_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1186_end_mask_0 = const()[name = string("op_1186_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1186_cast_fp16 = slice_by_index(begin = var_1186_begin_0, end = var_1186_end_0, end_mask = var_1186_end_mask_0, x = value_3_cast_fp16)[name = string("op_1186_cast_fp16")]; + tensor var_1190_begin_0 = const()[name = string("op_1190_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1190_end_0 = const()[name = string("op_1190_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_1190_end_mask_0 = const()[name = string("op_1190_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1190_cast_fp16 = slice_by_index(begin = var_1190_begin_0, end = var_1190_end_0, end_mask = var_1190_end_mask_0, x = value_3_cast_fp16)[name = string("op_1190_cast_fp16")]; + tensor var_1194_begin_0 = const()[name = string("op_1194_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1194_end_0 = const()[name = string("op_1194_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_1194_end_mask_0 = const()[name = string("op_1194_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1194_cast_fp16 = slice_by_index(begin = var_1194_begin_0, end = var_1194_end_0, end_mask = var_1194_end_mask_0, x = value_3_cast_fp16)[name = string("op_1194_cast_fp16")]; + string _SplitHeadsQ__mh_w_65_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_65_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_65_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_65_equation_0, values = (var_1136_cast_fp16, var_914_cast_fp16))[name = string("_SplitHeadsQ__mh_w_65_cast_fp16")]; + string _SplitHeadsQ__mh_w_67_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_67_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_67_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_67_equation_0, values = (var_1136_cast_fp16, var_921_cast_fp16))[name = string("_SplitHeadsQ__mh_w_67_cast_fp16")]; + string _SplitHeadsQ__mh_w_69_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_69_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_69_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_69_equation_0, values = (var_1136_cast_fp16, var_928_cast_fp16))[name = string("_SplitHeadsQ__mh_w_69_cast_fp16")]; + string _SplitHeadsQ__mh_w_71_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_71_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_71_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_71_equation_0, values = (var_1136_cast_fp16, var_935_cast_fp16))[name = string("_SplitHeadsQ__mh_w_71_cast_fp16")]; + string _SplitHeadsQ__mh_w_73_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_73_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_73_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_73_equation_0, values = (var_1140_cast_fp16, var_942_cast_fp16))[name = string("_SplitHeadsQ__mh_w_73_cast_fp16")]; + string _SplitHeadsQ__mh_w_75_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_75_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_75_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_75_equation_0, values = (var_1140_cast_fp16, var_949_cast_fp16))[name = string("_SplitHeadsQ__mh_w_75_cast_fp16")]; + string _SplitHeadsQ__mh_w_77_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_77_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_77_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_77_equation_0, values = (var_1140_cast_fp16, var_956_cast_fp16))[name = string("_SplitHeadsQ__mh_w_77_cast_fp16")]; + string _SplitHeadsQ__mh_w_79_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_79_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_79_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_79_equation_0, values = (var_1140_cast_fp16, var_963_cast_fp16))[name = string("_SplitHeadsQ__mh_w_79_cast_fp16")]; + string _SplitHeadsQ__mh_w_81_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_81_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_81_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_81_equation_0, values = (var_1144_cast_fp16, var_970_cast_fp16))[name = string("_SplitHeadsQ__mh_w_81_cast_fp16")]; + string _SplitHeadsQ__mh_w_83_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_83_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_83_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_83_equation_0, values = (var_1144_cast_fp16, var_977_cast_fp16))[name = string("_SplitHeadsQ__mh_w_83_cast_fp16")]; + string _SplitHeadsQ__mh_w_85_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_85_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_85_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_85_equation_0, values = (var_1144_cast_fp16, var_984_cast_fp16))[name = string("_SplitHeadsQ__mh_w_85_cast_fp16")]; + string _SplitHeadsQ__mh_w_87_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_87_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_87_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_87_equation_0, values = (var_1144_cast_fp16, var_991_cast_fp16))[name = string("_SplitHeadsQ__mh_w_87_cast_fp16")]; + string _SplitHeadsQ__mh_w_89_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_89_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_89_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_89_equation_0, values = (var_1148_cast_fp16, var_998_cast_fp16))[name = string("_SplitHeadsQ__mh_w_89_cast_fp16")]; + string _SplitHeadsQ__mh_w_91_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_91_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_91_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_91_equation_0, values = (var_1148_cast_fp16, var_1005_cast_fp16))[name = string("_SplitHeadsQ__mh_w_91_cast_fp16")]; + string _SplitHeadsQ__mh_w_93_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_93_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_93_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_93_equation_0, values = (var_1148_cast_fp16, var_1012_cast_fp16))[name = string("_SplitHeadsQ__mh_w_93_cast_fp16")]; + string _SplitHeadsQ__mh_w_95_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_95_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_95_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_95_equation_0, values = (var_1148_cast_fp16, var_1019_cast_fp16))[name = string("_SplitHeadsQ__mh_w_95_cast_fp16")]; + string _SplitHeadsQ__mh_w_97_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_97_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_97_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_97_equation_0, values = (var_1152_cast_fp16, var_1026_cast_fp16))[name = string("_SplitHeadsQ__mh_w_97_cast_fp16")]; + string _SplitHeadsQ__mh_w_99_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_99_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_99_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_99_equation_0, values = (var_1152_cast_fp16, var_1033_cast_fp16))[name = string("_SplitHeadsQ__mh_w_99_cast_fp16")]; + string _SplitHeadsQ__mh_w_101_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_101_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_101_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_101_equation_0, values = (var_1152_cast_fp16, var_1040_cast_fp16))[name = string("_SplitHeadsQ__mh_w_101_cast_fp16")]; + string _SplitHeadsQ__mh_w_103_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_103_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_103_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_103_equation_0, values = (var_1152_cast_fp16, var_1047_cast_fp16))[name = string("_SplitHeadsQ__mh_w_103_cast_fp16")]; + string _SplitHeadsQ__mh_w_105_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_105_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_105_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_105_equation_0, values = (var_1156_cast_fp16, var_1054_cast_fp16))[name = string("_SplitHeadsQ__mh_w_105_cast_fp16")]; + string _SplitHeadsQ__mh_w_107_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_107_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_107_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_107_equation_0, values = (var_1156_cast_fp16, var_1061_cast_fp16))[name = string("_SplitHeadsQ__mh_w_107_cast_fp16")]; + string _SplitHeadsQ__mh_w_109_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_109_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_109_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_109_equation_0, values = (var_1156_cast_fp16, var_1068_cast_fp16))[name = string("_SplitHeadsQ__mh_w_109_cast_fp16")]; + string _SplitHeadsQ__mh_w_111_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_111_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_111_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_111_equation_0, values = (var_1156_cast_fp16, var_1075_cast_fp16))[name = string("_SplitHeadsQ__mh_w_111_cast_fp16")]; + string _SplitHeadsQ__mh_w_113_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_113_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_113_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_113_equation_0, values = (var_1160_cast_fp16, var_1082_cast_fp16))[name = string("_SplitHeadsQ__mh_w_113_cast_fp16")]; + string _SplitHeadsQ__mh_w_115_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_115_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_115_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_115_equation_0, values = (var_1160_cast_fp16, var_1089_cast_fp16))[name = string("_SplitHeadsQ__mh_w_115_cast_fp16")]; + string _SplitHeadsQ__mh_w_117_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_117_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_117_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_117_equation_0, values = (var_1160_cast_fp16, var_1096_cast_fp16))[name = string("_SplitHeadsQ__mh_w_117_cast_fp16")]; + string _SplitHeadsQ__mh_w_119_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_119_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_119_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_119_equation_0, values = (var_1160_cast_fp16, var_1103_cast_fp16))[name = string("_SplitHeadsQ__mh_w_119_cast_fp16")]; + string _SplitHeadsQ__mh_w_121_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_121_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_121_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_121_equation_0, values = (var_1164_cast_fp16, var_1110_cast_fp16))[name = string("_SplitHeadsQ__mh_w_121_cast_fp16")]; + string _SplitHeadsQ__mh_w_123_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_123_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_123_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_123_equation_0, values = (var_1164_cast_fp16, var_1117_cast_fp16))[name = string("_SplitHeadsQ__mh_w_123_cast_fp16")]; + string _SplitHeadsQ__mh_w_125_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_125_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_125_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_125_equation_0, values = (var_1164_cast_fp16, var_1124_cast_fp16))[name = string("_SplitHeadsQ__mh_w_125_cast_fp16")]; + string _SplitHeadsQ__mh_w_127_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_127_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_127_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_127_equation_0, values = (var_1164_cast_fp16, var_1131_cast_fp16))[name = string("_SplitHeadsQ__mh_w_127_cast_fp16")]; + fp16 var_1261_to_fp16 = const()[name = string("op_1261_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_65_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_65_cast_fp16, y = var_1261_to_fp16)[name = string("aw_chunk_65_cast_fp16")]; + fp16 var_1263_to_fp16 = const()[name = string("op_1263_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_67_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_67_cast_fp16, y = var_1263_to_fp16)[name = string("aw_chunk_67_cast_fp16")]; + fp16 var_1265_to_fp16 = const()[name = string("op_1265_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_69_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_69_cast_fp16, y = var_1265_to_fp16)[name = string("aw_chunk_69_cast_fp16")]; + fp16 var_1267_to_fp16 = const()[name = string("op_1267_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_71_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_71_cast_fp16, y = var_1267_to_fp16)[name = string("aw_chunk_71_cast_fp16")]; + fp16 var_1269_to_fp16 = const()[name = string("op_1269_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_73_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_73_cast_fp16, y = var_1269_to_fp16)[name = string("aw_chunk_73_cast_fp16")]; + fp16 var_1271_to_fp16 = const()[name = string("op_1271_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_75_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_75_cast_fp16, y = var_1271_to_fp16)[name = string("aw_chunk_75_cast_fp16")]; + fp16 var_1273_to_fp16 = const()[name = string("op_1273_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_77_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_77_cast_fp16, y = var_1273_to_fp16)[name = string("aw_chunk_77_cast_fp16")]; + fp16 var_1275_to_fp16 = const()[name = string("op_1275_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_79_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_79_cast_fp16, y = var_1275_to_fp16)[name = string("aw_chunk_79_cast_fp16")]; + fp16 var_1277_to_fp16 = const()[name = string("op_1277_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_81_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_81_cast_fp16, y = var_1277_to_fp16)[name = string("aw_chunk_81_cast_fp16")]; + fp16 var_1279_to_fp16 = const()[name = string("op_1279_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_83_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_83_cast_fp16, y = var_1279_to_fp16)[name = string("aw_chunk_83_cast_fp16")]; + fp16 var_1281_to_fp16 = const()[name = string("op_1281_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_85_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_85_cast_fp16, y = var_1281_to_fp16)[name = string("aw_chunk_85_cast_fp16")]; + fp16 var_1283_to_fp16 = const()[name = string("op_1283_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_87_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_87_cast_fp16, y = var_1283_to_fp16)[name = string("aw_chunk_87_cast_fp16")]; + fp16 var_1285_to_fp16 = const()[name = string("op_1285_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_89_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_89_cast_fp16, y = var_1285_to_fp16)[name = string("aw_chunk_89_cast_fp16")]; + fp16 var_1287_to_fp16 = const()[name = string("op_1287_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_91_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_91_cast_fp16, y = var_1287_to_fp16)[name = string("aw_chunk_91_cast_fp16")]; + fp16 var_1289_to_fp16 = const()[name = string("op_1289_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_93_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_93_cast_fp16, y = var_1289_to_fp16)[name = string("aw_chunk_93_cast_fp16")]; + fp16 var_1291_to_fp16 = const()[name = string("op_1291_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_95_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_95_cast_fp16, y = var_1291_to_fp16)[name = string("aw_chunk_95_cast_fp16")]; + fp16 var_1293_to_fp16 = const()[name = string("op_1293_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_97_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_97_cast_fp16, y = var_1293_to_fp16)[name = string("aw_chunk_97_cast_fp16")]; + fp16 var_1295_to_fp16 = const()[name = string("op_1295_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_99_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_99_cast_fp16, y = var_1295_to_fp16)[name = string("aw_chunk_99_cast_fp16")]; + fp16 var_1297_to_fp16 = const()[name = string("op_1297_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_101_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_101_cast_fp16, y = var_1297_to_fp16)[name = string("aw_chunk_101_cast_fp16")]; + fp16 var_1299_to_fp16 = const()[name = string("op_1299_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_103_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_103_cast_fp16, y = var_1299_to_fp16)[name = string("aw_chunk_103_cast_fp16")]; + fp16 var_1301_to_fp16 = const()[name = string("op_1301_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_105_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_105_cast_fp16, y = var_1301_to_fp16)[name = string("aw_chunk_105_cast_fp16")]; + fp16 var_1303_to_fp16 = const()[name = string("op_1303_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_107_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_107_cast_fp16, y = var_1303_to_fp16)[name = string("aw_chunk_107_cast_fp16")]; + fp16 var_1305_to_fp16 = const()[name = string("op_1305_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_109_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_109_cast_fp16, y = var_1305_to_fp16)[name = string("aw_chunk_109_cast_fp16")]; + fp16 var_1307_to_fp16 = const()[name = string("op_1307_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_111_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_111_cast_fp16, y = var_1307_to_fp16)[name = string("aw_chunk_111_cast_fp16")]; + fp16 var_1309_to_fp16 = const()[name = string("op_1309_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_113_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_113_cast_fp16, y = var_1309_to_fp16)[name = string("aw_chunk_113_cast_fp16")]; + fp16 var_1311_to_fp16 = const()[name = string("op_1311_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_115_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_115_cast_fp16, y = var_1311_to_fp16)[name = string("aw_chunk_115_cast_fp16")]; + fp16 var_1313_to_fp16 = const()[name = string("op_1313_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_117_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_117_cast_fp16, y = var_1313_to_fp16)[name = string("aw_chunk_117_cast_fp16")]; + fp16 var_1315_to_fp16 = const()[name = string("op_1315_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_119_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_119_cast_fp16, y = var_1315_to_fp16)[name = string("aw_chunk_119_cast_fp16")]; + fp16 var_1317_to_fp16 = const()[name = string("op_1317_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_121_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_121_cast_fp16, y = var_1317_to_fp16)[name = string("aw_chunk_121_cast_fp16")]; + fp16 var_1319_to_fp16 = const()[name = string("op_1319_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_123_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_123_cast_fp16, y = var_1319_to_fp16)[name = string("aw_chunk_123_cast_fp16")]; + fp16 var_1321_to_fp16 = const()[name = string("op_1321_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_125_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_125_cast_fp16, y = var_1321_to_fp16)[name = string("aw_chunk_125_cast_fp16")]; + fp16 var_1323_to_fp16 = const()[name = string("op_1323_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_127_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_127_cast_fp16, y = var_1323_to_fp16)[name = string("aw_chunk_127_cast_fp16")]; + tensor var_1325_cast_fp16 = softmax(axis = var_822, x = aw_chunk_65_cast_fp16)[name = string("op_1325_cast_fp16")]; + tensor var_1326_cast_fp16 = softmax(axis = var_822, x = aw_chunk_67_cast_fp16)[name = string("op_1326_cast_fp16")]; + tensor var_1327_cast_fp16 = softmax(axis = var_822, x = aw_chunk_69_cast_fp16)[name = string("op_1327_cast_fp16")]; + tensor var_1328_cast_fp16 = softmax(axis = var_822, x = aw_chunk_71_cast_fp16)[name = string("op_1328_cast_fp16")]; + tensor var_1329_cast_fp16 = softmax(axis = var_822, x = aw_chunk_73_cast_fp16)[name = string("op_1329_cast_fp16")]; + tensor var_1330_cast_fp16 = softmax(axis = var_822, x = aw_chunk_75_cast_fp16)[name = string("op_1330_cast_fp16")]; + tensor var_1331_cast_fp16 = softmax(axis = var_822, x = aw_chunk_77_cast_fp16)[name = string("op_1331_cast_fp16")]; + tensor var_1332_cast_fp16 = softmax(axis = var_822, x = aw_chunk_79_cast_fp16)[name = string("op_1332_cast_fp16")]; + tensor var_1333_cast_fp16 = softmax(axis = var_822, x = aw_chunk_81_cast_fp16)[name = string("op_1333_cast_fp16")]; + tensor var_1334_cast_fp16 = softmax(axis = var_822, x = aw_chunk_83_cast_fp16)[name = string("op_1334_cast_fp16")]; + tensor var_1335_cast_fp16 = softmax(axis = var_822, x = aw_chunk_85_cast_fp16)[name = string("op_1335_cast_fp16")]; + tensor var_1336_cast_fp16 = softmax(axis = var_822, x = aw_chunk_87_cast_fp16)[name = string("op_1336_cast_fp16")]; + tensor var_1337_cast_fp16 = softmax(axis = var_822, x = aw_chunk_89_cast_fp16)[name = string("op_1337_cast_fp16")]; + tensor var_1338_cast_fp16 = softmax(axis = var_822, x = aw_chunk_91_cast_fp16)[name = string("op_1338_cast_fp16")]; + tensor var_1339_cast_fp16 = softmax(axis = var_822, x = aw_chunk_93_cast_fp16)[name = string("op_1339_cast_fp16")]; + tensor var_1340_cast_fp16 = softmax(axis = var_822, x = aw_chunk_95_cast_fp16)[name = string("op_1340_cast_fp16")]; + tensor var_1341_cast_fp16 = softmax(axis = var_822, x = aw_chunk_97_cast_fp16)[name = string("op_1341_cast_fp16")]; + tensor var_1342_cast_fp16 = softmax(axis = var_822, x = aw_chunk_99_cast_fp16)[name = string("op_1342_cast_fp16")]; + tensor var_1343_cast_fp16 = softmax(axis = var_822, x = aw_chunk_101_cast_fp16)[name = string("op_1343_cast_fp16")]; + tensor var_1344_cast_fp16 = softmax(axis = var_822, x = aw_chunk_103_cast_fp16)[name = string("op_1344_cast_fp16")]; + tensor var_1345_cast_fp16 = softmax(axis = var_822, x = aw_chunk_105_cast_fp16)[name = string("op_1345_cast_fp16")]; + tensor var_1346_cast_fp16 = softmax(axis = var_822, x = aw_chunk_107_cast_fp16)[name = string("op_1346_cast_fp16")]; + tensor var_1347_cast_fp16 = softmax(axis = var_822, x = aw_chunk_109_cast_fp16)[name = string("op_1347_cast_fp16")]; + tensor var_1348_cast_fp16 = softmax(axis = var_822, x = aw_chunk_111_cast_fp16)[name = string("op_1348_cast_fp16")]; + tensor var_1349_cast_fp16 = softmax(axis = var_822, x = aw_chunk_113_cast_fp16)[name = string("op_1349_cast_fp16")]; + tensor var_1350_cast_fp16 = softmax(axis = var_822, x = aw_chunk_115_cast_fp16)[name = string("op_1350_cast_fp16")]; + tensor var_1351_cast_fp16 = softmax(axis = var_822, x = aw_chunk_117_cast_fp16)[name = string("op_1351_cast_fp16")]; + tensor var_1352_cast_fp16 = softmax(axis = var_822, x = aw_chunk_119_cast_fp16)[name = string("op_1352_cast_fp16")]; + tensor var_1353_cast_fp16 = softmax(axis = var_822, x = aw_chunk_121_cast_fp16)[name = string("op_1353_cast_fp16")]; + tensor var_1354_cast_fp16 = softmax(axis = var_822, x = aw_chunk_123_cast_fp16)[name = string("op_1354_cast_fp16")]; + tensor var_1355_cast_fp16 = softmax(axis = var_822, x = aw_chunk_125_cast_fp16)[name = string("op_1355_cast_fp16")]; + tensor var_1356_cast_fp16 = softmax(axis = var_822, x = aw_chunk_127_cast_fp16)[name = string("op_1356_cast_fp16")]; + string var_1358_equation_0 = const()[name = string("op_1358_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1358_cast_fp16 = einsum(equation = var_1358_equation_0, values = (var_1166_cast_fp16, var_1325_cast_fp16))[name = string("op_1358_cast_fp16")]; + string var_1360_equation_0 = const()[name = string("op_1360_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1360_cast_fp16 = einsum(equation = var_1360_equation_0, values = (var_1166_cast_fp16, var_1326_cast_fp16))[name = string("op_1360_cast_fp16")]; + string var_1362_equation_0 = const()[name = string("op_1362_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1362_cast_fp16 = einsum(equation = var_1362_equation_0, values = (var_1166_cast_fp16, var_1327_cast_fp16))[name = string("op_1362_cast_fp16")]; + string var_1364_equation_0 = const()[name = string("op_1364_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1364_cast_fp16 = einsum(equation = var_1364_equation_0, values = (var_1166_cast_fp16, var_1328_cast_fp16))[name = string("op_1364_cast_fp16")]; + string var_1366_equation_0 = const()[name = string("op_1366_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1366_cast_fp16 = einsum(equation = var_1366_equation_0, values = (var_1170_cast_fp16, var_1329_cast_fp16))[name = string("op_1366_cast_fp16")]; + string var_1368_equation_0 = const()[name = string("op_1368_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1368_cast_fp16 = einsum(equation = var_1368_equation_0, values = (var_1170_cast_fp16, var_1330_cast_fp16))[name = string("op_1368_cast_fp16")]; + string var_1370_equation_0 = const()[name = string("op_1370_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1370_cast_fp16 = einsum(equation = var_1370_equation_0, values = (var_1170_cast_fp16, var_1331_cast_fp16))[name = string("op_1370_cast_fp16")]; + string var_1372_equation_0 = const()[name = string("op_1372_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1372_cast_fp16 = einsum(equation = var_1372_equation_0, values = (var_1170_cast_fp16, var_1332_cast_fp16))[name = string("op_1372_cast_fp16")]; + string var_1374_equation_0 = const()[name = string("op_1374_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1374_cast_fp16 = einsum(equation = var_1374_equation_0, values = (var_1174_cast_fp16, var_1333_cast_fp16))[name = string("op_1374_cast_fp16")]; + string var_1376_equation_0 = const()[name = string("op_1376_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1376_cast_fp16 = einsum(equation = var_1376_equation_0, values = (var_1174_cast_fp16, var_1334_cast_fp16))[name = string("op_1376_cast_fp16")]; + string var_1378_equation_0 = const()[name = string("op_1378_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1378_cast_fp16 = einsum(equation = var_1378_equation_0, values = (var_1174_cast_fp16, var_1335_cast_fp16))[name = string("op_1378_cast_fp16")]; + string var_1380_equation_0 = const()[name = string("op_1380_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1380_cast_fp16 = einsum(equation = var_1380_equation_0, values = (var_1174_cast_fp16, var_1336_cast_fp16))[name = string("op_1380_cast_fp16")]; + string var_1382_equation_0 = const()[name = string("op_1382_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1382_cast_fp16 = einsum(equation = var_1382_equation_0, values = (var_1178_cast_fp16, var_1337_cast_fp16))[name = string("op_1382_cast_fp16")]; + string var_1384_equation_0 = const()[name = string("op_1384_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1384_cast_fp16 = einsum(equation = var_1384_equation_0, values = (var_1178_cast_fp16, var_1338_cast_fp16))[name = string("op_1384_cast_fp16")]; + string var_1386_equation_0 = const()[name = string("op_1386_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1386_cast_fp16 = einsum(equation = var_1386_equation_0, values = (var_1178_cast_fp16, var_1339_cast_fp16))[name = string("op_1386_cast_fp16")]; + string var_1388_equation_0 = const()[name = string("op_1388_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1388_cast_fp16 = einsum(equation = var_1388_equation_0, values = (var_1178_cast_fp16, var_1340_cast_fp16))[name = string("op_1388_cast_fp16")]; + string var_1390_equation_0 = const()[name = string("op_1390_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1390_cast_fp16 = einsum(equation = var_1390_equation_0, values = (var_1182_cast_fp16, var_1341_cast_fp16))[name = string("op_1390_cast_fp16")]; + string var_1392_equation_0 = const()[name = string("op_1392_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1392_cast_fp16 = einsum(equation = var_1392_equation_0, values = (var_1182_cast_fp16, var_1342_cast_fp16))[name = string("op_1392_cast_fp16")]; + string var_1394_equation_0 = const()[name = string("op_1394_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1394_cast_fp16 = einsum(equation = var_1394_equation_0, values = (var_1182_cast_fp16, var_1343_cast_fp16))[name = string("op_1394_cast_fp16")]; + string var_1396_equation_0 = const()[name = string("op_1396_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1396_cast_fp16 = einsum(equation = var_1396_equation_0, values = (var_1182_cast_fp16, var_1344_cast_fp16))[name = string("op_1396_cast_fp16")]; + string var_1398_equation_0 = const()[name = string("op_1398_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1398_cast_fp16 = einsum(equation = var_1398_equation_0, values = (var_1186_cast_fp16, var_1345_cast_fp16))[name = string("op_1398_cast_fp16")]; + string var_1400_equation_0 = const()[name = string("op_1400_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1400_cast_fp16 = einsum(equation = var_1400_equation_0, values = (var_1186_cast_fp16, var_1346_cast_fp16))[name = string("op_1400_cast_fp16")]; + string var_1402_equation_0 = const()[name = string("op_1402_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1402_cast_fp16 = einsum(equation = var_1402_equation_0, values = (var_1186_cast_fp16, var_1347_cast_fp16))[name = string("op_1402_cast_fp16")]; + string var_1404_equation_0 = const()[name = string("op_1404_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1404_cast_fp16 = einsum(equation = var_1404_equation_0, values = (var_1186_cast_fp16, var_1348_cast_fp16))[name = string("op_1404_cast_fp16")]; + string var_1406_equation_0 = const()[name = string("op_1406_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1406_cast_fp16 = einsum(equation = var_1406_equation_0, values = (var_1190_cast_fp16, var_1349_cast_fp16))[name = string("op_1406_cast_fp16")]; + string var_1408_equation_0 = const()[name = string("op_1408_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1408_cast_fp16 = einsum(equation = var_1408_equation_0, values = (var_1190_cast_fp16, var_1350_cast_fp16))[name = string("op_1408_cast_fp16")]; + string var_1410_equation_0 = const()[name = string("op_1410_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1410_cast_fp16 = einsum(equation = var_1410_equation_0, values = (var_1190_cast_fp16, var_1351_cast_fp16))[name = string("op_1410_cast_fp16")]; + string var_1412_equation_0 = const()[name = string("op_1412_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1412_cast_fp16 = einsum(equation = var_1412_equation_0, values = (var_1190_cast_fp16, var_1352_cast_fp16))[name = string("op_1412_cast_fp16")]; + string var_1414_equation_0 = const()[name = string("op_1414_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1414_cast_fp16 = einsum(equation = var_1414_equation_0, values = (var_1194_cast_fp16, var_1353_cast_fp16))[name = string("op_1414_cast_fp16")]; + string var_1416_equation_0 = const()[name = string("op_1416_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1416_cast_fp16 = einsum(equation = var_1416_equation_0, values = (var_1194_cast_fp16, var_1354_cast_fp16))[name = string("op_1416_cast_fp16")]; + string var_1418_equation_0 = const()[name = string("op_1418_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1418_cast_fp16 = einsum(equation = var_1418_equation_0, values = (var_1194_cast_fp16, var_1355_cast_fp16))[name = string("op_1418_cast_fp16")]; + string var_1420_equation_0 = const()[name = string("op_1420_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_1420_cast_fp16 = einsum(equation = var_1420_equation_0, values = (var_1194_cast_fp16, var_1356_cast_fp16))[name = string("op_1420_cast_fp16")]; + bool var_1422_interleave_0 = const()[name = string("op_1422_interleave_0"), val = bool(false)]; + tensor var_1422_cast_fp16 = concat(axis = var_809, interleave = var_1422_interleave_0, values = (var_1358_cast_fp16, var_1360_cast_fp16, var_1362_cast_fp16, var_1364_cast_fp16))[name = string("op_1422_cast_fp16")]; + bool var_1424_interleave_0 = const()[name = string("op_1424_interleave_0"), val = bool(false)]; + tensor var_1424_cast_fp16 = concat(axis = var_809, interleave = var_1424_interleave_0, values = (var_1366_cast_fp16, var_1368_cast_fp16, var_1370_cast_fp16, var_1372_cast_fp16))[name = string("op_1424_cast_fp16")]; + bool var_1426_interleave_0 = const()[name = string("op_1426_interleave_0"), val = bool(false)]; + tensor var_1426_cast_fp16 = concat(axis = var_809, interleave = var_1426_interleave_0, values = (var_1374_cast_fp16, var_1376_cast_fp16, var_1378_cast_fp16, var_1380_cast_fp16))[name = string("op_1426_cast_fp16")]; + bool var_1428_interleave_0 = const()[name = string("op_1428_interleave_0"), val = bool(false)]; + tensor var_1428_cast_fp16 = concat(axis = var_809, interleave = var_1428_interleave_0, values = (var_1382_cast_fp16, var_1384_cast_fp16, var_1386_cast_fp16, var_1388_cast_fp16))[name = string("op_1428_cast_fp16")]; + bool var_1430_interleave_0 = const()[name = string("op_1430_interleave_0"), val = bool(false)]; + tensor var_1430_cast_fp16 = concat(axis = var_809, interleave = var_1430_interleave_0, values = (var_1390_cast_fp16, var_1392_cast_fp16, var_1394_cast_fp16, var_1396_cast_fp16))[name = string("op_1430_cast_fp16")]; + bool var_1432_interleave_0 = const()[name = string("op_1432_interleave_0"), val = bool(false)]; + tensor var_1432_cast_fp16 = concat(axis = var_809, interleave = var_1432_interleave_0, values = (var_1398_cast_fp16, var_1400_cast_fp16, var_1402_cast_fp16, var_1404_cast_fp16))[name = string("op_1432_cast_fp16")]; + bool var_1434_interleave_0 = const()[name = string("op_1434_interleave_0"), val = bool(false)]; + tensor var_1434_cast_fp16 = concat(axis = var_809, interleave = var_1434_interleave_0, values = (var_1406_cast_fp16, var_1408_cast_fp16, var_1410_cast_fp16, var_1412_cast_fp16))[name = string("op_1434_cast_fp16")]; + bool var_1436_interleave_0 = const()[name = string("op_1436_interleave_0"), val = bool(false)]; + tensor var_1436_cast_fp16 = concat(axis = var_809, interleave = var_1436_interleave_0, values = (var_1414_cast_fp16, var_1416_cast_fp16, var_1418_cast_fp16, var_1420_cast_fp16))[name = string("op_1436_cast_fp16")]; + bool input_9_interleave_0 = const()[name = string("input_9_interleave_0"), val = bool(false)]; + tensor input_9_cast_fp16 = concat(axis = var_822, interleave = input_9_interleave_0, values = (var_1422_cast_fp16, var_1424_cast_fp16, var_1426_cast_fp16, var_1428_cast_fp16, var_1430_cast_fp16, var_1432_cast_fp16, var_1434_cast_fp16, var_1436_cast_fp16))[name = string("input_9_cast_fp16")]; + string obj_7_pad_type_0 = const()[name = string("obj_7_pad_type_0"), val = string("valid")]; + tensor obj_7_strides_0 = const()[name = string("obj_7_strides_0"), val = tensor([1, 1])]; + tensor obj_7_pad_0 = const()[name = string("obj_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_7_dilations_0 = const()[name = string("obj_7_dilations_0"), val = tensor([1, 1])]; + int32 obj_7_groups_0 = const()[name = string("obj_7_groups_0"), val = int32(1)]; + tensor layers_1_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11241344)))]; + tensor layers_1_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_1_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11765696)))]; + tensor obj_7_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_bias_to_fp16, dilations = obj_7_dilations_0, groups = obj_7_groups_0, pad = obj_7_pad_0, pad_type = obj_7_pad_type_0, strides = obj_7_strides_0, weight = layers_1_self_attn_o_proj_weight_to_fp16, x = input_9_cast_fp16)[name = string("obj_7_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = obj_7_cast_fp16)[name = string("inputs_7_cast_fp16")]; + tensor out_7_axes_0 = const()[name = string("out_7_axes_0"), val = tensor([1])]; + fp16 var_1455_to_fp16 = const()[name = string("op_1455_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_1455_to_fp16, x = inputs_7_cast_fp16)[name = string("out_7_cast_fp16")]; + tensor input_11_gamma_0_to_fp16 = const()[name = string("input_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11766784)))]; + tensor input_11_beta_0_to_fp16 = const()[name = string("input_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11767872)))]; + fp16 input_11_epsilon_0_to_fp16 = const()[name = string("input_11_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_11_cast_fp16 = batch_norm(beta = input_11_beta_0_to_fp16, epsilon = input_11_epsilon_0_to_fp16, gamma = input_11_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_7_cast_fp16)[name = string("input_11_cast_fp16")]; + string input_13_pad_type_0 = const()[name = string("input_13_pad_type_0"), val = string("valid")]; + tensor input_13_strides_0 = const()[name = string("input_13_strides_0"), val = tensor([1, 1])]; + tensor input_13_pad_0 = const()[name = string("input_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_13_dilations_0 = const()[name = string("input_13_dilations_0"), val = tensor([1, 1])]; + int32 input_13_groups_0 = const()[name = string("input_13_groups_0"), val = int32(1)]; + tensor layers_1_fc1_weight_to_fp16 = const()[name = string("layers_1_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11768960)))]; + tensor layers_1_fc1_bias_to_fp16 = const()[name = string("layers_1_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13866176)))]; + tensor input_13_cast_fp16 = conv(bias = layers_1_fc1_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = layers_1_fc1_weight_to_fp16, x = input_11_cast_fp16)[name = string("input_13_cast_fp16")]; + string input_15_mode_0 = const()[name = string("input_15_mode_0"), val = string("EXACT")]; + tensor input_15_cast_fp16 = gelu(mode = input_15_mode_0, x = input_13_cast_fp16)[name = string("input_15_cast_fp16")]; + string hidden_states_7_pad_type_0 = const()[name = string("hidden_states_7_pad_type_0"), val = string("valid")]; + tensor hidden_states_7_strides_0 = const()[name = string("hidden_states_7_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_7_pad_0 = const()[name = string("hidden_states_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_7_dilations_0 = const()[name = string("hidden_states_7_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_7_groups_0 = const()[name = string("hidden_states_7_groups_0"), val = int32(1)]; + tensor layers_1_fc2_weight_to_fp16 = const()[name = string("layers_1_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(13870336)))]; + tensor layers_1_fc2_bias_to_fp16 = const()[name = string("layers_1_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15967552)))]; + tensor hidden_states_7_cast_fp16 = conv(bias = layers_1_fc2_bias_to_fp16, dilations = hidden_states_7_dilations_0, groups = hidden_states_7_groups_0, pad = hidden_states_7_pad_0, pad_type = hidden_states_7_pad_type_0, strides = hidden_states_7_strides_0, weight = layers_1_fc2_weight_to_fp16, x = input_15_cast_fp16)[name = string("hidden_states_7_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = hidden_states_7_cast_fp16)[name = string("inputs_9_cast_fp16")]; + int32 var_1484 = const()[name = string("op_1484"), val = int32(3)]; + int32 var_1497 = const()[name = string("op_1497"), val = int32(1)]; + tensor out_9_axes_0 = const()[name = string("out_9_axes_0"), val = tensor([1])]; + fp16 var_1514_to_fp16 = const()[name = string("op_1514_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_1514_to_fp16, x = inputs_9_cast_fp16)[name = string("out_9_cast_fp16")]; + tensor obj_9_gamma_0_to_fp16 = const()[name = string("obj_9_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15968640)))]; + tensor obj_9_beta_0_to_fp16 = const()[name = string("obj_9_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15969728)))]; + fp16 obj_9_epsilon_0_to_fp16 = const()[name = string("obj_9_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_9_cast_fp16 = batch_norm(beta = obj_9_beta_0_to_fp16, epsilon = obj_9_epsilon_0_to_fp16, gamma = obj_9_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_9_cast_fp16)[name = string("obj_9_cast_fp16")]; + string query_5_pad_type_0 = const()[name = string("query_5_pad_type_0"), val = string("valid")]; + tensor query_5_strides_0 = const()[name = string("query_5_strides_0"), val = tensor([1, 1])]; + tensor query_5_pad_0 = const()[name = string("query_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_5_dilations_0 = const()[name = string("query_5_dilations_0"), val = tensor([1, 1])]; + int32 query_5_groups_0 = const()[name = string("query_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15970816)))]; + tensor layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16495168)))]; + tensor query_5_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_bias_to_fp16, dilations = query_5_dilations_0, groups = query_5_groups_0, pad = query_5_pad_0, pad_type = query_5_pad_type_0, strides = query_5_strides_0, weight = layers_2_self_attn_q_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("query_5_cast_fp16")]; + string key_5_pad_type_0 = const()[name = string("key_5_pad_type_0"), val = string("valid")]; + tensor key_5_strides_0 = const()[name = string("key_5_strides_0"), val = tensor([1, 1])]; + tensor key_5_pad_0 = const()[name = string("key_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_5_dilations_0 = const()[name = string("key_5_dilations_0"), val = tensor([1, 1])]; + int32 key_5_groups_0 = const()[name = string("key_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16496256)))]; + tensor key_5_cast_fp16 = conv(dilations = key_5_dilations_0, groups = key_5_groups_0, pad = key_5_pad_0, pad_type = key_5_pad_type_0, strides = key_5_strides_0, weight = layers_2_self_attn_k_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("key_5_cast_fp16")]; + string value_5_pad_type_0 = const()[name = string("value_5_pad_type_0"), val = string("valid")]; + tensor value_5_strides_0 = const()[name = string("value_5_strides_0"), val = tensor([1, 1])]; + tensor value_5_pad_0 = const()[name = string("value_5_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_5_dilations_0 = const()[name = string("value_5_dilations_0"), val = tensor([1, 1])]; + int32 value_5_groups_0 = const()[name = string("value_5_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17020608)))]; + tensor layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17544960)))]; + tensor value_5_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_bias_to_fp16, dilations = value_5_dilations_0, groups = value_5_groups_0, pad = value_5_pad_0, pad_type = value_5_pad_type_0, strides = value_5_strides_0, weight = layers_2_self_attn_v_proj_weight_to_fp16, x = obj_9_cast_fp16)[name = string("value_5_cast_fp16")]; + tensor var_1552_begin_0 = const()[name = string("op_1552_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1552_end_0 = const()[name = string("op_1552_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1552_end_mask_0 = const()[name = string("op_1552_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1552_cast_fp16 = slice_by_index(begin = var_1552_begin_0, end = var_1552_end_0, end_mask = var_1552_end_mask_0, x = query_5_cast_fp16)[name = string("op_1552_cast_fp16")]; + tensor var_1556_begin_0 = const()[name = string("op_1556_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1556_end_0 = const()[name = string("op_1556_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_1556_end_mask_0 = const()[name = string("op_1556_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1556_cast_fp16 = slice_by_index(begin = var_1556_begin_0, end = var_1556_end_0, end_mask = var_1556_end_mask_0, x = query_5_cast_fp16)[name = string("op_1556_cast_fp16")]; + tensor var_1560_begin_0 = const()[name = string("op_1560_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1560_end_0 = const()[name = string("op_1560_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_1560_end_mask_0 = const()[name = string("op_1560_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1560_cast_fp16 = slice_by_index(begin = var_1560_begin_0, end = var_1560_end_0, end_mask = var_1560_end_mask_0, x = query_5_cast_fp16)[name = string("op_1560_cast_fp16")]; + tensor var_1564_begin_0 = const()[name = string("op_1564_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1564_end_0 = const()[name = string("op_1564_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_1564_end_mask_0 = const()[name = string("op_1564_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1564_cast_fp16 = slice_by_index(begin = var_1564_begin_0, end = var_1564_end_0, end_mask = var_1564_end_mask_0, x = query_5_cast_fp16)[name = string("op_1564_cast_fp16")]; + tensor var_1568_begin_0 = const()[name = string("op_1568_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1568_end_0 = const()[name = string("op_1568_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_1568_end_mask_0 = const()[name = string("op_1568_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1568_cast_fp16 = slice_by_index(begin = var_1568_begin_0, end = var_1568_end_0, end_mask = var_1568_end_mask_0, x = query_5_cast_fp16)[name = string("op_1568_cast_fp16")]; + tensor var_1572_begin_0 = const()[name = string("op_1572_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1572_end_0 = const()[name = string("op_1572_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1572_end_mask_0 = const()[name = string("op_1572_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1572_cast_fp16 = slice_by_index(begin = var_1572_begin_0, end = var_1572_end_0, end_mask = var_1572_end_mask_0, x = query_5_cast_fp16)[name = string("op_1572_cast_fp16")]; + tensor var_1576_begin_0 = const()[name = string("op_1576_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1576_end_0 = const()[name = string("op_1576_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_1576_end_mask_0 = const()[name = string("op_1576_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1576_cast_fp16 = slice_by_index(begin = var_1576_begin_0, end = var_1576_end_0, end_mask = var_1576_end_mask_0, x = query_5_cast_fp16)[name = string("op_1576_cast_fp16")]; + tensor var_1580_begin_0 = const()[name = string("op_1580_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1580_end_0 = const()[name = string("op_1580_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_1580_end_mask_0 = const()[name = string("op_1580_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1580_cast_fp16 = slice_by_index(begin = var_1580_begin_0, end = var_1580_end_0, end_mask = var_1580_end_mask_0, x = query_5_cast_fp16)[name = string("op_1580_cast_fp16")]; + tensor var_1589_begin_0 = const()[name = string("op_1589_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1589_end_0 = const()[name = string("op_1589_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1589_end_mask_0 = const()[name = string("op_1589_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1589_cast_fp16 = slice_by_index(begin = var_1589_begin_0, end = var_1589_end_0, end_mask = var_1589_end_mask_0, x = var_1552_cast_fp16)[name = string("op_1589_cast_fp16")]; + tensor var_1596_begin_0 = const()[name = string("op_1596_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1596_end_0 = const()[name = string("op_1596_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1596_end_mask_0 = const()[name = string("op_1596_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1596_cast_fp16 = slice_by_index(begin = var_1596_begin_0, end = var_1596_end_0, end_mask = var_1596_end_mask_0, x = var_1552_cast_fp16)[name = string("op_1596_cast_fp16")]; + tensor var_1603_begin_0 = const()[name = string("op_1603_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1603_end_0 = const()[name = string("op_1603_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1603_end_mask_0 = const()[name = string("op_1603_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1603_cast_fp16 = slice_by_index(begin = var_1603_begin_0, end = var_1603_end_0, end_mask = var_1603_end_mask_0, x = var_1552_cast_fp16)[name = string("op_1603_cast_fp16")]; + tensor var_1610_begin_0 = const()[name = string("op_1610_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1610_end_0 = const()[name = string("op_1610_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1610_end_mask_0 = const()[name = string("op_1610_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1610_cast_fp16 = slice_by_index(begin = var_1610_begin_0, end = var_1610_end_0, end_mask = var_1610_end_mask_0, x = var_1552_cast_fp16)[name = string("op_1610_cast_fp16")]; + tensor var_1617_begin_0 = const()[name = string("op_1617_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1617_end_0 = const()[name = string("op_1617_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1617_end_mask_0 = const()[name = string("op_1617_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1617_cast_fp16 = slice_by_index(begin = var_1617_begin_0, end = var_1617_end_0, end_mask = var_1617_end_mask_0, x = var_1556_cast_fp16)[name = string("op_1617_cast_fp16")]; + tensor var_1624_begin_0 = const()[name = string("op_1624_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1624_end_0 = const()[name = string("op_1624_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1624_end_mask_0 = const()[name = string("op_1624_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1624_cast_fp16 = slice_by_index(begin = var_1624_begin_0, end = var_1624_end_0, end_mask = var_1624_end_mask_0, x = var_1556_cast_fp16)[name = string("op_1624_cast_fp16")]; + tensor var_1631_begin_0 = const()[name = string("op_1631_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1631_end_0 = const()[name = string("op_1631_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1631_end_mask_0 = const()[name = string("op_1631_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1631_cast_fp16 = slice_by_index(begin = var_1631_begin_0, end = var_1631_end_0, end_mask = var_1631_end_mask_0, x = var_1556_cast_fp16)[name = string("op_1631_cast_fp16")]; + tensor var_1638_begin_0 = const()[name = string("op_1638_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1638_end_0 = const()[name = string("op_1638_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1638_end_mask_0 = const()[name = string("op_1638_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1638_cast_fp16 = slice_by_index(begin = var_1638_begin_0, end = var_1638_end_0, end_mask = var_1638_end_mask_0, x = var_1556_cast_fp16)[name = string("op_1638_cast_fp16")]; + tensor var_1645_begin_0 = const()[name = string("op_1645_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1645_end_0 = const()[name = string("op_1645_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1645_end_mask_0 = const()[name = string("op_1645_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1645_cast_fp16 = slice_by_index(begin = var_1645_begin_0, end = var_1645_end_0, end_mask = var_1645_end_mask_0, x = var_1560_cast_fp16)[name = string("op_1645_cast_fp16")]; + tensor var_1652_begin_0 = const()[name = string("op_1652_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1652_end_0 = const()[name = string("op_1652_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1652_end_mask_0 = const()[name = string("op_1652_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1652_cast_fp16 = slice_by_index(begin = var_1652_begin_0, end = var_1652_end_0, end_mask = var_1652_end_mask_0, x = var_1560_cast_fp16)[name = string("op_1652_cast_fp16")]; + tensor var_1659_begin_0 = const()[name = string("op_1659_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1659_end_0 = const()[name = string("op_1659_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1659_end_mask_0 = const()[name = string("op_1659_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1659_cast_fp16 = slice_by_index(begin = var_1659_begin_0, end = var_1659_end_0, end_mask = var_1659_end_mask_0, x = var_1560_cast_fp16)[name = string("op_1659_cast_fp16")]; + tensor var_1666_begin_0 = const()[name = string("op_1666_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1666_end_0 = const()[name = string("op_1666_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1666_end_mask_0 = const()[name = string("op_1666_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1666_cast_fp16 = slice_by_index(begin = var_1666_begin_0, end = var_1666_end_0, end_mask = var_1666_end_mask_0, x = var_1560_cast_fp16)[name = string("op_1666_cast_fp16")]; + tensor var_1673_begin_0 = const()[name = string("op_1673_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1673_end_0 = const()[name = string("op_1673_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1673_end_mask_0 = const()[name = string("op_1673_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1673_cast_fp16 = slice_by_index(begin = var_1673_begin_0, end = var_1673_end_0, end_mask = var_1673_end_mask_0, x = var_1564_cast_fp16)[name = string("op_1673_cast_fp16")]; + tensor var_1680_begin_0 = const()[name = string("op_1680_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1680_end_0 = const()[name = string("op_1680_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1680_end_mask_0 = const()[name = string("op_1680_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1680_cast_fp16 = slice_by_index(begin = var_1680_begin_0, end = var_1680_end_0, end_mask = var_1680_end_mask_0, x = var_1564_cast_fp16)[name = string("op_1680_cast_fp16")]; + tensor var_1687_begin_0 = const()[name = string("op_1687_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1687_end_0 = const()[name = string("op_1687_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1687_end_mask_0 = const()[name = string("op_1687_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1687_cast_fp16 = slice_by_index(begin = var_1687_begin_0, end = var_1687_end_0, end_mask = var_1687_end_mask_0, x = var_1564_cast_fp16)[name = string("op_1687_cast_fp16")]; + tensor var_1694_begin_0 = const()[name = string("op_1694_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1694_end_0 = const()[name = string("op_1694_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1694_end_mask_0 = const()[name = string("op_1694_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1694_cast_fp16 = slice_by_index(begin = var_1694_begin_0, end = var_1694_end_0, end_mask = var_1694_end_mask_0, x = var_1564_cast_fp16)[name = string("op_1694_cast_fp16")]; + tensor var_1701_begin_0 = const()[name = string("op_1701_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1701_end_0 = const()[name = string("op_1701_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1701_end_mask_0 = const()[name = string("op_1701_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1701_cast_fp16 = slice_by_index(begin = var_1701_begin_0, end = var_1701_end_0, end_mask = var_1701_end_mask_0, x = var_1568_cast_fp16)[name = string("op_1701_cast_fp16")]; + tensor var_1708_begin_0 = const()[name = string("op_1708_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1708_end_0 = const()[name = string("op_1708_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1708_end_mask_0 = const()[name = string("op_1708_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1708_cast_fp16 = slice_by_index(begin = var_1708_begin_0, end = var_1708_end_0, end_mask = var_1708_end_mask_0, x = var_1568_cast_fp16)[name = string("op_1708_cast_fp16")]; + tensor var_1715_begin_0 = const()[name = string("op_1715_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1715_end_0 = const()[name = string("op_1715_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1715_end_mask_0 = const()[name = string("op_1715_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1715_cast_fp16 = slice_by_index(begin = var_1715_begin_0, end = var_1715_end_0, end_mask = var_1715_end_mask_0, x = var_1568_cast_fp16)[name = string("op_1715_cast_fp16")]; + tensor var_1722_begin_0 = const()[name = string("op_1722_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1722_end_0 = const()[name = string("op_1722_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1722_end_mask_0 = const()[name = string("op_1722_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1722_cast_fp16 = slice_by_index(begin = var_1722_begin_0, end = var_1722_end_0, end_mask = var_1722_end_mask_0, x = var_1568_cast_fp16)[name = string("op_1722_cast_fp16")]; + tensor var_1729_begin_0 = const()[name = string("op_1729_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1729_end_0 = const()[name = string("op_1729_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1729_end_mask_0 = const()[name = string("op_1729_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1729_cast_fp16 = slice_by_index(begin = var_1729_begin_0, end = var_1729_end_0, end_mask = var_1729_end_mask_0, x = var_1572_cast_fp16)[name = string("op_1729_cast_fp16")]; + tensor var_1736_begin_0 = const()[name = string("op_1736_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1736_end_0 = const()[name = string("op_1736_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1736_end_mask_0 = const()[name = string("op_1736_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1736_cast_fp16 = slice_by_index(begin = var_1736_begin_0, end = var_1736_end_0, end_mask = var_1736_end_mask_0, x = var_1572_cast_fp16)[name = string("op_1736_cast_fp16")]; + tensor var_1743_begin_0 = const()[name = string("op_1743_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1743_end_0 = const()[name = string("op_1743_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1743_end_mask_0 = const()[name = string("op_1743_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1743_cast_fp16 = slice_by_index(begin = var_1743_begin_0, end = var_1743_end_0, end_mask = var_1743_end_mask_0, x = var_1572_cast_fp16)[name = string("op_1743_cast_fp16")]; + tensor var_1750_begin_0 = const()[name = string("op_1750_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1750_end_0 = const()[name = string("op_1750_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1750_end_mask_0 = const()[name = string("op_1750_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1750_cast_fp16 = slice_by_index(begin = var_1750_begin_0, end = var_1750_end_0, end_mask = var_1750_end_mask_0, x = var_1572_cast_fp16)[name = string("op_1750_cast_fp16")]; + tensor var_1757_begin_0 = const()[name = string("op_1757_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1757_end_0 = const()[name = string("op_1757_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1757_end_mask_0 = const()[name = string("op_1757_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1757_cast_fp16 = slice_by_index(begin = var_1757_begin_0, end = var_1757_end_0, end_mask = var_1757_end_mask_0, x = var_1576_cast_fp16)[name = string("op_1757_cast_fp16")]; + tensor var_1764_begin_0 = const()[name = string("op_1764_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1764_end_0 = const()[name = string("op_1764_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1764_end_mask_0 = const()[name = string("op_1764_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1764_cast_fp16 = slice_by_index(begin = var_1764_begin_0, end = var_1764_end_0, end_mask = var_1764_end_mask_0, x = var_1576_cast_fp16)[name = string("op_1764_cast_fp16")]; + tensor var_1771_begin_0 = const()[name = string("op_1771_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1771_end_0 = const()[name = string("op_1771_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1771_end_mask_0 = const()[name = string("op_1771_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1771_cast_fp16 = slice_by_index(begin = var_1771_begin_0, end = var_1771_end_0, end_mask = var_1771_end_mask_0, x = var_1576_cast_fp16)[name = string("op_1771_cast_fp16")]; + tensor var_1778_begin_0 = const()[name = string("op_1778_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1778_end_0 = const()[name = string("op_1778_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1778_end_mask_0 = const()[name = string("op_1778_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1778_cast_fp16 = slice_by_index(begin = var_1778_begin_0, end = var_1778_end_0, end_mask = var_1778_end_mask_0, x = var_1576_cast_fp16)[name = string("op_1778_cast_fp16")]; + tensor var_1785_begin_0 = const()[name = string("op_1785_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1785_end_0 = const()[name = string("op_1785_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_1785_end_mask_0 = const()[name = string("op_1785_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1785_cast_fp16 = slice_by_index(begin = var_1785_begin_0, end = var_1785_end_0, end_mask = var_1785_end_mask_0, x = var_1580_cast_fp16)[name = string("op_1785_cast_fp16")]; + tensor var_1792_begin_0 = const()[name = string("op_1792_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_1792_end_0 = const()[name = string("op_1792_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_1792_end_mask_0 = const()[name = string("op_1792_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1792_cast_fp16 = slice_by_index(begin = var_1792_begin_0, end = var_1792_end_0, end_mask = var_1792_end_mask_0, x = var_1580_cast_fp16)[name = string("op_1792_cast_fp16")]; + tensor var_1799_begin_0 = const()[name = string("op_1799_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_1799_end_0 = const()[name = string("op_1799_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_1799_end_mask_0 = const()[name = string("op_1799_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1799_cast_fp16 = slice_by_index(begin = var_1799_begin_0, end = var_1799_end_0, end_mask = var_1799_end_mask_0, x = var_1580_cast_fp16)[name = string("op_1799_cast_fp16")]; + tensor var_1806_begin_0 = const()[name = string("op_1806_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_1806_end_0 = const()[name = string("op_1806_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1806_end_mask_0 = const()[name = string("op_1806_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1806_cast_fp16 = slice_by_index(begin = var_1806_begin_0, end = var_1806_end_0, end_mask = var_1806_end_mask_0, x = var_1580_cast_fp16)[name = string("op_1806_cast_fp16")]; + tensor k_5_perm_0 = const()[name = string("k_5_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_1811_begin_0 = const()[name = string("op_1811_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1811_end_0 = const()[name = string("op_1811_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_1811_end_mask_0 = const()[name = string("op_1811_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_5_cast_fp16 = transpose(perm = k_5_perm_0, x = key_5_cast_fp16)[name = string("transpose_3")]; + tensor var_1811_cast_fp16 = slice_by_index(begin = var_1811_begin_0, end = var_1811_end_0, end_mask = var_1811_end_mask_0, x = k_5_cast_fp16)[name = string("op_1811_cast_fp16")]; + tensor var_1815_begin_0 = const()[name = string("op_1815_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_1815_end_0 = const()[name = string("op_1815_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_1815_end_mask_0 = const()[name = string("op_1815_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1815_cast_fp16 = slice_by_index(begin = var_1815_begin_0, end = var_1815_end_0, end_mask = var_1815_end_mask_0, x = k_5_cast_fp16)[name = string("op_1815_cast_fp16")]; + tensor var_1819_begin_0 = const()[name = string("op_1819_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_1819_end_0 = const()[name = string("op_1819_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_1819_end_mask_0 = const()[name = string("op_1819_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1819_cast_fp16 = slice_by_index(begin = var_1819_begin_0, end = var_1819_end_0, end_mask = var_1819_end_mask_0, x = k_5_cast_fp16)[name = string("op_1819_cast_fp16")]; + tensor var_1823_begin_0 = const()[name = string("op_1823_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_1823_end_0 = const()[name = string("op_1823_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_1823_end_mask_0 = const()[name = string("op_1823_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1823_cast_fp16 = slice_by_index(begin = var_1823_begin_0, end = var_1823_end_0, end_mask = var_1823_end_mask_0, x = k_5_cast_fp16)[name = string("op_1823_cast_fp16")]; + tensor var_1827_begin_0 = const()[name = string("op_1827_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_1827_end_0 = const()[name = string("op_1827_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_1827_end_mask_0 = const()[name = string("op_1827_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1827_cast_fp16 = slice_by_index(begin = var_1827_begin_0, end = var_1827_end_0, end_mask = var_1827_end_mask_0, x = k_5_cast_fp16)[name = string("op_1827_cast_fp16")]; + tensor var_1831_begin_0 = const()[name = string("op_1831_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_1831_end_0 = const()[name = string("op_1831_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_1831_end_mask_0 = const()[name = string("op_1831_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1831_cast_fp16 = slice_by_index(begin = var_1831_begin_0, end = var_1831_end_0, end_mask = var_1831_end_mask_0, x = k_5_cast_fp16)[name = string("op_1831_cast_fp16")]; + tensor var_1835_begin_0 = const()[name = string("op_1835_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_1835_end_0 = const()[name = string("op_1835_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_1835_end_mask_0 = const()[name = string("op_1835_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1835_cast_fp16 = slice_by_index(begin = var_1835_begin_0, end = var_1835_end_0, end_mask = var_1835_end_mask_0, x = k_5_cast_fp16)[name = string("op_1835_cast_fp16")]; + tensor var_1839_begin_0 = const()[name = string("op_1839_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_1839_end_0 = const()[name = string("op_1839_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_1839_end_mask_0 = const()[name = string("op_1839_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1839_cast_fp16 = slice_by_index(begin = var_1839_begin_0, end = var_1839_end_0, end_mask = var_1839_end_mask_0, x = k_5_cast_fp16)[name = string("op_1839_cast_fp16")]; + tensor var_1841_begin_0 = const()[name = string("op_1841_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1841_end_0 = const()[name = string("op_1841_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_1841_end_mask_0 = const()[name = string("op_1841_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1841_cast_fp16 = slice_by_index(begin = var_1841_begin_0, end = var_1841_end_0, end_mask = var_1841_end_mask_0, x = value_5_cast_fp16)[name = string("op_1841_cast_fp16")]; + tensor var_1845_begin_0 = const()[name = string("op_1845_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_1845_end_0 = const()[name = string("op_1845_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_1845_end_mask_0 = const()[name = string("op_1845_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1845_cast_fp16 = slice_by_index(begin = var_1845_begin_0, end = var_1845_end_0, end_mask = var_1845_end_mask_0, x = value_5_cast_fp16)[name = string("op_1845_cast_fp16")]; + tensor var_1849_begin_0 = const()[name = string("op_1849_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_1849_end_0 = const()[name = string("op_1849_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_1849_end_mask_0 = const()[name = string("op_1849_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1849_cast_fp16 = slice_by_index(begin = var_1849_begin_0, end = var_1849_end_0, end_mask = var_1849_end_mask_0, x = value_5_cast_fp16)[name = string("op_1849_cast_fp16")]; + tensor var_1853_begin_0 = const()[name = string("op_1853_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_1853_end_0 = const()[name = string("op_1853_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_1853_end_mask_0 = const()[name = string("op_1853_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1853_cast_fp16 = slice_by_index(begin = var_1853_begin_0, end = var_1853_end_0, end_mask = var_1853_end_mask_0, x = value_5_cast_fp16)[name = string("op_1853_cast_fp16")]; + tensor var_1857_begin_0 = const()[name = string("op_1857_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_1857_end_0 = const()[name = string("op_1857_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_1857_end_mask_0 = const()[name = string("op_1857_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1857_cast_fp16 = slice_by_index(begin = var_1857_begin_0, end = var_1857_end_0, end_mask = var_1857_end_mask_0, x = value_5_cast_fp16)[name = string("op_1857_cast_fp16")]; + tensor var_1861_begin_0 = const()[name = string("op_1861_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_1861_end_0 = const()[name = string("op_1861_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_1861_end_mask_0 = const()[name = string("op_1861_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1861_cast_fp16 = slice_by_index(begin = var_1861_begin_0, end = var_1861_end_0, end_mask = var_1861_end_mask_0, x = value_5_cast_fp16)[name = string("op_1861_cast_fp16")]; + tensor var_1865_begin_0 = const()[name = string("op_1865_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_1865_end_0 = const()[name = string("op_1865_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_1865_end_mask_0 = const()[name = string("op_1865_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1865_cast_fp16 = slice_by_index(begin = var_1865_begin_0, end = var_1865_end_0, end_mask = var_1865_end_mask_0, x = value_5_cast_fp16)[name = string("op_1865_cast_fp16")]; + tensor var_1869_begin_0 = const()[name = string("op_1869_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_1869_end_0 = const()[name = string("op_1869_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_1869_end_mask_0 = const()[name = string("op_1869_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_1869_cast_fp16 = slice_by_index(begin = var_1869_begin_0, end = var_1869_end_0, end_mask = var_1869_end_mask_0, x = value_5_cast_fp16)[name = string("op_1869_cast_fp16")]; + string _SplitHeadsQ__mh_w_129_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_129_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_129_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_129_equation_0, values = (var_1811_cast_fp16, var_1589_cast_fp16))[name = string("_SplitHeadsQ__mh_w_129_cast_fp16")]; + string _SplitHeadsQ__mh_w_131_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_131_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_131_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_131_equation_0, values = (var_1811_cast_fp16, var_1596_cast_fp16))[name = string("_SplitHeadsQ__mh_w_131_cast_fp16")]; + string _SplitHeadsQ__mh_w_133_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_133_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_133_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_133_equation_0, values = (var_1811_cast_fp16, var_1603_cast_fp16))[name = string("_SplitHeadsQ__mh_w_133_cast_fp16")]; + string _SplitHeadsQ__mh_w_135_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_135_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_135_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_135_equation_0, values = (var_1811_cast_fp16, var_1610_cast_fp16))[name = string("_SplitHeadsQ__mh_w_135_cast_fp16")]; + string _SplitHeadsQ__mh_w_137_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_137_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_137_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_137_equation_0, values = (var_1815_cast_fp16, var_1617_cast_fp16))[name = string("_SplitHeadsQ__mh_w_137_cast_fp16")]; + string _SplitHeadsQ__mh_w_139_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_139_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_139_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_139_equation_0, values = (var_1815_cast_fp16, var_1624_cast_fp16))[name = string("_SplitHeadsQ__mh_w_139_cast_fp16")]; + string _SplitHeadsQ__mh_w_141_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_141_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_141_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_141_equation_0, values = (var_1815_cast_fp16, var_1631_cast_fp16))[name = string("_SplitHeadsQ__mh_w_141_cast_fp16")]; + string _SplitHeadsQ__mh_w_143_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_143_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_143_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_143_equation_0, values = (var_1815_cast_fp16, var_1638_cast_fp16))[name = string("_SplitHeadsQ__mh_w_143_cast_fp16")]; + string _SplitHeadsQ__mh_w_145_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_145_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_145_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_145_equation_0, values = (var_1819_cast_fp16, var_1645_cast_fp16))[name = string("_SplitHeadsQ__mh_w_145_cast_fp16")]; + string _SplitHeadsQ__mh_w_147_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_147_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_147_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_147_equation_0, values = (var_1819_cast_fp16, var_1652_cast_fp16))[name = string("_SplitHeadsQ__mh_w_147_cast_fp16")]; + string _SplitHeadsQ__mh_w_149_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_149_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_149_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_149_equation_0, values = (var_1819_cast_fp16, var_1659_cast_fp16))[name = string("_SplitHeadsQ__mh_w_149_cast_fp16")]; + string _SplitHeadsQ__mh_w_151_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_151_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_151_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_151_equation_0, values = (var_1819_cast_fp16, var_1666_cast_fp16))[name = string("_SplitHeadsQ__mh_w_151_cast_fp16")]; + string _SplitHeadsQ__mh_w_153_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_153_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_153_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_153_equation_0, values = (var_1823_cast_fp16, var_1673_cast_fp16))[name = string("_SplitHeadsQ__mh_w_153_cast_fp16")]; + string _SplitHeadsQ__mh_w_155_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_155_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_155_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_155_equation_0, values = (var_1823_cast_fp16, var_1680_cast_fp16))[name = string("_SplitHeadsQ__mh_w_155_cast_fp16")]; + string _SplitHeadsQ__mh_w_157_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_157_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_157_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_157_equation_0, values = (var_1823_cast_fp16, var_1687_cast_fp16))[name = string("_SplitHeadsQ__mh_w_157_cast_fp16")]; + string _SplitHeadsQ__mh_w_159_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_159_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_159_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_159_equation_0, values = (var_1823_cast_fp16, var_1694_cast_fp16))[name = string("_SplitHeadsQ__mh_w_159_cast_fp16")]; + string _SplitHeadsQ__mh_w_161_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_161_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_161_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_161_equation_0, values = (var_1827_cast_fp16, var_1701_cast_fp16))[name = string("_SplitHeadsQ__mh_w_161_cast_fp16")]; + string _SplitHeadsQ__mh_w_163_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_163_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_163_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_163_equation_0, values = (var_1827_cast_fp16, var_1708_cast_fp16))[name = string("_SplitHeadsQ__mh_w_163_cast_fp16")]; + string _SplitHeadsQ__mh_w_165_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_165_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_165_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_165_equation_0, values = (var_1827_cast_fp16, var_1715_cast_fp16))[name = string("_SplitHeadsQ__mh_w_165_cast_fp16")]; + string _SplitHeadsQ__mh_w_167_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_167_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_167_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_167_equation_0, values = (var_1827_cast_fp16, var_1722_cast_fp16))[name = string("_SplitHeadsQ__mh_w_167_cast_fp16")]; + string _SplitHeadsQ__mh_w_169_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_169_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_169_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_169_equation_0, values = (var_1831_cast_fp16, var_1729_cast_fp16))[name = string("_SplitHeadsQ__mh_w_169_cast_fp16")]; + string _SplitHeadsQ__mh_w_171_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_171_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_171_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_171_equation_0, values = (var_1831_cast_fp16, var_1736_cast_fp16))[name = string("_SplitHeadsQ__mh_w_171_cast_fp16")]; + string _SplitHeadsQ__mh_w_173_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_173_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_173_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_173_equation_0, values = (var_1831_cast_fp16, var_1743_cast_fp16))[name = string("_SplitHeadsQ__mh_w_173_cast_fp16")]; + string _SplitHeadsQ__mh_w_175_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_175_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_175_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_175_equation_0, values = (var_1831_cast_fp16, var_1750_cast_fp16))[name = string("_SplitHeadsQ__mh_w_175_cast_fp16")]; + string _SplitHeadsQ__mh_w_177_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_177_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_177_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_177_equation_0, values = (var_1835_cast_fp16, var_1757_cast_fp16))[name = string("_SplitHeadsQ__mh_w_177_cast_fp16")]; + string _SplitHeadsQ__mh_w_179_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_179_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_179_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_179_equation_0, values = (var_1835_cast_fp16, var_1764_cast_fp16))[name = string("_SplitHeadsQ__mh_w_179_cast_fp16")]; + string _SplitHeadsQ__mh_w_181_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_181_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_181_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_181_equation_0, values = (var_1835_cast_fp16, var_1771_cast_fp16))[name = string("_SplitHeadsQ__mh_w_181_cast_fp16")]; + string _SplitHeadsQ__mh_w_183_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_183_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_183_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_183_equation_0, values = (var_1835_cast_fp16, var_1778_cast_fp16))[name = string("_SplitHeadsQ__mh_w_183_cast_fp16")]; + string _SplitHeadsQ__mh_w_185_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_185_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_185_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_185_equation_0, values = (var_1839_cast_fp16, var_1785_cast_fp16))[name = string("_SplitHeadsQ__mh_w_185_cast_fp16")]; + string _SplitHeadsQ__mh_w_187_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_187_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_187_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_187_equation_0, values = (var_1839_cast_fp16, var_1792_cast_fp16))[name = string("_SplitHeadsQ__mh_w_187_cast_fp16")]; + string _SplitHeadsQ__mh_w_189_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_189_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_189_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_189_equation_0, values = (var_1839_cast_fp16, var_1799_cast_fp16))[name = string("_SplitHeadsQ__mh_w_189_cast_fp16")]; + string _SplitHeadsQ__mh_w_191_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_191_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_191_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_191_equation_0, values = (var_1839_cast_fp16, var_1806_cast_fp16))[name = string("_SplitHeadsQ__mh_w_191_cast_fp16")]; + fp16 var_1936_to_fp16 = const()[name = string("op_1936_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_129_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_129_cast_fp16, y = var_1936_to_fp16)[name = string("aw_chunk_129_cast_fp16")]; + fp16 var_1938_to_fp16 = const()[name = string("op_1938_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_131_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_131_cast_fp16, y = var_1938_to_fp16)[name = string("aw_chunk_131_cast_fp16")]; + fp16 var_1940_to_fp16 = const()[name = string("op_1940_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_133_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_133_cast_fp16, y = var_1940_to_fp16)[name = string("aw_chunk_133_cast_fp16")]; + fp16 var_1942_to_fp16 = const()[name = string("op_1942_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_135_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_135_cast_fp16, y = var_1942_to_fp16)[name = string("aw_chunk_135_cast_fp16")]; + fp16 var_1944_to_fp16 = const()[name = string("op_1944_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_137_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_137_cast_fp16, y = var_1944_to_fp16)[name = string("aw_chunk_137_cast_fp16")]; + fp16 var_1946_to_fp16 = const()[name = string("op_1946_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_139_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_139_cast_fp16, y = var_1946_to_fp16)[name = string("aw_chunk_139_cast_fp16")]; + fp16 var_1948_to_fp16 = const()[name = string("op_1948_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_141_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_141_cast_fp16, y = var_1948_to_fp16)[name = string("aw_chunk_141_cast_fp16")]; + fp16 var_1950_to_fp16 = const()[name = string("op_1950_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_143_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_143_cast_fp16, y = var_1950_to_fp16)[name = string("aw_chunk_143_cast_fp16")]; + fp16 var_1952_to_fp16 = const()[name = string("op_1952_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_145_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_145_cast_fp16, y = var_1952_to_fp16)[name = string("aw_chunk_145_cast_fp16")]; + fp16 var_1954_to_fp16 = const()[name = string("op_1954_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_147_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_147_cast_fp16, y = var_1954_to_fp16)[name = string("aw_chunk_147_cast_fp16")]; + fp16 var_1956_to_fp16 = const()[name = string("op_1956_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_149_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_149_cast_fp16, y = var_1956_to_fp16)[name = string("aw_chunk_149_cast_fp16")]; + fp16 var_1958_to_fp16 = const()[name = string("op_1958_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_151_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_151_cast_fp16, y = var_1958_to_fp16)[name = string("aw_chunk_151_cast_fp16")]; + fp16 var_1960_to_fp16 = const()[name = string("op_1960_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_153_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_153_cast_fp16, y = var_1960_to_fp16)[name = string("aw_chunk_153_cast_fp16")]; + fp16 var_1962_to_fp16 = const()[name = string("op_1962_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_155_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_155_cast_fp16, y = var_1962_to_fp16)[name = string("aw_chunk_155_cast_fp16")]; + fp16 var_1964_to_fp16 = const()[name = string("op_1964_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_157_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_157_cast_fp16, y = var_1964_to_fp16)[name = string("aw_chunk_157_cast_fp16")]; + fp16 var_1966_to_fp16 = const()[name = string("op_1966_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_159_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_159_cast_fp16, y = var_1966_to_fp16)[name = string("aw_chunk_159_cast_fp16")]; + fp16 var_1968_to_fp16 = const()[name = string("op_1968_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_161_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_161_cast_fp16, y = var_1968_to_fp16)[name = string("aw_chunk_161_cast_fp16")]; + fp16 var_1970_to_fp16 = const()[name = string("op_1970_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_163_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_163_cast_fp16, y = var_1970_to_fp16)[name = string("aw_chunk_163_cast_fp16")]; + fp16 var_1972_to_fp16 = const()[name = string("op_1972_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_165_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_165_cast_fp16, y = var_1972_to_fp16)[name = string("aw_chunk_165_cast_fp16")]; + fp16 var_1974_to_fp16 = const()[name = string("op_1974_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_167_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_167_cast_fp16, y = var_1974_to_fp16)[name = string("aw_chunk_167_cast_fp16")]; + fp16 var_1976_to_fp16 = const()[name = string("op_1976_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_169_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_169_cast_fp16, y = var_1976_to_fp16)[name = string("aw_chunk_169_cast_fp16")]; + fp16 var_1978_to_fp16 = const()[name = string("op_1978_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_171_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_171_cast_fp16, y = var_1978_to_fp16)[name = string("aw_chunk_171_cast_fp16")]; + fp16 var_1980_to_fp16 = const()[name = string("op_1980_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_173_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_173_cast_fp16, y = var_1980_to_fp16)[name = string("aw_chunk_173_cast_fp16")]; + fp16 var_1982_to_fp16 = const()[name = string("op_1982_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_175_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_175_cast_fp16, y = var_1982_to_fp16)[name = string("aw_chunk_175_cast_fp16")]; + fp16 var_1984_to_fp16 = const()[name = string("op_1984_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_177_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_177_cast_fp16, y = var_1984_to_fp16)[name = string("aw_chunk_177_cast_fp16")]; + fp16 var_1986_to_fp16 = const()[name = string("op_1986_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_179_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_179_cast_fp16, y = var_1986_to_fp16)[name = string("aw_chunk_179_cast_fp16")]; + fp16 var_1988_to_fp16 = const()[name = string("op_1988_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_181_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_181_cast_fp16, y = var_1988_to_fp16)[name = string("aw_chunk_181_cast_fp16")]; + fp16 var_1990_to_fp16 = const()[name = string("op_1990_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_183_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_183_cast_fp16, y = var_1990_to_fp16)[name = string("aw_chunk_183_cast_fp16")]; + fp16 var_1992_to_fp16 = const()[name = string("op_1992_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_185_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_185_cast_fp16, y = var_1992_to_fp16)[name = string("aw_chunk_185_cast_fp16")]; + fp16 var_1994_to_fp16 = const()[name = string("op_1994_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_187_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_187_cast_fp16, y = var_1994_to_fp16)[name = string("aw_chunk_187_cast_fp16")]; + fp16 var_1996_to_fp16 = const()[name = string("op_1996_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_189_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_189_cast_fp16, y = var_1996_to_fp16)[name = string("aw_chunk_189_cast_fp16")]; + fp16 var_1998_to_fp16 = const()[name = string("op_1998_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_191_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_191_cast_fp16, y = var_1998_to_fp16)[name = string("aw_chunk_191_cast_fp16")]; + tensor var_2000_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_129_cast_fp16)[name = string("op_2000_cast_fp16")]; + tensor var_2001_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_131_cast_fp16)[name = string("op_2001_cast_fp16")]; + tensor var_2002_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_133_cast_fp16)[name = string("op_2002_cast_fp16")]; + tensor var_2003_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_135_cast_fp16)[name = string("op_2003_cast_fp16")]; + tensor var_2004_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_137_cast_fp16)[name = string("op_2004_cast_fp16")]; + tensor var_2005_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_139_cast_fp16)[name = string("op_2005_cast_fp16")]; + tensor var_2006_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_141_cast_fp16)[name = string("op_2006_cast_fp16")]; + tensor var_2007_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_143_cast_fp16)[name = string("op_2007_cast_fp16")]; + tensor var_2008_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_145_cast_fp16)[name = string("op_2008_cast_fp16")]; + tensor var_2009_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_147_cast_fp16)[name = string("op_2009_cast_fp16")]; + tensor var_2010_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_149_cast_fp16)[name = string("op_2010_cast_fp16")]; + tensor var_2011_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_151_cast_fp16)[name = string("op_2011_cast_fp16")]; + tensor var_2012_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_153_cast_fp16)[name = string("op_2012_cast_fp16")]; + tensor var_2013_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_155_cast_fp16)[name = string("op_2013_cast_fp16")]; + tensor var_2014_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_157_cast_fp16)[name = string("op_2014_cast_fp16")]; + tensor var_2015_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_159_cast_fp16)[name = string("op_2015_cast_fp16")]; + tensor var_2016_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_161_cast_fp16)[name = string("op_2016_cast_fp16")]; + tensor var_2017_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_163_cast_fp16)[name = string("op_2017_cast_fp16")]; + tensor var_2018_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_165_cast_fp16)[name = string("op_2018_cast_fp16")]; + tensor var_2019_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_167_cast_fp16)[name = string("op_2019_cast_fp16")]; + tensor var_2020_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_169_cast_fp16)[name = string("op_2020_cast_fp16")]; + tensor var_2021_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_171_cast_fp16)[name = string("op_2021_cast_fp16")]; + tensor var_2022_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_173_cast_fp16)[name = string("op_2022_cast_fp16")]; + tensor var_2023_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_175_cast_fp16)[name = string("op_2023_cast_fp16")]; + tensor var_2024_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_177_cast_fp16)[name = string("op_2024_cast_fp16")]; + tensor var_2025_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_179_cast_fp16)[name = string("op_2025_cast_fp16")]; + tensor var_2026_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_181_cast_fp16)[name = string("op_2026_cast_fp16")]; + tensor var_2027_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_183_cast_fp16)[name = string("op_2027_cast_fp16")]; + tensor var_2028_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_185_cast_fp16)[name = string("op_2028_cast_fp16")]; + tensor var_2029_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_187_cast_fp16)[name = string("op_2029_cast_fp16")]; + tensor var_2030_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_189_cast_fp16)[name = string("op_2030_cast_fp16")]; + tensor var_2031_cast_fp16 = softmax(axis = var_1497, x = aw_chunk_191_cast_fp16)[name = string("op_2031_cast_fp16")]; + string var_2033_equation_0 = const()[name = string("op_2033_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2033_cast_fp16 = einsum(equation = var_2033_equation_0, values = (var_1841_cast_fp16, var_2000_cast_fp16))[name = string("op_2033_cast_fp16")]; + string var_2035_equation_0 = const()[name = string("op_2035_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2035_cast_fp16 = einsum(equation = var_2035_equation_0, values = (var_1841_cast_fp16, var_2001_cast_fp16))[name = string("op_2035_cast_fp16")]; + string var_2037_equation_0 = const()[name = string("op_2037_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2037_cast_fp16 = einsum(equation = var_2037_equation_0, values = (var_1841_cast_fp16, var_2002_cast_fp16))[name = string("op_2037_cast_fp16")]; + string var_2039_equation_0 = const()[name = string("op_2039_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2039_cast_fp16 = einsum(equation = var_2039_equation_0, values = (var_1841_cast_fp16, var_2003_cast_fp16))[name = string("op_2039_cast_fp16")]; + string var_2041_equation_0 = const()[name = string("op_2041_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2041_cast_fp16 = einsum(equation = var_2041_equation_0, values = (var_1845_cast_fp16, var_2004_cast_fp16))[name = string("op_2041_cast_fp16")]; + string var_2043_equation_0 = const()[name = string("op_2043_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2043_cast_fp16 = einsum(equation = var_2043_equation_0, values = (var_1845_cast_fp16, var_2005_cast_fp16))[name = string("op_2043_cast_fp16")]; + string var_2045_equation_0 = const()[name = string("op_2045_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2045_cast_fp16 = einsum(equation = var_2045_equation_0, values = (var_1845_cast_fp16, var_2006_cast_fp16))[name = string("op_2045_cast_fp16")]; + string var_2047_equation_0 = const()[name = string("op_2047_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2047_cast_fp16 = einsum(equation = var_2047_equation_0, values = (var_1845_cast_fp16, var_2007_cast_fp16))[name = string("op_2047_cast_fp16")]; + string var_2049_equation_0 = const()[name = string("op_2049_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2049_cast_fp16 = einsum(equation = var_2049_equation_0, values = (var_1849_cast_fp16, var_2008_cast_fp16))[name = string("op_2049_cast_fp16")]; + string var_2051_equation_0 = const()[name = string("op_2051_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2051_cast_fp16 = einsum(equation = var_2051_equation_0, values = (var_1849_cast_fp16, var_2009_cast_fp16))[name = string("op_2051_cast_fp16")]; + string var_2053_equation_0 = const()[name = string("op_2053_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2053_cast_fp16 = einsum(equation = var_2053_equation_0, values = (var_1849_cast_fp16, var_2010_cast_fp16))[name = string("op_2053_cast_fp16")]; + string var_2055_equation_0 = const()[name = string("op_2055_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2055_cast_fp16 = einsum(equation = var_2055_equation_0, values = (var_1849_cast_fp16, var_2011_cast_fp16))[name = string("op_2055_cast_fp16")]; + string var_2057_equation_0 = const()[name = string("op_2057_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2057_cast_fp16 = einsum(equation = var_2057_equation_0, values = (var_1853_cast_fp16, var_2012_cast_fp16))[name = string("op_2057_cast_fp16")]; + string var_2059_equation_0 = const()[name = string("op_2059_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2059_cast_fp16 = einsum(equation = var_2059_equation_0, values = (var_1853_cast_fp16, var_2013_cast_fp16))[name = string("op_2059_cast_fp16")]; + string var_2061_equation_0 = const()[name = string("op_2061_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2061_cast_fp16 = einsum(equation = var_2061_equation_0, values = (var_1853_cast_fp16, var_2014_cast_fp16))[name = string("op_2061_cast_fp16")]; + string var_2063_equation_0 = const()[name = string("op_2063_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2063_cast_fp16 = einsum(equation = var_2063_equation_0, values = (var_1853_cast_fp16, var_2015_cast_fp16))[name = string("op_2063_cast_fp16")]; + string var_2065_equation_0 = const()[name = string("op_2065_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2065_cast_fp16 = einsum(equation = var_2065_equation_0, values = (var_1857_cast_fp16, var_2016_cast_fp16))[name = string("op_2065_cast_fp16")]; + string var_2067_equation_0 = const()[name = string("op_2067_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2067_cast_fp16 = einsum(equation = var_2067_equation_0, values = (var_1857_cast_fp16, var_2017_cast_fp16))[name = string("op_2067_cast_fp16")]; + string var_2069_equation_0 = const()[name = string("op_2069_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2069_cast_fp16 = einsum(equation = var_2069_equation_0, values = (var_1857_cast_fp16, var_2018_cast_fp16))[name = string("op_2069_cast_fp16")]; + string var_2071_equation_0 = const()[name = string("op_2071_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2071_cast_fp16 = einsum(equation = var_2071_equation_0, values = (var_1857_cast_fp16, var_2019_cast_fp16))[name = string("op_2071_cast_fp16")]; + string var_2073_equation_0 = const()[name = string("op_2073_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2073_cast_fp16 = einsum(equation = var_2073_equation_0, values = (var_1861_cast_fp16, var_2020_cast_fp16))[name = string("op_2073_cast_fp16")]; + string var_2075_equation_0 = const()[name = string("op_2075_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2075_cast_fp16 = einsum(equation = var_2075_equation_0, values = (var_1861_cast_fp16, var_2021_cast_fp16))[name = string("op_2075_cast_fp16")]; + string var_2077_equation_0 = const()[name = string("op_2077_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2077_cast_fp16 = einsum(equation = var_2077_equation_0, values = (var_1861_cast_fp16, var_2022_cast_fp16))[name = string("op_2077_cast_fp16")]; + string var_2079_equation_0 = const()[name = string("op_2079_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2079_cast_fp16 = einsum(equation = var_2079_equation_0, values = (var_1861_cast_fp16, var_2023_cast_fp16))[name = string("op_2079_cast_fp16")]; + string var_2081_equation_0 = const()[name = string("op_2081_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2081_cast_fp16 = einsum(equation = var_2081_equation_0, values = (var_1865_cast_fp16, var_2024_cast_fp16))[name = string("op_2081_cast_fp16")]; + string var_2083_equation_0 = const()[name = string("op_2083_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2083_cast_fp16 = einsum(equation = var_2083_equation_0, values = (var_1865_cast_fp16, var_2025_cast_fp16))[name = string("op_2083_cast_fp16")]; + string var_2085_equation_0 = const()[name = string("op_2085_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2085_cast_fp16 = einsum(equation = var_2085_equation_0, values = (var_1865_cast_fp16, var_2026_cast_fp16))[name = string("op_2085_cast_fp16")]; + string var_2087_equation_0 = const()[name = string("op_2087_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2087_cast_fp16 = einsum(equation = var_2087_equation_0, values = (var_1865_cast_fp16, var_2027_cast_fp16))[name = string("op_2087_cast_fp16")]; + string var_2089_equation_0 = const()[name = string("op_2089_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2089_cast_fp16 = einsum(equation = var_2089_equation_0, values = (var_1869_cast_fp16, var_2028_cast_fp16))[name = string("op_2089_cast_fp16")]; + string var_2091_equation_0 = const()[name = string("op_2091_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2091_cast_fp16 = einsum(equation = var_2091_equation_0, values = (var_1869_cast_fp16, var_2029_cast_fp16))[name = string("op_2091_cast_fp16")]; + string var_2093_equation_0 = const()[name = string("op_2093_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2093_cast_fp16 = einsum(equation = var_2093_equation_0, values = (var_1869_cast_fp16, var_2030_cast_fp16))[name = string("op_2093_cast_fp16")]; + string var_2095_equation_0 = const()[name = string("op_2095_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2095_cast_fp16 = einsum(equation = var_2095_equation_0, values = (var_1869_cast_fp16, var_2031_cast_fp16))[name = string("op_2095_cast_fp16")]; + bool var_2097_interleave_0 = const()[name = string("op_2097_interleave_0"), val = bool(false)]; + tensor var_2097_cast_fp16 = concat(axis = var_1484, interleave = var_2097_interleave_0, values = (var_2033_cast_fp16, var_2035_cast_fp16, var_2037_cast_fp16, var_2039_cast_fp16))[name = string("op_2097_cast_fp16")]; + bool var_2099_interleave_0 = const()[name = string("op_2099_interleave_0"), val = bool(false)]; + tensor var_2099_cast_fp16 = concat(axis = var_1484, interleave = var_2099_interleave_0, values = (var_2041_cast_fp16, var_2043_cast_fp16, var_2045_cast_fp16, var_2047_cast_fp16))[name = string("op_2099_cast_fp16")]; + bool var_2101_interleave_0 = const()[name = string("op_2101_interleave_0"), val = bool(false)]; + tensor var_2101_cast_fp16 = concat(axis = var_1484, interleave = var_2101_interleave_0, values = (var_2049_cast_fp16, var_2051_cast_fp16, var_2053_cast_fp16, var_2055_cast_fp16))[name = string("op_2101_cast_fp16")]; + bool var_2103_interleave_0 = const()[name = string("op_2103_interleave_0"), val = bool(false)]; + tensor var_2103_cast_fp16 = concat(axis = var_1484, interleave = var_2103_interleave_0, values = (var_2057_cast_fp16, var_2059_cast_fp16, var_2061_cast_fp16, var_2063_cast_fp16))[name = string("op_2103_cast_fp16")]; + bool var_2105_interleave_0 = const()[name = string("op_2105_interleave_0"), val = bool(false)]; + tensor var_2105_cast_fp16 = concat(axis = var_1484, interleave = var_2105_interleave_0, values = (var_2065_cast_fp16, var_2067_cast_fp16, var_2069_cast_fp16, var_2071_cast_fp16))[name = string("op_2105_cast_fp16")]; + bool var_2107_interleave_0 = const()[name = string("op_2107_interleave_0"), val = bool(false)]; + tensor var_2107_cast_fp16 = concat(axis = var_1484, interleave = var_2107_interleave_0, values = (var_2073_cast_fp16, var_2075_cast_fp16, var_2077_cast_fp16, var_2079_cast_fp16))[name = string("op_2107_cast_fp16")]; + bool var_2109_interleave_0 = const()[name = string("op_2109_interleave_0"), val = bool(false)]; + tensor var_2109_cast_fp16 = concat(axis = var_1484, interleave = var_2109_interleave_0, values = (var_2081_cast_fp16, var_2083_cast_fp16, var_2085_cast_fp16, var_2087_cast_fp16))[name = string("op_2109_cast_fp16")]; + bool var_2111_interleave_0 = const()[name = string("op_2111_interleave_0"), val = bool(false)]; + tensor var_2111_cast_fp16 = concat(axis = var_1484, interleave = var_2111_interleave_0, values = (var_2089_cast_fp16, var_2091_cast_fp16, var_2093_cast_fp16, var_2095_cast_fp16))[name = string("op_2111_cast_fp16")]; + bool input_17_interleave_0 = const()[name = string("input_17_interleave_0"), val = bool(false)]; + tensor input_17_cast_fp16 = concat(axis = var_1497, interleave = input_17_interleave_0, values = (var_2097_cast_fp16, var_2099_cast_fp16, var_2101_cast_fp16, var_2103_cast_fp16, var_2105_cast_fp16, var_2107_cast_fp16, var_2109_cast_fp16, var_2111_cast_fp16))[name = string("input_17_cast_fp16")]; + string obj_11_pad_type_0 = const()[name = string("obj_11_pad_type_0"), val = string("valid")]; + tensor obj_11_strides_0 = const()[name = string("obj_11_strides_0"), val = tensor([1, 1])]; + tensor obj_11_pad_0 = const()[name = string("obj_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_11_dilations_0 = const()[name = string("obj_11_dilations_0"), val = tensor([1, 1])]; + int32 obj_11_groups_0 = const()[name = string("obj_11_groups_0"), val = int32(1)]; + tensor layers_2_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(17546048)))]; + tensor layers_2_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_2_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18070400)))]; + tensor obj_11_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_bias_to_fp16, dilations = obj_11_dilations_0, groups = obj_11_groups_0, pad = obj_11_pad_0, pad_type = obj_11_pad_type_0, strides = obj_11_strides_0, weight = layers_2_self_attn_o_proj_weight_to_fp16, x = input_17_cast_fp16)[name = string("obj_11_cast_fp16")]; + tensor inputs_11_cast_fp16 = add(x = inputs_9_cast_fp16, y = obj_11_cast_fp16)[name = string("inputs_11_cast_fp16")]; + tensor out_11_axes_0 = const()[name = string("out_11_axes_0"), val = tensor([1])]; + fp16 var_2130_to_fp16 = const()[name = string("op_2130_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_2130_to_fp16, x = inputs_11_cast_fp16)[name = string("out_11_cast_fp16")]; + tensor input_19_gamma_0_to_fp16 = const()[name = string("input_19_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18071488)))]; + tensor input_19_beta_0_to_fp16 = const()[name = string("input_19_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18072576)))]; + fp16 input_19_epsilon_0_to_fp16 = const()[name = string("input_19_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_19_cast_fp16 = batch_norm(beta = input_19_beta_0_to_fp16, epsilon = input_19_epsilon_0_to_fp16, gamma = input_19_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_11_cast_fp16)[name = string("input_19_cast_fp16")]; + string input_21_pad_type_0 = const()[name = string("input_21_pad_type_0"), val = string("valid")]; + tensor input_21_strides_0 = const()[name = string("input_21_strides_0"), val = tensor([1, 1])]; + tensor input_21_pad_0 = const()[name = string("input_21_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_21_dilations_0 = const()[name = string("input_21_dilations_0"), val = tensor([1, 1])]; + int32 input_21_groups_0 = const()[name = string("input_21_groups_0"), val = int32(1)]; + tensor layers_2_fc1_weight_to_fp16 = const()[name = string("layers_2_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18073664)))]; + tensor layers_2_fc1_bias_to_fp16 = const()[name = string("layers_2_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20170880)))]; + tensor input_21_cast_fp16 = conv(bias = layers_2_fc1_bias_to_fp16, dilations = input_21_dilations_0, groups = input_21_groups_0, pad = input_21_pad_0, pad_type = input_21_pad_type_0, strides = input_21_strides_0, weight = layers_2_fc1_weight_to_fp16, x = input_19_cast_fp16)[name = string("input_21_cast_fp16")]; + string input_23_mode_0 = const()[name = string("input_23_mode_0"), val = string("EXACT")]; + tensor input_23_cast_fp16 = gelu(mode = input_23_mode_0, x = input_21_cast_fp16)[name = string("input_23_cast_fp16")]; + string hidden_states_9_pad_type_0 = const()[name = string("hidden_states_9_pad_type_0"), val = string("valid")]; + tensor hidden_states_9_strides_0 = const()[name = string("hidden_states_9_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_9_pad_0 = const()[name = string("hidden_states_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_9_dilations_0 = const()[name = string("hidden_states_9_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_9_groups_0 = const()[name = string("hidden_states_9_groups_0"), val = int32(1)]; + tensor layers_2_fc2_weight_to_fp16 = const()[name = string("layers_2_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20175040)))]; + tensor layers_2_fc2_bias_to_fp16 = const()[name = string("layers_2_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22272256)))]; + tensor hidden_states_9_cast_fp16 = conv(bias = layers_2_fc2_bias_to_fp16, dilations = hidden_states_9_dilations_0, groups = hidden_states_9_groups_0, pad = hidden_states_9_pad_0, pad_type = hidden_states_9_pad_type_0, strides = hidden_states_9_strides_0, weight = layers_2_fc2_weight_to_fp16, x = input_23_cast_fp16)[name = string("hidden_states_9_cast_fp16")]; + tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = hidden_states_9_cast_fp16)[name = string("inputs_13_cast_fp16")]; + int32 var_2159 = const()[name = string("op_2159"), val = int32(3)]; + int32 var_2172 = const()[name = string("op_2172"), val = int32(1)]; + tensor out_13_axes_0 = const()[name = string("out_13_axes_0"), val = tensor([1])]; + fp16 var_2189_to_fp16 = const()[name = string("op_2189_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_2189_to_fp16, x = inputs_13_cast_fp16)[name = string("out_13_cast_fp16")]; + tensor obj_13_gamma_0_to_fp16 = const()[name = string("obj_13_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22273344)))]; + tensor obj_13_beta_0_to_fp16 = const()[name = string("obj_13_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22274432)))]; + fp16 obj_13_epsilon_0_to_fp16 = const()[name = string("obj_13_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_13_cast_fp16 = batch_norm(beta = obj_13_beta_0_to_fp16, epsilon = obj_13_epsilon_0_to_fp16, gamma = obj_13_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_13_cast_fp16)[name = string("obj_13_cast_fp16")]; + string query_7_pad_type_0 = const()[name = string("query_7_pad_type_0"), val = string("valid")]; + tensor query_7_strides_0 = const()[name = string("query_7_strides_0"), val = tensor([1, 1])]; + tensor query_7_pad_0 = const()[name = string("query_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_7_dilations_0 = const()[name = string("query_7_dilations_0"), val = tensor([1, 1])]; + int32 query_7_groups_0 = const()[name = string("query_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22275520)))]; + tensor layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22799872)))]; + tensor query_7_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_bias_to_fp16, dilations = query_7_dilations_0, groups = query_7_groups_0, pad = query_7_pad_0, pad_type = query_7_pad_type_0, strides = query_7_strides_0, weight = layers_3_self_attn_q_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("query_7_cast_fp16")]; + string key_7_pad_type_0 = const()[name = string("key_7_pad_type_0"), val = string("valid")]; + tensor key_7_strides_0 = const()[name = string("key_7_strides_0"), val = tensor([1, 1])]; + tensor key_7_pad_0 = const()[name = string("key_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_7_dilations_0 = const()[name = string("key_7_dilations_0"), val = tensor([1, 1])]; + int32 key_7_groups_0 = const()[name = string("key_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22800960)))]; + tensor key_7_cast_fp16 = conv(dilations = key_7_dilations_0, groups = key_7_groups_0, pad = key_7_pad_0, pad_type = key_7_pad_type_0, strides = key_7_strides_0, weight = layers_3_self_attn_k_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("key_7_cast_fp16")]; + string value_7_pad_type_0 = const()[name = string("value_7_pad_type_0"), val = string("valid")]; + tensor value_7_strides_0 = const()[name = string("value_7_strides_0"), val = tensor([1, 1])]; + tensor value_7_pad_0 = const()[name = string("value_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_7_dilations_0 = const()[name = string("value_7_dilations_0"), val = tensor([1, 1])]; + int32 value_7_groups_0 = const()[name = string("value_7_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23325312)))]; + tensor layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23849664)))]; + tensor value_7_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_bias_to_fp16, dilations = value_7_dilations_0, groups = value_7_groups_0, pad = value_7_pad_0, pad_type = value_7_pad_type_0, strides = value_7_strides_0, weight = layers_3_self_attn_v_proj_weight_to_fp16, x = obj_13_cast_fp16)[name = string("value_7_cast_fp16")]; + tensor var_2227_begin_0 = const()[name = string("op_2227_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2227_end_0 = const()[name = string("op_2227_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2227_end_mask_0 = const()[name = string("op_2227_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2227_cast_fp16 = slice_by_index(begin = var_2227_begin_0, end = var_2227_end_0, end_mask = var_2227_end_mask_0, x = query_7_cast_fp16)[name = string("op_2227_cast_fp16")]; + tensor var_2231_begin_0 = const()[name = string("op_2231_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2231_end_0 = const()[name = string("op_2231_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_2231_end_mask_0 = const()[name = string("op_2231_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2231_cast_fp16 = slice_by_index(begin = var_2231_begin_0, end = var_2231_end_0, end_mask = var_2231_end_mask_0, x = query_7_cast_fp16)[name = string("op_2231_cast_fp16")]; + tensor var_2235_begin_0 = const()[name = string("op_2235_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2235_end_0 = const()[name = string("op_2235_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_2235_end_mask_0 = const()[name = string("op_2235_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2235_cast_fp16 = slice_by_index(begin = var_2235_begin_0, end = var_2235_end_0, end_mask = var_2235_end_mask_0, x = query_7_cast_fp16)[name = string("op_2235_cast_fp16")]; + tensor var_2239_begin_0 = const()[name = string("op_2239_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2239_end_0 = const()[name = string("op_2239_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_2239_end_mask_0 = const()[name = string("op_2239_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2239_cast_fp16 = slice_by_index(begin = var_2239_begin_0, end = var_2239_end_0, end_mask = var_2239_end_mask_0, x = query_7_cast_fp16)[name = string("op_2239_cast_fp16")]; + tensor var_2243_begin_0 = const()[name = string("op_2243_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2243_end_0 = const()[name = string("op_2243_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_2243_end_mask_0 = const()[name = string("op_2243_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2243_cast_fp16 = slice_by_index(begin = var_2243_begin_0, end = var_2243_end_0, end_mask = var_2243_end_mask_0, x = query_7_cast_fp16)[name = string("op_2243_cast_fp16")]; + tensor var_2247_begin_0 = const()[name = string("op_2247_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2247_end_0 = const()[name = string("op_2247_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_2247_end_mask_0 = const()[name = string("op_2247_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2247_cast_fp16 = slice_by_index(begin = var_2247_begin_0, end = var_2247_end_0, end_mask = var_2247_end_mask_0, x = query_7_cast_fp16)[name = string("op_2247_cast_fp16")]; + tensor var_2251_begin_0 = const()[name = string("op_2251_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2251_end_0 = const()[name = string("op_2251_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_2251_end_mask_0 = const()[name = string("op_2251_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2251_cast_fp16 = slice_by_index(begin = var_2251_begin_0, end = var_2251_end_0, end_mask = var_2251_end_mask_0, x = query_7_cast_fp16)[name = string("op_2251_cast_fp16")]; + tensor var_2255_begin_0 = const()[name = string("op_2255_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2255_end_0 = const()[name = string("op_2255_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_2255_end_mask_0 = const()[name = string("op_2255_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2255_cast_fp16 = slice_by_index(begin = var_2255_begin_0, end = var_2255_end_0, end_mask = var_2255_end_mask_0, x = query_7_cast_fp16)[name = string("op_2255_cast_fp16")]; + tensor var_2264_begin_0 = const()[name = string("op_2264_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2264_end_0 = const()[name = string("op_2264_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2264_end_mask_0 = const()[name = string("op_2264_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2264_cast_fp16 = slice_by_index(begin = var_2264_begin_0, end = var_2264_end_0, end_mask = var_2264_end_mask_0, x = var_2227_cast_fp16)[name = string("op_2264_cast_fp16")]; + tensor var_2271_begin_0 = const()[name = string("op_2271_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2271_end_0 = const()[name = string("op_2271_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2271_end_mask_0 = const()[name = string("op_2271_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2271_cast_fp16 = slice_by_index(begin = var_2271_begin_0, end = var_2271_end_0, end_mask = var_2271_end_mask_0, x = var_2227_cast_fp16)[name = string("op_2271_cast_fp16")]; + tensor var_2278_begin_0 = const()[name = string("op_2278_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2278_end_0 = const()[name = string("op_2278_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2278_end_mask_0 = const()[name = string("op_2278_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2278_cast_fp16 = slice_by_index(begin = var_2278_begin_0, end = var_2278_end_0, end_mask = var_2278_end_mask_0, x = var_2227_cast_fp16)[name = string("op_2278_cast_fp16")]; + tensor var_2285_begin_0 = const()[name = string("op_2285_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2285_end_0 = const()[name = string("op_2285_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2285_end_mask_0 = const()[name = string("op_2285_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2285_cast_fp16 = slice_by_index(begin = var_2285_begin_0, end = var_2285_end_0, end_mask = var_2285_end_mask_0, x = var_2227_cast_fp16)[name = string("op_2285_cast_fp16")]; + tensor var_2292_begin_0 = const()[name = string("op_2292_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2292_end_0 = const()[name = string("op_2292_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2292_end_mask_0 = const()[name = string("op_2292_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2292_cast_fp16 = slice_by_index(begin = var_2292_begin_0, end = var_2292_end_0, end_mask = var_2292_end_mask_0, x = var_2231_cast_fp16)[name = string("op_2292_cast_fp16")]; + tensor var_2299_begin_0 = const()[name = string("op_2299_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2299_end_0 = const()[name = string("op_2299_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2299_end_mask_0 = const()[name = string("op_2299_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2299_cast_fp16 = slice_by_index(begin = var_2299_begin_0, end = var_2299_end_0, end_mask = var_2299_end_mask_0, x = var_2231_cast_fp16)[name = string("op_2299_cast_fp16")]; + tensor var_2306_begin_0 = const()[name = string("op_2306_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2306_end_0 = const()[name = string("op_2306_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2306_end_mask_0 = const()[name = string("op_2306_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2306_cast_fp16 = slice_by_index(begin = var_2306_begin_0, end = var_2306_end_0, end_mask = var_2306_end_mask_0, x = var_2231_cast_fp16)[name = string("op_2306_cast_fp16")]; + tensor var_2313_begin_0 = const()[name = string("op_2313_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2313_end_0 = const()[name = string("op_2313_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2313_end_mask_0 = const()[name = string("op_2313_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2313_cast_fp16 = slice_by_index(begin = var_2313_begin_0, end = var_2313_end_0, end_mask = var_2313_end_mask_0, x = var_2231_cast_fp16)[name = string("op_2313_cast_fp16")]; + tensor var_2320_begin_0 = const()[name = string("op_2320_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2320_end_0 = const()[name = string("op_2320_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2320_end_mask_0 = const()[name = string("op_2320_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2320_cast_fp16 = slice_by_index(begin = var_2320_begin_0, end = var_2320_end_0, end_mask = var_2320_end_mask_0, x = var_2235_cast_fp16)[name = string("op_2320_cast_fp16")]; + tensor var_2327_begin_0 = const()[name = string("op_2327_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2327_end_0 = const()[name = string("op_2327_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2327_end_mask_0 = const()[name = string("op_2327_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2327_cast_fp16 = slice_by_index(begin = var_2327_begin_0, end = var_2327_end_0, end_mask = var_2327_end_mask_0, x = var_2235_cast_fp16)[name = string("op_2327_cast_fp16")]; + tensor var_2334_begin_0 = const()[name = string("op_2334_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2334_end_0 = const()[name = string("op_2334_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2334_end_mask_0 = const()[name = string("op_2334_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2334_cast_fp16 = slice_by_index(begin = var_2334_begin_0, end = var_2334_end_0, end_mask = var_2334_end_mask_0, x = var_2235_cast_fp16)[name = string("op_2334_cast_fp16")]; + tensor var_2341_begin_0 = const()[name = string("op_2341_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2341_end_0 = const()[name = string("op_2341_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2341_end_mask_0 = const()[name = string("op_2341_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2341_cast_fp16 = slice_by_index(begin = var_2341_begin_0, end = var_2341_end_0, end_mask = var_2341_end_mask_0, x = var_2235_cast_fp16)[name = string("op_2341_cast_fp16")]; + tensor var_2348_begin_0 = const()[name = string("op_2348_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2348_end_0 = const()[name = string("op_2348_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2348_end_mask_0 = const()[name = string("op_2348_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2348_cast_fp16 = slice_by_index(begin = var_2348_begin_0, end = var_2348_end_0, end_mask = var_2348_end_mask_0, x = var_2239_cast_fp16)[name = string("op_2348_cast_fp16")]; + tensor var_2355_begin_0 = const()[name = string("op_2355_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2355_end_0 = const()[name = string("op_2355_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2355_end_mask_0 = const()[name = string("op_2355_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2355_cast_fp16 = slice_by_index(begin = var_2355_begin_0, end = var_2355_end_0, end_mask = var_2355_end_mask_0, x = var_2239_cast_fp16)[name = string("op_2355_cast_fp16")]; + tensor var_2362_begin_0 = const()[name = string("op_2362_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2362_end_0 = const()[name = string("op_2362_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2362_end_mask_0 = const()[name = string("op_2362_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2362_cast_fp16 = slice_by_index(begin = var_2362_begin_0, end = var_2362_end_0, end_mask = var_2362_end_mask_0, x = var_2239_cast_fp16)[name = string("op_2362_cast_fp16")]; + tensor var_2369_begin_0 = const()[name = string("op_2369_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2369_end_0 = const()[name = string("op_2369_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2369_end_mask_0 = const()[name = string("op_2369_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2369_cast_fp16 = slice_by_index(begin = var_2369_begin_0, end = var_2369_end_0, end_mask = var_2369_end_mask_0, x = var_2239_cast_fp16)[name = string("op_2369_cast_fp16")]; + tensor var_2376_begin_0 = const()[name = string("op_2376_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2376_end_0 = const()[name = string("op_2376_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2376_end_mask_0 = const()[name = string("op_2376_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2376_cast_fp16 = slice_by_index(begin = var_2376_begin_0, end = var_2376_end_0, end_mask = var_2376_end_mask_0, x = var_2243_cast_fp16)[name = string("op_2376_cast_fp16")]; + tensor var_2383_begin_0 = const()[name = string("op_2383_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2383_end_0 = const()[name = string("op_2383_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2383_end_mask_0 = const()[name = string("op_2383_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2383_cast_fp16 = slice_by_index(begin = var_2383_begin_0, end = var_2383_end_0, end_mask = var_2383_end_mask_0, x = var_2243_cast_fp16)[name = string("op_2383_cast_fp16")]; + tensor var_2390_begin_0 = const()[name = string("op_2390_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2390_end_0 = const()[name = string("op_2390_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2390_end_mask_0 = const()[name = string("op_2390_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2390_cast_fp16 = slice_by_index(begin = var_2390_begin_0, end = var_2390_end_0, end_mask = var_2390_end_mask_0, x = var_2243_cast_fp16)[name = string("op_2390_cast_fp16")]; + tensor var_2397_begin_0 = const()[name = string("op_2397_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2397_end_0 = const()[name = string("op_2397_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2397_end_mask_0 = const()[name = string("op_2397_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2397_cast_fp16 = slice_by_index(begin = var_2397_begin_0, end = var_2397_end_0, end_mask = var_2397_end_mask_0, x = var_2243_cast_fp16)[name = string("op_2397_cast_fp16")]; + tensor var_2404_begin_0 = const()[name = string("op_2404_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2404_end_0 = const()[name = string("op_2404_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2404_end_mask_0 = const()[name = string("op_2404_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2404_cast_fp16 = slice_by_index(begin = var_2404_begin_0, end = var_2404_end_0, end_mask = var_2404_end_mask_0, x = var_2247_cast_fp16)[name = string("op_2404_cast_fp16")]; + tensor var_2411_begin_0 = const()[name = string("op_2411_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2411_end_0 = const()[name = string("op_2411_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2411_end_mask_0 = const()[name = string("op_2411_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2411_cast_fp16 = slice_by_index(begin = var_2411_begin_0, end = var_2411_end_0, end_mask = var_2411_end_mask_0, x = var_2247_cast_fp16)[name = string("op_2411_cast_fp16")]; + tensor var_2418_begin_0 = const()[name = string("op_2418_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2418_end_0 = const()[name = string("op_2418_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2418_end_mask_0 = const()[name = string("op_2418_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2418_cast_fp16 = slice_by_index(begin = var_2418_begin_0, end = var_2418_end_0, end_mask = var_2418_end_mask_0, x = var_2247_cast_fp16)[name = string("op_2418_cast_fp16")]; + tensor var_2425_begin_0 = const()[name = string("op_2425_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2425_end_0 = const()[name = string("op_2425_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2425_end_mask_0 = const()[name = string("op_2425_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2425_cast_fp16 = slice_by_index(begin = var_2425_begin_0, end = var_2425_end_0, end_mask = var_2425_end_mask_0, x = var_2247_cast_fp16)[name = string("op_2425_cast_fp16")]; + tensor var_2432_begin_0 = const()[name = string("op_2432_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2432_end_0 = const()[name = string("op_2432_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2432_end_mask_0 = const()[name = string("op_2432_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2432_cast_fp16 = slice_by_index(begin = var_2432_begin_0, end = var_2432_end_0, end_mask = var_2432_end_mask_0, x = var_2251_cast_fp16)[name = string("op_2432_cast_fp16")]; + tensor var_2439_begin_0 = const()[name = string("op_2439_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2439_end_0 = const()[name = string("op_2439_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2439_end_mask_0 = const()[name = string("op_2439_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2439_cast_fp16 = slice_by_index(begin = var_2439_begin_0, end = var_2439_end_0, end_mask = var_2439_end_mask_0, x = var_2251_cast_fp16)[name = string("op_2439_cast_fp16")]; + tensor var_2446_begin_0 = const()[name = string("op_2446_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2446_end_0 = const()[name = string("op_2446_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2446_end_mask_0 = const()[name = string("op_2446_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2446_cast_fp16 = slice_by_index(begin = var_2446_begin_0, end = var_2446_end_0, end_mask = var_2446_end_mask_0, x = var_2251_cast_fp16)[name = string("op_2446_cast_fp16")]; + tensor var_2453_begin_0 = const()[name = string("op_2453_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2453_end_0 = const()[name = string("op_2453_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2453_end_mask_0 = const()[name = string("op_2453_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2453_cast_fp16 = slice_by_index(begin = var_2453_begin_0, end = var_2453_end_0, end_mask = var_2453_end_mask_0, x = var_2251_cast_fp16)[name = string("op_2453_cast_fp16")]; + tensor var_2460_begin_0 = const()[name = string("op_2460_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2460_end_0 = const()[name = string("op_2460_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2460_end_mask_0 = const()[name = string("op_2460_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2460_cast_fp16 = slice_by_index(begin = var_2460_begin_0, end = var_2460_end_0, end_mask = var_2460_end_mask_0, x = var_2255_cast_fp16)[name = string("op_2460_cast_fp16")]; + tensor var_2467_begin_0 = const()[name = string("op_2467_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2467_end_0 = const()[name = string("op_2467_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2467_end_mask_0 = const()[name = string("op_2467_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2467_cast_fp16 = slice_by_index(begin = var_2467_begin_0, end = var_2467_end_0, end_mask = var_2467_end_mask_0, x = var_2255_cast_fp16)[name = string("op_2467_cast_fp16")]; + tensor var_2474_begin_0 = const()[name = string("op_2474_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2474_end_0 = const()[name = string("op_2474_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2474_end_mask_0 = const()[name = string("op_2474_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2474_cast_fp16 = slice_by_index(begin = var_2474_begin_0, end = var_2474_end_0, end_mask = var_2474_end_mask_0, x = var_2255_cast_fp16)[name = string("op_2474_cast_fp16")]; + tensor var_2481_begin_0 = const()[name = string("op_2481_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2481_end_0 = const()[name = string("op_2481_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2481_end_mask_0 = const()[name = string("op_2481_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2481_cast_fp16 = slice_by_index(begin = var_2481_begin_0, end = var_2481_end_0, end_mask = var_2481_end_mask_0, x = var_2255_cast_fp16)[name = string("op_2481_cast_fp16")]; + tensor k_7_perm_0 = const()[name = string("k_7_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_2486_begin_0 = const()[name = string("op_2486_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2486_end_0 = const()[name = string("op_2486_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_2486_end_mask_0 = const()[name = string("op_2486_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_7_cast_fp16 = transpose(perm = k_7_perm_0, x = key_7_cast_fp16)[name = string("transpose_2")]; + tensor var_2486_cast_fp16 = slice_by_index(begin = var_2486_begin_0, end = var_2486_end_0, end_mask = var_2486_end_mask_0, x = k_7_cast_fp16)[name = string("op_2486_cast_fp16")]; + tensor var_2490_begin_0 = const()[name = string("op_2490_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_2490_end_0 = const()[name = string("op_2490_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_2490_end_mask_0 = const()[name = string("op_2490_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2490_cast_fp16 = slice_by_index(begin = var_2490_begin_0, end = var_2490_end_0, end_mask = var_2490_end_mask_0, x = k_7_cast_fp16)[name = string("op_2490_cast_fp16")]; + tensor var_2494_begin_0 = const()[name = string("op_2494_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_2494_end_0 = const()[name = string("op_2494_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_2494_end_mask_0 = const()[name = string("op_2494_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2494_cast_fp16 = slice_by_index(begin = var_2494_begin_0, end = var_2494_end_0, end_mask = var_2494_end_mask_0, x = k_7_cast_fp16)[name = string("op_2494_cast_fp16")]; + tensor var_2498_begin_0 = const()[name = string("op_2498_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_2498_end_0 = const()[name = string("op_2498_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_2498_end_mask_0 = const()[name = string("op_2498_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2498_cast_fp16 = slice_by_index(begin = var_2498_begin_0, end = var_2498_end_0, end_mask = var_2498_end_mask_0, x = k_7_cast_fp16)[name = string("op_2498_cast_fp16")]; + tensor var_2502_begin_0 = const()[name = string("op_2502_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_2502_end_0 = const()[name = string("op_2502_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_2502_end_mask_0 = const()[name = string("op_2502_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2502_cast_fp16 = slice_by_index(begin = var_2502_begin_0, end = var_2502_end_0, end_mask = var_2502_end_mask_0, x = k_7_cast_fp16)[name = string("op_2502_cast_fp16")]; + tensor var_2506_begin_0 = const()[name = string("op_2506_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_2506_end_0 = const()[name = string("op_2506_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_2506_end_mask_0 = const()[name = string("op_2506_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2506_cast_fp16 = slice_by_index(begin = var_2506_begin_0, end = var_2506_end_0, end_mask = var_2506_end_mask_0, x = k_7_cast_fp16)[name = string("op_2506_cast_fp16")]; + tensor var_2510_begin_0 = const()[name = string("op_2510_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_2510_end_0 = const()[name = string("op_2510_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_2510_end_mask_0 = const()[name = string("op_2510_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2510_cast_fp16 = slice_by_index(begin = var_2510_begin_0, end = var_2510_end_0, end_mask = var_2510_end_mask_0, x = k_7_cast_fp16)[name = string("op_2510_cast_fp16")]; + tensor var_2514_begin_0 = const()[name = string("op_2514_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_2514_end_0 = const()[name = string("op_2514_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_2514_end_mask_0 = const()[name = string("op_2514_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2514_cast_fp16 = slice_by_index(begin = var_2514_begin_0, end = var_2514_end_0, end_mask = var_2514_end_mask_0, x = k_7_cast_fp16)[name = string("op_2514_cast_fp16")]; + tensor var_2516_begin_0 = const()[name = string("op_2516_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2516_end_0 = const()[name = string("op_2516_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2516_end_mask_0 = const()[name = string("op_2516_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2516_cast_fp16 = slice_by_index(begin = var_2516_begin_0, end = var_2516_end_0, end_mask = var_2516_end_mask_0, x = value_7_cast_fp16)[name = string("op_2516_cast_fp16")]; + tensor var_2520_begin_0 = const()[name = string("op_2520_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2520_end_0 = const()[name = string("op_2520_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_2520_end_mask_0 = const()[name = string("op_2520_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2520_cast_fp16 = slice_by_index(begin = var_2520_begin_0, end = var_2520_end_0, end_mask = var_2520_end_mask_0, x = value_7_cast_fp16)[name = string("op_2520_cast_fp16")]; + tensor var_2524_begin_0 = const()[name = string("op_2524_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2524_end_0 = const()[name = string("op_2524_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_2524_end_mask_0 = const()[name = string("op_2524_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2524_cast_fp16 = slice_by_index(begin = var_2524_begin_0, end = var_2524_end_0, end_mask = var_2524_end_mask_0, x = value_7_cast_fp16)[name = string("op_2524_cast_fp16")]; + tensor var_2528_begin_0 = const()[name = string("op_2528_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2528_end_0 = const()[name = string("op_2528_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_2528_end_mask_0 = const()[name = string("op_2528_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2528_cast_fp16 = slice_by_index(begin = var_2528_begin_0, end = var_2528_end_0, end_mask = var_2528_end_mask_0, x = value_7_cast_fp16)[name = string("op_2528_cast_fp16")]; + tensor var_2532_begin_0 = const()[name = string("op_2532_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2532_end_0 = const()[name = string("op_2532_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_2532_end_mask_0 = const()[name = string("op_2532_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2532_cast_fp16 = slice_by_index(begin = var_2532_begin_0, end = var_2532_end_0, end_mask = var_2532_end_mask_0, x = value_7_cast_fp16)[name = string("op_2532_cast_fp16")]; + tensor var_2536_begin_0 = const()[name = string("op_2536_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2536_end_0 = const()[name = string("op_2536_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_2536_end_mask_0 = const()[name = string("op_2536_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2536_cast_fp16 = slice_by_index(begin = var_2536_begin_0, end = var_2536_end_0, end_mask = var_2536_end_mask_0, x = value_7_cast_fp16)[name = string("op_2536_cast_fp16")]; + tensor var_2540_begin_0 = const()[name = string("op_2540_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2540_end_0 = const()[name = string("op_2540_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_2540_end_mask_0 = const()[name = string("op_2540_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2540_cast_fp16 = slice_by_index(begin = var_2540_begin_0, end = var_2540_end_0, end_mask = var_2540_end_mask_0, x = value_7_cast_fp16)[name = string("op_2540_cast_fp16")]; + tensor var_2544_begin_0 = const()[name = string("op_2544_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2544_end_0 = const()[name = string("op_2544_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_2544_end_mask_0 = const()[name = string("op_2544_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2544_cast_fp16 = slice_by_index(begin = var_2544_begin_0, end = var_2544_end_0, end_mask = var_2544_end_mask_0, x = value_7_cast_fp16)[name = string("op_2544_cast_fp16")]; + string _SplitHeadsQ__mh_w_193_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_193_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_193_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_193_equation_0, values = (var_2486_cast_fp16, var_2264_cast_fp16))[name = string("_SplitHeadsQ__mh_w_193_cast_fp16")]; + string _SplitHeadsQ__mh_w_195_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_195_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_195_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_195_equation_0, values = (var_2486_cast_fp16, var_2271_cast_fp16))[name = string("_SplitHeadsQ__mh_w_195_cast_fp16")]; + string _SplitHeadsQ__mh_w_197_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_197_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_197_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_197_equation_0, values = (var_2486_cast_fp16, var_2278_cast_fp16))[name = string("_SplitHeadsQ__mh_w_197_cast_fp16")]; + string _SplitHeadsQ__mh_w_199_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_199_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_199_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_199_equation_0, values = (var_2486_cast_fp16, var_2285_cast_fp16))[name = string("_SplitHeadsQ__mh_w_199_cast_fp16")]; + string _SplitHeadsQ__mh_w_201_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_201_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_201_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_201_equation_0, values = (var_2490_cast_fp16, var_2292_cast_fp16))[name = string("_SplitHeadsQ__mh_w_201_cast_fp16")]; + string _SplitHeadsQ__mh_w_203_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_203_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_203_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_203_equation_0, values = (var_2490_cast_fp16, var_2299_cast_fp16))[name = string("_SplitHeadsQ__mh_w_203_cast_fp16")]; + string _SplitHeadsQ__mh_w_205_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_205_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_205_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_205_equation_0, values = (var_2490_cast_fp16, var_2306_cast_fp16))[name = string("_SplitHeadsQ__mh_w_205_cast_fp16")]; + string _SplitHeadsQ__mh_w_207_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_207_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_207_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_207_equation_0, values = (var_2490_cast_fp16, var_2313_cast_fp16))[name = string("_SplitHeadsQ__mh_w_207_cast_fp16")]; + string _SplitHeadsQ__mh_w_209_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_209_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_209_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_209_equation_0, values = (var_2494_cast_fp16, var_2320_cast_fp16))[name = string("_SplitHeadsQ__mh_w_209_cast_fp16")]; + string _SplitHeadsQ__mh_w_211_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_211_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_211_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_211_equation_0, values = (var_2494_cast_fp16, var_2327_cast_fp16))[name = string("_SplitHeadsQ__mh_w_211_cast_fp16")]; + string _SplitHeadsQ__mh_w_213_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_213_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_213_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_213_equation_0, values = (var_2494_cast_fp16, var_2334_cast_fp16))[name = string("_SplitHeadsQ__mh_w_213_cast_fp16")]; + string _SplitHeadsQ__mh_w_215_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_215_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_215_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_215_equation_0, values = (var_2494_cast_fp16, var_2341_cast_fp16))[name = string("_SplitHeadsQ__mh_w_215_cast_fp16")]; + string _SplitHeadsQ__mh_w_217_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_217_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_217_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_217_equation_0, values = (var_2498_cast_fp16, var_2348_cast_fp16))[name = string("_SplitHeadsQ__mh_w_217_cast_fp16")]; + string _SplitHeadsQ__mh_w_219_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_219_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_219_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_219_equation_0, values = (var_2498_cast_fp16, var_2355_cast_fp16))[name = string("_SplitHeadsQ__mh_w_219_cast_fp16")]; + string _SplitHeadsQ__mh_w_221_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_221_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_221_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_221_equation_0, values = (var_2498_cast_fp16, var_2362_cast_fp16))[name = string("_SplitHeadsQ__mh_w_221_cast_fp16")]; + string _SplitHeadsQ__mh_w_223_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_223_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_223_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_223_equation_0, values = (var_2498_cast_fp16, var_2369_cast_fp16))[name = string("_SplitHeadsQ__mh_w_223_cast_fp16")]; + string _SplitHeadsQ__mh_w_225_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_225_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_225_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_225_equation_0, values = (var_2502_cast_fp16, var_2376_cast_fp16))[name = string("_SplitHeadsQ__mh_w_225_cast_fp16")]; + string _SplitHeadsQ__mh_w_227_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_227_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_227_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_227_equation_0, values = (var_2502_cast_fp16, var_2383_cast_fp16))[name = string("_SplitHeadsQ__mh_w_227_cast_fp16")]; + string _SplitHeadsQ__mh_w_229_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_229_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_229_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_229_equation_0, values = (var_2502_cast_fp16, var_2390_cast_fp16))[name = string("_SplitHeadsQ__mh_w_229_cast_fp16")]; + string _SplitHeadsQ__mh_w_231_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_231_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_231_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_231_equation_0, values = (var_2502_cast_fp16, var_2397_cast_fp16))[name = string("_SplitHeadsQ__mh_w_231_cast_fp16")]; + string _SplitHeadsQ__mh_w_233_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_233_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_233_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_233_equation_0, values = (var_2506_cast_fp16, var_2404_cast_fp16))[name = string("_SplitHeadsQ__mh_w_233_cast_fp16")]; + string _SplitHeadsQ__mh_w_235_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_235_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_235_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_235_equation_0, values = (var_2506_cast_fp16, var_2411_cast_fp16))[name = string("_SplitHeadsQ__mh_w_235_cast_fp16")]; + string _SplitHeadsQ__mh_w_237_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_237_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_237_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_237_equation_0, values = (var_2506_cast_fp16, var_2418_cast_fp16))[name = string("_SplitHeadsQ__mh_w_237_cast_fp16")]; + string _SplitHeadsQ__mh_w_239_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_239_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_239_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_239_equation_0, values = (var_2506_cast_fp16, var_2425_cast_fp16))[name = string("_SplitHeadsQ__mh_w_239_cast_fp16")]; + string _SplitHeadsQ__mh_w_241_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_241_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_241_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_241_equation_0, values = (var_2510_cast_fp16, var_2432_cast_fp16))[name = string("_SplitHeadsQ__mh_w_241_cast_fp16")]; + string _SplitHeadsQ__mh_w_243_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_243_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_243_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_243_equation_0, values = (var_2510_cast_fp16, var_2439_cast_fp16))[name = string("_SplitHeadsQ__mh_w_243_cast_fp16")]; + string _SplitHeadsQ__mh_w_245_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_245_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_245_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_245_equation_0, values = (var_2510_cast_fp16, var_2446_cast_fp16))[name = string("_SplitHeadsQ__mh_w_245_cast_fp16")]; + string _SplitHeadsQ__mh_w_247_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_247_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_247_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_247_equation_0, values = (var_2510_cast_fp16, var_2453_cast_fp16))[name = string("_SplitHeadsQ__mh_w_247_cast_fp16")]; + string _SplitHeadsQ__mh_w_249_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_249_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_249_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_249_equation_0, values = (var_2514_cast_fp16, var_2460_cast_fp16))[name = string("_SplitHeadsQ__mh_w_249_cast_fp16")]; + string _SplitHeadsQ__mh_w_251_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_251_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_251_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_251_equation_0, values = (var_2514_cast_fp16, var_2467_cast_fp16))[name = string("_SplitHeadsQ__mh_w_251_cast_fp16")]; + string _SplitHeadsQ__mh_w_253_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_253_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_253_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_253_equation_0, values = (var_2514_cast_fp16, var_2474_cast_fp16))[name = string("_SplitHeadsQ__mh_w_253_cast_fp16")]; + string _SplitHeadsQ__mh_w_255_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_255_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_255_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_255_equation_0, values = (var_2514_cast_fp16, var_2481_cast_fp16))[name = string("_SplitHeadsQ__mh_w_255_cast_fp16")]; + fp16 var_2611_to_fp16 = const()[name = string("op_2611_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_193_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_193_cast_fp16, y = var_2611_to_fp16)[name = string("aw_chunk_193_cast_fp16")]; + fp16 var_2613_to_fp16 = const()[name = string("op_2613_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_195_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_195_cast_fp16, y = var_2613_to_fp16)[name = string("aw_chunk_195_cast_fp16")]; + fp16 var_2615_to_fp16 = const()[name = string("op_2615_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_197_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_197_cast_fp16, y = var_2615_to_fp16)[name = string("aw_chunk_197_cast_fp16")]; + fp16 var_2617_to_fp16 = const()[name = string("op_2617_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_199_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_199_cast_fp16, y = var_2617_to_fp16)[name = string("aw_chunk_199_cast_fp16")]; + fp16 var_2619_to_fp16 = const()[name = string("op_2619_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_201_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_201_cast_fp16, y = var_2619_to_fp16)[name = string("aw_chunk_201_cast_fp16")]; + fp16 var_2621_to_fp16 = const()[name = string("op_2621_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_203_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_203_cast_fp16, y = var_2621_to_fp16)[name = string("aw_chunk_203_cast_fp16")]; + fp16 var_2623_to_fp16 = const()[name = string("op_2623_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_205_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_205_cast_fp16, y = var_2623_to_fp16)[name = string("aw_chunk_205_cast_fp16")]; + fp16 var_2625_to_fp16 = const()[name = string("op_2625_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_207_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_207_cast_fp16, y = var_2625_to_fp16)[name = string("aw_chunk_207_cast_fp16")]; + fp16 var_2627_to_fp16 = const()[name = string("op_2627_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_209_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_209_cast_fp16, y = var_2627_to_fp16)[name = string("aw_chunk_209_cast_fp16")]; + fp16 var_2629_to_fp16 = const()[name = string("op_2629_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_211_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_211_cast_fp16, y = var_2629_to_fp16)[name = string("aw_chunk_211_cast_fp16")]; + fp16 var_2631_to_fp16 = const()[name = string("op_2631_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_213_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_213_cast_fp16, y = var_2631_to_fp16)[name = string("aw_chunk_213_cast_fp16")]; + fp16 var_2633_to_fp16 = const()[name = string("op_2633_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_215_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_215_cast_fp16, y = var_2633_to_fp16)[name = string("aw_chunk_215_cast_fp16")]; + fp16 var_2635_to_fp16 = const()[name = string("op_2635_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_217_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_217_cast_fp16, y = var_2635_to_fp16)[name = string("aw_chunk_217_cast_fp16")]; + fp16 var_2637_to_fp16 = const()[name = string("op_2637_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_219_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_219_cast_fp16, y = var_2637_to_fp16)[name = string("aw_chunk_219_cast_fp16")]; + fp16 var_2639_to_fp16 = const()[name = string("op_2639_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_221_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_221_cast_fp16, y = var_2639_to_fp16)[name = string("aw_chunk_221_cast_fp16")]; + fp16 var_2641_to_fp16 = const()[name = string("op_2641_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_223_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_223_cast_fp16, y = var_2641_to_fp16)[name = string("aw_chunk_223_cast_fp16")]; + fp16 var_2643_to_fp16 = const()[name = string("op_2643_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_225_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_225_cast_fp16, y = var_2643_to_fp16)[name = string("aw_chunk_225_cast_fp16")]; + fp16 var_2645_to_fp16 = const()[name = string("op_2645_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_227_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_227_cast_fp16, y = var_2645_to_fp16)[name = string("aw_chunk_227_cast_fp16")]; + fp16 var_2647_to_fp16 = const()[name = string("op_2647_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_229_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_229_cast_fp16, y = var_2647_to_fp16)[name = string("aw_chunk_229_cast_fp16")]; + fp16 var_2649_to_fp16 = const()[name = string("op_2649_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_231_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_231_cast_fp16, y = var_2649_to_fp16)[name = string("aw_chunk_231_cast_fp16")]; + fp16 var_2651_to_fp16 = const()[name = string("op_2651_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_233_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_233_cast_fp16, y = var_2651_to_fp16)[name = string("aw_chunk_233_cast_fp16")]; + fp16 var_2653_to_fp16 = const()[name = string("op_2653_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_235_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_235_cast_fp16, y = var_2653_to_fp16)[name = string("aw_chunk_235_cast_fp16")]; + fp16 var_2655_to_fp16 = const()[name = string("op_2655_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_237_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_237_cast_fp16, y = var_2655_to_fp16)[name = string("aw_chunk_237_cast_fp16")]; + fp16 var_2657_to_fp16 = const()[name = string("op_2657_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_239_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_239_cast_fp16, y = var_2657_to_fp16)[name = string("aw_chunk_239_cast_fp16")]; + fp16 var_2659_to_fp16 = const()[name = string("op_2659_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_241_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_241_cast_fp16, y = var_2659_to_fp16)[name = string("aw_chunk_241_cast_fp16")]; + fp16 var_2661_to_fp16 = const()[name = string("op_2661_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_243_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_243_cast_fp16, y = var_2661_to_fp16)[name = string("aw_chunk_243_cast_fp16")]; + fp16 var_2663_to_fp16 = const()[name = string("op_2663_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_245_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_245_cast_fp16, y = var_2663_to_fp16)[name = string("aw_chunk_245_cast_fp16")]; + fp16 var_2665_to_fp16 = const()[name = string("op_2665_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_247_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_247_cast_fp16, y = var_2665_to_fp16)[name = string("aw_chunk_247_cast_fp16")]; + fp16 var_2667_to_fp16 = const()[name = string("op_2667_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_249_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_249_cast_fp16, y = var_2667_to_fp16)[name = string("aw_chunk_249_cast_fp16")]; + fp16 var_2669_to_fp16 = const()[name = string("op_2669_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_251_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_251_cast_fp16, y = var_2669_to_fp16)[name = string("aw_chunk_251_cast_fp16")]; + fp16 var_2671_to_fp16 = const()[name = string("op_2671_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_253_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_253_cast_fp16, y = var_2671_to_fp16)[name = string("aw_chunk_253_cast_fp16")]; + fp16 var_2673_to_fp16 = const()[name = string("op_2673_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_255_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_255_cast_fp16, y = var_2673_to_fp16)[name = string("aw_chunk_255_cast_fp16")]; + tensor var_2675_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_193_cast_fp16)[name = string("op_2675_cast_fp16")]; + tensor var_2676_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_195_cast_fp16)[name = string("op_2676_cast_fp16")]; + tensor var_2677_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_197_cast_fp16)[name = string("op_2677_cast_fp16")]; + tensor var_2678_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_199_cast_fp16)[name = string("op_2678_cast_fp16")]; + tensor var_2679_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_201_cast_fp16)[name = string("op_2679_cast_fp16")]; + tensor var_2680_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_203_cast_fp16)[name = string("op_2680_cast_fp16")]; + tensor var_2681_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_205_cast_fp16)[name = string("op_2681_cast_fp16")]; + tensor var_2682_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_207_cast_fp16)[name = string("op_2682_cast_fp16")]; + tensor var_2683_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_209_cast_fp16)[name = string("op_2683_cast_fp16")]; + tensor var_2684_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_211_cast_fp16)[name = string("op_2684_cast_fp16")]; + tensor var_2685_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_213_cast_fp16)[name = string("op_2685_cast_fp16")]; + tensor var_2686_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_215_cast_fp16)[name = string("op_2686_cast_fp16")]; + tensor var_2687_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_217_cast_fp16)[name = string("op_2687_cast_fp16")]; + tensor var_2688_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_219_cast_fp16)[name = string("op_2688_cast_fp16")]; + tensor var_2689_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_221_cast_fp16)[name = string("op_2689_cast_fp16")]; + tensor var_2690_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_223_cast_fp16)[name = string("op_2690_cast_fp16")]; + tensor var_2691_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_225_cast_fp16)[name = string("op_2691_cast_fp16")]; + tensor var_2692_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_227_cast_fp16)[name = string("op_2692_cast_fp16")]; + tensor var_2693_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_229_cast_fp16)[name = string("op_2693_cast_fp16")]; + tensor var_2694_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_231_cast_fp16)[name = string("op_2694_cast_fp16")]; + tensor var_2695_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_233_cast_fp16)[name = string("op_2695_cast_fp16")]; + tensor var_2696_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_235_cast_fp16)[name = string("op_2696_cast_fp16")]; + tensor var_2697_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_237_cast_fp16)[name = string("op_2697_cast_fp16")]; + tensor var_2698_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_239_cast_fp16)[name = string("op_2698_cast_fp16")]; + tensor var_2699_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_241_cast_fp16)[name = string("op_2699_cast_fp16")]; + tensor var_2700_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_243_cast_fp16)[name = string("op_2700_cast_fp16")]; + tensor var_2701_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_245_cast_fp16)[name = string("op_2701_cast_fp16")]; + tensor var_2702_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_247_cast_fp16)[name = string("op_2702_cast_fp16")]; + tensor var_2703_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_249_cast_fp16)[name = string("op_2703_cast_fp16")]; + tensor var_2704_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_251_cast_fp16)[name = string("op_2704_cast_fp16")]; + tensor var_2705_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_253_cast_fp16)[name = string("op_2705_cast_fp16")]; + tensor var_2706_cast_fp16 = softmax(axis = var_2172, x = aw_chunk_255_cast_fp16)[name = string("op_2706_cast_fp16")]; + string var_2708_equation_0 = const()[name = string("op_2708_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2708_cast_fp16 = einsum(equation = var_2708_equation_0, values = (var_2516_cast_fp16, var_2675_cast_fp16))[name = string("op_2708_cast_fp16")]; + string var_2710_equation_0 = const()[name = string("op_2710_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2710_cast_fp16 = einsum(equation = var_2710_equation_0, values = (var_2516_cast_fp16, var_2676_cast_fp16))[name = string("op_2710_cast_fp16")]; + string var_2712_equation_0 = const()[name = string("op_2712_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2712_cast_fp16 = einsum(equation = var_2712_equation_0, values = (var_2516_cast_fp16, var_2677_cast_fp16))[name = string("op_2712_cast_fp16")]; + string var_2714_equation_0 = const()[name = string("op_2714_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2714_cast_fp16 = einsum(equation = var_2714_equation_0, values = (var_2516_cast_fp16, var_2678_cast_fp16))[name = string("op_2714_cast_fp16")]; + string var_2716_equation_0 = const()[name = string("op_2716_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2716_cast_fp16 = einsum(equation = var_2716_equation_0, values = (var_2520_cast_fp16, var_2679_cast_fp16))[name = string("op_2716_cast_fp16")]; + string var_2718_equation_0 = const()[name = string("op_2718_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2718_cast_fp16 = einsum(equation = var_2718_equation_0, values = (var_2520_cast_fp16, var_2680_cast_fp16))[name = string("op_2718_cast_fp16")]; + string var_2720_equation_0 = const()[name = string("op_2720_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2720_cast_fp16 = einsum(equation = var_2720_equation_0, values = (var_2520_cast_fp16, var_2681_cast_fp16))[name = string("op_2720_cast_fp16")]; + string var_2722_equation_0 = const()[name = string("op_2722_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2722_cast_fp16 = einsum(equation = var_2722_equation_0, values = (var_2520_cast_fp16, var_2682_cast_fp16))[name = string("op_2722_cast_fp16")]; + string var_2724_equation_0 = const()[name = string("op_2724_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2724_cast_fp16 = einsum(equation = var_2724_equation_0, values = (var_2524_cast_fp16, var_2683_cast_fp16))[name = string("op_2724_cast_fp16")]; + string var_2726_equation_0 = const()[name = string("op_2726_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2726_cast_fp16 = einsum(equation = var_2726_equation_0, values = (var_2524_cast_fp16, var_2684_cast_fp16))[name = string("op_2726_cast_fp16")]; + string var_2728_equation_0 = const()[name = string("op_2728_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2728_cast_fp16 = einsum(equation = var_2728_equation_0, values = (var_2524_cast_fp16, var_2685_cast_fp16))[name = string("op_2728_cast_fp16")]; + string var_2730_equation_0 = const()[name = string("op_2730_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2730_cast_fp16 = einsum(equation = var_2730_equation_0, values = (var_2524_cast_fp16, var_2686_cast_fp16))[name = string("op_2730_cast_fp16")]; + string var_2732_equation_0 = const()[name = string("op_2732_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2732_cast_fp16 = einsum(equation = var_2732_equation_0, values = (var_2528_cast_fp16, var_2687_cast_fp16))[name = string("op_2732_cast_fp16")]; + string var_2734_equation_0 = const()[name = string("op_2734_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2734_cast_fp16 = einsum(equation = var_2734_equation_0, values = (var_2528_cast_fp16, var_2688_cast_fp16))[name = string("op_2734_cast_fp16")]; + string var_2736_equation_0 = const()[name = string("op_2736_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2736_cast_fp16 = einsum(equation = var_2736_equation_0, values = (var_2528_cast_fp16, var_2689_cast_fp16))[name = string("op_2736_cast_fp16")]; + string var_2738_equation_0 = const()[name = string("op_2738_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2738_cast_fp16 = einsum(equation = var_2738_equation_0, values = (var_2528_cast_fp16, var_2690_cast_fp16))[name = string("op_2738_cast_fp16")]; + string var_2740_equation_0 = const()[name = string("op_2740_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2740_cast_fp16 = einsum(equation = var_2740_equation_0, values = (var_2532_cast_fp16, var_2691_cast_fp16))[name = string("op_2740_cast_fp16")]; + string var_2742_equation_0 = const()[name = string("op_2742_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2742_cast_fp16 = einsum(equation = var_2742_equation_0, values = (var_2532_cast_fp16, var_2692_cast_fp16))[name = string("op_2742_cast_fp16")]; + string var_2744_equation_0 = const()[name = string("op_2744_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2744_cast_fp16 = einsum(equation = var_2744_equation_0, values = (var_2532_cast_fp16, var_2693_cast_fp16))[name = string("op_2744_cast_fp16")]; + string var_2746_equation_0 = const()[name = string("op_2746_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2746_cast_fp16 = einsum(equation = var_2746_equation_0, values = (var_2532_cast_fp16, var_2694_cast_fp16))[name = string("op_2746_cast_fp16")]; + string var_2748_equation_0 = const()[name = string("op_2748_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2748_cast_fp16 = einsum(equation = var_2748_equation_0, values = (var_2536_cast_fp16, var_2695_cast_fp16))[name = string("op_2748_cast_fp16")]; + string var_2750_equation_0 = const()[name = string("op_2750_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2750_cast_fp16 = einsum(equation = var_2750_equation_0, values = (var_2536_cast_fp16, var_2696_cast_fp16))[name = string("op_2750_cast_fp16")]; + string var_2752_equation_0 = const()[name = string("op_2752_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2752_cast_fp16 = einsum(equation = var_2752_equation_0, values = (var_2536_cast_fp16, var_2697_cast_fp16))[name = string("op_2752_cast_fp16")]; + string var_2754_equation_0 = const()[name = string("op_2754_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2754_cast_fp16 = einsum(equation = var_2754_equation_0, values = (var_2536_cast_fp16, var_2698_cast_fp16))[name = string("op_2754_cast_fp16")]; + string var_2756_equation_0 = const()[name = string("op_2756_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2756_cast_fp16 = einsum(equation = var_2756_equation_0, values = (var_2540_cast_fp16, var_2699_cast_fp16))[name = string("op_2756_cast_fp16")]; + string var_2758_equation_0 = const()[name = string("op_2758_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2758_cast_fp16 = einsum(equation = var_2758_equation_0, values = (var_2540_cast_fp16, var_2700_cast_fp16))[name = string("op_2758_cast_fp16")]; + string var_2760_equation_0 = const()[name = string("op_2760_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2760_cast_fp16 = einsum(equation = var_2760_equation_0, values = (var_2540_cast_fp16, var_2701_cast_fp16))[name = string("op_2760_cast_fp16")]; + string var_2762_equation_0 = const()[name = string("op_2762_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2762_cast_fp16 = einsum(equation = var_2762_equation_0, values = (var_2540_cast_fp16, var_2702_cast_fp16))[name = string("op_2762_cast_fp16")]; + string var_2764_equation_0 = const()[name = string("op_2764_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2764_cast_fp16 = einsum(equation = var_2764_equation_0, values = (var_2544_cast_fp16, var_2703_cast_fp16))[name = string("op_2764_cast_fp16")]; + string var_2766_equation_0 = const()[name = string("op_2766_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2766_cast_fp16 = einsum(equation = var_2766_equation_0, values = (var_2544_cast_fp16, var_2704_cast_fp16))[name = string("op_2766_cast_fp16")]; + string var_2768_equation_0 = const()[name = string("op_2768_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2768_cast_fp16 = einsum(equation = var_2768_equation_0, values = (var_2544_cast_fp16, var_2705_cast_fp16))[name = string("op_2768_cast_fp16")]; + string var_2770_equation_0 = const()[name = string("op_2770_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_2770_cast_fp16 = einsum(equation = var_2770_equation_0, values = (var_2544_cast_fp16, var_2706_cast_fp16))[name = string("op_2770_cast_fp16")]; + bool var_2772_interleave_0 = const()[name = string("op_2772_interleave_0"), val = bool(false)]; + tensor var_2772_cast_fp16 = concat(axis = var_2159, interleave = var_2772_interleave_0, values = (var_2708_cast_fp16, var_2710_cast_fp16, var_2712_cast_fp16, var_2714_cast_fp16))[name = string("op_2772_cast_fp16")]; + bool var_2774_interleave_0 = const()[name = string("op_2774_interleave_0"), val = bool(false)]; + tensor var_2774_cast_fp16 = concat(axis = var_2159, interleave = var_2774_interleave_0, values = (var_2716_cast_fp16, var_2718_cast_fp16, var_2720_cast_fp16, var_2722_cast_fp16))[name = string("op_2774_cast_fp16")]; + bool var_2776_interleave_0 = const()[name = string("op_2776_interleave_0"), val = bool(false)]; + tensor var_2776_cast_fp16 = concat(axis = var_2159, interleave = var_2776_interleave_0, values = (var_2724_cast_fp16, var_2726_cast_fp16, var_2728_cast_fp16, var_2730_cast_fp16))[name = string("op_2776_cast_fp16")]; + bool var_2778_interleave_0 = const()[name = string("op_2778_interleave_0"), val = bool(false)]; + tensor var_2778_cast_fp16 = concat(axis = var_2159, interleave = var_2778_interleave_0, values = (var_2732_cast_fp16, var_2734_cast_fp16, var_2736_cast_fp16, var_2738_cast_fp16))[name = string("op_2778_cast_fp16")]; + bool var_2780_interleave_0 = const()[name = string("op_2780_interleave_0"), val = bool(false)]; + tensor var_2780_cast_fp16 = concat(axis = var_2159, interleave = var_2780_interleave_0, values = (var_2740_cast_fp16, var_2742_cast_fp16, var_2744_cast_fp16, var_2746_cast_fp16))[name = string("op_2780_cast_fp16")]; + bool var_2782_interleave_0 = const()[name = string("op_2782_interleave_0"), val = bool(false)]; + tensor var_2782_cast_fp16 = concat(axis = var_2159, interleave = var_2782_interleave_0, values = (var_2748_cast_fp16, var_2750_cast_fp16, var_2752_cast_fp16, var_2754_cast_fp16))[name = string("op_2782_cast_fp16")]; + bool var_2784_interleave_0 = const()[name = string("op_2784_interleave_0"), val = bool(false)]; + tensor var_2784_cast_fp16 = concat(axis = var_2159, interleave = var_2784_interleave_0, values = (var_2756_cast_fp16, var_2758_cast_fp16, var_2760_cast_fp16, var_2762_cast_fp16))[name = string("op_2784_cast_fp16")]; + bool var_2786_interleave_0 = const()[name = string("op_2786_interleave_0"), val = bool(false)]; + tensor var_2786_cast_fp16 = concat(axis = var_2159, interleave = var_2786_interleave_0, values = (var_2764_cast_fp16, var_2766_cast_fp16, var_2768_cast_fp16, var_2770_cast_fp16))[name = string("op_2786_cast_fp16")]; + bool input_25_interleave_0 = const()[name = string("input_25_interleave_0"), val = bool(false)]; + tensor input_25_cast_fp16 = concat(axis = var_2172, interleave = input_25_interleave_0, values = (var_2772_cast_fp16, var_2774_cast_fp16, var_2776_cast_fp16, var_2778_cast_fp16, var_2780_cast_fp16, var_2782_cast_fp16, var_2784_cast_fp16, var_2786_cast_fp16))[name = string("input_25_cast_fp16")]; + string obj_15_pad_type_0 = const()[name = string("obj_15_pad_type_0"), val = string("valid")]; + tensor obj_15_strides_0 = const()[name = string("obj_15_strides_0"), val = tensor([1, 1])]; + tensor obj_15_pad_0 = const()[name = string("obj_15_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_15_dilations_0 = const()[name = string("obj_15_dilations_0"), val = tensor([1, 1])]; + int32 obj_15_groups_0 = const()[name = string("obj_15_groups_0"), val = int32(1)]; + tensor layers_3_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23850752)))]; + tensor layers_3_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_3_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24375104)))]; + tensor obj_15_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_bias_to_fp16, dilations = obj_15_dilations_0, groups = obj_15_groups_0, pad = obj_15_pad_0, pad_type = obj_15_pad_type_0, strides = obj_15_strides_0, weight = layers_3_self_attn_o_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("obj_15_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_15_cast_fp16)[name = string("inputs_15_cast_fp16")]; + tensor out_15_axes_0 = const()[name = string("out_15_axes_0"), val = tensor([1])]; + fp16 var_2805_to_fp16 = const()[name = string("op_2805_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_2805_to_fp16, x = inputs_15_cast_fp16)[name = string("out_15_cast_fp16")]; + tensor input_27_gamma_0_to_fp16 = const()[name = string("input_27_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24376192)))]; + tensor input_27_beta_0_to_fp16 = const()[name = string("input_27_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24377280)))]; + fp16 input_27_epsilon_0_to_fp16 = const()[name = string("input_27_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_27_cast_fp16 = batch_norm(beta = input_27_beta_0_to_fp16, epsilon = input_27_epsilon_0_to_fp16, gamma = input_27_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_15_cast_fp16)[name = string("input_27_cast_fp16")]; + string input_29_pad_type_0 = const()[name = string("input_29_pad_type_0"), val = string("valid")]; + tensor input_29_strides_0 = const()[name = string("input_29_strides_0"), val = tensor([1, 1])]; + tensor input_29_pad_0 = const()[name = string("input_29_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_29_dilations_0 = const()[name = string("input_29_dilations_0"), val = tensor([1, 1])]; + int32 input_29_groups_0 = const()[name = string("input_29_groups_0"), val = int32(1)]; + tensor layers_3_fc1_weight_to_fp16 = const()[name = string("layers_3_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(24378368)))]; + tensor layers_3_fc1_bias_to_fp16 = const()[name = string("layers_3_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26475584)))]; + tensor input_29_cast_fp16 = conv(bias = layers_3_fc1_bias_to_fp16, dilations = input_29_dilations_0, groups = input_29_groups_0, pad = input_29_pad_0, pad_type = input_29_pad_type_0, strides = input_29_strides_0, weight = layers_3_fc1_weight_to_fp16, x = input_27_cast_fp16)[name = string("input_29_cast_fp16")]; + string input_31_mode_0 = const()[name = string("input_31_mode_0"), val = string("EXACT")]; + tensor input_31_cast_fp16 = gelu(mode = input_31_mode_0, x = input_29_cast_fp16)[name = string("input_31_cast_fp16")]; + string hidden_states_11_pad_type_0 = const()[name = string("hidden_states_11_pad_type_0"), val = string("valid")]; + tensor hidden_states_11_strides_0 = const()[name = string("hidden_states_11_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_11_pad_0 = const()[name = string("hidden_states_11_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_11_dilations_0 = const()[name = string("hidden_states_11_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_11_groups_0 = const()[name = string("hidden_states_11_groups_0"), val = int32(1)]; + tensor layers_3_fc2_weight_to_fp16 = const()[name = string("layers_3_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26479744)))]; + tensor layers_3_fc2_bias_to_fp16 = const()[name = string("layers_3_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28576960)))]; + tensor hidden_states_11_cast_fp16 = conv(bias = layers_3_fc2_bias_to_fp16, dilations = hidden_states_11_dilations_0, groups = hidden_states_11_groups_0, pad = hidden_states_11_pad_0, pad_type = hidden_states_11_pad_type_0, strides = hidden_states_11_strides_0, weight = layers_3_fc2_weight_to_fp16, x = input_31_cast_fp16)[name = string("hidden_states_11_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = hidden_states_11_cast_fp16)[name = string("inputs_17_cast_fp16")]; + int32 var_2834 = const()[name = string("op_2834"), val = int32(3)]; + int32 var_2847 = const()[name = string("op_2847"), val = int32(1)]; + tensor out_17_axes_0 = const()[name = string("out_17_axes_0"), val = tensor([1])]; + fp16 var_2864_to_fp16 = const()[name = string("op_2864_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_2864_to_fp16, x = inputs_17_cast_fp16)[name = string("out_17_cast_fp16")]; + tensor obj_17_gamma_0_to_fp16 = const()[name = string("obj_17_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28578048)))]; + tensor obj_17_beta_0_to_fp16 = const()[name = string("obj_17_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28579136)))]; + fp16 obj_17_epsilon_0_to_fp16 = const()[name = string("obj_17_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_17_cast_fp16 = batch_norm(beta = obj_17_beta_0_to_fp16, epsilon = obj_17_epsilon_0_to_fp16, gamma = obj_17_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_17_cast_fp16)[name = string("obj_17_cast_fp16")]; + string query_9_pad_type_0 = const()[name = string("query_9_pad_type_0"), val = string("valid")]; + tensor query_9_strides_0 = const()[name = string("query_9_strides_0"), val = tensor([1, 1])]; + tensor query_9_pad_0 = const()[name = string("query_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_9_dilations_0 = const()[name = string("query_9_dilations_0"), val = tensor([1, 1])]; + int32 query_9_groups_0 = const()[name = string("query_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28580224)))]; + tensor layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29104576)))]; + tensor query_9_cast_fp16 = conv(bias = layers_4_self_attn_q_proj_bias_to_fp16, dilations = query_9_dilations_0, groups = query_9_groups_0, pad = query_9_pad_0, pad_type = query_9_pad_type_0, strides = query_9_strides_0, weight = layers_4_self_attn_q_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("query_9_cast_fp16")]; + string key_9_pad_type_0 = const()[name = string("key_9_pad_type_0"), val = string("valid")]; + tensor key_9_strides_0 = const()[name = string("key_9_strides_0"), val = tensor([1, 1])]; + tensor key_9_pad_0 = const()[name = string("key_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_9_dilations_0 = const()[name = string("key_9_dilations_0"), val = tensor([1, 1])]; + int32 key_9_groups_0 = const()[name = string("key_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29105664)))]; + tensor key_9_cast_fp16 = conv(dilations = key_9_dilations_0, groups = key_9_groups_0, pad = key_9_pad_0, pad_type = key_9_pad_type_0, strides = key_9_strides_0, weight = layers_4_self_attn_k_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("key_9_cast_fp16")]; + string value_9_pad_type_0 = const()[name = string("value_9_pad_type_0"), val = string("valid")]; + tensor value_9_strides_0 = const()[name = string("value_9_strides_0"), val = tensor([1, 1])]; + tensor value_9_pad_0 = const()[name = string("value_9_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_9_dilations_0 = const()[name = string("value_9_dilations_0"), val = tensor([1, 1])]; + int32 value_9_groups_0 = const()[name = string("value_9_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29630016)))]; + tensor layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30154368)))]; + tensor value_9_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_bias_to_fp16, dilations = value_9_dilations_0, groups = value_9_groups_0, pad = value_9_pad_0, pad_type = value_9_pad_type_0, strides = value_9_strides_0, weight = layers_4_self_attn_v_proj_weight_to_fp16, x = obj_17_cast_fp16)[name = string("value_9_cast_fp16")]; + tensor var_2902_begin_0 = const()[name = string("op_2902_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2902_end_0 = const()[name = string("op_2902_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2902_end_mask_0 = const()[name = string("op_2902_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2902_cast_fp16 = slice_by_index(begin = var_2902_begin_0, end = var_2902_end_0, end_mask = var_2902_end_mask_0, x = query_9_cast_fp16)[name = string("op_2902_cast_fp16")]; + tensor var_2906_begin_0 = const()[name = string("op_2906_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_2906_end_0 = const()[name = string("op_2906_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_2906_end_mask_0 = const()[name = string("op_2906_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2906_cast_fp16 = slice_by_index(begin = var_2906_begin_0, end = var_2906_end_0, end_mask = var_2906_end_mask_0, x = query_9_cast_fp16)[name = string("op_2906_cast_fp16")]; + tensor var_2910_begin_0 = const()[name = string("op_2910_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_2910_end_0 = const()[name = string("op_2910_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_2910_end_mask_0 = const()[name = string("op_2910_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2910_cast_fp16 = slice_by_index(begin = var_2910_begin_0, end = var_2910_end_0, end_mask = var_2910_end_mask_0, x = query_9_cast_fp16)[name = string("op_2910_cast_fp16")]; + tensor var_2914_begin_0 = const()[name = string("op_2914_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_2914_end_0 = const()[name = string("op_2914_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_2914_end_mask_0 = const()[name = string("op_2914_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2914_cast_fp16 = slice_by_index(begin = var_2914_begin_0, end = var_2914_end_0, end_mask = var_2914_end_mask_0, x = query_9_cast_fp16)[name = string("op_2914_cast_fp16")]; + tensor var_2918_begin_0 = const()[name = string("op_2918_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_2918_end_0 = const()[name = string("op_2918_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_2918_end_mask_0 = const()[name = string("op_2918_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2918_cast_fp16 = slice_by_index(begin = var_2918_begin_0, end = var_2918_end_0, end_mask = var_2918_end_mask_0, x = query_9_cast_fp16)[name = string("op_2918_cast_fp16")]; + tensor var_2922_begin_0 = const()[name = string("op_2922_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_2922_end_0 = const()[name = string("op_2922_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_2922_end_mask_0 = const()[name = string("op_2922_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2922_cast_fp16 = slice_by_index(begin = var_2922_begin_0, end = var_2922_end_0, end_mask = var_2922_end_mask_0, x = query_9_cast_fp16)[name = string("op_2922_cast_fp16")]; + tensor var_2926_begin_0 = const()[name = string("op_2926_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_2926_end_0 = const()[name = string("op_2926_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_2926_end_mask_0 = const()[name = string("op_2926_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2926_cast_fp16 = slice_by_index(begin = var_2926_begin_0, end = var_2926_end_0, end_mask = var_2926_end_mask_0, x = query_9_cast_fp16)[name = string("op_2926_cast_fp16")]; + tensor var_2930_begin_0 = const()[name = string("op_2930_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_2930_end_0 = const()[name = string("op_2930_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_2930_end_mask_0 = const()[name = string("op_2930_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_2930_cast_fp16 = slice_by_index(begin = var_2930_begin_0, end = var_2930_end_0, end_mask = var_2930_end_mask_0, x = query_9_cast_fp16)[name = string("op_2930_cast_fp16")]; + tensor var_2939_begin_0 = const()[name = string("op_2939_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2939_end_0 = const()[name = string("op_2939_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2939_end_mask_0 = const()[name = string("op_2939_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2939_cast_fp16 = slice_by_index(begin = var_2939_begin_0, end = var_2939_end_0, end_mask = var_2939_end_mask_0, x = var_2902_cast_fp16)[name = string("op_2939_cast_fp16")]; + tensor var_2946_begin_0 = const()[name = string("op_2946_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2946_end_0 = const()[name = string("op_2946_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2946_end_mask_0 = const()[name = string("op_2946_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2946_cast_fp16 = slice_by_index(begin = var_2946_begin_0, end = var_2946_end_0, end_mask = var_2946_end_mask_0, x = var_2902_cast_fp16)[name = string("op_2946_cast_fp16")]; + tensor var_2953_begin_0 = const()[name = string("op_2953_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2953_end_0 = const()[name = string("op_2953_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2953_end_mask_0 = const()[name = string("op_2953_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2953_cast_fp16 = slice_by_index(begin = var_2953_begin_0, end = var_2953_end_0, end_mask = var_2953_end_mask_0, x = var_2902_cast_fp16)[name = string("op_2953_cast_fp16")]; + tensor var_2960_begin_0 = const()[name = string("op_2960_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2960_end_0 = const()[name = string("op_2960_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2960_end_mask_0 = const()[name = string("op_2960_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2960_cast_fp16 = slice_by_index(begin = var_2960_begin_0, end = var_2960_end_0, end_mask = var_2960_end_mask_0, x = var_2902_cast_fp16)[name = string("op_2960_cast_fp16")]; + tensor var_2967_begin_0 = const()[name = string("op_2967_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2967_end_0 = const()[name = string("op_2967_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2967_end_mask_0 = const()[name = string("op_2967_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2967_cast_fp16 = slice_by_index(begin = var_2967_begin_0, end = var_2967_end_0, end_mask = var_2967_end_mask_0, x = var_2906_cast_fp16)[name = string("op_2967_cast_fp16")]; + tensor var_2974_begin_0 = const()[name = string("op_2974_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_2974_end_0 = const()[name = string("op_2974_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_2974_end_mask_0 = const()[name = string("op_2974_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2974_cast_fp16 = slice_by_index(begin = var_2974_begin_0, end = var_2974_end_0, end_mask = var_2974_end_mask_0, x = var_2906_cast_fp16)[name = string("op_2974_cast_fp16")]; + tensor var_2981_begin_0 = const()[name = string("op_2981_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_2981_end_0 = const()[name = string("op_2981_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_2981_end_mask_0 = const()[name = string("op_2981_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2981_cast_fp16 = slice_by_index(begin = var_2981_begin_0, end = var_2981_end_0, end_mask = var_2981_end_mask_0, x = var_2906_cast_fp16)[name = string("op_2981_cast_fp16")]; + tensor var_2988_begin_0 = const()[name = string("op_2988_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_2988_end_0 = const()[name = string("op_2988_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_2988_end_mask_0 = const()[name = string("op_2988_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2988_cast_fp16 = slice_by_index(begin = var_2988_begin_0, end = var_2988_end_0, end_mask = var_2988_end_mask_0, x = var_2906_cast_fp16)[name = string("op_2988_cast_fp16")]; + tensor var_2995_begin_0 = const()[name = string("op_2995_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2995_end_0 = const()[name = string("op_2995_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_2995_end_mask_0 = const()[name = string("op_2995_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2995_cast_fp16 = slice_by_index(begin = var_2995_begin_0, end = var_2995_end_0, end_mask = var_2995_end_mask_0, x = var_2910_cast_fp16)[name = string("op_2995_cast_fp16")]; + tensor var_3002_begin_0 = const()[name = string("op_3002_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3002_end_0 = const()[name = string("op_3002_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3002_end_mask_0 = const()[name = string("op_3002_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3002_cast_fp16 = slice_by_index(begin = var_3002_begin_0, end = var_3002_end_0, end_mask = var_3002_end_mask_0, x = var_2910_cast_fp16)[name = string("op_3002_cast_fp16")]; + tensor var_3009_begin_0 = const()[name = string("op_3009_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3009_end_0 = const()[name = string("op_3009_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3009_end_mask_0 = const()[name = string("op_3009_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3009_cast_fp16 = slice_by_index(begin = var_3009_begin_0, end = var_3009_end_0, end_mask = var_3009_end_mask_0, x = var_2910_cast_fp16)[name = string("op_3009_cast_fp16")]; + tensor var_3016_begin_0 = const()[name = string("op_3016_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3016_end_0 = const()[name = string("op_3016_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3016_end_mask_0 = const()[name = string("op_3016_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3016_cast_fp16 = slice_by_index(begin = var_3016_begin_0, end = var_3016_end_0, end_mask = var_3016_end_mask_0, x = var_2910_cast_fp16)[name = string("op_3016_cast_fp16")]; + tensor var_3023_begin_0 = const()[name = string("op_3023_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3023_end_0 = const()[name = string("op_3023_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3023_end_mask_0 = const()[name = string("op_3023_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3023_cast_fp16 = slice_by_index(begin = var_3023_begin_0, end = var_3023_end_0, end_mask = var_3023_end_mask_0, x = var_2914_cast_fp16)[name = string("op_3023_cast_fp16")]; + tensor var_3030_begin_0 = const()[name = string("op_3030_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3030_end_0 = const()[name = string("op_3030_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3030_end_mask_0 = const()[name = string("op_3030_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3030_cast_fp16 = slice_by_index(begin = var_3030_begin_0, end = var_3030_end_0, end_mask = var_3030_end_mask_0, x = var_2914_cast_fp16)[name = string("op_3030_cast_fp16")]; + tensor var_3037_begin_0 = const()[name = string("op_3037_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3037_end_0 = const()[name = string("op_3037_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3037_end_mask_0 = const()[name = string("op_3037_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3037_cast_fp16 = slice_by_index(begin = var_3037_begin_0, end = var_3037_end_0, end_mask = var_3037_end_mask_0, x = var_2914_cast_fp16)[name = string("op_3037_cast_fp16")]; + tensor var_3044_begin_0 = const()[name = string("op_3044_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3044_end_0 = const()[name = string("op_3044_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3044_end_mask_0 = const()[name = string("op_3044_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3044_cast_fp16 = slice_by_index(begin = var_3044_begin_0, end = var_3044_end_0, end_mask = var_3044_end_mask_0, x = var_2914_cast_fp16)[name = string("op_3044_cast_fp16")]; + tensor var_3051_begin_0 = const()[name = string("op_3051_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3051_end_0 = const()[name = string("op_3051_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3051_end_mask_0 = const()[name = string("op_3051_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3051_cast_fp16 = slice_by_index(begin = var_3051_begin_0, end = var_3051_end_0, end_mask = var_3051_end_mask_0, x = var_2918_cast_fp16)[name = string("op_3051_cast_fp16")]; + tensor var_3058_begin_0 = const()[name = string("op_3058_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3058_end_0 = const()[name = string("op_3058_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3058_end_mask_0 = const()[name = string("op_3058_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3058_cast_fp16 = slice_by_index(begin = var_3058_begin_0, end = var_3058_end_0, end_mask = var_3058_end_mask_0, x = var_2918_cast_fp16)[name = string("op_3058_cast_fp16")]; + tensor var_3065_begin_0 = const()[name = string("op_3065_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3065_end_0 = const()[name = string("op_3065_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3065_end_mask_0 = const()[name = string("op_3065_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3065_cast_fp16 = slice_by_index(begin = var_3065_begin_0, end = var_3065_end_0, end_mask = var_3065_end_mask_0, x = var_2918_cast_fp16)[name = string("op_3065_cast_fp16")]; + tensor var_3072_begin_0 = const()[name = string("op_3072_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3072_end_0 = const()[name = string("op_3072_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3072_end_mask_0 = const()[name = string("op_3072_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3072_cast_fp16 = slice_by_index(begin = var_3072_begin_0, end = var_3072_end_0, end_mask = var_3072_end_mask_0, x = var_2918_cast_fp16)[name = string("op_3072_cast_fp16")]; + tensor var_3079_begin_0 = const()[name = string("op_3079_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3079_end_0 = const()[name = string("op_3079_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3079_end_mask_0 = const()[name = string("op_3079_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3079_cast_fp16 = slice_by_index(begin = var_3079_begin_0, end = var_3079_end_0, end_mask = var_3079_end_mask_0, x = var_2922_cast_fp16)[name = string("op_3079_cast_fp16")]; + tensor var_3086_begin_0 = const()[name = string("op_3086_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3086_end_0 = const()[name = string("op_3086_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3086_end_mask_0 = const()[name = string("op_3086_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3086_cast_fp16 = slice_by_index(begin = var_3086_begin_0, end = var_3086_end_0, end_mask = var_3086_end_mask_0, x = var_2922_cast_fp16)[name = string("op_3086_cast_fp16")]; + tensor var_3093_begin_0 = const()[name = string("op_3093_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3093_end_0 = const()[name = string("op_3093_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3093_end_mask_0 = const()[name = string("op_3093_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3093_cast_fp16 = slice_by_index(begin = var_3093_begin_0, end = var_3093_end_0, end_mask = var_3093_end_mask_0, x = var_2922_cast_fp16)[name = string("op_3093_cast_fp16")]; + tensor var_3100_begin_0 = const()[name = string("op_3100_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3100_end_0 = const()[name = string("op_3100_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3100_end_mask_0 = const()[name = string("op_3100_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3100_cast_fp16 = slice_by_index(begin = var_3100_begin_0, end = var_3100_end_0, end_mask = var_3100_end_mask_0, x = var_2922_cast_fp16)[name = string("op_3100_cast_fp16")]; + tensor var_3107_begin_0 = const()[name = string("op_3107_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3107_end_0 = const()[name = string("op_3107_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3107_end_mask_0 = const()[name = string("op_3107_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3107_cast_fp16 = slice_by_index(begin = var_3107_begin_0, end = var_3107_end_0, end_mask = var_3107_end_mask_0, x = var_2926_cast_fp16)[name = string("op_3107_cast_fp16")]; + tensor var_3114_begin_0 = const()[name = string("op_3114_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3114_end_0 = const()[name = string("op_3114_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3114_end_mask_0 = const()[name = string("op_3114_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3114_cast_fp16 = slice_by_index(begin = var_3114_begin_0, end = var_3114_end_0, end_mask = var_3114_end_mask_0, x = var_2926_cast_fp16)[name = string("op_3114_cast_fp16")]; + tensor var_3121_begin_0 = const()[name = string("op_3121_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3121_end_0 = const()[name = string("op_3121_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3121_end_mask_0 = const()[name = string("op_3121_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3121_cast_fp16 = slice_by_index(begin = var_3121_begin_0, end = var_3121_end_0, end_mask = var_3121_end_mask_0, x = var_2926_cast_fp16)[name = string("op_3121_cast_fp16")]; + tensor var_3128_begin_0 = const()[name = string("op_3128_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3128_end_0 = const()[name = string("op_3128_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3128_end_mask_0 = const()[name = string("op_3128_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3128_cast_fp16 = slice_by_index(begin = var_3128_begin_0, end = var_3128_end_0, end_mask = var_3128_end_mask_0, x = var_2926_cast_fp16)[name = string("op_3128_cast_fp16")]; + tensor var_3135_begin_0 = const()[name = string("op_3135_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3135_end_0 = const()[name = string("op_3135_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3135_end_mask_0 = const()[name = string("op_3135_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3135_cast_fp16 = slice_by_index(begin = var_3135_begin_0, end = var_3135_end_0, end_mask = var_3135_end_mask_0, x = var_2930_cast_fp16)[name = string("op_3135_cast_fp16")]; + tensor var_3142_begin_0 = const()[name = string("op_3142_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3142_end_0 = const()[name = string("op_3142_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3142_end_mask_0 = const()[name = string("op_3142_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3142_cast_fp16 = slice_by_index(begin = var_3142_begin_0, end = var_3142_end_0, end_mask = var_3142_end_mask_0, x = var_2930_cast_fp16)[name = string("op_3142_cast_fp16")]; + tensor var_3149_begin_0 = const()[name = string("op_3149_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3149_end_0 = const()[name = string("op_3149_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3149_end_mask_0 = const()[name = string("op_3149_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3149_cast_fp16 = slice_by_index(begin = var_3149_begin_0, end = var_3149_end_0, end_mask = var_3149_end_mask_0, x = var_2930_cast_fp16)[name = string("op_3149_cast_fp16")]; + tensor var_3156_begin_0 = const()[name = string("op_3156_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3156_end_0 = const()[name = string("op_3156_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3156_end_mask_0 = const()[name = string("op_3156_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3156_cast_fp16 = slice_by_index(begin = var_3156_begin_0, end = var_3156_end_0, end_mask = var_3156_end_mask_0, x = var_2930_cast_fp16)[name = string("op_3156_cast_fp16")]; + tensor k_9_perm_0 = const()[name = string("k_9_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_3161_begin_0 = const()[name = string("op_3161_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3161_end_0 = const()[name = string("op_3161_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_3161_end_mask_0 = const()[name = string("op_3161_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_9_cast_fp16 = transpose(perm = k_9_perm_0, x = key_9_cast_fp16)[name = string("transpose_1")]; + tensor var_3161_cast_fp16 = slice_by_index(begin = var_3161_begin_0, end = var_3161_end_0, end_mask = var_3161_end_mask_0, x = k_9_cast_fp16)[name = string("op_3161_cast_fp16")]; + tensor var_3165_begin_0 = const()[name = string("op_3165_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_3165_end_0 = const()[name = string("op_3165_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_3165_end_mask_0 = const()[name = string("op_3165_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3165_cast_fp16 = slice_by_index(begin = var_3165_begin_0, end = var_3165_end_0, end_mask = var_3165_end_mask_0, x = k_9_cast_fp16)[name = string("op_3165_cast_fp16")]; + tensor var_3169_begin_0 = const()[name = string("op_3169_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_3169_end_0 = const()[name = string("op_3169_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_3169_end_mask_0 = const()[name = string("op_3169_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3169_cast_fp16 = slice_by_index(begin = var_3169_begin_0, end = var_3169_end_0, end_mask = var_3169_end_mask_0, x = k_9_cast_fp16)[name = string("op_3169_cast_fp16")]; + tensor var_3173_begin_0 = const()[name = string("op_3173_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_3173_end_0 = const()[name = string("op_3173_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_3173_end_mask_0 = const()[name = string("op_3173_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3173_cast_fp16 = slice_by_index(begin = var_3173_begin_0, end = var_3173_end_0, end_mask = var_3173_end_mask_0, x = k_9_cast_fp16)[name = string("op_3173_cast_fp16")]; + tensor var_3177_begin_0 = const()[name = string("op_3177_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_3177_end_0 = const()[name = string("op_3177_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_3177_end_mask_0 = const()[name = string("op_3177_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3177_cast_fp16 = slice_by_index(begin = var_3177_begin_0, end = var_3177_end_0, end_mask = var_3177_end_mask_0, x = k_9_cast_fp16)[name = string("op_3177_cast_fp16")]; + tensor var_3181_begin_0 = const()[name = string("op_3181_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_3181_end_0 = const()[name = string("op_3181_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_3181_end_mask_0 = const()[name = string("op_3181_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3181_cast_fp16 = slice_by_index(begin = var_3181_begin_0, end = var_3181_end_0, end_mask = var_3181_end_mask_0, x = k_9_cast_fp16)[name = string("op_3181_cast_fp16")]; + tensor var_3185_begin_0 = const()[name = string("op_3185_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_3185_end_0 = const()[name = string("op_3185_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_3185_end_mask_0 = const()[name = string("op_3185_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3185_cast_fp16 = slice_by_index(begin = var_3185_begin_0, end = var_3185_end_0, end_mask = var_3185_end_mask_0, x = k_9_cast_fp16)[name = string("op_3185_cast_fp16")]; + tensor var_3189_begin_0 = const()[name = string("op_3189_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_3189_end_0 = const()[name = string("op_3189_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_3189_end_mask_0 = const()[name = string("op_3189_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3189_cast_fp16 = slice_by_index(begin = var_3189_begin_0, end = var_3189_end_0, end_mask = var_3189_end_mask_0, x = k_9_cast_fp16)[name = string("op_3189_cast_fp16")]; + tensor var_3191_begin_0 = const()[name = string("op_3191_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3191_end_0 = const()[name = string("op_3191_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3191_end_mask_0 = const()[name = string("op_3191_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3191_cast_fp16 = slice_by_index(begin = var_3191_begin_0, end = var_3191_end_0, end_mask = var_3191_end_mask_0, x = value_9_cast_fp16)[name = string("op_3191_cast_fp16")]; + tensor var_3195_begin_0 = const()[name = string("op_3195_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3195_end_0 = const()[name = string("op_3195_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_3195_end_mask_0 = const()[name = string("op_3195_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3195_cast_fp16 = slice_by_index(begin = var_3195_begin_0, end = var_3195_end_0, end_mask = var_3195_end_mask_0, x = value_9_cast_fp16)[name = string("op_3195_cast_fp16")]; + tensor var_3199_begin_0 = const()[name = string("op_3199_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3199_end_0 = const()[name = string("op_3199_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_3199_end_mask_0 = const()[name = string("op_3199_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3199_cast_fp16 = slice_by_index(begin = var_3199_begin_0, end = var_3199_end_0, end_mask = var_3199_end_mask_0, x = value_9_cast_fp16)[name = string("op_3199_cast_fp16")]; + tensor var_3203_begin_0 = const()[name = string("op_3203_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3203_end_0 = const()[name = string("op_3203_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_3203_end_mask_0 = const()[name = string("op_3203_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3203_cast_fp16 = slice_by_index(begin = var_3203_begin_0, end = var_3203_end_0, end_mask = var_3203_end_mask_0, x = value_9_cast_fp16)[name = string("op_3203_cast_fp16")]; + tensor var_3207_begin_0 = const()[name = string("op_3207_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3207_end_0 = const()[name = string("op_3207_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_3207_end_mask_0 = const()[name = string("op_3207_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3207_cast_fp16 = slice_by_index(begin = var_3207_begin_0, end = var_3207_end_0, end_mask = var_3207_end_mask_0, x = value_9_cast_fp16)[name = string("op_3207_cast_fp16")]; + tensor var_3211_begin_0 = const()[name = string("op_3211_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3211_end_0 = const()[name = string("op_3211_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_3211_end_mask_0 = const()[name = string("op_3211_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3211_cast_fp16 = slice_by_index(begin = var_3211_begin_0, end = var_3211_end_0, end_mask = var_3211_end_mask_0, x = value_9_cast_fp16)[name = string("op_3211_cast_fp16")]; + tensor var_3215_begin_0 = const()[name = string("op_3215_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3215_end_0 = const()[name = string("op_3215_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_3215_end_mask_0 = const()[name = string("op_3215_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3215_cast_fp16 = slice_by_index(begin = var_3215_begin_0, end = var_3215_end_0, end_mask = var_3215_end_mask_0, x = value_9_cast_fp16)[name = string("op_3215_cast_fp16")]; + tensor var_3219_begin_0 = const()[name = string("op_3219_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3219_end_0 = const()[name = string("op_3219_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_3219_end_mask_0 = const()[name = string("op_3219_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3219_cast_fp16 = slice_by_index(begin = var_3219_begin_0, end = var_3219_end_0, end_mask = var_3219_end_mask_0, x = value_9_cast_fp16)[name = string("op_3219_cast_fp16")]; + string _SplitHeadsQ__mh_w_257_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_257_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_257_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_257_equation_0, values = (var_3161_cast_fp16, var_2939_cast_fp16))[name = string("_SplitHeadsQ__mh_w_257_cast_fp16")]; + string _SplitHeadsQ__mh_w_259_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_259_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_259_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_259_equation_0, values = (var_3161_cast_fp16, var_2946_cast_fp16))[name = string("_SplitHeadsQ__mh_w_259_cast_fp16")]; + string _SplitHeadsQ__mh_w_261_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_261_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_261_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_261_equation_0, values = (var_3161_cast_fp16, var_2953_cast_fp16))[name = string("_SplitHeadsQ__mh_w_261_cast_fp16")]; + string _SplitHeadsQ__mh_w_263_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_263_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_263_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_263_equation_0, values = (var_3161_cast_fp16, var_2960_cast_fp16))[name = string("_SplitHeadsQ__mh_w_263_cast_fp16")]; + string _SplitHeadsQ__mh_w_265_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_265_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_265_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_265_equation_0, values = (var_3165_cast_fp16, var_2967_cast_fp16))[name = string("_SplitHeadsQ__mh_w_265_cast_fp16")]; + string _SplitHeadsQ__mh_w_267_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_267_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_267_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_267_equation_0, values = (var_3165_cast_fp16, var_2974_cast_fp16))[name = string("_SplitHeadsQ__mh_w_267_cast_fp16")]; + string _SplitHeadsQ__mh_w_269_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_269_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_269_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_269_equation_0, values = (var_3165_cast_fp16, var_2981_cast_fp16))[name = string("_SplitHeadsQ__mh_w_269_cast_fp16")]; + string _SplitHeadsQ__mh_w_271_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_271_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_271_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_271_equation_0, values = (var_3165_cast_fp16, var_2988_cast_fp16))[name = string("_SplitHeadsQ__mh_w_271_cast_fp16")]; + string _SplitHeadsQ__mh_w_273_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_273_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_273_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_273_equation_0, values = (var_3169_cast_fp16, var_2995_cast_fp16))[name = string("_SplitHeadsQ__mh_w_273_cast_fp16")]; + string _SplitHeadsQ__mh_w_275_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_275_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_275_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_275_equation_0, values = (var_3169_cast_fp16, var_3002_cast_fp16))[name = string("_SplitHeadsQ__mh_w_275_cast_fp16")]; + string _SplitHeadsQ__mh_w_277_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_277_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_277_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_277_equation_0, values = (var_3169_cast_fp16, var_3009_cast_fp16))[name = string("_SplitHeadsQ__mh_w_277_cast_fp16")]; + string _SplitHeadsQ__mh_w_279_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_279_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_279_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_279_equation_0, values = (var_3169_cast_fp16, var_3016_cast_fp16))[name = string("_SplitHeadsQ__mh_w_279_cast_fp16")]; + string _SplitHeadsQ__mh_w_281_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_281_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_281_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_281_equation_0, values = (var_3173_cast_fp16, var_3023_cast_fp16))[name = string("_SplitHeadsQ__mh_w_281_cast_fp16")]; + string _SplitHeadsQ__mh_w_283_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_283_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_283_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_283_equation_0, values = (var_3173_cast_fp16, var_3030_cast_fp16))[name = string("_SplitHeadsQ__mh_w_283_cast_fp16")]; + string _SplitHeadsQ__mh_w_285_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_285_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_285_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_285_equation_0, values = (var_3173_cast_fp16, var_3037_cast_fp16))[name = string("_SplitHeadsQ__mh_w_285_cast_fp16")]; + string _SplitHeadsQ__mh_w_287_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_287_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_287_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_287_equation_0, values = (var_3173_cast_fp16, var_3044_cast_fp16))[name = string("_SplitHeadsQ__mh_w_287_cast_fp16")]; + string _SplitHeadsQ__mh_w_289_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_289_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_289_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_289_equation_0, values = (var_3177_cast_fp16, var_3051_cast_fp16))[name = string("_SplitHeadsQ__mh_w_289_cast_fp16")]; + string _SplitHeadsQ__mh_w_291_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_291_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_291_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_291_equation_0, values = (var_3177_cast_fp16, var_3058_cast_fp16))[name = string("_SplitHeadsQ__mh_w_291_cast_fp16")]; + string _SplitHeadsQ__mh_w_293_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_293_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_293_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_293_equation_0, values = (var_3177_cast_fp16, var_3065_cast_fp16))[name = string("_SplitHeadsQ__mh_w_293_cast_fp16")]; + string _SplitHeadsQ__mh_w_295_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_295_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_295_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_295_equation_0, values = (var_3177_cast_fp16, var_3072_cast_fp16))[name = string("_SplitHeadsQ__mh_w_295_cast_fp16")]; + string _SplitHeadsQ__mh_w_297_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_297_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_297_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_297_equation_0, values = (var_3181_cast_fp16, var_3079_cast_fp16))[name = string("_SplitHeadsQ__mh_w_297_cast_fp16")]; + string _SplitHeadsQ__mh_w_299_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_299_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_299_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_299_equation_0, values = (var_3181_cast_fp16, var_3086_cast_fp16))[name = string("_SplitHeadsQ__mh_w_299_cast_fp16")]; + string _SplitHeadsQ__mh_w_301_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_301_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_301_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_301_equation_0, values = (var_3181_cast_fp16, var_3093_cast_fp16))[name = string("_SplitHeadsQ__mh_w_301_cast_fp16")]; + string _SplitHeadsQ__mh_w_303_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_303_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_303_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_303_equation_0, values = (var_3181_cast_fp16, var_3100_cast_fp16))[name = string("_SplitHeadsQ__mh_w_303_cast_fp16")]; + string _SplitHeadsQ__mh_w_305_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_305_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_305_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_305_equation_0, values = (var_3185_cast_fp16, var_3107_cast_fp16))[name = string("_SplitHeadsQ__mh_w_305_cast_fp16")]; + string _SplitHeadsQ__mh_w_307_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_307_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_307_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_307_equation_0, values = (var_3185_cast_fp16, var_3114_cast_fp16))[name = string("_SplitHeadsQ__mh_w_307_cast_fp16")]; + string _SplitHeadsQ__mh_w_309_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_309_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_309_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_309_equation_0, values = (var_3185_cast_fp16, var_3121_cast_fp16))[name = string("_SplitHeadsQ__mh_w_309_cast_fp16")]; + string _SplitHeadsQ__mh_w_311_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_311_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_311_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_311_equation_0, values = (var_3185_cast_fp16, var_3128_cast_fp16))[name = string("_SplitHeadsQ__mh_w_311_cast_fp16")]; + string _SplitHeadsQ__mh_w_313_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_313_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_313_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_313_equation_0, values = (var_3189_cast_fp16, var_3135_cast_fp16))[name = string("_SplitHeadsQ__mh_w_313_cast_fp16")]; + string _SplitHeadsQ__mh_w_315_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_315_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_315_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_315_equation_0, values = (var_3189_cast_fp16, var_3142_cast_fp16))[name = string("_SplitHeadsQ__mh_w_315_cast_fp16")]; + string _SplitHeadsQ__mh_w_317_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_317_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_317_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_317_equation_0, values = (var_3189_cast_fp16, var_3149_cast_fp16))[name = string("_SplitHeadsQ__mh_w_317_cast_fp16")]; + string _SplitHeadsQ__mh_w_319_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_319_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_319_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_319_equation_0, values = (var_3189_cast_fp16, var_3156_cast_fp16))[name = string("_SplitHeadsQ__mh_w_319_cast_fp16")]; + fp16 var_3286_to_fp16 = const()[name = string("op_3286_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_257_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_257_cast_fp16, y = var_3286_to_fp16)[name = string("aw_chunk_257_cast_fp16")]; + fp16 var_3288_to_fp16 = const()[name = string("op_3288_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_259_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_259_cast_fp16, y = var_3288_to_fp16)[name = string("aw_chunk_259_cast_fp16")]; + fp16 var_3290_to_fp16 = const()[name = string("op_3290_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_261_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_261_cast_fp16, y = var_3290_to_fp16)[name = string("aw_chunk_261_cast_fp16")]; + fp16 var_3292_to_fp16 = const()[name = string("op_3292_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_263_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_263_cast_fp16, y = var_3292_to_fp16)[name = string("aw_chunk_263_cast_fp16")]; + fp16 var_3294_to_fp16 = const()[name = string("op_3294_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_265_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_265_cast_fp16, y = var_3294_to_fp16)[name = string("aw_chunk_265_cast_fp16")]; + fp16 var_3296_to_fp16 = const()[name = string("op_3296_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_267_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_267_cast_fp16, y = var_3296_to_fp16)[name = string("aw_chunk_267_cast_fp16")]; + fp16 var_3298_to_fp16 = const()[name = string("op_3298_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_269_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_269_cast_fp16, y = var_3298_to_fp16)[name = string("aw_chunk_269_cast_fp16")]; + fp16 var_3300_to_fp16 = const()[name = string("op_3300_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_271_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_271_cast_fp16, y = var_3300_to_fp16)[name = string("aw_chunk_271_cast_fp16")]; + fp16 var_3302_to_fp16 = const()[name = string("op_3302_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_273_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_273_cast_fp16, y = var_3302_to_fp16)[name = string("aw_chunk_273_cast_fp16")]; + fp16 var_3304_to_fp16 = const()[name = string("op_3304_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_275_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_275_cast_fp16, y = var_3304_to_fp16)[name = string("aw_chunk_275_cast_fp16")]; + fp16 var_3306_to_fp16 = const()[name = string("op_3306_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_277_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_277_cast_fp16, y = var_3306_to_fp16)[name = string("aw_chunk_277_cast_fp16")]; + fp16 var_3308_to_fp16 = const()[name = string("op_3308_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_279_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_279_cast_fp16, y = var_3308_to_fp16)[name = string("aw_chunk_279_cast_fp16")]; + fp16 var_3310_to_fp16 = const()[name = string("op_3310_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_281_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_281_cast_fp16, y = var_3310_to_fp16)[name = string("aw_chunk_281_cast_fp16")]; + fp16 var_3312_to_fp16 = const()[name = string("op_3312_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_283_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_283_cast_fp16, y = var_3312_to_fp16)[name = string("aw_chunk_283_cast_fp16")]; + fp16 var_3314_to_fp16 = const()[name = string("op_3314_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_285_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_285_cast_fp16, y = var_3314_to_fp16)[name = string("aw_chunk_285_cast_fp16")]; + fp16 var_3316_to_fp16 = const()[name = string("op_3316_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_287_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_287_cast_fp16, y = var_3316_to_fp16)[name = string("aw_chunk_287_cast_fp16")]; + fp16 var_3318_to_fp16 = const()[name = string("op_3318_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_289_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_289_cast_fp16, y = var_3318_to_fp16)[name = string("aw_chunk_289_cast_fp16")]; + fp16 var_3320_to_fp16 = const()[name = string("op_3320_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_291_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_291_cast_fp16, y = var_3320_to_fp16)[name = string("aw_chunk_291_cast_fp16")]; + fp16 var_3322_to_fp16 = const()[name = string("op_3322_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_293_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_293_cast_fp16, y = var_3322_to_fp16)[name = string("aw_chunk_293_cast_fp16")]; + fp16 var_3324_to_fp16 = const()[name = string("op_3324_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_295_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_295_cast_fp16, y = var_3324_to_fp16)[name = string("aw_chunk_295_cast_fp16")]; + fp16 var_3326_to_fp16 = const()[name = string("op_3326_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_297_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_297_cast_fp16, y = var_3326_to_fp16)[name = string("aw_chunk_297_cast_fp16")]; + fp16 var_3328_to_fp16 = const()[name = string("op_3328_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_299_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_299_cast_fp16, y = var_3328_to_fp16)[name = string("aw_chunk_299_cast_fp16")]; + fp16 var_3330_to_fp16 = const()[name = string("op_3330_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_301_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_301_cast_fp16, y = var_3330_to_fp16)[name = string("aw_chunk_301_cast_fp16")]; + fp16 var_3332_to_fp16 = const()[name = string("op_3332_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_303_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_303_cast_fp16, y = var_3332_to_fp16)[name = string("aw_chunk_303_cast_fp16")]; + fp16 var_3334_to_fp16 = const()[name = string("op_3334_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_305_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_305_cast_fp16, y = var_3334_to_fp16)[name = string("aw_chunk_305_cast_fp16")]; + fp16 var_3336_to_fp16 = const()[name = string("op_3336_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_307_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_307_cast_fp16, y = var_3336_to_fp16)[name = string("aw_chunk_307_cast_fp16")]; + fp16 var_3338_to_fp16 = const()[name = string("op_3338_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_309_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_309_cast_fp16, y = var_3338_to_fp16)[name = string("aw_chunk_309_cast_fp16")]; + fp16 var_3340_to_fp16 = const()[name = string("op_3340_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_311_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_311_cast_fp16, y = var_3340_to_fp16)[name = string("aw_chunk_311_cast_fp16")]; + fp16 var_3342_to_fp16 = const()[name = string("op_3342_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_313_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_313_cast_fp16, y = var_3342_to_fp16)[name = string("aw_chunk_313_cast_fp16")]; + fp16 var_3344_to_fp16 = const()[name = string("op_3344_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_315_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_315_cast_fp16, y = var_3344_to_fp16)[name = string("aw_chunk_315_cast_fp16")]; + fp16 var_3346_to_fp16 = const()[name = string("op_3346_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_317_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_317_cast_fp16, y = var_3346_to_fp16)[name = string("aw_chunk_317_cast_fp16")]; + fp16 var_3348_to_fp16 = const()[name = string("op_3348_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_319_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_319_cast_fp16, y = var_3348_to_fp16)[name = string("aw_chunk_319_cast_fp16")]; + tensor var_3350_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_257_cast_fp16)[name = string("op_3350_cast_fp16")]; + tensor var_3351_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_259_cast_fp16)[name = string("op_3351_cast_fp16")]; + tensor var_3352_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_261_cast_fp16)[name = string("op_3352_cast_fp16")]; + tensor var_3353_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_263_cast_fp16)[name = string("op_3353_cast_fp16")]; + tensor var_3354_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_265_cast_fp16)[name = string("op_3354_cast_fp16")]; + tensor var_3355_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_267_cast_fp16)[name = string("op_3355_cast_fp16")]; + tensor var_3356_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_269_cast_fp16)[name = string("op_3356_cast_fp16")]; + tensor var_3357_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_271_cast_fp16)[name = string("op_3357_cast_fp16")]; + tensor var_3358_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_273_cast_fp16)[name = string("op_3358_cast_fp16")]; + tensor var_3359_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_275_cast_fp16)[name = string("op_3359_cast_fp16")]; + tensor var_3360_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_277_cast_fp16)[name = string("op_3360_cast_fp16")]; + tensor var_3361_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_279_cast_fp16)[name = string("op_3361_cast_fp16")]; + tensor var_3362_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_281_cast_fp16)[name = string("op_3362_cast_fp16")]; + tensor var_3363_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_283_cast_fp16)[name = string("op_3363_cast_fp16")]; + tensor var_3364_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_285_cast_fp16)[name = string("op_3364_cast_fp16")]; + tensor var_3365_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_287_cast_fp16)[name = string("op_3365_cast_fp16")]; + tensor var_3366_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_289_cast_fp16)[name = string("op_3366_cast_fp16")]; + tensor var_3367_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_291_cast_fp16)[name = string("op_3367_cast_fp16")]; + tensor var_3368_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_293_cast_fp16)[name = string("op_3368_cast_fp16")]; + tensor var_3369_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_295_cast_fp16)[name = string("op_3369_cast_fp16")]; + tensor var_3370_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_297_cast_fp16)[name = string("op_3370_cast_fp16")]; + tensor var_3371_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_299_cast_fp16)[name = string("op_3371_cast_fp16")]; + tensor var_3372_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_301_cast_fp16)[name = string("op_3372_cast_fp16")]; + tensor var_3373_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_303_cast_fp16)[name = string("op_3373_cast_fp16")]; + tensor var_3374_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_305_cast_fp16)[name = string("op_3374_cast_fp16")]; + tensor var_3375_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_307_cast_fp16)[name = string("op_3375_cast_fp16")]; + tensor var_3376_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_309_cast_fp16)[name = string("op_3376_cast_fp16")]; + tensor var_3377_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_311_cast_fp16)[name = string("op_3377_cast_fp16")]; + tensor var_3378_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_313_cast_fp16)[name = string("op_3378_cast_fp16")]; + tensor var_3379_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_315_cast_fp16)[name = string("op_3379_cast_fp16")]; + tensor var_3380_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_317_cast_fp16)[name = string("op_3380_cast_fp16")]; + tensor var_3381_cast_fp16 = softmax(axis = var_2847, x = aw_chunk_319_cast_fp16)[name = string("op_3381_cast_fp16")]; + string var_3383_equation_0 = const()[name = string("op_3383_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3383_cast_fp16 = einsum(equation = var_3383_equation_0, values = (var_3191_cast_fp16, var_3350_cast_fp16))[name = string("op_3383_cast_fp16")]; + string var_3385_equation_0 = const()[name = string("op_3385_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3385_cast_fp16 = einsum(equation = var_3385_equation_0, values = (var_3191_cast_fp16, var_3351_cast_fp16))[name = string("op_3385_cast_fp16")]; + string var_3387_equation_0 = const()[name = string("op_3387_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3387_cast_fp16 = einsum(equation = var_3387_equation_0, values = (var_3191_cast_fp16, var_3352_cast_fp16))[name = string("op_3387_cast_fp16")]; + string var_3389_equation_0 = const()[name = string("op_3389_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3389_cast_fp16 = einsum(equation = var_3389_equation_0, values = (var_3191_cast_fp16, var_3353_cast_fp16))[name = string("op_3389_cast_fp16")]; + string var_3391_equation_0 = const()[name = string("op_3391_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3391_cast_fp16 = einsum(equation = var_3391_equation_0, values = (var_3195_cast_fp16, var_3354_cast_fp16))[name = string("op_3391_cast_fp16")]; + string var_3393_equation_0 = const()[name = string("op_3393_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3393_cast_fp16 = einsum(equation = var_3393_equation_0, values = (var_3195_cast_fp16, var_3355_cast_fp16))[name = string("op_3393_cast_fp16")]; + string var_3395_equation_0 = const()[name = string("op_3395_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3395_cast_fp16 = einsum(equation = var_3395_equation_0, values = (var_3195_cast_fp16, var_3356_cast_fp16))[name = string("op_3395_cast_fp16")]; + string var_3397_equation_0 = const()[name = string("op_3397_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3397_cast_fp16 = einsum(equation = var_3397_equation_0, values = (var_3195_cast_fp16, var_3357_cast_fp16))[name = string("op_3397_cast_fp16")]; + string var_3399_equation_0 = const()[name = string("op_3399_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3399_cast_fp16 = einsum(equation = var_3399_equation_0, values = (var_3199_cast_fp16, var_3358_cast_fp16))[name = string("op_3399_cast_fp16")]; + string var_3401_equation_0 = const()[name = string("op_3401_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3401_cast_fp16 = einsum(equation = var_3401_equation_0, values = (var_3199_cast_fp16, var_3359_cast_fp16))[name = string("op_3401_cast_fp16")]; + string var_3403_equation_0 = const()[name = string("op_3403_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3403_cast_fp16 = einsum(equation = var_3403_equation_0, values = (var_3199_cast_fp16, var_3360_cast_fp16))[name = string("op_3403_cast_fp16")]; + string var_3405_equation_0 = const()[name = string("op_3405_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3405_cast_fp16 = einsum(equation = var_3405_equation_0, values = (var_3199_cast_fp16, var_3361_cast_fp16))[name = string("op_3405_cast_fp16")]; + string var_3407_equation_0 = const()[name = string("op_3407_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3407_cast_fp16 = einsum(equation = var_3407_equation_0, values = (var_3203_cast_fp16, var_3362_cast_fp16))[name = string("op_3407_cast_fp16")]; + string var_3409_equation_0 = const()[name = string("op_3409_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3409_cast_fp16 = einsum(equation = var_3409_equation_0, values = (var_3203_cast_fp16, var_3363_cast_fp16))[name = string("op_3409_cast_fp16")]; + string var_3411_equation_0 = const()[name = string("op_3411_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3411_cast_fp16 = einsum(equation = var_3411_equation_0, values = (var_3203_cast_fp16, var_3364_cast_fp16))[name = string("op_3411_cast_fp16")]; + string var_3413_equation_0 = const()[name = string("op_3413_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3413_cast_fp16 = einsum(equation = var_3413_equation_0, values = (var_3203_cast_fp16, var_3365_cast_fp16))[name = string("op_3413_cast_fp16")]; + string var_3415_equation_0 = const()[name = string("op_3415_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3415_cast_fp16 = einsum(equation = var_3415_equation_0, values = (var_3207_cast_fp16, var_3366_cast_fp16))[name = string("op_3415_cast_fp16")]; + string var_3417_equation_0 = const()[name = string("op_3417_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3417_cast_fp16 = einsum(equation = var_3417_equation_0, values = (var_3207_cast_fp16, var_3367_cast_fp16))[name = string("op_3417_cast_fp16")]; + string var_3419_equation_0 = const()[name = string("op_3419_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3419_cast_fp16 = einsum(equation = var_3419_equation_0, values = (var_3207_cast_fp16, var_3368_cast_fp16))[name = string("op_3419_cast_fp16")]; + string var_3421_equation_0 = const()[name = string("op_3421_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3421_cast_fp16 = einsum(equation = var_3421_equation_0, values = (var_3207_cast_fp16, var_3369_cast_fp16))[name = string("op_3421_cast_fp16")]; + string var_3423_equation_0 = const()[name = string("op_3423_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3423_cast_fp16 = einsum(equation = var_3423_equation_0, values = (var_3211_cast_fp16, var_3370_cast_fp16))[name = string("op_3423_cast_fp16")]; + string var_3425_equation_0 = const()[name = string("op_3425_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3425_cast_fp16 = einsum(equation = var_3425_equation_0, values = (var_3211_cast_fp16, var_3371_cast_fp16))[name = string("op_3425_cast_fp16")]; + string var_3427_equation_0 = const()[name = string("op_3427_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3427_cast_fp16 = einsum(equation = var_3427_equation_0, values = (var_3211_cast_fp16, var_3372_cast_fp16))[name = string("op_3427_cast_fp16")]; + string var_3429_equation_0 = const()[name = string("op_3429_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3429_cast_fp16 = einsum(equation = var_3429_equation_0, values = (var_3211_cast_fp16, var_3373_cast_fp16))[name = string("op_3429_cast_fp16")]; + string var_3431_equation_0 = const()[name = string("op_3431_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3431_cast_fp16 = einsum(equation = var_3431_equation_0, values = (var_3215_cast_fp16, var_3374_cast_fp16))[name = string("op_3431_cast_fp16")]; + string var_3433_equation_0 = const()[name = string("op_3433_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3433_cast_fp16 = einsum(equation = var_3433_equation_0, values = (var_3215_cast_fp16, var_3375_cast_fp16))[name = string("op_3433_cast_fp16")]; + string var_3435_equation_0 = const()[name = string("op_3435_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3435_cast_fp16 = einsum(equation = var_3435_equation_0, values = (var_3215_cast_fp16, var_3376_cast_fp16))[name = string("op_3435_cast_fp16")]; + string var_3437_equation_0 = const()[name = string("op_3437_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3437_cast_fp16 = einsum(equation = var_3437_equation_0, values = (var_3215_cast_fp16, var_3377_cast_fp16))[name = string("op_3437_cast_fp16")]; + string var_3439_equation_0 = const()[name = string("op_3439_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3439_cast_fp16 = einsum(equation = var_3439_equation_0, values = (var_3219_cast_fp16, var_3378_cast_fp16))[name = string("op_3439_cast_fp16")]; + string var_3441_equation_0 = const()[name = string("op_3441_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3441_cast_fp16 = einsum(equation = var_3441_equation_0, values = (var_3219_cast_fp16, var_3379_cast_fp16))[name = string("op_3441_cast_fp16")]; + string var_3443_equation_0 = const()[name = string("op_3443_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3443_cast_fp16 = einsum(equation = var_3443_equation_0, values = (var_3219_cast_fp16, var_3380_cast_fp16))[name = string("op_3443_cast_fp16")]; + string var_3445_equation_0 = const()[name = string("op_3445_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_3445_cast_fp16 = einsum(equation = var_3445_equation_0, values = (var_3219_cast_fp16, var_3381_cast_fp16))[name = string("op_3445_cast_fp16")]; + bool var_3447_interleave_0 = const()[name = string("op_3447_interleave_0"), val = bool(false)]; + tensor var_3447_cast_fp16 = concat(axis = var_2834, interleave = var_3447_interleave_0, values = (var_3383_cast_fp16, var_3385_cast_fp16, var_3387_cast_fp16, var_3389_cast_fp16))[name = string("op_3447_cast_fp16")]; + bool var_3449_interleave_0 = const()[name = string("op_3449_interleave_0"), val = bool(false)]; + tensor var_3449_cast_fp16 = concat(axis = var_2834, interleave = var_3449_interleave_0, values = (var_3391_cast_fp16, var_3393_cast_fp16, var_3395_cast_fp16, var_3397_cast_fp16))[name = string("op_3449_cast_fp16")]; + bool var_3451_interleave_0 = const()[name = string("op_3451_interleave_0"), val = bool(false)]; + tensor var_3451_cast_fp16 = concat(axis = var_2834, interleave = var_3451_interleave_0, values = (var_3399_cast_fp16, var_3401_cast_fp16, var_3403_cast_fp16, var_3405_cast_fp16))[name = string("op_3451_cast_fp16")]; + bool var_3453_interleave_0 = const()[name = string("op_3453_interleave_0"), val = bool(false)]; + tensor var_3453_cast_fp16 = concat(axis = var_2834, interleave = var_3453_interleave_0, values = (var_3407_cast_fp16, var_3409_cast_fp16, var_3411_cast_fp16, var_3413_cast_fp16))[name = string("op_3453_cast_fp16")]; + bool var_3455_interleave_0 = const()[name = string("op_3455_interleave_0"), val = bool(false)]; + tensor var_3455_cast_fp16 = concat(axis = var_2834, interleave = var_3455_interleave_0, values = (var_3415_cast_fp16, var_3417_cast_fp16, var_3419_cast_fp16, var_3421_cast_fp16))[name = string("op_3455_cast_fp16")]; + bool var_3457_interleave_0 = const()[name = string("op_3457_interleave_0"), val = bool(false)]; + tensor var_3457_cast_fp16 = concat(axis = var_2834, interleave = var_3457_interleave_0, values = (var_3423_cast_fp16, var_3425_cast_fp16, var_3427_cast_fp16, var_3429_cast_fp16))[name = string("op_3457_cast_fp16")]; + bool var_3459_interleave_0 = const()[name = string("op_3459_interleave_0"), val = bool(false)]; + tensor var_3459_cast_fp16 = concat(axis = var_2834, interleave = var_3459_interleave_0, values = (var_3431_cast_fp16, var_3433_cast_fp16, var_3435_cast_fp16, var_3437_cast_fp16))[name = string("op_3459_cast_fp16")]; + bool var_3461_interleave_0 = const()[name = string("op_3461_interleave_0"), val = bool(false)]; + tensor var_3461_cast_fp16 = concat(axis = var_2834, interleave = var_3461_interleave_0, values = (var_3439_cast_fp16, var_3441_cast_fp16, var_3443_cast_fp16, var_3445_cast_fp16))[name = string("op_3461_cast_fp16")]; + bool input_33_interleave_0 = const()[name = string("input_33_interleave_0"), val = bool(false)]; + tensor input_33_cast_fp16 = concat(axis = var_2847, interleave = input_33_interleave_0, values = (var_3447_cast_fp16, var_3449_cast_fp16, var_3451_cast_fp16, var_3453_cast_fp16, var_3455_cast_fp16, var_3457_cast_fp16, var_3459_cast_fp16, var_3461_cast_fp16))[name = string("input_33_cast_fp16")]; + string obj_19_pad_type_0 = const()[name = string("obj_19_pad_type_0"), val = string("valid")]; + tensor obj_19_strides_0 = const()[name = string("obj_19_strides_0"), val = tensor([1, 1])]; + tensor obj_19_pad_0 = const()[name = string("obj_19_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_19_dilations_0 = const()[name = string("obj_19_dilations_0"), val = tensor([1, 1])]; + int32 obj_19_groups_0 = const()[name = string("obj_19_groups_0"), val = int32(1)]; + tensor layers_4_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30155456)))]; + tensor layers_4_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_4_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30679808)))]; + tensor obj_19_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_bias_to_fp16, dilations = obj_19_dilations_0, groups = obj_19_groups_0, pad = obj_19_pad_0, pad_type = obj_19_pad_type_0, strides = obj_19_strides_0, weight = layers_4_self_attn_o_proj_weight_to_fp16, x = input_33_cast_fp16)[name = string("obj_19_cast_fp16")]; + tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = obj_19_cast_fp16)[name = string("inputs_19_cast_fp16")]; + tensor out_19_axes_0 = const()[name = string("out_19_axes_0"), val = tensor([1])]; + fp16 var_3480_to_fp16 = const()[name = string("op_3480_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_3480_to_fp16, x = inputs_19_cast_fp16)[name = string("out_19_cast_fp16")]; + tensor input_35_gamma_0_to_fp16 = const()[name = string("input_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30680896)))]; + tensor input_35_beta_0_to_fp16 = const()[name = string("input_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30681984)))]; + fp16 input_35_epsilon_0_to_fp16 = const()[name = string("input_35_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_35_cast_fp16 = batch_norm(beta = input_35_beta_0_to_fp16, epsilon = input_35_epsilon_0_to_fp16, gamma = input_35_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_19_cast_fp16)[name = string("input_35_cast_fp16")]; + string input_37_pad_type_0 = const()[name = string("input_37_pad_type_0"), val = string("valid")]; + tensor input_37_strides_0 = const()[name = string("input_37_strides_0"), val = tensor([1, 1])]; + tensor input_37_pad_0 = const()[name = string("input_37_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_37_dilations_0 = const()[name = string("input_37_dilations_0"), val = tensor([1, 1])]; + int32 input_37_groups_0 = const()[name = string("input_37_groups_0"), val = int32(1)]; + tensor layers_4_fc1_weight_to_fp16 = const()[name = string("layers_4_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(30683072)))]; + tensor layers_4_fc1_bias_to_fp16 = const()[name = string("layers_4_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32780288)))]; + tensor input_37_cast_fp16 = conv(bias = layers_4_fc1_bias_to_fp16, dilations = input_37_dilations_0, groups = input_37_groups_0, pad = input_37_pad_0, pad_type = input_37_pad_type_0, strides = input_37_strides_0, weight = layers_4_fc1_weight_to_fp16, x = input_35_cast_fp16)[name = string("input_37_cast_fp16")]; + string input_39_mode_0 = const()[name = string("input_39_mode_0"), val = string("EXACT")]; + tensor input_39_cast_fp16 = gelu(mode = input_39_mode_0, x = input_37_cast_fp16)[name = string("input_39_cast_fp16")]; + string hidden_states_13_pad_type_0 = const()[name = string("hidden_states_13_pad_type_0"), val = string("valid")]; + tensor hidden_states_13_strides_0 = const()[name = string("hidden_states_13_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_13_pad_0 = const()[name = string("hidden_states_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_13_dilations_0 = const()[name = string("hidden_states_13_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_13_groups_0 = const()[name = string("hidden_states_13_groups_0"), val = int32(1)]; + tensor layers_4_fc2_weight_to_fp16 = const()[name = string("layers_4_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(32784448)))]; + tensor layers_4_fc2_bias_to_fp16 = const()[name = string("layers_4_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34881664)))]; + tensor hidden_states_13_cast_fp16 = conv(bias = layers_4_fc2_bias_to_fp16, dilations = hidden_states_13_dilations_0, groups = hidden_states_13_groups_0, pad = hidden_states_13_pad_0, pad_type = hidden_states_13_pad_type_0, strides = hidden_states_13_strides_0, weight = layers_4_fc2_weight_to_fp16, x = input_39_cast_fp16)[name = string("hidden_states_13_cast_fp16")]; + tensor inputs_21_cast_fp16 = add(x = inputs_19_cast_fp16, y = hidden_states_13_cast_fp16)[name = string("inputs_21_cast_fp16")]; + int32 var_3509 = const()[name = string("op_3509"), val = int32(3)]; + int32 var_3522 = const()[name = string("op_3522"), val = int32(1)]; + tensor out_21_axes_0 = const()[name = string("out_21_axes_0"), val = tensor([1])]; + fp16 var_3539_to_fp16 = const()[name = string("op_3539_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_3539_to_fp16, x = inputs_21_cast_fp16)[name = string("out_21_cast_fp16")]; + tensor obj_21_gamma_0_to_fp16 = const()[name = string("obj_21_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34882752)))]; + tensor obj_21_beta_0_to_fp16 = const()[name = string("obj_21_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34883840)))]; + fp16 obj_21_epsilon_0_to_fp16 = const()[name = string("obj_21_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor obj_21_cast_fp16 = batch_norm(beta = obj_21_beta_0_to_fp16, epsilon = obj_21_epsilon_0_to_fp16, gamma = obj_21_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_21_cast_fp16)[name = string("obj_21_cast_fp16")]; + string query_pad_type_0 = const()[name = string("query_pad_type_0"), val = string("valid")]; + tensor query_strides_0 = const()[name = string("query_strides_0"), val = tensor([1, 1])]; + tensor query_pad_0 = const()[name = string("query_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor query_dilations_0 = const()[name = string("query_dilations_0"), val = tensor([1, 1])]; + int32 query_groups_0 = const()[name = string("query_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34884928)))]; + tensor layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35409280)))]; + tensor query_cast_fp16 = conv(bias = layers_5_self_attn_q_proj_bias_to_fp16, dilations = query_dilations_0, groups = query_groups_0, pad = query_pad_0, pad_type = query_pad_type_0, strides = query_strides_0, weight = layers_5_self_attn_q_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("query_cast_fp16")]; + string key_pad_type_0 = const()[name = string("key_pad_type_0"), val = string("valid")]; + tensor key_strides_0 = const()[name = string("key_strides_0"), val = tensor([1, 1])]; + tensor key_pad_0 = const()[name = string("key_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor key_dilations_0 = const()[name = string("key_dilations_0"), val = tensor([1, 1])]; + int32 key_groups_0 = const()[name = string("key_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35410368)))]; + tensor key_cast_fp16 = conv(dilations = key_dilations_0, groups = key_groups_0, pad = key_pad_0, pad_type = key_pad_type_0, strides = key_strides_0, weight = layers_5_self_attn_k_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("key_cast_fp16")]; + string value_pad_type_0 = const()[name = string("value_pad_type_0"), val = string("valid")]; + tensor value_strides_0 = const()[name = string("value_strides_0"), val = tensor([1, 1])]; + tensor value_pad_0 = const()[name = string("value_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor value_dilations_0 = const()[name = string("value_dilations_0"), val = tensor([1, 1])]; + int32 value_groups_0 = const()[name = string("value_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35934720)))]; + tensor layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36459072)))]; + tensor value_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_bias_to_fp16, dilations = value_dilations_0, groups = value_groups_0, pad = value_pad_0, pad_type = value_pad_type_0, strides = value_strides_0, weight = layers_5_self_attn_v_proj_weight_to_fp16, x = obj_21_cast_fp16)[name = string("value_cast_fp16")]; + tensor var_3577_begin_0 = const()[name = string("op_3577_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3577_end_0 = const()[name = string("op_3577_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3577_end_mask_0 = const()[name = string("op_3577_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3577_cast_fp16 = slice_by_index(begin = var_3577_begin_0, end = var_3577_end_0, end_mask = var_3577_end_mask_0, x = query_cast_fp16)[name = string("op_3577_cast_fp16")]; + tensor var_3581_begin_0 = const()[name = string("op_3581_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3581_end_0 = const()[name = string("op_3581_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_3581_end_mask_0 = const()[name = string("op_3581_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3581_cast_fp16 = slice_by_index(begin = var_3581_begin_0, end = var_3581_end_0, end_mask = var_3581_end_mask_0, x = query_cast_fp16)[name = string("op_3581_cast_fp16")]; + tensor var_3585_begin_0 = const()[name = string("op_3585_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3585_end_0 = const()[name = string("op_3585_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_3585_end_mask_0 = const()[name = string("op_3585_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3585_cast_fp16 = slice_by_index(begin = var_3585_begin_0, end = var_3585_end_0, end_mask = var_3585_end_mask_0, x = query_cast_fp16)[name = string("op_3585_cast_fp16")]; + tensor var_3589_begin_0 = const()[name = string("op_3589_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3589_end_0 = const()[name = string("op_3589_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_3589_end_mask_0 = const()[name = string("op_3589_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3589_cast_fp16 = slice_by_index(begin = var_3589_begin_0, end = var_3589_end_0, end_mask = var_3589_end_mask_0, x = query_cast_fp16)[name = string("op_3589_cast_fp16")]; + tensor var_3593_begin_0 = const()[name = string("op_3593_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3593_end_0 = const()[name = string("op_3593_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_3593_end_mask_0 = const()[name = string("op_3593_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3593_cast_fp16 = slice_by_index(begin = var_3593_begin_0, end = var_3593_end_0, end_mask = var_3593_end_mask_0, x = query_cast_fp16)[name = string("op_3593_cast_fp16")]; + tensor var_3597_begin_0 = const()[name = string("op_3597_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3597_end_0 = const()[name = string("op_3597_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_3597_end_mask_0 = const()[name = string("op_3597_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3597_cast_fp16 = slice_by_index(begin = var_3597_begin_0, end = var_3597_end_0, end_mask = var_3597_end_mask_0, x = query_cast_fp16)[name = string("op_3597_cast_fp16")]; + tensor var_3601_begin_0 = const()[name = string("op_3601_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3601_end_0 = const()[name = string("op_3601_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_3601_end_mask_0 = const()[name = string("op_3601_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3601_cast_fp16 = slice_by_index(begin = var_3601_begin_0, end = var_3601_end_0, end_mask = var_3601_end_mask_0, x = query_cast_fp16)[name = string("op_3601_cast_fp16")]; + tensor var_3605_begin_0 = const()[name = string("op_3605_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3605_end_0 = const()[name = string("op_3605_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_3605_end_mask_0 = const()[name = string("op_3605_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3605_cast_fp16 = slice_by_index(begin = var_3605_begin_0, end = var_3605_end_0, end_mask = var_3605_end_mask_0, x = query_cast_fp16)[name = string("op_3605_cast_fp16")]; + tensor var_3614_begin_0 = const()[name = string("op_3614_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3614_end_0 = const()[name = string("op_3614_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3614_end_mask_0 = const()[name = string("op_3614_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3614_cast_fp16 = slice_by_index(begin = var_3614_begin_0, end = var_3614_end_0, end_mask = var_3614_end_mask_0, x = var_3577_cast_fp16)[name = string("op_3614_cast_fp16")]; + tensor var_3621_begin_0 = const()[name = string("op_3621_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3621_end_0 = const()[name = string("op_3621_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3621_end_mask_0 = const()[name = string("op_3621_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3621_cast_fp16 = slice_by_index(begin = var_3621_begin_0, end = var_3621_end_0, end_mask = var_3621_end_mask_0, x = var_3577_cast_fp16)[name = string("op_3621_cast_fp16")]; + tensor var_3628_begin_0 = const()[name = string("op_3628_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3628_end_0 = const()[name = string("op_3628_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3628_end_mask_0 = const()[name = string("op_3628_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3628_cast_fp16 = slice_by_index(begin = var_3628_begin_0, end = var_3628_end_0, end_mask = var_3628_end_mask_0, x = var_3577_cast_fp16)[name = string("op_3628_cast_fp16")]; + tensor var_3635_begin_0 = const()[name = string("op_3635_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3635_end_0 = const()[name = string("op_3635_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3635_end_mask_0 = const()[name = string("op_3635_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3635_cast_fp16 = slice_by_index(begin = var_3635_begin_0, end = var_3635_end_0, end_mask = var_3635_end_mask_0, x = var_3577_cast_fp16)[name = string("op_3635_cast_fp16")]; + tensor var_3642_begin_0 = const()[name = string("op_3642_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3642_end_0 = const()[name = string("op_3642_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3642_end_mask_0 = const()[name = string("op_3642_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3642_cast_fp16 = slice_by_index(begin = var_3642_begin_0, end = var_3642_end_0, end_mask = var_3642_end_mask_0, x = var_3581_cast_fp16)[name = string("op_3642_cast_fp16")]; + tensor var_3649_begin_0 = const()[name = string("op_3649_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3649_end_0 = const()[name = string("op_3649_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3649_end_mask_0 = const()[name = string("op_3649_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3649_cast_fp16 = slice_by_index(begin = var_3649_begin_0, end = var_3649_end_0, end_mask = var_3649_end_mask_0, x = var_3581_cast_fp16)[name = string("op_3649_cast_fp16")]; + tensor var_3656_begin_0 = const()[name = string("op_3656_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3656_end_0 = const()[name = string("op_3656_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3656_end_mask_0 = const()[name = string("op_3656_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3656_cast_fp16 = slice_by_index(begin = var_3656_begin_0, end = var_3656_end_0, end_mask = var_3656_end_mask_0, x = var_3581_cast_fp16)[name = string("op_3656_cast_fp16")]; + tensor var_3663_begin_0 = const()[name = string("op_3663_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3663_end_0 = const()[name = string("op_3663_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3663_end_mask_0 = const()[name = string("op_3663_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3663_cast_fp16 = slice_by_index(begin = var_3663_begin_0, end = var_3663_end_0, end_mask = var_3663_end_mask_0, x = var_3581_cast_fp16)[name = string("op_3663_cast_fp16")]; + tensor var_3670_begin_0 = const()[name = string("op_3670_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3670_end_0 = const()[name = string("op_3670_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3670_end_mask_0 = const()[name = string("op_3670_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3670_cast_fp16 = slice_by_index(begin = var_3670_begin_0, end = var_3670_end_0, end_mask = var_3670_end_mask_0, x = var_3585_cast_fp16)[name = string("op_3670_cast_fp16")]; + tensor var_3677_begin_0 = const()[name = string("op_3677_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3677_end_0 = const()[name = string("op_3677_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3677_end_mask_0 = const()[name = string("op_3677_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3677_cast_fp16 = slice_by_index(begin = var_3677_begin_0, end = var_3677_end_0, end_mask = var_3677_end_mask_0, x = var_3585_cast_fp16)[name = string("op_3677_cast_fp16")]; + tensor var_3684_begin_0 = const()[name = string("op_3684_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3684_end_0 = const()[name = string("op_3684_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3684_end_mask_0 = const()[name = string("op_3684_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3684_cast_fp16 = slice_by_index(begin = var_3684_begin_0, end = var_3684_end_0, end_mask = var_3684_end_mask_0, x = var_3585_cast_fp16)[name = string("op_3684_cast_fp16")]; + tensor var_3691_begin_0 = const()[name = string("op_3691_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3691_end_0 = const()[name = string("op_3691_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3691_end_mask_0 = const()[name = string("op_3691_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3691_cast_fp16 = slice_by_index(begin = var_3691_begin_0, end = var_3691_end_0, end_mask = var_3691_end_mask_0, x = var_3585_cast_fp16)[name = string("op_3691_cast_fp16")]; + tensor var_3698_begin_0 = const()[name = string("op_3698_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3698_end_0 = const()[name = string("op_3698_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3698_end_mask_0 = const()[name = string("op_3698_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3698_cast_fp16 = slice_by_index(begin = var_3698_begin_0, end = var_3698_end_0, end_mask = var_3698_end_mask_0, x = var_3589_cast_fp16)[name = string("op_3698_cast_fp16")]; + tensor var_3705_begin_0 = const()[name = string("op_3705_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3705_end_0 = const()[name = string("op_3705_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3705_end_mask_0 = const()[name = string("op_3705_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3705_cast_fp16 = slice_by_index(begin = var_3705_begin_0, end = var_3705_end_0, end_mask = var_3705_end_mask_0, x = var_3589_cast_fp16)[name = string("op_3705_cast_fp16")]; + tensor var_3712_begin_0 = const()[name = string("op_3712_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3712_end_0 = const()[name = string("op_3712_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3712_end_mask_0 = const()[name = string("op_3712_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3712_cast_fp16 = slice_by_index(begin = var_3712_begin_0, end = var_3712_end_0, end_mask = var_3712_end_mask_0, x = var_3589_cast_fp16)[name = string("op_3712_cast_fp16")]; + tensor var_3719_begin_0 = const()[name = string("op_3719_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3719_end_0 = const()[name = string("op_3719_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3719_end_mask_0 = const()[name = string("op_3719_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3719_cast_fp16 = slice_by_index(begin = var_3719_begin_0, end = var_3719_end_0, end_mask = var_3719_end_mask_0, x = var_3589_cast_fp16)[name = string("op_3719_cast_fp16")]; + tensor var_3726_begin_0 = const()[name = string("op_3726_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3726_end_0 = const()[name = string("op_3726_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3726_end_mask_0 = const()[name = string("op_3726_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3726_cast_fp16 = slice_by_index(begin = var_3726_begin_0, end = var_3726_end_0, end_mask = var_3726_end_mask_0, x = var_3593_cast_fp16)[name = string("op_3726_cast_fp16")]; + tensor var_3733_begin_0 = const()[name = string("op_3733_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3733_end_0 = const()[name = string("op_3733_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3733_end_mask_0 = const()[name = string("op_3733_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3733_cast_fp16 = slice_by_index(begin = var_3733_begin_0, end = var_3733_end_0, end_mask = var_3733_end_mask_0, x = var_3593_cast_fp16)[name = string("op_3733_cast_fp16")]; + tensor var_3740_begin_0 = const()[name = string("op_3740_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3740_end_0 = const()[name = string("op_3740_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3740_end_mask_0 = const()[name = string("op_3740_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3740_cast_fp16 = slice_by_index(begin = var_3740_begin_0, end = var_3740_end_0, end_mask = var_3740_end_mask_0, x = var_3593_cast_fp16)[name = string("op_3740_cast_fp16")]; + tensor var_3747_begin_0 = const()[name = string("op_3747_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3747_end_0 = const()[name = string("op_3747_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3747_end_mask_0 = const()[name = string("op_3747_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3747_cast_fp16 = slice_by_index(begin = var_3747_begin_0, end = var_3747_end_0, end_mask = var_3747_end_mask_0, x = var_3593_cast_fp16)[name = string("op_3747_cast_fp16")]; + tensor var_3754_begin_0 = const()[name = string("op_3754_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3754_end_0 = const()[name = string("op_3754_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3754_end_mask_0 = const()[name = string("op_3754_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3754_cast_fp16 = slice_by_index(begin = var_3754_begin_0, end = var_3754_end_0, end_mask = var_3754_end_mask_0, x = var_3597_cast_fp16)[name = string("op_3754_cast_fp16")]; + tensor var_3761_begin_0 = const()[name = string("op_3761_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3761_end_0 = const()[name = string("op_3761_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3761_end_mask_0 = const()[name = string("op_3761_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3761_cast_fp16 = slice_by_index(begin = var_3761_begin_0, end = var_3761_end_0, end_mask = var_3761_end_mask_0, x = var_3597_cast_fp16)[name = string("op_3761_cast_fp16")]; + tensor var_3768_begin_0 = const()[name = string("op_3768_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3768_end_0 = const()[name = string("op_3768_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3768_end_mask_0 = const()[name = string("op_3768_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3768_cast_fp16 = slice_by_index(begin = var_3768_begin_0, end = var_3768_end_0, end_mask = var_3768_end_mask_0, x = var_3597_cast_fp16)[name = string("op_3768_cast_fp16")]; + tensor var_3775_begin_0 = const()[name = string("op_3775_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3775_end_0 = const()[name = string("op_3775_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3775_end_mask_0 = const()[name = string("op_3775_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3775_cast_fp16 = slice_by_index(begin = var_3775_begin_0, end = var_3775_end_0, end_mask = var_3775_end_mask_0, x = var_3597_cast_fp16)[name = string("op_3775_cast_fp16")]; + tensor var_3782_begin_0 = const()[name = string("op_3782_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3782_end_0 = const()[name = string("op_3782_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3782_end_mask_0 = const()[name = string("op_3782_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3782_cast_fp16 = slice_by_index(begin = var_3782_begin_0, end = var_3782_end_0, end_mask = var_3782_end_mask_0, x = var_3601_cast_fp16)[name = string("op_3782_cast_fp16")]; + tensor var_3789_begin_0 = const()[name = string("op_3789_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3789_end_0 = const()[name = string("op_3789_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3789_end_mask_0 = const()[name = string("op_3789_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3789_cast_fp16 = slice_by_index(begin = var_3789_begin_0, end = var_3789_end_0, end_mask = var_3789_end_mask_0, x = var_3601_cast_fp16)[name = string("op_3789_cast_fp16")]; + tensor var_3796_begin_0 = const()[name = string("op_3796_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3796_end_0 = const()[name = string("op_3796_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3796_end_mask_0 = const()[name = string("op_3796_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3796_cast_fp16 = slice_by_index(begin = var_3796_begin_0, end = var_3796_end_0, end_mask = var_3796_end_mask_0, x = var_3601_cast_fp16)[name = string("op_3796_cast_fp16")]; + tensor var_3803_begin_0 = const()[name = string("op_3803_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3803_end_0 = const()[name = string("op_3803_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3803_end_mask_0 = const()[name = string("op_3803_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3803_cast_fp16 = slice_by_index(begin = var_3803_begin_0, end = var_3803_end_0, end_mask = var_3803_end_mask_0, x = var_3601_cast_fp16)[name = string("op_3803_cast_fp16")]; + tensor var_3810_begin_0 = const()[name = string("op_3810_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3810_end_0 = const()[name = string("op_3810_end_0"), val = tensor([1, 64, 1, 375])]; + tensor var_3810_end_mask_0 = const()[name = string("op_3810_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3810_cast_fp16 = slice_by_index(begin = var_3810_begin_0, end = var_3810_end_0, end_mask = var_3810_end_mask_0, x = var_3605_cast_fp16)[name = string("op_3810_cast_fp16")]; + tensor var_3817_begin_0 = const()[name = string("op_3817_begin_0"), val = tensor([0, 0, 0, 375])]; + tensor var_3817_end_0 = const()[name = string("op_3817_end_0"), val = tensor([1, 64, 1, 750])]; + tensor var_3817_end_mask_0 = const()[name = string("op_3817_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3817_cast_fp16 = slice_by_index(begin = var_3817_begin_0, end = var_3817_end_0, end_mask = var_3817_end_mask_0, x = var_3605_cast_fp16)[name = string("op_3817_cast_fp16")]; + tensor var_3824_begin_0 = const()[name = string("op_3824_begin_0"), val = tensor([0, 0, 0, 750])]; + tensor var_3824_end_0 = const()[name = string("op_3824_end_0"), val = tensor([1, 64, 1, 1125])]; + tensor var_3824_end_mask_0 = const()[name = string("op_3824_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3824_cast_fp16 = slice_by_index(begin = var_3824_begin_0, end = var_3824_end_0, end_mask = var_3824_end_mask_0, x = var_3605_cast_fp16)[name = string("op_3824_cast_fp16")]; + tensor var_3831_begin_0 = const()[name = string("op_3831_begin_0"), val = tensor([0, 0, 0, 1125])]; + tensor var_3831_end_0 = const()[name = string("op_3831_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3831_end_mask_0 = const()[name = string("op_3831_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3831_cast_fp16 = slice_by_index(begin = var_3831_begin_0, end = var_3831_end_0, end_mask = var_3831_end_mask_0, x = var_3605_cast_fp16)[name = string("op_3831_cast_fp16")]; + tensor k_11_perm_0 = const()[name = string("k_11_perm_0"), val = tensor([0, 3, 2, 1])]; + tensor var_3836_begin_0 = const()[name = string("op_3836_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3836_end_0 = const()[name = string("op_3836_end_0"), val = tensor([1, 1500, 1, 64])]; + tensor var_3836_end_mask_0 = const()[name = string("op_3836_end_mask_0"), val = tensor([true, true, true, false])]; + tensor k_11_cast_fp16 = transpose(perm = k_11_perm_0, x = key_cast_fp16)[name = string("transpose_0")]; + tensor var_3836_cast_fp16 = slice_by_index(begin = var_3836_begin_0, end = var_3836_end_0, end_mask = var_3836_end_mask_0, x = k_11_cast_fp16)[name = string("op_3836_cast_fp16")]; + tensor var_3840_begin_0 = const()[name = string("op_3840_begin_0"), val = tensor([0, 0, 0, 64])]; + tensor var_3840_end_0 = const()[name = string("op_3840_end_0"), val = tensor([1, 1500, 1, 128])]; + tensor var_3840_end_mask_0 = const()[name = string("op_3840_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3840_cast_fp16 = slice_by_index(begin = var_3840_begin_0, end = var_3840_end_0, end_mask = var_3840_end_mask_0, x = k_11_cast_fp16)[name = string("op_3840_cast_fp16")]; + tensor var_3844_begin_0 = const()[name = string("op_3844_begin_0"), val = tensor([0, 0, 0, 128])]; + tensor var_3844_end_0 = const()[name = string("op_3844_end_0"), val = tensor([1, 1500, 1, 192])]; + tensor var_3844_end_mask_0 = const()[name = string("op_3844_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3844_cast_fp16 = slice_by_index(begin = var_3844_begin_0, end = var_3844_end_0, end_mask = var_3844_end_mask_0, x = k_11_cast_fp16)[name = string("op_3844_cast_fp16")]; + tensor var_3848_begin_0 = const()[name = string("op_3848_begin_0"), val = tensor([0, 0, 0, 192])]; + tensor var_3848_end_0 = const()[name = string("op_3848_end_0"), val = tensor([1, 1500, 1, 256])]; + tensor var_3848_end_mask_0 = const()[name = string("op_3848_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3848_cast_fp16 = slice_by_index(begin = var_3848_begin_0, end = var_3848_end_0, end_mask = var_3848_end_mask_0, x = k_11_cast_fp16)[name = string("op_3848_cast_fp16")]; + tensor var_3852_begin_0 = const()[name = string("op_3852_begin_0"), val = tensor([0, 0, 0, 256])]; + tensor var_3852_end_0 = const()[name = string("op_3852_end_0"), val = tensor([1, 1500, 1, 320])]; + tensor var_3852_end_mask_0 = const()[name = string("op_3852_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3852_cast_fp16 = slice_by_index(begin = var_3852_begin_0, end = var_3852_end_0, end_mask = var_3852_end_mask_0, x = k_11_cast_fp16)[name = string("op_3852_cast_fp16")]; + tensor var_3856_begin_0 = const()[name = string("op_3856_begin_0"), val = tensor([0, 0, 0, 320])]; + tensor var_3856_end_0 = const()[name = string("op_3856_end_0"), val = tensor([1, 1500, 1, 384])]; + tensor var_3856_end_mask_0 = const()[name = string("op_3856_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3856_cast_fp16 = slice_by_index(begin = var_3856_begin_0, end = var_3856_end_0, end_mask = var_3856_end_mask_0, x = k_11_cast_fp16)[name = string("op_3856_cast_fp16")]; + tensor var_3860_begin_0 = const()[name = string("op_3860_begin_0"), val = tensor([0, 0, 0, 384])]; + tensor var_3860_end_0 = const()[name = string("op_3860_end_0"), val = tensor([1, 1500, 1, 448])]; + tensor var_3860_end_mask_0 = const()[name = string("op_3860_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3860_cast_fp16 = slice_by_index(begin = var_3860_begin_0, end = var_3860_end_0, end_mask = var_3860_end_mask_0, x = k_11_cast_fp16)[name = string("op_3860_cast_fp16")]; + tensor var_3864_begin_0 = const()[name = string("op_3864_begin_0"), val = tensor([0, 0, 0, 448])]; + tensor var_3864_end_0 = const()[name = string("op_3864_end_0"), val = tensor([1, 1500, 1, 512])]; + tensor var_3864_end_mask_0 = const()[name = string("op_3864_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3864_cast_fp16 = slice_by_index(begin = var_3864_begin_0, end = var_3864_end_0, end_mask = var_3864_end_mask_0, x = k_11_cast_fp16)[name = string("op_3864_cast_fp16")]; + tensor var_3866_begin_0 = const()[name = string("op_3866_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3866_end_0 = const()[name = string("op_3866_end_0"), val = tensor([1, 64, 1, 1500])]; + tensor var_3866_end_mask_0 = const()[name = string("op_3866_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3866_cast_fp16 = slice_by_index(begin = var_3866_begin_0, end = var_3866_end_0, end_mask = var_3866_end_mask_0, x = value_cast_fp16)[name = string("op_3866_cast_fp16")]; + tensor var_3870_begin_0 = const()[name = string("op_3870_begin_0"), val = tensor([0, 64, 0, 0])]; + tensor var_3870_end_0 = const()[name = string("op_3870_end_0"), val = tensor([1, 128, 1, 1500])]; + tensor var_3870_end_mask_0 = const()[name = string("op_3870_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3870_cast_fp16 = slice_by_index(begin = var_3870_begin_0, end = var_3870_end_0, end_mask = var_3870_end_mask_0, x = value_cast_fp16)[name = string("op_3870_cast_fp16")]; + tensor var_3874_begin_0 = const()[name = string("op_3874_begin_0"), val = tensor([0, 128, 0, 0])]; + tensor var_3874_end_0 = const()[name = string("op_3874_end_0"), val = tensor([1, 192, 1, 1500])]; + tensor var_3874_end_mask_0 = const()[name = string("op_3874_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3874_cast_fp16 = slice_by_index(begin = var_3874_begin_0, end = var_3874_end_0, end_mask = var_3874_end_mask_0, x = value_cast_fp16)[name = string("op_3874_cast_fp16")]; + tensor var_3878_begin_0 = const()[name = string("op_3878_begin_0"), val = tensor([0, 192, 0, 0])]; + tensor var_3878_end_0 = const()[name = string("op_3878_end_0"), val = tensor([1, 256, 1, 1500])]; + tensor var_3878_end_mask_0 = const()[name = string("op_3878_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3878_cast_fp16 = slice_by_index(begin = var_3878_begin_0, end = var_3878_end_0, end_mask = var_3878_end_mask_0, x = value_cast_fp16)[name = string("op_3878_cast_fp16")]; + tensor var_3882_begin_0 = const()[name = string("op_3882_begin_0"), val = tensor([0, 256, 0, 0])]; + tensor var_3882_end_0 = const()[name = string("op_3882_end_0"), val = tensor([1, 320, 1, 1500])]; + tensor var_3882_end_mask_0 = const()[name = string("op_3882_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3882_cast_fp16 = slice_by_index(begin = var_3882_begin_0, end = var_3882_end_0, end_mask = var_3882_end_mask_0, x = value_cast_fp16)[name = string("op_3882_cast_fp16")]; + tensor var_3886_begin_0 = const()[name = string("op_3886_begin_0"), val = tensor([0, 320, 0, 0])]; + tensor var_3886_end_0 = const()[name = string("op_3886_end_0"), val = tensor([1, 384, 1, 1500])]; + tensor var_3886_end_mask_0 = const()[name = string("op_3886_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3886_cast_fp16 = slice_by_index(begin = var_3886_begin_0, end = var_3886_end_0, end_mask = var_3886_end_mask_0, x = value_cast_fp16)[name = string("op_3886_cast_fp16")]; + tensor var_3890_begin_0 = const()[name = string("op_3890_begin_0"), val = tensor([0, 384, 0, 0])]; + tensor var_3890_end_0 = const()[name = string("op_3890_end_0"), val = tensor([1, 448, 1, 1500])]; + tensor var_3890_end_mask_0 = const()[name = string("op_3890_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3890_cast_fp16 = slice_by_index(begin = var_3890_begin_0, end = var_3890_end_0, end_mask = var_3890_end_mask_0, x = value_cast_fp16)[name = string("op_3890_cast_fp16")]; + tensor var_3894_begin_0 = const()[name = string("op_3894_begin_0"), val = tensor([0, 448, 0, 0])]; + tensor var_3894_end_0 = const()[name = string("op_3894_end_0"), val = tensor([1, 512, 1, 1500])]; + tensor var_3894_end_mask_0 = const()[name = string("op_3894_end_mask_0"), val = tensor([true, false, true, true])]; + tensor var_3894_cast_fp16 = slice_by_index(begin = var_3894_begin_0, end = var_3894_end_0, end_mask = var_3894_end_mask_0, x = value_cast_fp16)[name = string("op_3894_cast_fp16")]; + string _SplitHeadsQ__mh_w_321_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_321_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_321_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_321_equation_0, values = (var_3836_cast_fp16, var_3614_cast_fp16))[name = string("_SplitHeadsQ__mh_w_321_cast_fp16")]; + string _SplitHeadsQ__mh_w_323_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_323_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_323_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_323_equation_0, values = (var_3836_cast_fp16, var_3621_cast_fp16))[name = string("_SplitHeadsQ__mh_w_323_cast_fp16")]; + string _SplitHeadsQ__mh_w_325_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_325_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_325_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_325_equation_0, values = (var_3836_cast_fp16, var_3628_cast_fp16))[name = string("_SplitHeadsQ__mh_w_325_cast_fp16")]; + string _SplitHeadsQ__mh_w_327_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_327_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_327_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_327_equation_0, values = (var_3836_cast_fp16, var_3635_cast_fp16))[name = string("_SplitHeadsQ__mh_w_327_cast_fp16")]; + string _SplitHeadsQ__mh_w_329_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_329_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_329_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_329_equation_0, values = (var_3840_cast_fp16, var_3642_cast_fp16))[name = string("_SplitHeadsQ__mh_w_329_cast_fp16")]; + string _SplitHeadsQ__mh_w_331_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_331_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_331_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_331_equation_0, values = (var_3840_cast_fp16, var_3649_cast_fp16))[name = string("_SplitHeadsQ__mh_w_331_cast_fp16")]; + string _SplitHeadsQ__mh_w_333_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_333_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_333_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_333_equation_0, values = (var_3840_cast_fp16, var_3656_cast_fp16))[name = string("_SplitHeadsQ__mh_w_333_cast_fp16")]; + string _SplitHeadsQ__mh_w_335_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_335_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_335_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_335_equation_0, values = (var_3840_cast_fp16, var_3663_cast_fp16))[name = string("_SplitHeadsQ__mh_w_335_cast_fp16")]; + string _SplitHeadsQ__mh_w_337_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_337_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_337_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_337_equation_0, values = (var_3844_cast_fp16, var_3670_cast_fp16))[name = string("_SplitHeadsQ__mh_w_337_cast_fp16")]; + string _SplitHeadsQ__mh_w_339_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_339_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_339_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_339_equation_0, values = (var_3844_cast_fp16, var_3677_cast_fp16))[name = string("_SplitHeadsQ__mh_w_339_cast_fp16")]; + string _SplitHeadsQ__mh_w_341_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_341_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_341_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_341_equation_0, values = (var_3844_cast_fp16, var_3684_cast_fp16))[name = string("_SplitHeadsQ__mh_w_341_cast_fp16")]; + string _SplitHeadsQ__mh_w_343_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_343_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_343_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_343_equation_0, values = (var_3844_cast_fp16, var_3691_cast_fp16))[name = string("_SplitHeadsQ__mh_w_343_cast_fp16")]; + string _SplitHeadsQ__mh_w_345_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_345_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_345_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_345_equation_0, values = (var_3848_cast_fp16, var_3698_cast_fp16))[name = string("_SplitHeadsQ__mh_w_345_cast_fp16")]; + string _SplitHeadsQ__mh_w_347_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_347_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_347_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_347_equation_0, values = (var_3848_cast_fp16, var_3705_cast_fp16))[name = string("_SplitHeadsQ__mh_w_347_cast_fp16")]; + string _SplitHeadsQ__mh_w_349_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_349_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_349_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_349_equation_0, values = (var_3848_cast_fp16, var_3712_cast_fp16))[name = string("_SplitHeadsQ__mh_w_349_cast_fp16")]; + string _SplitHeadsQ__mh_w_351_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_351_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_351_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_351_equation_0, values = (var_3848_cast_fp16, var_3719_cast_fp16))[name = string("_SplitHeadsQ__mh_w_351_cast_fp16")]; + string _SplitHeadsQ__mh_w_353_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_353_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_353_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_353_equation_0, values = (var_3852_cast_fp16, var_3726_cast_fp16))[name = string("_SplitHeadsQ__mh_w_353_cast_fp16")]; + string _SplitHeadsQ__mh_w_355_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_355_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_355_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_355_equation_0, values = (var_3852_cast_fp16, var_3733_cast_fp16))[name = string("_SplitHeadsQ__mh_w_355_cast_fp16")]; + string _SplitHeadsQ__mh_w_357_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_357_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_357_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_357_equation_0, values = (var_3852_cast_fp16, var_3740_cast_fp16))[name = string("_SplitHeadsQ__mh_w_357_cast_fp16")]; + string _SplitHeadsQ__mh_w_359_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_359_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_359_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_359_equation_0, values = (var_3852_cast_fp16, var_3747_cast_fp16))[name = string("_SplitHeadsQ__mh_w_359_cast_fp16")]; + string _SplitHeadsQ__mh_w_361_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_361_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_361_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_361_equation_0, values = (var_3856_cast_fp16, var_3754_cast_fp16))[name = string("_SplitHeadsQ__mh_w_361_cast_fp16")]; + string _SplitHeadsQ__mh_w_363_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_363_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_363_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_363_equation_0, values = (var_3856_cast_fp16, var_3761_cast_fp16))[name = string("_SplitHeadsQ__mh_w_363_cast_fp16")]; + string _SplitHeadsQ__mh_w_365_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_365_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_365_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_365_equation_0, values = (var_3856_cast_fp16, var_3768_cast_fp16))[name = string("_SplitHeadsQ__mh_w_365_cast_fp16")]; + string _SplitHeadsQ__mh_w_367_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_367_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_367_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_367_equation_0, values = (var_3856_cast_fp16, var_3775_cast_fp16))[name = string("_SplitHeadsQ__mh_w_367_cast_fp16")]; + string _SplitHeadsQ__mh_w_369_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_369_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_369_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_369_equation_0, values = (var_3860_cast_fp16, var_3782_cast_fp16))[name = string("_SplitHeadsQ__mh_w_369_cast_fp16")]; + string _SplitHeadsQ__mh_w_371_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_371_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_371_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_371_equation_0, values = (var_3860_cast_fp16, var_3789_cast_fp16))[name = string("_SplitHeadsQ__mh_w_371_cast_fp16")]; + string _SplitHeadsQ__mh_w_373_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_373_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_373_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_373_equation_0, values = (var_3860_cast_fp16, var_3796_cast_fp16))[name = string("_SplitHeadsQ__mh_w_373_cast_fp16")]; + string _SplitHeadsQ__mh_w_375_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_375_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_375_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_375_equation_0, values = (var_3860_cast_fp16, var_3803_cast_fp16))[name = string("_SplitHeadsQ__mh_w_375_cast_fp16")]; + string _SplitHeadsQ__mh_w_377_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_377_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_377_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_377_equation_0, values = (var_3864_cast_fp16, var_3810_cast_fp16))[name = string("_SplitHeadsQ__mh_w_377_cast_fp16")]; + string _SplitHeadsQ__mh_w_379_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_379_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_379_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_379_equation_0, values = (var_3864_cast_fp16, var_3817_cast_fp16))[name = string("_SplitHeadsQ__mh_w_379_cast_fp16")]; + string _SplitHeadsQ__mh_w_381_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_381_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_381_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_381_equation_0, values = (var_3864_cast_fp16, var_3824_cast_fp16))[name = string("_SplitHeadsQ__mh_w_381_cast_fp16")]; + string _SplitHeadsQ__mh_w_equation_0 = const()[name = string("_SplitHeadsQ__mh_w_equation_0"), val = string("bkhc,bchq->bkhq")]; + tensor _SplitHeadsQ__mh_w_cast_fp16 = einsum(equation = _SplitHeadsQ__mh_w_equation_0, values = (var_3864_cast_fp16, var_3831_cast_fp16))[name = string("_SplitHeadsQ__mh_w_cast_fp16")]; + fp16 var_3961_to_fp16 = const()[name = string("op_3961_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_321_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_321_cast_fp16, y = var_3961_to_fp16)[name = string("aw_chunk_321_cast_fp16")]; + fp16 var_3963_to_fp16 = const()[name = string("op_3963_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_323_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_323_cast_fp16, y = var_3963_to_fp16)[name = string("aw_chunk_323_cast_fp16")]; + fp16 var_3965_to_fp16 = const()[name = string("op_3965_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_325_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_325_cast_fp16, y = var_3965_to_fp16)[name = string("aw_chunk_325_cast_fp16")]; + fp16 var_3967_to_fp16 = const()[name = string("op_3967_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_327_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_327_cast_fp16, y = var_3967_to_fp16)[name = string("aw_chunk_327_cast_fp16")]; + fp16 var_3969_to_fp16 = const()[name = string("op_3969_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_329_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_329_cast_fp16, y = var_3969_to_fp16)[name = string("aw_chunk_329_cast_fp16")]; + fp16 var_3971_to_fp16 = const()[name = string("op_3971_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_331_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_331_cast_fp16, y = var_3971_to_fp16)[name = string("aw_chunk_331_cast_fp16")]; + fp16 var_3973_to_fp16 = const()[name = string("op_3973_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_333_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_333_cast_fp16, y = var_3973_to_fp16)[name = string("aw_chunk_333_cast_fp16")]; + fp16 var_3975_to_fp16 = const()[name = string("op_3975_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_335_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_335_cast_fp16, y = var_3975_to_fp16)[name = string("aw_chunk_335_cast_fp16")]; + fp16 var_3977_to_fp16 = const()[name = string("op_3977_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_337_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_337_cast_fp16, y = var_3977_to_fp16)[name = string("aw_chunk_337_cast_fp16")]; + fp16 var_3979_to_fp16 = const()[name = string("op_3979_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_339_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_339_cast_fp16, y = var_3979_to_fp16)[name = string("aw_chunk_339_cast_fp16")]; + fp16 var_3981_to_fp16 = const()[name = string("op_3981_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_341_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_341_cast_fp16, y = var_3981_to_fp16)[name = string("aw_chunk_341_cast_fp16")]; + fp16 var_3983_to_fp16 = const()[name = string("op_3983_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_343_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_343_cast_fp16, y = var_3983_to_fp16)[name = string("aw_chunk_343_cast_fp16")]; + fp16 var_3985_to_fp16 = const()[name = string("op_3985_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_345_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_345_cast_fp16, y = var_3985_to_fp16)[name = string("aw_chunk_345_cast_fp16")]; + fp16 var_3987_to_fp16 = const()[name = string("op_3987_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_347_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_347_cast_fp16, y = var_3987_to_fp16)[name = string("aw_chunk_347_cast_fp16")]; + fp16 var_3989_to_fp16 = const()[name = string("op_3989_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_349_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_349_cast_fp16, y = var_3989_to_fp16)[name = string("aw_chunk_349_cast_fp16")]; + fp16 var_3991_to_fp16 = const()[name = string("op_3991_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_351_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_351_cast_fp16, y = var_3991_to_fp16)[name = string("aw_chunk_351_cast_fp16")]; + fp16 var_3993_to_fp16 = const()[name = string("op_3993_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_353_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_353_cast_fp16, y = var_3993_to_fp16)[name = string("aw_chunk_353_cast_fp16")]; + fp16 var_3995_to_fp16 = const()[name = string("op_3995_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_355_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_355_cast_fp16, y = var_3995_to_fp16)[name = string("aw_chunk_355_cast_fp16")]; + fp16 var_3997_to_fp16 = const()[name = string("op_3997_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_357_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_357_cast_fp16, y = var_3997_to_fp16)[name = string("aw_chunk_357_cast_fp16")]; + fp16 var_3999_to_fp16 = const()[name = string("op_3999_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_359_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_359_cast_fp16, y = var_3999_to_fp16)[name = string("aw_chunk_359_cast_fp16")]; + fp16 var_4001_to_fp16 = const()[name = string("op_4001_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_361_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_361_cast_fp16, y = var_4001_to_fp16)[name = string("aw_chunk_361_cast_fp16")]; + fp16 var_4003_to_fp16 = const()[name = string("op_4003_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_363_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_363_cast_fp16, y = var_4003_to_fp16)[name = string("aw_chunk_363_cast_fp16")]; + fp16 var_4005_to_fp16 = const()[name = string("op_4005_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_365_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_365_cast_fp16, y = var_4005_to_fp16)[name = string("aw_chunk_365_cast_fp16")]; + fp16 var_4007_to_fp16 = const()[name = string("op_4007_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_367_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_367_cast_fp16, y = var_4007_to_fp16)[name = string("aw_chunk_367_cast_fp16")]; + fp16 var_4009_to_fp16 = const()[name = string("op_4009_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_369_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_369_cast_fp16, y = var_4009_to_fp16)[name = string("aw_chunk_369_cast_fp16")]; + fp16 var_4011_to_fp16 = const()[name = string("op_4011_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_371_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_371_cast_fp16, y = var_4011_to_fp16)[name = string("aw_chunk_371_cast_fp16")]; + fp16 var_4013_to_fp16 = const()[name = string("op_4013_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_373_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_373_cast_fp16, y = var_4013_to_fp16)[name = string("aw_chunk_373_cast_fp16")]; + fp16 var_4015_to_fp16 = const()[name = string("op_4015_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_375_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_375_cast_fp16, y = var_4015_to_fp16)[name = string("aw_chunk_375_cast_fp16")]; + fp16 var_4017_to_fp16 = const()[name = string("op_4017_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_377_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_377_cast_fp16, y = var_4017_to_fp16)[name = string("aw_chunk_377_cast_fp16")]; + fp16 var_4019_to_fp16 = const()[name = string("op_4019_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_379_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_379_cast_fp16, y = var_4019_to_fp16)[name = string("aw_chunk_379_cast_fp16")]; + fp16 var_4021_to_fp16 = const()[name = string("op_4021_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_381_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_381_cast_fp16, y = var_4021_to_fp16)[name = string("aw_chunk_381_cast_fp16")]; + fp16 var_4023_to_fp16 = const()[name = string("op_4023_to_fp16"), val = fp16(0x1p-3)]; + tensor aw_chunk_cast_fp16 = mul(x = _SplitHeadsQ__mh_w_cast_fp16, y = var_4023_to_fp16)[name = string("aw_chunk_cast_fp16")]; + tensor var_4025_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_321_cast_fp16)[name = string("op_4025_cast_fp16")]; + tensor var_4026_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_323_cast_fp16)[name = string("op_4026_cast_fp16")]; + tensor var_4027_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_325_cast_fp16)[name = string("op_4027_cast_fp16")]; + tensor var_4028_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_327_cast_fp16)[name = string("op_4028_cast_fp16")]; + tensor var_4029_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_329_cast_fp16)[name = string("op_4029_cast_fp16")]; + tensor var_4030_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_331_cast_fp16)[name = string("op_4030_cast_fp16")]; + tensor var_4031_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_333_cast_fp16)[name = string("op_4031_cast_fp16")]; + tensor var_4032_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_335_cast_fp16)[name = string("op_4032_cast_fp16")]; + tensor var_4033_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_337_cast_fp16)[name = string("op_4033_cast_fp16")]; + tensor var_4034_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_339_cast_fp16)[name = string("op_4034_cast_fp16")]; + tensor var_4035_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_341_cast_fp16)[name = string("op_4035_cast_fp16")]; + tensor var_4036_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_343_cast_fp16)[name = string("op_4036_cast_fp16")]; + tensor var_4037_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_345_cast_fp16)[name = string("op_4037_cast_fp16")]; + tensor var_4038_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_347_cast_fp16)[name = string("op_4038_cast_fp16")]; + tensor var_4039_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_349_cast_fp16)[name = string("op_4039_cast_fp16")]; + tensor var_4040_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_351_cast_fp16)[name = string("op_4040_cast_fp16")]; + tensor var_4041_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_353_cast_fp16)[name = string("op_4041_cast_fp16")]; + tensor var_4042_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_355_cast_fp16)[name = string("op_4042_cast_fp16")]; + tensor var_4043_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_357_cast_fp16)[name = string("op_4043_cast_fp16")]; + tensor var_4044_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_359_cast_fp16)[name = string("op_4044_cast_fp16")]; + tensor var_4045_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_361_cast_fp16)[name = string("op_4045_cast_fp16")]; + tensor var_4046_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_363_cast_fp16)[name = string("op_4046_cast_fp16")]; + tensor var_4047_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_365_cast_fp16)[name = string("op_4047_cast_fp16")]; + tensor var_4048_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_367_cast_fp16)[name = string("op_4048_cast_fp16")]; + tensor var_4049_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_369_cast_fp16)[name = string("op_4049_cast_fp16")]; + tensor var_4050_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_371_cast_fp16)[name = string("op_4050_cast_fp16")]; + tensor var_4051_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_373_cast_fp16)[name = string("op_4051_cast_fp16")]; + tensor var_4052_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_375_cast_fp16)[name = string("op_4052_cast_fp16")]; + tensor var_4053_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_377_cast_fp16)[name = string("op_4053_cast_fp16")]; + tensor var_4054_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_379_cast_fp16)[name = string("op_4054_cast_fp16")]; + tensor var_4055_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_381_cast_fp16)[name = string("op_4055_cast_fp16")]; + tensor var_4056_cast_fp16 = softmax(axis = var_3522, x = aw_chunk_cast_fp16)[name = string("op_4056_cast_fp16")]; + string var_4058_equation_0 = const()[name = string("op_4058_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4058_cast_fp16 = einsum(equation = var_4058_equation_0, values = (var_3866_cast_fp16, var_4025_cast_fp16))[name = string("op_4058_cast_fp16")]; + string var_4060_equation_0 = const()[name = string("op_4060_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4060_cast_fp16 = einsum(equation = var_4060_equation_0, values = (var_3866_cast_fp16, var_4026_cast_fp16))[name = string("op_4060_cast_fp16")]; + string var_4062_equation_0 = const()[name = string("op_4062_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4062_cast_fp16 = einsum(equation = var_4062_equation_0, values = (var_3866_cast_fp16, var_4027_cast_fp16))[name = string("op_4062_cast_fp16")]; + string var_4064_equation_0 = const()[name = string("op_4064_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4064_cast_fp16 = einsum(equation = var_4064_equation_0, values = (var_3866_cast_fp16, var_4028_cast_fp16))[name = string("op_4064_cast_fp16")]; + string var_4066_equation_0 = const()[name = string("op_4066_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4066_cast_fp16 = einsum(equation = var_4066_equation_0, values = (var_3870_cast_fp16, var_4029_cast_fp16))[name = string("op_4066_cast_fp16")]; + string var_4068_equation_0 = const()[name = string("op_4068_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4068_cast_fp16 = einsum(equation = var_4068_equation_0, values = (var_3870_cast_fp16, var_4030_cast_fp16))[name = string("op_4068_cast_fp16")]; + string var_4070_equation_0 = const()[name = string("op_4070_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4070_cast_fp16 = einsum(equation = var_4070_equation_0, values = (var_3870_cast_fp16, var_4031_cast_fp16))[name = string("op_4070_cast_fp16")]; + string var_4072_equation_0 = const()[name = string("op_4072_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4072_cast_fp16 = einsum(equation = var_4072_equation_0, values = (var_3870_cast_fp16, var_4032_cast_fp16))[name = string("op_4072_cast_fp16")]; + string var_4074_equation_0 = const()[name = string("op_4074_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4074_cast_fp16 = einsum(equation = var_4074_equation_0, values = (var_3874_cast_fp16, var_4033_cast_fp16))[name = string("op_4074_cast_fp16")]; + string var_4076_equation_0 = const()[name = string("op_4076_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4076_cast_fp16 = einsum(equation = var_4076_equation_0, values = (var_3874_cast_fp16, var_4034_cast_fp16))[name = string("op_4076_cast_fp16")]; + string var_4078_equation_0 = const()[name = string("op_4078_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4078_cast_fp16 = einsum(equation = var_4078_equation_0, values = (var_3874_cast_fp16, var_4035_cast_fp16))[name = string("op_4078_cast_fp16")]; + string var_4080_equation_0 = const()[name = string("op_4080_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4080_cast_fp16 = einsum(equation = var_4080_equation_0, values = (var_3874_cast_fp16, var_4036_cast_fp16))[name = string("op_4080_cast_fp16")]; + string var_4082_equation_0 = const()[name = string("op_4082_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4082_cast_fp16 = einsum(equation = var_4082_equation_0, values = (var_3878_cast_fp16, var_4037_cast_fp16))[name = string("op_4082_cast_fp16")]; + string var_4084_equation_0 = const()[name = string("op_4084_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4084_cast_fp16 = einsum(equation = var_4084_equation_0, values = (var_3878_cast_fp16, var_4038_cast_fp16))[name = string("op_4084_cast_fp16")]; + string var_4086_equation_0 = const()[name = string("op_4086_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4086_cast_fp16 = einsum(equation = var_4086_equation_0, values = (var_3878_cast_fp16, var_4039_cast_fp16))[name = string("op_4086_cast_fp16")]; + string var_4088_equation_0 = const()[name = string("op_4088_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4088_cast_fp16 = einsum(equation = var_4088_equation_0, values = (var_3878_cast_fp16, var_4040_cast_fp16))[name = string("op_4088_cast_fp16")]; + string var_4090_equation_0 = const()[name = string("op_4090_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4090_cast_fp16 = einsum(equation = var_4090_equation_0, values = (var_3882_cast_fp16, var_4041_cast_fp16))[name = string("op_4090_cast_fp16")]; + string var_4092_equation_0 = const()[name = string("op_4092_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4092_cast_fp16 = einsum(equation = var_4092_equation_0, values = (var_3882_cast_fp16, var_4042_cast_fp16))[name = string("op_4092_cast_fp16")]; + string var_4094_equation_0 = const()[name = string("op_4094_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4094_cast_fp16 = einsum(equation = var_4094_equation_0, values = (var_3882_cast_fp16, var_4043_cast_fp16))[name = string("op_4094_cast_fp16")]; + string var_4096_equation_0 = const()[name = string("op_4096_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4096_cast_fp16 = einsum(equation = var_4096_equation_0, values = (var_3882_cast_fp16, var_4044_cast_fp16))[name = string("op_4096_cast_fp16")]; + string var_4098_equation_0 = const()[name = string("op_4098_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4098_cast_fp16 = einsum(equation = var_4098_equation_0, values = (var_3886_cast_fp16, var_4045_cast_fp16))[name = string("op_4098_cast_fp16")]; + string var_4100_equation_0 = const()[name = string("op_4100_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4100_cast_fp16 = einsum(equation = var_4100_equation_0, values = (var_3886_cast_fp16, var_4046_cast_fp16))[name = string("op_4100_cast_fp16")]; + string var_4102_equation_0 = const()[name = string("op_4102_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4102_cast_fp16 = einsum(equation = var_4102_equation_0, values = (var_3886_cast_fp16, var_4047_cast_fp16))[name = string("op_4102_cast_fp16")]; + string var_4104_equation_0 = const()[name = string("op_4104_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4104_cast_fp16 = einsum(equation = var_4104_equation_0, values = (var_3886_cast_fp16, var_4048_cast_fp16))[name = string("op_4104_cast_fp16")]; + string var_4106_equation_0 = const()[name = string("op_4106_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4106_cast_fp16 = einsum(equation = var_4106_equation_0, values = (var_3890_cast_fp16, var_4049_cast_fp16))[name = string("op_4106_cast_fp16")]; + string var_4108_equation_0 = const()[name = string("op_4108_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4108_cast_fp16 = einsum(equation = var_4108_equation_0, values = (var_3890_cast_fp16, var_4050_cast_fp16))[name = string("op_4108_cast_fp16")]; + string var_4110_equation_0 = const()[name = string("op_4110_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4110_cast_fp16 = einsum(equation = var_4110_equation_0, values = (var_3890_cast_fp16, var_4051_cast_fp16))[name = string("op_4110_cast_fp16")]; + string var_4112_equation_0 = const()[name = string("op_4112_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4112_cast_fp16 = einsum(equation = var_4112_equation_0, values = (var_3890_cast_fp16, var_4052_cast_fp16))[name = string("op_4112_cast_fp16")]; + string var_4114_equation_0 = const()[name = string("op_4114_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4114_cast_fp16 = einsum(equation = var_4114_equation_0, values = (var_3894_cast_fp16, var_4053_cast_fp16))[name = string("op_4114_cast_fp16")]; + string var_4116_equation_0 = const()[name = string("op_4116_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4116_cast_fp16 = einsum(equation = var_4116_equation_0, values = (var_3894_cast_fp16, var_4054_cast_fp16))[name = string("op_4116_cast_fp16")]; + string var_4118_equation_0 = const()[name = string("op_4118_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4118_cast_fp16 = einsum(equation = var_4118_equation_0, values = (var_3894_cast_fp16, var_4055_cast_fp16))[name = string("op_4118_cast_fp16")]; + string var_4120_equation_0 = const()[name = string("op_4120_equation_0"), val = string("bchk,bkhq->bchq")]; + tensor var_4120_cast_fp16 = einsum(equation = var_4120_equation_0, values = (var_3894_cast_fp16, var_4056_cast_fp16))[name = string("op_4120_cast_fp16")]; + bool var_4122_interleave_0 = const()[name = string("op_4122_interleave_0"), val = bool(false)]; + tensor var_4122_cast_fp16 = concat(axis = var_3509, interleave = var_4122_interleave_0, values = (var_4058_cast_fp16, var_4060_cast_fp16, var_4062_cast_fp16, var_4064_cast_fp16))[name = string("op_4122_cast_fp16")]; + bool var_4124_interleave_0 = const()[name = string("op_4124_interleave_0"), val = bool(false)]; + tensor var_4124_cast_fp16 = concat(axis = var_3509, interleave = var_4124_interleave_0, values = (var_4066_cast_fp16, var_4068_cast_fp16, var_4070_cast_fp16, var_4072_cast_fp16))[name = string("op_4124_cast_fp16")]; + bool var_4126_interleave_0 = const()[name = string("op_4126_interleave_0"), val = bool(false)]; + tensor var_4126_cast_fp16 = concat(axis = var_3509, interleave = var_4126_interleave_0, values = (var_4074_cast_fp16, var_4076_cast_fp16, var_4078_cast_fp16, var_4080_cast_fp16))[name = string("op_4126_cast_fp16")]; + bool var_4128_interleave_0 = const()[name = string("op_4128_interleave_0"), val = bool(false)]; + tensor var_4128_cast_fp16 = concat(axis = var_3509, interleave = var_4128_interleave_0, values = (var_4082_cast_fp16, var_4084_cast_fp16, var_4086_cast_fp16, var_4088_cast_fp16))[name = string("op_4128_cast_fp16")]; + bool var_4130_interleave_0 = const()[name = string("op_4130_interleave_0"), val = bool(false)]; + tensor var_4130_cast_fp16 = concat(axis = var_3509, interleave = var_4130_interleave_0, values = (var_4090_cast_fp16, var_4092_cast_fp16, var_4094_cast_fp16, var_4096_cast_fp16))[name = string("op_4130_cast_fp16")]; + bool var_4132_interleave_0 = const()[name = string("op_4132_interleave_0"), val = bool(false)]; + tensor var_4132_cast_fp16 = concat(axis = var_3509, interleave = var_4132_interleave_0, values = (var_4098_cast_fp16, var_4100_cast_fp16, var_4102_cast_fp16, var_4104_cast_fp16))[name = string("op_4132_cast_fp16")]; + bool var_4134_interleave_0 = const()[name = string("op_4134_interleave_0"), val = bool(false)]; + tensor var_4134_cast_fp16 = concat(axis = var_3509, interleave = var_4134_interleave_0, values = (var_4106_cast_fp16, var_4108_cast_fp16, var_4110_cast_fp16, var_4112_cast_fp16))[name = string("op_4134_cast_fp16")]; + bool var_4136_interleave_0 = const()[name = string("op_4136_interleave_0"), val = bool(false)]; + tensor var_4136_cast_fp16 = concat(axis = var_3509, interleave = var_4136_interleave_0, values = (var_4114_cast_fp16, var_4116_cast_fp16, var_4118_cast_fp16, var_4120_cast_fp16))[name = string("op_4136_cast_fp16")]; + bool input_41_interleave_0 = const()[name = string("input_41_interleave_0"), val = bool(false)]; + tensor input_41_cast_fp16 = concat(axis = var_3522, interleave = input_41_interleave_0, values = (var_4122_cast_fp16, var_4124_cast_fp16, var_4126_cast_fp16, var_4128_cast_fp16, var_4130_cast_fp16, var_4132_cast_fp16, var_4134_cast_fp16, var_4136_cast_fp16))[name = string("input_41_cast_fp16")]; + string obj_pad_type_0 = const()[name = string("obj_pad_type_0"), val = string("valid")]; + tensor obj_strides_0 = const()[name = string("obj_strides_0"), val = tensor([1, 1])]; + tensor obj_pad_0 = const()[name = string("obj_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor obj_dilations_0 = const()[name = string("obj_dilations_0"), val = tensor([1, 1])]; + int32 obj_groups_0 = const()[name = string("obj_groups_0"), val = int32(1)]; + tensor layers_5_self_attn_o_proj_weight_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36460160)))]; + tensor layers_5_self_attn_o_proj_bias_to_fp16 = const()[name = string("layers_5_self_attn_o_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36984512)))]; + tensor obj_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_bias_to_fp16, dilations = obj_dilations_0, groups = obj_groups_0, pad = obj_pad_0, pad_type = obj_pad_type_0, strides = obj_strides_0, weight = layers_5_self_attn_o_proj_weight_to_fp16, x = input_41_cast_fp16)[name = string("obj_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = obj_cast_fp16)[name = string("inputs_23_cast_fp16")]; + tensor out_23_axes_0 = const()[name = string("out_23_axes_0"), val = tensor([1])]; + fp16 var_4155_to_fp16 = const()[name = string("op_4155_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_4155_to_fp16, x = inputs_23_cast_fp16)[name = string("out_23_cast_fp16")]; + tensor input_43_gamma_0_to_fp16 = const()[name = string("input_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36985600)))]; + tensor input_43_beta_0_to_fp16 = const()[name = string("input_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36986688)))]; + fp16 input_43_epsilon_0_to_fp16 = const()[name = string("input_43_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor input_43_cast_fp16 = batch_norm(beta = input_43_beta_0_to_fp16, epsilon = input_43_epsilon_0_to_fp16, gamma = input_43_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_23_cast_fp16)[name = string("input_43_cast_fp16")]; + string input_45_pad_type_0 = const()[name = string("input_45_pad_type_0"), val = string("valid")]; + tensor input_45_strides_0 = const()[name = string("input_45_strides_0"), val = tensor([1, 1])]; + tensor input_45_pad_0 = const()[name = string("input_45_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_45_dilations_0 = const()[name = string("input_45_dilations_0"), val = tensor([1, 1])]; + int32 input_45_groups_0 = const()[name = string("input_45_groups_0"), val = int32(1)]; + tensor layers_5_fc1_weight_to_fp16 = const()[name = string("layers_5_fc1_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36987776)))]; + tensor layers_5_fc1_bias_to_fp16 = const()[name = string("layers_5_fc1_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39084992)))]; + tensor input_45_cast_fp16 = conv(bias = layers_5_fc1_bias_to_fp16, dilations = input_45_dilations_0, groups = input_45_groups_0, pad = input_45_pad_0, pad_type = input_45_pad_type_0, strides = input_45_strides_0, weight = layers_5_fc1_weight_to_fp16, x = input_43_cast_fp16)[name = string("input_45_cast_fp16")]; + string input_47_mode_0 = const()[name = string("input_47_mode_0"), val = string("EXACT")]; + tensor input_47_cast_fp16 = gelu(mode = input_47_mode_0, x = input_45_cast_fp16)[name = string("input_47_cast_fp16")]; + string hidden_states_pad_type_0 = const()[name = string("hidden_states_pad_type_0"), val = string("valid")]; + tensor hidden_states_strides_0 = const()[name = string("hidden_states_strides_0"), val = tensor([1, 1])]; + tensor hidden_states_pad_0 = const()[name = string("hidden_states_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor hidden_states_dilations_0 = const()[name = string("hidden_states_dilations_0"), val = tensor([1, 1])]; + int32 hidden_states_groups_0 = const()[name = string("hidden_states_groups_0"), val = int32(1)]; + tensor layers_5_fc2_weight_to_fp16 = const()[name = string("layers_5_fc2_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39089152)))]; + tensor layers_5_fc2_bias_to_fp16 = const()[name = string("layers_5_fc2_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41186368)))]; + tensor hidden_states_cast_fp16 = conv(bias = layers_5_fc2_bias_to_fp16, dilations = hidden_states_dilations_0, groups = hidden_states_groups_0, pad = hidden_states_pad_0, pad_type = hidden_states_pad_type_0, strides = hidden_states_strides_0, weight = layers_5_fc2_weight_to_fp16, x = input_47_cast_fp16)[name = string("hidden_states_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_23_cast_fp16, y = hidden_states_cast_fp16)[name = string("inputs_cast_fp16")]; + tensor out_axes_0 = const()[name = string("out_axes_0"), val = tensor([1])]; + fp16 var_4193_to_fp16 = const()[name = string("op_4193_to_fp16"), val = fp16(0x1.5p-17)]; + tensor out_cast_fp16 = layer_norm(axes = out_axes_0, epsilon = var_4193_to_fp16, x = inputs_cast_fp16)[name = string("out_cast_fp16")]; + tensor encoder_output_embeds_type_fp32_gamma_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41187456)))]; + tensor encoder_output_embeds_type_fp32_beta_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_beta_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41188544)))]; + fp16 encoder_output_embeds_type_fp32_epsilon_0_to_fp16 = const()[name = string("encoder_output_embeds_type_fp32_epsilon_0_to_fp16"), val = fp16(0x1.5p-17)]; + tensor encoder_output_embeds = batch_norm(beta = encoder_output_embeds_type_fp32_beta_0_to_fp16, epsilon = encoder_output_embeds_type_fp32_epsilon_0_to_fp16, gamma = encoder_output_embeds_type_fp32_gamma_0_to_fp16, mean = obj_1_mean_0_to_fp16, variance = obj_1_variance_0_to_fp16, x = out_cast_fp16)[name = string("encoder_output_embeds_type_fp32_cast_fp16")]; + string var_4217_pad_type_0 = const()[name = string("op_4217_pad_type_0"), val = string("valid")]; + tensor var_4217_strides_0 = const()[name = string("op_4217_strides_0"), val = tensor([1, 1])]; + tensor var_4217_pad_0 = const()[name = string("op_4217_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4217_dilations_0 = const()[name = string("op_4217_dilations_0"), val = tensor([1, 1])]; + int32 var_4217_groups_0 = const()[name = string("op_4217_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41189632)))]; + tensor var_4217_cast_fp16 = conv(dilations = var_4217_dilations_0, groups = var_4217_groups_0, pad = var_4217_pad_0, pad_type = var_4217_pad_type_0, strides = var_4217_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4217_cast_fp16")]; + string var_4224_pad_type_0 = const()[name = string("op_4224_pad_type_0"), val = string("valid")]; + tensor var_4224_strides_0 = const()[name = string("op_4224_strides_0"), val = tensor([1, 1])]; + tensor var_4224_pad_0 = const()[name = string("op_4224_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4224_dilations_0 = const()[name = string("op_4224_dilations_0"), val = tensor([1, 1])]; + int32 var_4224_groups_0 = const()[name = string("op_4224_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41713984)))]; + tensor decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42238336)))]; + tensor var_4224_cast_fp16 = conv(bias = decoder_kv_cache_prep_0_encoder_attn_v_proj_bias_to_fp16, dilations = var_4224_dilations_0, groups = var_4224_groups_0, pad = var_4224_pad_0, pad_type = var_4224_pad_type_0, strides = var_4224_strides_0, weight = decoder_kv_cache_prep_0_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4224_cast_fp16")]; + string var_4242_pad_type_0 = const()[name = string("op_4242_pad_type_0"), val = string("valid")]; + tensor var_4242_strides_0 = const()[name = string("op_4242_strides_0"), val = tensor([1, 1])]; + tensor var_4242_pad_0 = const()[name = string("op_4242_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4242_dilations_0 = const()[name = string("op_4242_dilations_0"), val = tensor([1, 1])]; + int32 var_4242_groups_0 = const()[name = string("op_4242_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42239424)))]; + tensor var_4242_cast_fp16 = conv(dilations = var_4242_dilations_0, groups = var_4242_groups_0, pad = var_4242_pad_0, pad_type = var_4242_pad_type_0, strides = var_4242_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4242_cast_fp16")]; + string var_4249_pad_type_0 = const()[name = string("op_4249_pad_type_0"), val = string("valid")]; + tensor var_4249_strides_0 = const()[name = string("op_4249_strides_0"), val = tensor([1, 1])]; + tensor var_4249_pad_0 = const()[name = string("op_4249_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4249_dilations_0 = const()[name = string("op_4249_dilations_0"), val = tensor([1, 1])]; + int32 var_4249_groups_0 = const()[name = string("op_4249_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42763776)))]; + tensor decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43288128)))]; + tensor var_4249_cast_fp16 = conv(bias = decoder_kv_cache_prep_1_encoder_attn_v_proj_bias_to_fp16, dilations = var_4249_dilations_0, groups = var_4249_groups_0, pad = var_4249_pad_0, pad_type = var_4249_pad_type_0, strides = var_4249_strides_0, weight = decoder_kv_cache_prep_1_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4249_cast_fp16")]; + string var_4267_pad_type_0 = const()[name = string("op_4267_pad_type_0"), val = string("valid")]; + tensor var_4267_strides_0 = const()[name = string("op_4267_strides_0"), val = tensor([1, 1])]; + tensor var_4267_pad_0 = const()[name = string("op_4267_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4267_dilations_0 = const()[name = string("op_4267_dilations_0"), val = tensor([1, 1])]; + int32 var_4267_groups_0 = const()[name = string("op_4267_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43289216)))]; + tensor var_4267_cast_fp16 = conv(dilations = var_4267_dilations_0, groups = var_4267_groups_0, pad = var_4267_pad_0, pad_type = var_4267_pad_type_0, strides = var_4267_strides_0, weight = decoder_kv_cache_prep_2_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4267_cast_fp16")]; + string var_4274_pad_type_0 = const()[name = string("op_4274_pad_type_0"), val = string("valid")]; + tensor var_4274_strides_0 = const()[name = string("op_4274_strides_0"), val = tensor([1, 1])]; + tensor var_4274_pad_0 = const()[name = string("op_4274_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4274_dilations_0 = const()[name = string("op_4274_dilations_0"), val = tensor([1, 1])]; + int32 var_4274_groups_0 = const()[name = string("op_4274_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43813568)))]; + tensor decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44337920)))]; + tensor var_4274_cast_fp16 = conv(bias = decoder_kv_cache_prep_2_encoder_attn_v_proj_bias_to_fp16, dilations = var_4274_dilations_0, groups = var_4274_groups_0, pad = var_4274_pad_0, pad_type = var_4274_pad_type_0, strides = var_4274_strides_0, weight = decoder_kv_cache_prep_2_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4274_cast_fp16")]; + string var_4292_pad_type_0 = const()[name = string("op_4292_pad_type_0"), val = string("valid")]; + tensor var_4292_strides_0 = const()[name = string("op_4292_strides_0"), val = tensor([1, 1])]; + tensor var_4292_pad_0 = const()[name = string("op_4292_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4292_dilations_0 = const()[name = string("op_4292_dilations_0"), val = tensor([1, 1])]; + int32 var_4292_groups_0 = const()[name = string("op_4292_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44339008)))]; + tensor var_4292_cast_fp16 = conv(dilations = var_4292_dilations_0, groups = var_4292_groups_0, pad = var_4292_pad_0, pad_type = var_4292_pad_type_0, strides = var_4292_strides_0, weight = decoder_kv_cache_prep_3_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4292_cast_fp16")]; + string var_4299_pad_type_0 = const()[name = string("op_4299_pad_type_0"), val = string("valid")]; + tensor var_4299_strides_0 = const()[name = string("op_4299_strides_0"), val = tensor([1, 1])]; + tensor var_4299_pad_0 = const()[name = string("op_4299_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4299_dilations_0 = const()[name = string("op_4299_dilations_0"), val = tensor([1, 1])]; + int32 var_4299_groups_0 = const()[name = string("op_4299_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44863360)))]; + tensor decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45387712)))]; + tensor var_4299_cast_fp16 = conv(bias = decoder_kv_cache_prep_3_encoder_attn_v_proj_bias_to_fp16, dilations = var_4299_dilations_0, groups = var_4299_groups_0, pad = var_4299_pad_0, pad_type = var_4299_pad_type_0, strides = var_4299_strides_0, weight = decoder_kv_cache_prep_3_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4299_cast_fp16")]; + string var_4317_pad_type_0 = const()[name = string("op_4317_pad_type_0"), val = string("valid")]; + tensor var_4317_strides_0 = const()[name = string("op_4317_strides_0"), val = tensor([1, 1])]; + tensor var_4317_pad_0 = const()[name = string("op_4317_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4317_dilations_0 = const()[name = string("op_4317_dilations_0"), val = tensor([1, 1])]; + int32 var_4317_groups_0 = const()[name = string("op_4317_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_4_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_4_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45388800)))]; + tensor var_4317_cast_fp16 = conv(dilations = var_4317_dilations_0, groups = var_4317_groups_0, pad = var_4317_pad_0, pad_type = var_4317_pad_type_0, strides = var_4317_strides_0, weight = decoder_kv_cache_prep_4_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4317_cast_fp16")]; + string var_4324_pad_type_0 = const()[name = string("op_4324_pad_type_0"), val = string("valid")]; + tensor var_4324_strides_0 = const()[name = string("op_4324_strides_0"), val = tensor([1, 1])]; + tensor var_4324_pad_0 = const()[name = string("op_4324_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4324_dilations_0 = const()[name = string("op_4324_dilations_0"), val = tensor([1, 1])]; + int32 var_4324_groups_0 = const()[name = string("op_4324_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_4_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_4_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(45913152)))]; + tensor decoder_kv_cache_prep_4_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_4_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46437504)))]; + tensor var_4324_cast_fp16 = conv(bias = decoder_kv_cache_prep_4_encoder_attn_v_proj_bias_to_fp16, dilations = var_4324_dilations_0, groups = var_4324_groups_0, pad = var_4324_pad_0, pad_type = var_4324_pad_type_0, strides = var_4324_strides_0, weight = decoder_kv_cache_prep_4_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("op_4324_cast_fp16")]; + string k_pad_type_0 = const()[name = string("k_pad_type_0"), val = string("valid")]; + tensor k_strides_0 = const()[name = string("k_strides_0"), val = tensor([1, 1])]; + tensor k_pad_0 = const()[name = string("k_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor k_dilations_0 = const()[name = string("k_dilations_0"), val = tensor([1, 1])]; + int32 k_groups_0 = const()[name = string("k_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_5_encoder_attn_k_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_5_encoder_attn_k_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46438592)))]; + tensor k_cast_fp16 = conv(dilations = k_dilations_0, groups = k_groups_0, pad = k_pad_0, pad_type = k_pad_type_0, strides = k_strides_0, weight = decoder_kv_cache_prep_5_encoder_attn_k_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("k_cast_fp16")]; + string v_pad_type_0 = const()[name = string("v_pad_type_0"), val = string("valid")]; + tensor v_strides_0 = const()[name = string("v_strides_0"), val = tensor([1, 1])]; + tensor v_pad_0 = const()[name = string("v_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor v_dilations_0 = const()[name = string("v_dilations_0"), val = tensor([1, 1])]; + int32 v_groups_0 = const()[name = string("v_groups_0"), val = int32(1)]; + tensor decoder_kv_cache_prep_5_encoder_attn_v_proj_weight_to_fp16 = const()[name = string("decoder_kv_cache_prep_5_encoder_attn_v_proj_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46962944)))]; + tensor decoder_kv_cache_prep_5_encoder_attn_v_proj_bias_to_fp16 = const()[name = string("decoder_kv_cache_prep_5_encoder_attn_v_proj_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47487296)))]; + tensor v_cast_fp16 = conv(bias = decoder_kv_cache_prep_5_encoder_attn_v_proj_bias_to_fp16, dilations = v_dilations_0, groups = v_groups_0, pad = v_pad_0, pad_type = v_pad_type_0, strides = v_strides_0, weight = decoder_kv_cache_prep_5_encoder_attn_v_proj_weight_to_fp16, x = encoder_output_embeds)[name = string("v_cast_fp16")]; + int32 var_4354 = const()[name = string("op_4354"), val = int32(0)]; + bool input_51_interleave_0 = const()[name = string("input_51_interleave_0"), val = bool(false)]; + tensor input_51_cast_fp16 = concat(axis = var_4354, interleave = input_51_interleave_0, values = (var_4217_cast_fp16, var_4242_cast_fp16, var_4267_cast_fp16, var_4292_cast_fp16, var_4317_cast_fp16, k_cast_fp16))[name = string("input_51_cast_fp16")]; + int32 var_4357 = const()[name = string("op_4357"), val = int32(0)]; + bool input_interleave_0 = const()[name = string("input_interleave_0"), val = bool(false)]; + tensor input_cast_fp16 = concat(axis = var_4357, interleave = input_interleave_0, values = (var_4224_cast_fp16, var_4249_cast_fp16, var_4274_cast_fp16, var_4299_cast_fp16, var_4324_cast_fp16, v_cast_fp16))[name = string("input_cast_fp16")]; + tensor var_4364_pad_0 = const()[name = string("op_4364_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; + string var_4364_mode_0 = const()[name = string("op_4364_mode_0"), val = string("constant")]; + fp16 const_7_to_fp16 = const()[name = string("const_7_to_fp16"), val = fp16(0x0p+0)]; + tensor encoder_attn_key_cache = pad(constant_val = const_7_to_fp16, mode = var_4364_mode_0, pad = var_4364_pad_0, x = input_51_cast_fp16)[name = string("op_4364_cast_fp16")]; + tensor var_4370_pad_0 = const()[name = string("op_4370_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 0, 36])]; + string var_4370_mode_0 = const()[name = string("op_4370_mode_0"), val = string("constant")]; + fp16 const_8_to_fp16 = const()[name = string("const_8_to_fp16"), val = fp16(0x0p+0)]; + tensor encoder_attn_value_cache = pad(constant_val = const_8_to_fp16, mode = var_4370_mode_0, pad = var_4370_pad_0, x = input_cast_fp16)[name = string("op_4370_cast_fp16")]; + } -> (encoder_output_embeds, encoder_attn_key_cache, encoder_attn_value_cache); +} \ No newline at end of file