diff --git "a/ParakeetEncoder.mlmodelc/model.mil" "b/ParakeetEncoder.mlmodelc/model.mil"
new file mode 100644--- /dev/null
+++ "b/ParakeetEncoder.mlmodelc/model.mil"
@@ -0,0 +1,3330 @@
+program(1.0)
+[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3405.2.1"}, {"coremlc-version", "3405.2.1"}, {"coremltools-component-torch", "2.4.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "7.2"}})]
+{
+    func main<ios15>(tensor<fp32, [1, 128, 1001]> audio_signal, tensor<int32, [1]> length) {
+            tensor<int32, []> var_5 = const()[name = tensor<string, []>("op_5"), val = tensor<int32, []>(1024)];
+            tensor<int32, []> var_22 = const()[name = tensor<string, []>("op_22"), val = tensor<int32, []>(256)];
+            tensor<int32, []> var_25 = const()[name = tensor<string, []>("op_25"), val = tensor<int32, []>(-1)];
+            tensor<int32, []> var_27 = const()[name = tensor<string, []>("op_27"), val = tensor<int32, []>(1)];
+            tensor<int32, [3]> x_1_perm_0 = const()[name = tensor<string, []>("x_1_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<string, []> audio_signal_to_fp16_dtype_0 = const()[name = tensor<string, []>("audio_signal_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
+            tensor<string, []> var_81_to_fp16_dtype_0 = const()[name = tensor<string, []>("op_81_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
+            tensor<fp16, []> var_82_promoted_to_fp16 = const()[name = tensor<string, []>("op_82_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
+            tensor<fp16, [1]> cast_229 = cast(dtype = var_81_to_fp16_dtype_0, x = length)[name = tensor<string, []>("cast_229")];
+            tensor<fp16, [1]> var_83_cast_fp16 = add(x = cast_229, y = var_82_promoted_to_fp16)[name = tensor<string, []>("op_83_cast_fp16")];
+            tensor<fp16, []> _inversed_85_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_85_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1]> _inversed_85_cast_fp16 = mul(x = var_83_cast_fp16, y = _inversed_85_y_0_to_fp16)[name = tensor<string, []>("_inversed_85_cast_fp16")];
+            tensor<fp16, []> var_86_to_fp16 = const()[name = tensor<string, []>("op_86_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
+            tensor<fp16, [1]> lengths_1_cast_fp16 = add(x = _inversed_85_cast_fp16, y = var_86_to_fp16)[name = tensor<string, []>("lengths_1_cast_fp16")];
+            tensor<fp16, [1]> lengths_3_cast_fp16 = floor(x = lengths_1_cast_fp16)[name = tensor<string, []>("lengths_3_cast_fp16")];
+            tensor<fp16, []> var_90_promoted_to_fp16 = const()[name = tensor<string, []>("op_90_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
+            tensor<fp16, [1]> var_91_cast_fp16 = add(x = lengths_3_cast_fp16, y = var_90_promoted_to_fp16)[name = tensor<string, []>("op_91_cast_fp16")];
+            tensor<fp16, []> _inversed_93_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_93_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1]> _inversed_93_cast_fp16 = mul(x = var_91_cast_fp16, y = _inversed_93_y_0_to_fp16)[name = tensor<string, []>("_inversed_93_cast_fp16")];
+            tensor<fp16, []> var_94_to_fp16 = const()[name = tensor<string, []>("op_94_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
+            tensor<fp16, [1]> lengths_7_cast_fp16 = add(x = _inversed_93_cast_fp16, y = var_94_to_fp16)[name = tensor<string, []>("lengths_7_cast_fp16")];
+            tensor<fp16, [1]> lengths_9_cast_fp16 = floor(x = lengths_7_cast_fp16)[name = tensor<string, []>("lengths_9_cast_fp16")];
+            tensor<fp16, []> var_98_promoted_to_fp16 = const()[name = tensor<string, []>("op_98_promoted_to_fp16"), val = tensor<fp16, []>(-0x1p+0)];
+            tensor<fp16, [1]> var_99_cast_fp16 = add(x = lengths_9_cast_fp16, y = var_98_promoted_to_fp16)[name = tensor<string, []>("op_99_cast_fp16")];
+            tensor<fp16, []> _inversed_101_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_101_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1]> _inversed_101_cast_fp16 = mul(x = var_99_cast_fp16, y = _inversed_101_y_0_to_fp16)[name = tensor<string, []>("_inversed_101_cast_fp16")];
+            tensor<fp16, []> var_102_to_fp16 = const()[name = tensor<string, []>("op_102_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
+            tensor<fp16, [1]> lengths_13_cast_fp16 = add(x = _inversed_101_cast_fp16, y = var_102_to_fp16)[name = tensor<string, []>("lengths_13_cast_fp16")];
+            tensor<fp16, [1]> lengths_cast_fp16 = floor(x = lengths_13_cast_fp16)[name = tensor<string, []>("lengths_cast_fp16")];
+            tensor<int32, [1]> input_1_axes_0 = const()[name = tensor<string, []>("input_1_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<fp16, [1, 128, 1001]> cast_230 = cast(dtype = audio_signal_to_fp16_dtype_0, x = audio_signal)[name = tensor<string, []>("cast_230")];
+            tensor<fp16, [1, 1001, 128]> transpose_291 = transpose(perm = x_1_perm_0, x = cast_230)[name = tensor<string, []>("transpose_291")];
+            tensor<fp16, [1, 1, 1001, 128]> input_1_cast_fp16 = expand_dims(axes = input_1_axes_0, x = transpose_291)[name = tensor<string, []>("input_1_cast_fp16")];
+            tensor<int32, [2]> var_114 = const()[name = tensor<string, []>("op_114"), val = tensor<int32, [2]>([2, 2])];
+            tensor<int32, [2]> var_116 = const()[name = tensor<string, []>("op_116"), val = tensor<int32, [2]>([1, 1])];
+            tensor<string, []> input_3_pad_type_0 = const()[name = tensor<string, []>("input_3_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [4]> input_3_pad_0 = const()[name = tensor<string, []>("input_3_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
+            tensor<fp16, [256, 1, 3, 3]> model_pre_encode_conv_0_weight_to_fp16 = const()[name = tensor<string, []>("model_pre_encode_conv_0_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))];
+            tensor<fp16, [256]> model_pre_encode_conv_0_bias_to_fp16 = const()[name = tensor<string, []>("model_pre_encode_conv_0_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4736)))];
+            tensor<fp16, [1, 256, 501, 64]> input_3_cast_fp16 = conv(bias = model_pre_encode_conv_0_bias_to_fp16, dilations = var_116, groups = var_27, pad = input_3_pad_0, pad_type = input_3_pad_type_0, strides = var_114, weight = model_pre_encode_conv_0_weight_to_fp16, x = input_1_cast_fp16)[name = tensor<string, []>("input_3_cast_fp16")];
+            tensor<fp16, [1, 256, 501, 64]> input_5_cast_fp16 = relu(x = input_3_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")];
+            tensor<int32, [2]> var_122 = const()[name = tensor<string, []>("op_122"), val = tensor<int32, [2]>([2, 2])];
+            tensor<int32, [2]> var_124 = const()[name = tensor<string, []>("op_124"), val = tensor<int32, [2]>([1, 1])];
+            tensor<string, []> input_7_pad_type_0 = const()[name = tensor<string, []>("input_7_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [4]> input_7_pad_0 = const()[name = tensor<string, []>("input_7_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
+            tensor<fp16, [256, 1, 3, 3]> model_pre_encode_conv_2_weight_to_fp16 = const()[name = tensor<string, []>("model_pre_encode_conv_2_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5312)))];
+            tensor<fp16, [256]> model_pre_encode_conv_2_bias_to_fp16 = const()[name = tensor<string, []>("model_pre_encode_conv_2_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9984)))];
+            tensor<fp16, [1, 256, 251, 32]> input_7_cast_fp16 = conv(bias = model_pre_encode_conv_2_bias_to_fp16, dilations = var_124, groups = var_22, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = var_122, weight = model_pre_encode_conv_2_weight_to_fp16, x = input_5_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")];
+            tensor<int32, [2]> var_129 = const()[name = tensor<string, []>("op_129"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [2]> var_131 = const()[name = tensor<string, []>("op_131"), val = tensor<int32, [2]>([1, 1])];
+            tensor<string, []> input_9_pad_type_0 = const()[name = tensor<string, []>("input_9_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [4]> input_9_pad_0 = const()[name = tensor<string, []>("input_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<fp16, [256, 256, 1, 1]> model_pre_encode_conv_3_weight_to_fp16 = const()[name = tensor<string, []>("model_pre_encode_conv_3_weight_to_fp16"), val = tensor<fp16, [256, 256, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10560)))];
+            tensor<fp16, [256]> model_pre_encode_conv_3_bias_to_fp16 = const()[name = tensor<string, []>("model_pre_encode_conv_3_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(141696)))];
+            tensor<fp16, [1, 256, 251, 32]> input_9_cast_fp16 = conv(bias = model_pre_encode_conv_3_bias_to_fp16, dilations = var_131, groups = var_27, pad = input_9_pad_0, pad_type = input_9_pad_type_0, strides = var_129, weight = model_pre_encode_conv_3_weight_to_fp16, x = input_7_cast_fp16)[name = tensor<string, []>("input_9_cast_fp16")];
+            tensor<fp16, [1, 256, 251, 32]> input_11_cast_fp16 = relu(x = input_9_cast_fp16)[name = tensor<string, []>("input_11_cast_fp16")];
+            tensor<int32, [2]> var_137 = const()[name = tensor<string, []>("op_137"), val = tensor<int32, [2]>([2, 2])];
+            tensor<int32, [2]> var_139 = const()[name = tensor<string, []>("op_139"), val = tensor<int32, [2]>([1, 1])];
+            tensor<string, []> input_13_pad_type_0 = const()[name = tensor<string, []>("input_13_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [4]> input_13_pad_0 = const()[name = tensor<string, []>("input_13_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
+            tensor<fp16, [256, 1, 3, 3]> model_pre_encode_conv_5_weight_to_fp16 = const()[name = tensor<string, []>("model_pre_encode_conv_5_weight_to_fp16"), val = tensor<fp16, [256, 1, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142272)))];
+            tensor<fp16, [256]> model_pre_encode_conv_5_bias_to_fp16 = const()[name = tensor<string, []>("model_pre_encode_conv_5_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146944)))];
+            tensor<fp16, [1, 256, 126, 16]> input_13_cast_fp16 = conv(bias = model_pre_encode_conv_5_bias_to_fp16, dilations = var_139, groups = var_22, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = var_137, weight = model_pre_encode_conv_5_weight_to_fp16, x = input_11_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")];
+            tensor<int32, [2]> var_144 = const()[name = tensor<string, []>("op_144"), val = tensor<int32, [2]>([1, 1])];
+            tensor<int32, [2]> var_146 = const()[name = tensor<string, []>("op_146"), val = tensor<int32, [2]>([1, 1])];
+            tensor<string, []> input_15_pad_type_0 = const()[name = tensor<string, []>("input_15_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [4]> input_15_pad_0 = const()[name = tensor<string, []>("input_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<fp16, [256, 256, 1, 1]> model_pre_encode_conv_6_weight_to_fp16 = const()[name = tensor<string, []>("model_pre_encode_conv_6_weight_to_fp16"), val = tensor<fp16, [256, 256, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147520)))];
+            tensor<fp16, [256]> model_pre_encode_conv_6_bias_to_fp16 = const()[name = tensor<string, []>("model_pre_encode_conv_6_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(278656)))];
+            tensor<fp16, [1, 256, 126, 16]> input_15_cast_fp16 = conv(bias = model_pre_encode_conv_6_bias_to_fp16, dilations = var_146, groups = var_27, pad = input_15_pad_0, pad_type = input_15_pad_type_0, strides = var_144, weight = model_pre_encode_conv_6_weight_to_fp16, x = input_13_cast_fp16)[name = tensor<string, []>("input_15_cast_fp16")];
+            tensor<fp16, [1, 256, 126, 16]> x_3_cast_fp16 = relu(x = input_15_cast_fp16)[name = tensor<string, []>("x_3_cast_fp16")];
+            tensor<int32, [4]> var_152_perm_0 = const()[name = tensor<string, []>("op_152_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_153 = const()[name = tensor<string, []>("op_153"), val = tensor<int32, [3]>([1, 126, -1])];
+            tensor<fp16, [1, 126, 256, 16]> transpose_290 = transpose(perm = var_152_perm_0, x = x_3_cast_fp16)[name = tensor<string, []>("transpose_290")];
+            tensor<fp16, [1, 126, 4096]> input_17_cast_fp16 = reshape(shape = var_153, x = transpose_290)[name = tensor<string, []>("input_17_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_pre_encode_out_weight_to_fp16 = const()[name = tensor<string, []>("model_pre_encode_out_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(279232)))];
+            tensor<fp16, [1024]> model_pre_encode_out_bias_to_fp16 = const()[name = tensor<string, []>("model_pre_encode_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8667904)))];
+            tensor<fp16, [1, 126, 1024]> linear_0_cast_fp16 = linear(bias = model_pre_encode_out_bias_to_fp16, weight = model_pre_encode_out_weight_to_fp16, x = input_17_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")];
+            tensor<string, []> padding_length_dtype_0 = const()[name = tensor<string, []>("padding_length_dtype_0"), val = tensor<string, []>("int32")];
+            tensor<int32, [1, 126]> expand_dims_0 = const()[name = tensor<string, []>("expand_dims_0"), val = tensor<int32, [1, 126]>([[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73, 74, 75, 76, 77, 78, 79, 80, 81, 82, 83, 84, 85, 86, 87, 88, 89, 90, 91, 92, 93, 94, 95, 96, 97, 98, 99, 100, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111, 112, 113, 114, 115, 116, 117, 118, 119, 120, 121, 122, 123, 124, 125]])];
+            tensor<int32, [1]> var_191_axes_0 = const()[name = tensor<string, []>("op_191_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<int32, [1]> encoder_output_length = cast(dtype = padding_length_dtype_0, x = lengths_cast_fp16)[name = tensor<string, []>("cast_228")];
+            tensor<int32, [1, 1]> var_191 = expand_dims(axes = var_191_axes_0, x = encoder_output_length)[name = tensor<string, []>("op_191")];
+            tensor<bool, [1, 126]> pad_mask_1 = less(x = expand_dims_0, y = var_191)[name = tensor<string, []>("pad_mask_1")];
+            tensor<int32, [1]> var_193_axes_0 = const()[name = tensor<string, []>("op_193_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<bool, [1, 1, 126]> var_193 = expand_dims(axes = var_193_axes_0, x = pad_mask_1)[name = tensor<string, []>("op_193")];
+            tensor<int32, [3]> var_194 = const()[name = tensor<string, []>("op_194"), val = tensor<int32, [3]>([1, 126, 1])];
+            tensor<bool, [1, 126, 126]> pad_mask_for_att_mask_1 = tile(reps = var_194, x = var_193)[name = tensor<string, []>("pad_mask_for_att_mask_1")];
+            tensor<int32, [3]> var_196_perm_0 = const()[name = tensor<string, []>("op_196_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<bool, [1, 126, 126]> transpose_289 = transpose(perm = var_196_perm_0, x = pad_mask_for_att_mask_1)[name = tensor<string, []>("transpose_289")];
+            tensor<bool, [1, 126, 126]> pad_mask_for_att_mask = logical_and(x = pad_mask_for_att_mask_1, y = transpose_289)[name = tensor<string, []>("pad_mask_for_att_mask")];
+            tensor<bool, [1, 126, 126]> att_mask_5 = const()[name = tensor<string, []>("att_mask_5"), val = tensor<bool, [1, 126, 126]>([[[true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true]]])];
+            tensor<bool, [1, 126, 126]> att_mask = logical_and(x = pad_mask_for_att_mask, y = att_mask_5)[name = tensor<string, []>("att_mask")];
+            tensor<bool, [1, 126, 126]> mask_1 = logical_not(x = att_mask)[name = tensor<string, []>("mask_1")];
+            tensor<bool, [1, 126]> pad_mask = logical_not(x = pad_mask_1)[name = tensor<string, []>("pad_mask")];
+            tensor<int32, [1]> input_21_axes_0 = const()[name = tensor<string, []>("input_21_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_0_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_0_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8670016)))];
+            tensor<fp16, [1024]> model_layers_0_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_0_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8672128)))];
+            tensor<fp16, []> var_4_to_fp16 = const()[name = tensor<string, []>("op_4_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
+            tensor<fp16, [1, 126, 1024]> input_21_cast_fp16 = layer_norm(axes = input_21_axes_0, beta = model_layers_0_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_0_norm_feed_forward1_weight_to_fp16, x = linear_0_cast_fp16)[name = tensor<string, []>("input_21_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_0_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_0_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8674240)))];
+            tensor<fp16, [4096]> linear_1_bias_0_to_fp16 = const()[name = tensor<string, []>("linear_1_bias_0_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17062912)))];
+            tensor<fp16, [1, 126, 4096]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_0_feed_forward1_linear1_weight_to_fp16, x = input_21_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_25_cast_fp16 = silu(x = linear_1_cast_fp16)[name = tensor<string, []>("input_25_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_0_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_0_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17071168)))];
+            tensor<fp16, [1024]> linear_2_bias_0_to_fp16 = const()[name = tensor<string, []>("linear_2_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25459840)))];
+            tensor<fp16, [1, 126, 1024]> linear_2_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_0_feed_forward1_linear2_weight_to_fp16, x = input_25_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")];
+            tensor<fp16, []> var_227_to_fp16 = const()[name = tensor<string, []>("op_227_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_228_cast_fp16 = mul(x = linear_2_cast_fp16, y = var_227_to_fp16)[name = tensor<string, []>("op_228_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_31_cast_fp16 = add(x = linear_0_cast_fp16, y = var_228_cast_fp16)[name = tensor<string, []>("input_31_cast_fp16")];
+            tensor<int32, [1]> query_1_axes_0 = const()[name = tensor<string, []>("query_1_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_0_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_0_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25461952)))];
+            tensor<fp16, [1024]> model_layers_0_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_0_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25464064)))];
+            tensor<fp16, [1, 126, 1024]> query_1_cast_fp16 = layer_norm(axes = query_1_axes_0, beta = model_layers_0_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_0_norm_self_att_weight_to_fp16, x = input_31_cast_fp16)[name = tensor<string, []>("query_1_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_0_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_0_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25466176)))];
+            tensor<fp16, [1, 126, 1024]> linear_3_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_0_self_attn_linear_q_weight_to_fp16, x = query_1_cast_fp16)[name = tensor<string, []>("linear_3_cast_fp16")];
+            tensor<int32, [4]> var_244 = const()[name = tensor<string, []>("op_244"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_1_cast_fp16 = reshape(shape = var_244, x = linear_3_cast_fp16)[name = tensor<string, []>("q_1_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_0_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_0_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27563392)))];
+            tensor<fp16, [1, 126, 1024]> linear_4_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_0_self_attn_linear_k_weight_to_fp16, x = query_1_cast_fp16)[name = tensor<string, []>("linear_4_cast_fp16")];
+            tensor<int32, [4]> var_248 = const()[name = tensor<string, []>("op_248"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_1_cast_fp16 = reshape(shape = var_248, x = linear_4_cast_fp16)[name = tensor<string, []>("k_1_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_0_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_0_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29660608)))];
+            tensor<fp16, [1, 126, 1024]> linear_5_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_0_self_attn_linear_v_weight_to_fp16, x = query_1_cast_fp16)[name = tensor<string, []>("linear_5_cast_fp16")];
+            tensor<int32, [4]> var_252 = const()[name = tensor<string, []>("op_252"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_1_cast_fp16 = reshape(shape = var_252, x = linear_5_cast_fp16)[name = tensor<string, []>("v_1_cast_fp16")];
+            tensor<int32, [4]> value_1_perm_0 = const()[name = tensor<string, []>("value_1_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_0_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_0_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31757824)))];
+            tensor<fp16, [1, 126, 8, 128]> var_264_cast_fp16 = add(x = q_1_cast_fp16, y = model_layers_0_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_264_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_0_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_0_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31759936)))];
+            tensor<fp16, [1, 126, 8, 128]> var_266_cast_fp16 = add(x = q_1_cast_fp16, y = model_layers_0_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_266_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_1_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_1_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_7_transpose_x_0 = const()[name = tensor<string, []>("x_7_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_7_transpose_y_0 = const()[name = tensor<string, []>("x_7_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_268_to_fp16 = const()[name = tensor<string, []>("op_268_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31762048)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_287 = transpose(perm = q_with_bias_v_1_perm_0, x = var_266_cast_fp16)[name = tensor<string, []>("transpose_287")];
+            tensor<fp16, [1, 8, 126, 251]> x_7_cast_fp16 = matmul(transpose_x = x_7_transpose_x_0, transpose_y = x_7_transpose_y_0, x = transpose_287, y = var_268_to_fp16)[name = tensor<string, []>("x_7_cast_fp16")];
+            tensor<int32, [8]> x_9_pad_0 = const()[name = tensor<string, []>("x_9_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_9_mode_0 = const()[name = tensor<string, []>("x_9_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_13_to_fp16 = const()[name = tensor<string, []>("const_13_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_9_cast_fp16 = pad(constant_val = const_13_to_fp16, mode = x_9_mode_0, pad = x_9_pad_0, x = x_7_cast_fp16)[name = tensor<string, []>("x_9_cast_fp16")];
+            tensor<int32, [4]> var_276 = const()[name = tensor<string, []>("op_276"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_11_cast_fp16 = reshape(shape = var_276, x = x_9_cast_fp16)[name = tensor<string, []>("x_11_cast_fp16")];
+            tensor<int32, [4]> var_280_begin_0 = const()[name = tensor<string, []>("op_280_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_280_end_0 = const()[name = tensor<string, []>("op_280_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_280_end_mask_0 = const()[name = tensor<string, []>("op_280_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_280_cast_fp16 = slice_by_index(begin = var_280_begin_0, end = var_280_end_0, end_mask = var_280_end_mask_0, x = x_11_cast_fp16)[name = tensor<string, []>("op_280_cast_fp16")];
+            tensor<int32, [4]> var_281 = const()[name = tensor<string, []>("op_281"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_1_cast_fp16 = reshape(shape = var_281, x = var_280_cast_fp16)[name = tensor<string, []>("matrix_bd_1_cast_fp16")];
+            tensor<bool, []> matrix_ac_1_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_1_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_1_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_1_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_72_perm_0 = const()[name = tensor<string, []>("transpose_72_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_73_perm_0 = const()[name = tensor<string, []>("transpose_73_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_285 = transpose(perm = transpose_73_perm_0, x = k_1_cast_fp16)[name = tensor<string, []>("transpose_285")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_286 = transpose(perm = transpose_72_perm_0, x = var_264_cast_fp16)[name = tensor<string, []>("transpose_286")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_1_cast_fp16 = matmul(transpose_x = matrix_ac_1_transpose_x_0, transpose_y = matrix_ac_1_transpose_y_0, x = transpose_286, y = transpose_285)[name = tensor<string, []>("matrix_ac_1_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_3_begin_0 = const()[name = tensor<string, []>("matrix_bd_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_3_end_0 = const()[name = tensor<string, []>("matrix_bd_3_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_3_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_3_cast_fp16 = slice_by_index(begin = matrix_bd_3_begin_0, end = matrix_bd_3_end_0, end_mask = matrix_bd_3_end_mask_0, x = matrix_bd_1_cast_fp16)[name = tensor<string, []>("matrix_bd_3_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_290_cast_fp16 = add(x = matrix_ac_1_cast_fp16, y = matrix_bd_3_cast_fp16)[name = tensor<string, []>("op_290_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_1_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_1_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_1_cast_fp16 = mul(x = var_290_cast_fp16, y = _inversed_scores_1_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_1_cast_fp16")];
+            tensor<int32, [1]> mask_3_axes_0 = const()[name = tensor<string, []>("mask_3_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<bool, [1, 1, 126, 126]> mask_3 = expand_dims(axes = mask_3_axes_0, x = mask_1)[name = tensor<string, []>("mask_3")];
+            tensor<fp16, []> var_7_to_fp16 = const()[name = tensor<string, []>("op_7_to_fp16"), val = tensor<fp16, []>(-0x1.388p+13)];
+            tensor<fp16, [1, 8, 126, 126]> scores_3_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_1_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_3_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_296_cast_fp16 = softmax(axis = var_25, x = scores_3_cast_fp16)[name = tensor<string, []>("op_296_cast_fp16")];
+            tensor<fp16, []> var_6_to_fp16 = const()[name = tensor<string, []>("op_6_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 126]> input_33_cast_fp16 = select(a = var_6_to_fp16, b = var_296_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_33_cast_fp16")];
+            tensor<bool, []> x_13_transpose_x_0 = const()[name = tensor<string, []>("x_13_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_13_transpose_y_0 = const()[name = tensor<string, []>("x_13_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_288 = transpose(perm = value_1_perm_0, x = v_1_cast_fp16)[name = tensor<string, []>("transpose_288")];
+            tensor<fp16, [1, 8, 126, 128]> x_13_cast_fp16 = matmul(transpose_x = x_13_transpose_x_0, transpose_y = x_13_transpose_y_0, x = input_33_cast_fp16, y = transpose_288)[name = tensor<string, []>("x_13_cast_fp16")];
+            tensor<int32, [4]> var_300_perm_0 = const()[name = tensor<string, []>("op_300_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_301 = const()[name = tensor<string, []>("op_301"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_284 = transpose(perm = var_300_perm_0, x = x_13_cast_fp16)[name = tensor<string, []>("transpose_284")];
+            tensor<fp16, [1, 126, 1024]> input_35_cast_fp16 = reshape(shape = var_301, x = transpose_284)[name = tensor<string, []>("input_35_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_0_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_0_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32276160)))];
+            tensor<fp16, [1, 126, 1024]> linear_7_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_0_self_attn_linear_out_weight_to_fp16, x = input_35_cast_fp16)[name = tensor<string, []>("linear_7_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_39_cast_fp16 = add(x = input_31_cast_fp16, y = linear_7_cast_fp16)[name = tensor<string, []>("input_39_cast_fp16")];
+            tensor<int32, [1]> x_17_axes_0 = const()[name = tensor<string, []>("x_17_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_0_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_0_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34373376)))];
+            tensor<fp16, [1024]> model_layers_0_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_0_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34375488)))];
+            tensor<fp16, [1, 126, 1024]> x_17_cast_fp16 = layer_norm(axes = x_17_axes_0, beta = model_layers_0_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_0_norm_conv_weight_to_fp16, x = input_39_cast_fp16)[name = tensor<string, []>("x_17_cast_fp16")];
+            tensor<int32, [3]> input_41_perm_0 = const()[name = tensor<string, []>("input_41_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_317 = const()[name = tensor<string, []>("op_317"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_319 = const()[name = tensor<string, []>("op_319"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_43_pad_type_0 = const()[name = tensor<string, []>("input_43_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_43_pad_0 = const()[name = tensor<string, []>("input_43_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_0_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_0_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34377600)))];
+            tensor<fp16, [1, 1024, 126]> transpose_283 = transpose(perm = input_41_perm_0, x = x_17_cast_fp16)[name = tensor<string, []>("transpose_283")];
+            tensor<fp16, [1, 2048, 126]> input_43_cast_fp16 = conv(dilations = var_319, groups = var_27, pad = input_43_pad_0, pad_type = input_43_pad_type_0, strides = var_317, weight = model_layers_0_conv_pointwise_conv1_weight_to_fp16, x = transpose_283)[name = tensor<string, []>("input_43_cast_fp16")];
+            tensor<int32, []> x_19_split_num_splits_0 = const()[name = tensor<string, []>("x_19_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_19_split_axis_0 = const()[name = tensor<string, []>("x_19_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_19_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_19_split_cast_fp16_1 = split(axis = x_19_split_axis_0, num_splits = x_19_split_num_splits_0, x = input_43_cast_fp16)[name = tensor<string, []>("x_19_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_19_split_1_sigmoid_cast_fp16 = sigmoid(x = x_19_split_cast_fp16_1)[name = tensor<string, []>("x_19_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_19_cast_fp16 = mul(x = x_19_split_cast_fp16_0, y = x_19_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_19_cast_fp16")];
+            tensor<int32, [1]> var_323_axes_0 = const()[name = tensor<string, []>("op_323_axes_0"), val = tensor<int32, [1]>([1])];
+            tensor<bool, [1, 1, 126]> var_323 = expand_dims(axes = var_323_axes_0, x = pad_mask)[name = tensor<string, []>("op_323")];
+            tensor<fp16, [1, 1024, 126]> input_45_cast_fp16 = select(a = var_6_to_fp16, b = x_19_cast_fp16, cond = var_323)[name = tensor<string, []>("input_45_cast_fp16")];
+            tensor<int32, [6]> input_47_pad_0 = const()[name = tensor<string, []>("input_47_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_47_mode_0 = const()[name = tensor<string, []>("input_47_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_16_to_fp16 = const()[name = tensor<string, []>("const_16_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_47_cast_fp16 = pad(constant_val = const_16_to_fp16, mode = input_47_mode_0, pad = input_47_pad_0, x = input_45_cast_fp16)[name = tensor<string, []>("input_47_cast_fp16")];
+            tensor<int32, [1]> var_328 = const()[name = tensor<string, []>("op_328"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_330 = const()[name = tensor<string, []>("op_330"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_49_pad_type_0 = const()[name = tensor<string, []>("input_49_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_49_pad_0 = const()[name = tensor<string, []>("input_49_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_51_weight_0_to_fp16 = const()[name = tensor<string, []>("input_51_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38571968)))];
+            tensor<fp16, [1024]> input_51_bias_0_to_fp16 = const()[name = tensor<string, []>("input_51_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38590464)))];
+            tensor<fp16, [1, 1024, 126]> input_51_cast_fp16 = conv(bias = input_51_bias_0_to_fp16, dilations = var_330, groups = var_5, pad = input_49_pad_0, pad_type = input_49_pad_type_0, strides = var_328, weight = input_51_weight_0_to_fp16, x = input_47_cast_fp16)[name = tensor<string, []>("input_51_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_53_cast_fp16 = silu(x = input_51_cast_fp16)[name = tensor<string, []>("input_53_cast_fp16")];
+            tensor<int32, [1]> var_340 = const()[name = tensor<string, []>("op_340"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_342 = const()[name = tensor<string, []>("op_342"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_21_pad_type_0 = const()[name = tensor<string, []>("x_21_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_21_pad_0 = const()[name = tensor<string, []>("x_21_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_0_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_0_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38592576)))];
+            tensor<fp16, [1, 1024, 126]> x_21_cast_fp16 = conv(dilations = var_342, groups = var_27, pad = x_21_pad_0, pad_type = x_21_pad_type_0, strides = var_340, weight = model_layers_0_conv_pointwise_conv2_weight_to_fp16, x = input_53_cast_fp16)[name = tensor<string, []>("x_21_cast_fp16")];
+            tensor<int32, [3]> input_55_perm_0 = const()[name = tensor<string, []>("input_55_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_282 = transpose(perm = input_55_perm_0, x = x_21_cast_fp16)[name = tensor<string, []>("transpose_282")];
+            tensor<fp16, [1, 126, 1024]> input_57_cast_fp16 = add(x = input_39_cast_fp16, y = transpose_282)[name = tensor<string, []>("input_57_cast_fp16")];
+            tensor<int32, [1]> input_59_axes_0 = const()[name = tensor<string, []>("input_59_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_0_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_0_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40689792)))];
+            tensor<fp16, [1024]> model_layers_0_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_0_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40691904)))];
+            tensor<fp16, [1, 126, 1024]> input_59_cast_fp16 = layer_norm(axes = input_59_axes_0, beta = model_layers_0_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_0_norm_feed_forward2_weight_to_fp16, x = input_57_cast_fp16)[name = tensor<string, []>("input_59_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_0_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_0_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40694016)))];
+            tensor<fp16, [1, 126, 4096]> linear_8_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_0_feed_forward2_linear1_weight_to_fp16, x = input_59_cast_fp16)[name = tensor<string, []>("linear_8_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_63_cast_fp16 = silu(x = linear_8_cast_fp16)[name = tensor<string, []>("input_63_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_0_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_0_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49082688)))];
+            tensor<fp16, [1, 126, 1024]> linear_9_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_0_feed_forward2_linear2_weight_to_fp16, x = input_63_cast_fp16)[name = tensor<string, []>("linear_9_cast_fp16")];
+            tensor<fp16, []> var_361_to_fp16 = const()[name = tensor<string, []>("op_361_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_362_cast_fp16 = mul(x = linear_9_cast_fp16, y = var_361_to_fp16)[name = tensor<string, []>("op_362_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_69_cast_fp16 = add(x = input_57_cast_fp16, y = var_362_cast_fp16)[name = tensor<string, []>("input_69_cast_fp16")];
+            tensor<int32, [1]> input_71_axes_0 = const()[name = tensor<string, []>("input_71_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_0_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_0_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57471360)))];
+            tensor<fp16, [1024]> model_layers_0_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_0_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57473472)))];
+            tensor<fp16, [1, 126, 1024]> input_71_cast_fp16 = layer_norm(axes = input_71_axes_0, beta = model_layers_0_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_0_norm_out_weight_to_fp16, x = input_69_cast_fp16)[name = tensor<string, []>("input_71_cast_fp16")];
+            tensor<int32, [1]> input_73_axes_0 = const()[name = tensor<string, []>("input_73_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_1_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_1_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57475584)))];
+            tensor<fp16, [1024]> model_layers_1_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_1_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57477696)))];
+            tensor<fp16, [1, 126, 1024]> input_73_cast_fp16 = layer_norm(axes = input_73_axes_0, beta = model_layers_1_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_1_norm_feed_forward1_weight_to_fp16, x = input_71_cast_fp16)[name = tensor<string, []>("input_73_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_1_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_1_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57479808)))];
+            tensor<fp16, [1, 126, 4096]> linear_10_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_1_feed_forward1_linear1_weight_to_fp16, x = input_73_cast_fp16)[name = tensor<string, []>("linear_10_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_77_cast_fp16 = silu(x = linear_10_cast_fp16)[name = tensor<string, []>("input_77_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_1_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_1_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65868480)))];
+            tensor<fp16, [1, 126, 1024]> linear_11_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_1_feed_forward1_linear2_weight_to_fp16, x = input_77_cast_fp16)[name = tensor<string, []>("linear_11_cast_fp16")];
+            tensor<fp16, []> var_390_to_fp16 = const()[name = tensor<string, []>("op_390_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_391_cast_fp16 = mul(x = linear_11_cast_fp16, y = var_390_to_fp16)[name = tensor<string, []>("op_391_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_83_cast_fp16 = add(x = input_71_cast_fp16, y = var_391_cast_fp16)[name = tensor<string, []>("input_83_cast_fp16")];
+            tensor<int32, [1]> query_3_axes_0 = const()[name = tensor<string, []>("query_3_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_1_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_1_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74257152)))];
+            tensor<fp16, [1024]> model_layers_1_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_1_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74259264)))];
+            tensor<fp16, [1, 126, 1024]> query_3_cast_fp16 = layer_norm(axes = query_3_axes_0, beta = model_layers_1_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_1_norm_self_att_weight_to_fp16, x = input_83_cast_fp16)[name = tensor<string, []>("query_3_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_1_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_1_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74261376)))];
+            tensor<fp16, [1, 126, 1024]> linear_12_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_1_self_attn_linear_q_weight_to_fp16, x = query_3_cast_fp16)[name = tensor<string, []>("linear_12_cast_fp16")];
+            tensor<int32, [4]> var_407 = const()[name = tensor<string, []>("op_407"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_7_cast_fp16 = reshape(shape = var_407, x = linear_12_cast_fp16)[name = tensor<string, []>("q_7_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_1_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_1_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76358592)))];
+            tensor<fp16, [1, 126, 1024]> linear_13_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_1_self_attn_linear_k_weight_to_fp16, x = query_3_cast_fp16)[name = tensor<string, []>("linear_13_cast_fp16")];
+            tensor<int32, [4]> var_411 = const()[name = tensor<string, []>("op_411"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_5_cast_fp16 = reshape(shape = var_411, x = linear_13_cast_fp16)[name = tensor<string, []>("k_5_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_1_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_1_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78455808)))];
+            tensor<fp16, [1, 126, 1024]> linear_14_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_1_self_attn_linear_v_weight_to_fp16, x = query_3_cast_fp16)[name = tensor<string, []>("linear_14_cast_fp16")];
+            tensor<int32, [4]> var_415 = const()[name = tensor<string, []>("op_415"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_3_cast_fp16 = reshape(shape = var_415, x = linear_14_cast_fp16)[name = tensor<string, []>("v_3_cast_fp16")];
+            tensor<int32, [4]> value_3_perm_0 = const()[name = tensor<string, []>("value_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_1_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_1_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80553024)))];
+            tensor<fp16, [1, 126, 8, 128]> var_427_cast_fp16 = add(x = q_7_cast_fp16, y = model_layers_1_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_427_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_1_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_1_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80555136)))];
+            tensor<fp16, [1, 126, 8, 128]> var_429_cast_fp16 = add(x = q_7_cast_fp16, y = model_layers_1_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_429_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_3_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_3_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_29_transpose_x_0 = const()[name = tensor<string, []>("x_29_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_29_transpose_y_0 = const()[name = tensor<string, []>("x_29_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_431_to_fp16 = const()[name = tensor<string, []>("op_431_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80557248)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_280 = transpose(perm = q_with_bias_v_3_perm_0, x = var_429_cast_fp16)[name = tensor<string, []>("transpose_280")];
+            tensor<fp16, [1, 8, 126, 251]> x_29_cast_fp16 = matmul(transpose_x = x_29_transpose_x_0, transpose_y = x_29_transpose_y_0, x = transpose_280, y = var_431_to_fp16)[name = tensor<string, []>("x_29_cast_fp16")];
+            tensor<int32, [8]> x_31_pad_0 = const()[name = tensor<string, []>("x_31_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_31_mode_0 = const()[name = tensor<string, []>("x_31_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_23_to_fp16 = const()[name = tensor<string, []>("const_23_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_31_cast_fp16 = pad(constant_val = const_23_to_fp16, mode = x_31_mode_0, pad = x_31_pad_0, x = x_29_cast_fp16)[name = tensor<string, []>("x_31_cast_fp16")];
+            tensor<int32, [4]> var_439 = const()[name = tensor<string, []>("op_439"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_33_cast_fp16 = reshape(shape = var_439, x = x_31_cast_fp16)[name = tensor<string, []>("x_33_cast_fp16")];
+            tensor<int32, [4]> var_443_begin_0 = const()[name = tensor<string, []>("op_443_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_443_end_0 = const()[name = tensor<string, []>("op_443_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_443_end_mask_0 = const()[name = tensor<string, []>("op_443_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_443_cast_fp16 = slice_by_index(begin = var_443_begin_0, end = var_443_end_0, end_mask = var_443_end_mask_0, x = x_33_cast_fp16)[name = tensor<string, []>("op_443_cast_fp16")];
+            tensor<int32, [4]> var_444 = const()[name = tensor<string, []>("op_444"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_5_cast_fp16 = reshape(shape = var_444, x = var_443_cast_fp16)[name = tensor<string, []>("matrix_bd_5_cast_fp16")];
+            tensor<bool, []> matrix_ac_3_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_3_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_3_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_3_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_74_perm_0 = const()[name = tensor<string, []>("transpose_74_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_75_perm_0 = const()[name = tensor<string, []>("transpose_75_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_278 = transpose(perm = transpose_75_perm_0, x = k_5_cast_fp16)[name = tensor<string, []>("transpose_278")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_279 = transpose(perm = transpose_74_perm_0, x = var_427_cast_fp16)[name = tensor<string, []>("transpose_279")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_3_cast_fp16 = matmul(transpose_x = matrix_ac_3_transpose_x_0, transpose_y = matrix_ac_3_transpose_y_0, x = transpose_279, y = transpose_278)[name = tensor<string, []>("matrix_ac_3_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_7_begin_0 = const()[name = tensor<string, []>("matrix_bd_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_7_end_0 = const()[name = tensor<string, []>("matrix_bd_7_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_7_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_7_cast_fp16 = slice_by_index(begin = matrix_bd_7_begin_0, end = matrix_bd_7_end_0, end_mask = matrix_bd_7_end_mask_0, x = matrix_bd_5_cast_fp16)[name = tensor<string, []>("matrix_bd_7_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_453_cast_fp16 = add(x = matrix_ac_3_cast_fp16, y = matrix_bd_7_cast_fp16)[name = tensor<string, []>("op_453_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_5_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_5_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_5_cast_fp16 = mul(x = var_453_cast_fp16, y = _inversed_scores_5_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_5_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_7_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_5_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_7_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_459_cast_fp16 = softmax(axis = var_25, x = scores_7_cast_fp16)[name = tensor<string, []>("op_459_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_85_cast_fp16 = select(a = var_6_to_fp16, b = var_459_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_85_cast_fp16")];
+            tensor<bool, []> x_35_transpose_x_0 = const()[name = tensor<string, []>("x_35_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_35_transpose_y_0 = const()[name = tensor<string, []>("x_35_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_281 = transpose(perm = value_3_perm_0, x = v_3_cast_fp16)[name = tensor<string, []>("transpose_281")];
+            tensor<fp16, [1, 8, 126, 128]> x_35_cast_fp16 = matmul(transpose_x = x_35_transpose_x_0, transpose_y = x_35_transpose_y_0, x = input_85_cast_fp16, y = transpose_281)[name = tensor<string, []>("x_35_cast_fp16")];
+            tensor<int32, [4]> var_463_perm_0 = const()[name = tensor<string, []>("op_463_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_464 = const()[name = tensor<string, []>("op_464"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_277 = transpose(perm = var_463_perm_0, x = x_35_cast_fp16)[name = tensor<string, []>("transpose_277")];
+            tensor<fp16, [1, 126, 1024]> input_87_cast_fp16 = reshape(shape = var_464, x = transpose_277)[name = tensor<string, []>("input_87_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_1_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_1_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81071360)))];
+            tensor<fp16, [1, 126, 1024]> linear_16_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_1_self_attn_linear_out_weight_to_fp16, x = input_87_cast_fp16)[name = tensor<string, []>("linear_16_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_91_cast_fp16 = add(x = input_83_cast_fp16, y = linear_16_cast_fp16)[name = tensor<string, []>("input_91_cast_fp16")];
+            tensor<int32, [1]> x_39_axes_0 = const()[name = tensor<string, []>("x_39_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_1_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_1_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83168576)))];
+            tensor<fp16, [1024]> model_layers_1_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_1_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83170688)))];
+            tensor<fp16, [1, 126, 1024]> x_39_cast_fp16 = layer_norm(axes = x_39_axes_0, beta = model_layers_1_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_1_norm_conv_weight_to_fp16, x = input_91_cast_fp16)[name = tensor<string, []>("x_39_cast_fp16")];
+            tensor<int32, [3]> input_93_perm_0 = const()[name = tensor<string, []>("input_93_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_480 = const()[name = tensor<string, []>("op_480"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_482 = const()[name = tensor<string, []>("op_482"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_95_pad_type_0 = const()[name = tensor<string, []>("input_95_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_95_pad_0 = const()[name = tensor<string, []>("input_95_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_1_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_1_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83172800)))];
+            tensor<fp16, [1, 1024, 126]> transpose_276 = transpose(perm = input_93_perm_0, x = x_39_cast_fp16)[name = tensor<string, []>("transpose_276")];
+            tensor<fp16, [1, 2048, 126]> input_95_cast_fp16 = conv(dilations = var_482, groups = var_27, pad = input_95_pad_0, pad_type = input_95_pad_type_0, strides = var_480, weight = model_layers_1_conv_pointwise_conv1_weight_to_fp16, x = transpose_276)[name = tensor<string, []>("input_95_cast_fp16")];
+            tensor<int32, []> x_41_split_num_splits_0 = const()[name = tensor<string, []>("x_41_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_41_split_axis_0 = const()[name = tensor<string, []>("x_41_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_41_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_41_split_cast_fp16_1 = split(axis = x_41_split_axis_0, num_splits = x_41_split_num_splits_0, x = input_95_cast_fp16)[name = tensor<string, []>("x_41_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_41_split_1_sigmoid_cast_fp16 = sigmoid(x = x_41_split_cast_fp16_1)[name = tensor<string, []>("x_41_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_41_cast_fp16 = mul(x = x_41_split_cast_fp16_0, y = x_41_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_41_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_97_cast_fp16 = select(a = var_6_to_fp16, b = x_41_cast_fp16, cond = var_323)[name = tensor<string, []>("input_97_cast_fp16")];
+            tensor<int32, [6]> input_99_pad_0 = const()[name = tensor<string, []>("input_99_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_99_mode_0 = const()[name = tensor<string, []>("input_99_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_26_to_fp16 = const()[name = tensor<string, []>("const_26_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_99_cast_fp16 = pad(constant_val = const_26_to_fp16, mode = input_99_mode_0, pad = input_99_pad_0, x = input_97_cast_fp16)[name = tensor<string, []>("input_99_cast_fp16")];
+            tensor<int32, [1]> var_491 = const()[name = tensor<string, []>("op_491"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_493 = const()[name = tensor<string, []>("op_493"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_101_pad_type_0 = const()[name = tensor<string, []>("input_101_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_101_pad_0 = const()[name = tensor<string, []>("input_101_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_103_weight_0_to_fp16 = const()[name = tensor<string, []>("input_103_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87367168)))];
+            tensor<fp16, [1024]> input_103_bias_0_to_fp16 = const()[name = tensor<string, []>("input_103_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87385664)))];
+            tensor<fp16, [1, 1024, 126]> input_103_cast_fp16 = conv(bias = input_103_bias_0_to_fp16, dilations = var_493, groups = var_5, pad = input_101_pad_0, pad_type = input_101_pad_type_0, strides = var_491, weight = input_103_weight_0_to_fp16, x = input_99_cast_fp16)[name = tensor<string, []>("input_103_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_105_cast_fp16 = silu(x = input_103_cast_fp16)[name = tensor<string, []>("input_105_cast_fp16")];
+            tensor<int32, [1]> var_503 = const()[name = tensor<string, []>("op_503"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_505 = const()[name = tensor<string, []>("op_505"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_43_pad_type_0 = const()[name = tensor<string, []>("x_43_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_43_pad_0 = const()[name = tensor<string, []>("x_43_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_1_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_1_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87387776)))];
+            tensor<fp16, [1, 1024, 126]> x_43_cast_fp16 = conv(dilations = var_505, groups = var_27, pad = x_43_pad_0, pad_type = x_43_pad_type_0, strides = var_503, weight = model_layers_1_conv_pointwise_conv2_weight_to_fp16, x = input_105_cast_fp16)[name = tensor<string, []>("x_43_cast_fp16")];
+            tensor<int32, [3]> input_107_perm_0 = const()[name = tensor<string, []>("input_107_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_275 = transpose(perm = input_107_perm_0, x = x_43_cast_fp16)[name = tensor<string, []>("transpose_275")];
+            tensor<fp16, [1, 126, 1024]> input_109_cast_fp16 = add(x = input_91_cast_fp16, y = transpose_275)[name = tensor<string, []>("input_109_cast_fp16")];
+            tensor<int32, [1]> input_111_axes_0 = const()[name = tensor<string, []>("input_111_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_1_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_1_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89484992)))];
+            tensor<fp16, [1024]> model_layers_1_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_1_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89487104)))];
+            tensor<fp16, [1, 126, 1024]> input_111_cast_fp16 = layer_norm(axes = input_111_axes_0, beta = model_layers_1_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_1_norm_feed_forward2_weight_to_fp16, x = input_109_cast_fp16)[name = tensor<string, []>("input_111_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_1_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_1_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89489216)))];
+            tensor<fp16, [1, 126, 4096]> linear_17_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_1_feed_forward2_linear1_weight_to_fp16, x = input_111_cast_fp16)[name = tensor<string, []>("linear_17_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_115_cast_fp16 = silu(x = linear_17_cast_fp16)[name = tensor<string, []>("input_115_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_1_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_1_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97877888)))];
+            tensor<fp16, [1, 126, 1024]> linear_18_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_1_feed_forward2_linear2_weight_to_fp16, x = input_115_cast_fp16)[name = tensor<string, []>("linear_18_cast_fp16")];
+            tensor<fp16, []> var_524_to_fp16 = const()[name = tensor<string, []>("op_524_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_525_cast_fp16 = mul(x = linear_18_cast_fp16, y = var_524_to_fp16)[name = tensor<string, []>("op_525_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_121_cast_fp16 = add(x = input_109_cast_fp16, y = var_525_cast_fp16)[name = tensor<string, []>("input_121_cast_fp16")];
+            tensor<int32, [1]> input_123_axes_0 = const()[name = tensor<string, []>("input_123_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_1_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_1_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106266560)))];
+            tensor<fp16, [1024]> model_layers_1_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_1_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106268672)))];
+            tensor<fp16, [1, 126, 1024]> input_123_cast_fp16 = layer_norm(axes = input_123_axes_0, beta = model_layers_1_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_1_norm_out_weight_to_fp16, x = input_121_cast_fp16)[name = tensor<string, []>("input_123_cast_fp16")];
+            tensor<int32, [1]> input_125_axes_0 = const()[name = tensor<string, []>("input_125_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_2_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_2_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106270784)))];
+            tensor<fp16, [1024]> model_layers_2_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_2_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106272896)))];
+            tensor<fp16, [1, 126, 1024]> input_125_cast_fp16 = layer_norm(axes = input_125_axes_0, beta = model_layers_2_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_2_norm_feed_forward1_weight_to_fp16, x = input_123_cast_fp16)[name = tensor<string, []>("input_125_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_2_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_2_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106275008)))];
+            tensor<fp16, [1, 126, 4096]> linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_2_feed_forward1_linear1_weight_to_fp16, x = input_125_cast_fp16)[name = tensor<string, []>("linear_19_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_129_cast_fp16 = silu(x = linear_19_cast_fp16)[name = tensor<string, []>("input_129_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_2_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_2_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114663680)))];
+            tensor<fp16, [1, 126, 1024]> linear_20_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_2_feed_forward1_linear2_weight_to_fp16, x = input_129_cast_fp16)[name = tensor<string, []>("linear_20_cast_fp16")];
+            tensor<fp16, []> var_553_to_fp16 = const()[name = tensor<string, []>("op_553_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_554_cast_fp16 = mul(x = linear_20_cast_fp16, y = var_553_to_fp16)[name = tensor<string, []>("op_554_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_135_cast_fp16 = add(x = input_123_cast_fp16, y = var_554_cast_fp16)[name = tensor<string, []>("input_135_cast_fp16")];
+            tensor<int32, [1]> query_5_axes_0 = const()[name = tensor<string, []>("query_5_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_2_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_2_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123052352)))];
+            tensor<fp16, [1024]> model_layers_2_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_2_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123054464)))];
+            tensor<fp16, [1, 126, 1024]> query_5_cast_fp16 = layer_norm(axes = query_5_axes_0, beta = model_layers_2_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_2_norm_self_att_weight_to_fp16, x = input_135_cast_fp16)[name = tensor<string, []>("query_5_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_2_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_2_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123056576)))];
+            tensor<fp16, [1, 126, 1024]> linear_21_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_2_self_attn_linear_q_weight_to_fp16, x = query_5_cast_fp16)[name = tensor<string, []>("linear_21_cast_fp16")];
+            tensor<int32, [4]> var_570 = const()[name = tensor<string, []>("op_570"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_13_cast_fp16 = reshape(shape = var_570, x = linear_21_cast_fp16)[name = tensor<string, []>("q_13_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_2_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_2_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(125153792)))];
+            tensor<fp16, [1, 126, 1024]> linear_22_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_2_self_attn_linear_k_weight_to_fp16, x = query_5_cast_fp16)[name = tensor<string, []>("linear_22_cast_fp16")];
+            tensor<int32, [4]> var_574 = const()[name = tensor<string, []>("op_574"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_9_cast_fp16 = reshape(shape = var_574, x = linear_22_cast_fp16)[name = tensor<string, []>("k_9_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_2_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_2_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(127251008)))];
+            tensor<fp16, [1, 126, 1024]> linear_23_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_2_self_attn_linear_v_weight_to_fp16, x = query_5_cast_fp16)[name = tensor<string, []>("linear_23_cast_fp16")];
+            tensor<int32, [4]> var_578 = const()[name = tensor<string, []>("op_578"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_5_cast_fp16 = reshape(shape = var_578, x = linear_23_cast_fp16)[name = tensor<string, []>("v_5_cast_fp16")];
+            tensor<int32, [4]> value_5_perm_0 = const()[name = tensor<string, []>("value_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_2_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_2_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129348224)))];
+            tensor<fp16, [1, 126, 8, 128]> var_590_cast_fp16 = add(x = q_13_cast_fp16, y = model_layers_2_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_590_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_2_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_2_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129350336)))];
+            tensor<fp16, [1, 126, 8, 128]> var_592_cast_fp16 = add(x = q_13_cast_fp16, y = model_layers_2_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_592_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_5_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_5_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_51_transpose_x_0 = const()[name = tensor<string, []>("x_51_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_51_transpose_y_0 = const()[name = tensor<string, []>("x_51_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_594_to_fp16 = const()[name = tensor<string, []>("op_594_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129352448)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_273 = transpose(perm = q_with_bias_v_5_perm_0, x = var_592_cast_fp16)[name = tensor<string, []>("transpose_273")];
+            tensor<fp16, [1, 8, 126, 251]> x_51_cast_fp16 = matmul(transpose_x = x_51_transpose_x_0, transpose_y = x_51_transpose_y_0, x = transpose_273, y = var_594_to_fp16)[name = tensor<string, []>("x_51_cast_fp16")];
+            tensor<int32, [8]> x_53_pad_0 = const()[name = tensor<string, []>("x_53_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_53_mode_0 = const()[name = tensor<string, []>("x_53_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_33_to_fp16 = const()[name = tensor<string, []>("const_33_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_53_cast_fp16 = pad(constant_val = const_33_to_fp16, mode = x_53_mode_0, pad = x_53_pad_0, x = x_51_cast_fp16)[name = tensor<string, []>("x_53_cast_fp16")];
+            tensor<int32, [4]> var_602 = const()[name = tensor<string, []>("op_602"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_55_cast_fp16 = reshape(shape = var_602, x = x_53_cast_fp16)[name = tensor<string, []>("x_55_cast_fp16")];
+            tensor<int32, [4]> var_606_begin_0 = const()[name = tensor<string, []>("op_606_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_606_end_0 = const()[name = tensor<string, []>("op_606_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_606_end_mask_0 = const()[name = tensor<string, []>("op_606_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_606_cast_fp16 = slice_by_index(begin = var_606_begin_0, end = var_606_end_0, end_mask = var_606_end_mask_0, x = x_55_cast_fp16)[name = tensor<string, []>("op_606_cast_fp16")];
+            tensor<int32, [4]> var_607 = const()[name = tensor<string, []>("op_607"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_9_cast_fp16 = reshape(shape = var_607, x = var_606_cast_fp16)[name = tensor<string, []>("matrix_bd_9_cast_fp16")];
+            tensor<bool, []> matrix_ac_5_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_5_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_5_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_5_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_76_perm_0 = const()[name = tensor<string, []>("transpose_76_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_77_perm_0 = const()[name = tensor<string, []>("transpose_77_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_271 = transpose(perm = transpose_77_perm_0, x = k_9_cast_fp16)[name = tensor<string, []>("transpose_271")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_272 = transpose(perm = transpose_76_perm_0, x = var_590_cast_fp16)[name = tensor<string, []>("transpose_272")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_5_cast_fp16 = matmul(transpose_x = matrix_ac_5_transpose_x_0, transpose_y = matrix_ac_5_transpose_y_0, x = transpose_272, y = transpose_271)[name = tensor<string, []>("matrix_ac_5_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_11_begin_0 = const()[name = tensor<string, []>("matrix_bd_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_11_end_0 = const()[name = tensor<string, []>("matrix_bd_11_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_11_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_11_cast_fp16 = slice_by_index(begin = matrix_bd_11_begin_0, end = matrix_bd_11_end_0, end_mask = matrix_bd_11_end_mask_0, x = matrix_bd_9_cast_fp16)[name = tensor<string, []>("matrix_bd_11_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_616_cast_fp16 = add(x = matrix_ac_5_cast_fp16, y = matrix_bd_11_cast_fp16)[name = tensor<string, []>("op_616_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_9_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_9_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_9_cast_fp16 = mul(x = var_616_cast_fp16, y = _inversed_scores_9_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_9_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_11_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_9_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_11_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_622_cast_fp16 = softmax(axis = var_25, x = scores_11_cast_fp16)[name = tensor<string, []>("op_622_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_137_cast_fp16 = select(a = var_6_to_fp16, b = var_622_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_137_cast_fp16")];
+            tensor<bool, []> x_57_transpose_x_0 = const()[name = tensor<string, []>("x_57_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_57_transpose_y_0 = const()[name = tensor<string, []>("x_57_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_274 = transpose(perm = value_5_perm_0, x = v_5_cast_fp16)[name = tensor<string, []>("transpose_274")];
+            tensor<fp16, [1, 8, 126, 128]> x_57_cast_fp16 = matmul(transpose_x = x_57_transpose_x_0, transpose_y = x_57_transpose_y_0, x = input_137_cast_fp16, y = transpose_274)[name = tensor<string, []>("x_57_cast_fp16")];
+            tensor<int32, [4]> var_626_perm_0 = const()[name = tensor<string, []>("op_626_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_627 = const()[name = tensor<string, []>("op_627"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_270 = transpose(perm = var_626_perm_0, x = x_57_cast_fp16)[name = tensor<string, []>("transpose_270")];
+            tensor<fp16, [1, 126, 1024]> input_139_cast_fp16 = reshape(shape = var_627, x = transpose_270)[name = tensor<string, []>("input_139_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_2_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_2_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129866560)))];
+            tensor<fp16, [1, 126, 1024]> linear_25_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_2_self_attn_linear_out_weight_to_fp16, x = input_139_cast_fp16)[name = tensor<string, []>("linear_25_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_143_cast_fp16 = add(x = input_135_cast_fp16, y = linear_25_cast_fp16)[name = tensor<string, []>("input_143_cast_fp16")];
+            tensor<int32, [1]> x_61_axes_0 = const()[name = tensor<string, []>("x_61_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_2_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_2_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131963776)))];
+            tensor<fp16, [1024]> model_layers_2_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_2_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131965888)))];
+            tensor<fp16, [1, 126, 1024]> x_61_cast_fp16 = layer_norm(axes = x_61_axes_0, beta = model_layers_2_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_2_norm_conv_weight_to_fp16, x = input_143_cast_fp16)[name = tensor<string, []>("x_61_cast_fp16")];
+            tensor<int32, [3]> input_145_perm_0 = const()[name = tensor<string, []>("input_145_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_643 = const()[name = tensor<string, []>("op_643"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_645 = const()[name = tensor<string, []>("op_645"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_147_pad_type_0 = const()[name = tensor<string, []>("input_147_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_147_pad_0 = const()[name = tensor<string, []>("input_147_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_2_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_2_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(131968000)))];
+            tensor<fp16, [1, 1024, 126]> transpose_269 = transpose(perm = input_145_perm_0, x = x_61_cast_fp16)[name = tensor<string, []>("transpose_269")];
+            tensor<fp16, [1, 2048, 126]> input_147_cast_fp16 = conv(dilations = var_645, groups = var_27, pad = input_147_pad_0, pad_type = input_147_pad_type_0, strides = var_643, weight = model_layers_2_conv_pointwise_conv1_weight_to_fp16, x = transpose_269)[name = tensor<string, []>("input_147_cast_fp16")];
+            tensor<int32, []> x_63_split_num_splits_0 = const()[name = tensor<string, []>("x_63_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_63_split_axis_0 = const()[name = tensor<string, []>("x_63_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_63_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_63_split_cast_fp16_1 = split(axis = x_63_split_axis_0, num_splits = x_63_split_num_splits_0, x = input_147_cast_fp16)[name = tensor<string, []>("x_63_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_63_split_1_sigmoid_cast_fp16 = sigmoid(x = x_63_split_cast_fp16_1)[name = tensor<string, []>("x_63_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_63_cast_fp16 = mul(x = x_63_split_cast_fp16_0, y = x_63_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_63_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_149_cast_fp16 = select(a = var_6_to_fp16, b = x_63_cast_fp16, cond = var_323)[name = tensor<string, []>("input_149_cast_fp16")];
+            tensor<int32, [6]> input_151_pad_0 = const()[name = tensor<string, []>("input_151_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_151_mode_0 = const()[name = tensor<string, []>("input_151_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_36_to_fp16 = const()[name = tensor<string, []>("const_36_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_151_cast_fp16 = pad(constant_val = const_36_to_fp16, mode = input_151_mode_0, pad = input_151_pad_0, x = input_149_cast_fp16)[name = tensor<string, []>("input_151_cast_fp16")];
+            tensor<int32, [1]> var_654 = const()[name = tensor<string, []>("op_654"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_656 = const()[name = tensor<string, []>("op_656"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_153_pad_type_0 = const()[name = tensor<string, []>("input_153_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_153_pad_0 = const()[name = tensor<string, []>("input_153_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_155_weight_0_to_fp16 = const()[name = tensor<string, []>("input_155_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136162368)))];
+            tensor<fp16, [1024]> input_155_bias_0_to_fp16 = const()[name = tensor<string, []>("input_155_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136180864)))];
+            tensor<fp16, [1, 1024, 126]> input_155_cast_fp16 = conv(bias = input_155_bias_0_to_fp16, dilations = var_656, groups = var_5, pad = input_153_pad_0, pad_type = input_153_pad_type_0, strides = var_654, weight = input_155_weight_0_to_fp16, x = input_151_cast_fp16)[name = tensor<string, []>("input_155_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_157_cast_fp16 = silu(x = input_155_cast_fp16)[name = tensor<string, []>("input_157_cast_fp16")];
+            tensor<int32, [1]> var_666 = const()[name = tensor<string, []>("op_666"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_668 = const()[name = tensor<string, []>("op_668"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_65_pad_type_0 = const()[name = tensor<string, []>("x_65_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_65_pad_0 = const()[name = tensor<string, []>("x_65_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_2_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_2_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136182976)))];
+            tensor<fp16, [1, 1024, 126]> x_65_cast_fp16 = conv(dilations = var_668, groups = var_27, pad = x_65_pad_0, pad_type = x_65_pad_type_0, strides = var_666, weight = model_layers_2_conv_pointwise_conv2_weight_to_fp16, x = input_157_cast_fp16)[name = tensor<string, []>("x_65_cast_fp16")];
+            tensor<int32, [3]> input_159_perm_0 = const()[name = tensor<string, []>("input_159_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_268 = transpose(perm = input_159_perm_0, x = x_65_cast_fp16)[name = tensor<string, []>("transpose_268")];
+            tensor<fp16, [1, 126, 1024]> input_161_cast_fp16 = add(x = input_143_cast_fp16, y = transpose_268)[name = tensor<string, []>("input_161_cast_fp16")];
+            tensor<int32, [1]> input_163_axes_0 = const()[name = tensor<string, []>("input_163_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_2_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_2_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138280192)))];
+            tensor<fp16, [1024]> model_layers_2_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_2_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138282304)))];
+            tensor<fp16, [1, 126, 1024]> input_163_cast_fp16 = layer_norm(axes = input_163_axes_0, beta = model_layers_2_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_2_norm_feed_forward2_weight_to_fp16, x = input_161_cast_fp16)[name = tensor<string, []>("input_163_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_2_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_2_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138284416)))];
+            tensor<fp16, [1, 126, 4096]> linear_26_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_2_feed_forward2_linear1_weight_to_fp16, x = input_163_cast_fp16)[name = tensor<string, []>("linear_26_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_167_cast_fp16 = silu(x = linear_26_cast_fp16)[name = tensor<string, []>("input_167_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_2_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_2_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(146673088)))];
+            tensor<fp16, [1, 126, 1024]> linear_27_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_2_feed_forward2_linear2_weight_to_fp16, x = input_167_cast_fp16)[name = tensor<string, []>("linear_27_cast_fp16")];
+            tensor<fp16, []> var_687_to_fp16 = const()[name = tensor<string, []>("op_687_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_688_cast_fp16 = mul(x = linear_27_cast_fp16, y = var_687_to_fp16)[name = tensor<string, []>("op_688_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_173_cast_fp16 = add(x = input_161_cast_fp16, y = var_688_cast_fp16)[name = tensor<string, []>("input_173_cast_fp16")];
+            tensor<int32, [1]> input_175_axes_0 = const()[name = tensor<string, []>("input_175_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_2_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_2_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155061760)))];
+            tensor<fp16, [1024]> model_layers_2_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_2_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155063872)))];
+            tensor<fp16, [1, 126, 1024]> input_175_cast_fp16 = layer_norm(axes = input_175_axes_0, beta = model_layers_2_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_2_norm_out_weight_to_fp16, x = input_173_cast_fp16)[name = tensor<string, []>("input_175_cast_fp16")];
+            tensor<int32, [1]> input_177_axes_0 = const()[name = tensor<string, []>("input_177_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_3_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_3_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155065984)))];
+            tensor<fp16, [1024]> model_layers_3_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_3_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155068096)))];
+            tensor<fp16, [1, 126, 1024]> input_177_cast_fp16 = layer_norm(axes = input_177_axes_0, beta = model_layers_3_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_3_norm_feed_forward1_weight_to_fp16, x = input_175_cast_fp16)[name = tensor<string, []>("input_177_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_3_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_3_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(155070208)))];
+            tensor<fp16, [1, 126, 4096]> linear_28_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_3_feed_forward1_linear1_weight_to_fp16, x = input_177_cast_fp16)[name = tensor<string, []>("linear_28_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_181_cast_fp16 = silu(x = linear_28_cast_fp16)[name = tensor<string, []>("input_181_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_3_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_3_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163458880)))];
+            tensor<fp16, [1, 126, 1024]> linear_29_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_3_feed_forward1_linear2_weight_to_fp16, x = input_181_cast_fp16)[name = tensor<string, []>("linear_29_cast_fp16")];
+            tensor<fp16, []> var_716_to_fp16 = const()[name = tensor<string, []>("op_716_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_717_cast_fp16 = mul(x = linear_29_cast_fp16, y = var_716_to_fp16)[name = tensor<string, []>("op_717_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_187_cast_fp16 = add(x = input_175_cast_fp16, y = var_717_cast_fp16)[name = tensor<string, []>("input_187_cast_fp16")];
+            tensor<int32, [1]> query_7_axes_0 = const()[name = tensor<string, []>("query_7_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_3_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_3_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(171847552)))];
+            tensor<fp16, [1024]> model_layers_3_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_3_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(171849664)))];
+            tensor<fp16, [1, 126, 1024]> query_7_cast_fp16 = layer_norm(axes = query_7_axes_0, beta = model_layers_3_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_3_norm_self_att_weight_to_fp16, x = input_187_cast_fp16)[name = tensor<string, []>("query_7_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_3_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_3_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(171851776)))];
+            tensor<fp16, [1, 126, 1024]> linear_30_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_3_self_attn_linear_q_weight_to_fp16, x = query_7_cast_fp16)[name = tensor<string, []>("linear_30_cast_fp16")];
+            tensor<int32, [4]> var_733 = const()[name = tensor<string, []>("op_733"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_19_cast_fp16 = reshape(shape = var_733, x = linear_30_cast_fp16)[name = tensor<string, []>("q_19_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_3_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_3_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(173948992)))];
+            tensor<fp16, [1, 126, 1024]> linear_31_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_3_self_attn_linear_k_weight_to_fp16, x = query_7_cast_fp16)[name = tensor<string, []>("linear_31_cast_fp16")];
+            tensor<int32, [4]> var_737 = const()[name = tensor<string, []>("op_737"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_13_cast_fp16 = reshape(shape = var_737, x = linear_31_cast_fp16)[name = tensor<string, []>("k_13_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_3_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_3_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(176046208)))];
+            tensor<fp16, [1, 126, 1024]> linear_32_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_3_self_attn_linear_v_weight_to_fp16, x = query_7_cast_fp16)[name = tensor<string, []>("linear_32_cast_fp16")];
+            tensor<int32, [4]> var_741 = const()[name = tensor<string, []>("op_741"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_7_cast_fp16 = reshape(shape = var_741, x = linear_32_cast_fp16)[name = tensor<string, []>("v_7_cast_fp16")];
+            tensor<int32, [4]> value_7_perm_0 = const()[name = tensor<string, []>("value_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_3_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_3_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178143424)))];
+            tensor<fp16, [1, 126, 8, 128]> var_753_cast_fp16 = add(x = q_19_cast_fp16, y = model_layers_3_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_753_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_3_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_3_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178145536)))];
+            tensor<fp16, [1, 126, 8, 128]> var_755_cast_fp16 = add(x = q_19_cast_fp16, y = model_layers_3_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_755_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_7_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_7_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_73_transpose_x_0 = const()[name = tensor<string, []>("x_73_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_73_transpose_y_0 = const()[name = tensor<string, []>("x_73_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_757_to_fp16 = const()[name = tensor<string, []>("op_757_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178147648)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_266 = transpose(perm = q_with_bias_v_7_perm_0, x = var_755_cast_fp16)[name = tensor<string, []>("transpose_266")];
+            tensor<fp16, [1, 8, 126, 251]> x_73_cast_fp16 = matmul(transpose_x = x_73_transpose_x_0, transpose_y = x_73_transpose_y_0, x = transpose_266, y = var_757_to_fp16)[name = tensor<string, []>("x_73_cast_fp16")];
+            tensor<int32, [8]> x_75_pad_0 = const()[name = tensor<string, []>("x_75_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_75_mode_0 = const()[name = tensor<string, []>("x_75_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_43_to_fp16 = const()[name = tensor<string, []>("const_43_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_75_cast_fp16 = pad(constant_val = const_43_to_fp16, mode = x_75_mode_0, pad = x_75_pad_0, x = x_73_cast_fp16)[name = tensor<string, []>("x_75_cast_fp16")];
+            tensor<int32, [4]> var_765 = const()[name = tensor<string, []>("op_765"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_77_cast_fp16 = reshape(shape = var_765, x = x_75_cast_fp16)[name = tensor<string, []>("x_77_cast_fp16")];
+            tensor<int32, [4]> var_769_begin_0 = const()[name = tensor<string, []>("op_769_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_769_end_0 = const()[name = tensor<string, []>("op_769_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_769_end_mask_0 = const()[name = tensor<string, []>("op_769_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_769_cast_fp16 = slice_by_index(begin = var_769_begin_0, end = var_769_end_0, end_mask = var_769_end_mask_0, x = x_77_cast_fp16)[name = tensor<string, []>("op_769_cast_fp16")];
+            tensor<int32, [4]> var_770 = const()[name = tensor<string, []>("op_770"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_13_cast_fp16 = reshape(shape = var_770, x = var_769_cast_fp16)[name = tensor<string, []>("matrix_bd_13_cast_fp16")];
+            tensor<bool, []> matrix_ac_7_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_7_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_7_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_7_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_78_perm_0 = const()[name = tensor<string, []>("transpose_78_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_79_perm_0 = const()[name = tensor<string, []>("transpose_79_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_264 = transpose(perm = transpose_79_perm_0, x = k_13_cast_fp16)[name = tensor<string, []>("transpose_264")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_265 = transpose(perm = transpose_78_perm_0, x = var_753_cast_fp16)[name = tensor<string, []>("transpose_265")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_7_cast_fp16 = matmul(transpose_x = matrix_ac_7_transpose_x_0, transpose_y = matrix_ac_7_transpose_y_0, x = transpose_265, y = transpose_264)[name = tensor<string, []>("matrix_ac_7_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_15_begin_0 = const()[name = tensor<string, []>("matrix_bd_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_15_end_0 = const()[name = tensor<string, []>("matrix_bd_15_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_15_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_15_cast_fp16 = slice_by_index(begin = matrix_bd_15_begin_0, end = matrix_bd_15_end_0, end_mask = matrix_bd_15_end_mask_0, x = matrix_bd_13_cast_fp16)[name = tensor<string, []>("matrix_bd_15_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_779_cast_fp16 = add(x = matrix_ac_7_cast_fp16, y = matrix_bd_15_cast_fp16)[name = tensor<string, []>("op_779_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_13_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_13_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_13_cast_fp16 = mul(x = var_779_cast_fp16, y = _inversed_scores_13_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_13_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_15_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_13_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_15_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_785_cast_fp16 = softmax(axis = var_25, x = scores_15_cast_fp16)[name = tensor<string, []>("op_785_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_189_cast_fp16 = select(a = var_6_to_fp16, b = var_785_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_189_cast_fp16")];
+            tensor<bool, []> x_79_transpose_x_0 = const()[name = tensor<string, []>("x_79_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_79_transpose_y_0 = const()[name = tensor<string, []>("x_79_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_267 = transpose(perm = value_7_perm_0, x = v_7_cast_fp16)[name = tensor<string, []>("transpose_267")];
+            tensor<fp16, [1, 8, 126, 128]> x_79_cast_fp16 = matmul(transpose_x = x_79_transpose_x_0, transpose_y = x_79_transpose_y_0, x = input_189_cast_fp16, y = transpose_267)[name = tensor<string, []>("x_79_cast_fp16")];
+            tensor<int32, [4]> var_789_perm_0 = const()[name = tensor<string, []>("op_789_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_790 = const()[name = tensor<string, []>("op_790"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_263 = transpose(perm = var_789_perm_0, x = x_79_cast_fp16)[name = tensor<string, []>("transpose_263")];
+            tensor<fp16, [1, 126, 1024]> input_191_cast_fp16 = reshape(shape = var_790, x = transpose_263)[name = tensor<string, []>("input_191_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_3_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_3_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178661760)))];
+            tensor<fp16, [1, 126, 1024]> linear_34_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_3_self_attn_linear_out_weight_to_fp16, x = input_191_cast_fp16)[name = tensor<string, []>("linear_34_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_195_cast_fp16 = add(x = input_187_cast_fp16, y = linear_34_cast_fp16)[name = tensor<string, []>("input_195_cast_fp16")];
+            tensor<int32, [1]> x_83_axes_0 = const()[name = tensor<string, []>("x_83_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_3_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_3_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(180758976)))];
+            tensor<fp16, [1024]> model_layers_3_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_3_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(180761088)))];
+            tensor<fp16, [1, 126, 1024]> x_83_cast_fp16 = layer_norm(axes = x_83_axes_0, beta = model_layers_3_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_3_norm_conv_weight_to_fp16, x = input_195_cast_fp16)[name = tensor<string, []>("x_83_cast_fp16")];
+            tensor<int32, [3]> input_197_perm_0 = const()[name = tensor<string, []>("input_197_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_806 = const()[name = tensor<string, []>("op_806"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_808 = const()[name = tensor<string, []>("op_808"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_199_pad_type_0 = const()[name = tensor<string, []>("input_199_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_199_pad_0 = const()[name = tensor<string, []>("input_199_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_3_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_3_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(180763200)))];
+            tensor<fp16, [1, 1024, 126]> transpose_262 = transpose(perm = input_197_perm_0, x = x_83_cast_fp16)[name = tensor<string, []>("transpose_262")];
+            tensor<fp16, [1, 2048, 126]> input_199_cast_fp16 = conv(dilations = var_808, groups = var_27, pad = input_199_pad_0, pad_type = input_199_pad_type_0, strides = var_806, weight = model_layers_3_conv_pointwise_conv1_weight_to_fp16, x = transpose_262)[name = tensor<string, []>("input_199_cast_fp16")];
+            tensor<int32, []> x_85_split_num_splits_0 = const()[name = tensor<string, []>("x_85_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_85_split_axis_0 = const()[name = tensor<string, []>("x_85_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_85_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_85_split_cast_fp16_1 = split(axis = x_85_split_axis_0, num_splits = x_85_split_num_splits_0, x = input_199_cast_fp16)[name = tensor<string, []>("x_85_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_85_split_1_sigmoid_cast_fp16 = sigmoid(x = x_85_split_cast_fp16_1)[name = tensor<string, []>("x_85_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_85_cast_fp16 = mul(x = x_85_split_cast_fp16_0, y = x_85_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_85_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_201_cast_fp16 = select(a = var_6_to_fp16, b = x_85_cast_fp16, cond = var_323)[name = tensor<string, []>("input_201_cast_fp16")];
+            tensor<int32, [6]> input_203_pad_0 = const()[name = tensor<string, []>("input_203_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_203_mode_0 = const()[name = tensor<string, []>("input_203_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_46_to_fp16 = const()[name = tensor<string, []>("const_46_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_203_cast_fp16 = pad(constant_val = const_46_to_fp16, mode = input_203_mode_0, pad = input_203_pad_0, x = input_201_cast_fp16)[name = tensor<string, []>("input_203_cast_fp16")];
+            tensor<int32, [1]> var_817 = const()[name = tensor<string, []>("op_817"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_819 = const()[name = tensor<string, []>("op_819"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_205_pad_type_0 = const()[name = tensor<string, []>("input_205_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_205_pad_0 = const()[name = tensor<string, []>("input_205_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_207_weight_0_to_fp16 = const()[name = tensor<string, []>("input_207_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(184957568)))];
+            tensor<fp16, [1024]> input_207_bias_0_to_fp16 = const()[name = tensor<string, []>("input_207_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(184976064)))];
+            tensor<fp16, [1, 1024, 126]> input_207_cast_fp16 = conv(bias = input_207_bias_0_to_fp16, dilations = var_819, groups = var_5, pad = input_205_pad_0, pad_type = input_205_pad_type_0, strides = var_817, weight = input_207_weight_0_to_fp16, x = input_203_cast_fp16)[name = tensor<string, []>("input_207_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_209_cast_fp16 = silu(x = input_207_cast_fp16)[name = tensor<string, []>("input_209_cast_fp16")];
+            tensor<int32, [1]> var_829 = const()[name = tensor<string, []>("op_829"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_831 = const()[name = tensor<string, []>("op_831"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_87_pad_type_0 = const()[name = tensor<string, []>("x_87_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_87_pad_0 = const()[name = tensor<string, []>("x_87_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_3_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_3_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(184978176)))];
+            tensor<fp16, [1, 1024, 126]> x_87_cast_fp16 = conv(dilations = var_831, groups = var_27, pad = x_87_pad_0, pad_type = x_87_pad_type_0, strides = var_829, weight = model_layers_3_conv_pointwise_conv2_weight_to_fp16, x = input_209_cast_fp16)[name = tensor<string, []>("x_87_cast_fp16")];
+            tensor<int32, [3]> input_211_perm_0 = const()[name = tensor<string, []>("input_211_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_261 = transpose(perm = input_211_perm_0, x = x_87_cast_fp16)[name = tensor<string, []>("transpose_261")];
+            tensor<fp16, [1, 126, 1024]> input_213_cast_fp16 = add(x = input_195_cast_fp16, y = transpose_261)[name = tensor<string, []>("input_213_cast_fp16")];
+            tensor<int32, [1]> input_215_axes_0 = const()[name = tensor<string, []>("input_215_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_3_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_3_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187075392)))];
+            tensor<fp16, [1024]> model_layers_3_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_3_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187077504)))];
+            tensor<fp16, [1, 126, 1024]> input_215_cast_fp16 = layer_norm(axes = input_215_axes_0, beta = model_layers_3_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_3_norm_feed_forward2_weight_to_fp16, x = input_213_cast_fp16)[name = tensor<string, []>("input_215_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_3_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_3_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(187079616)))];
+            tensor<fp16, [1, 126, 4096]> linear_35_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_3_feed_forward2_linear1_weight_to_fp16, x = input_215_cast_fp16)[name = tensor<string, []>("linear_35_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_219_cast_fp16 = silu(x = linear_35_cast_fp16)[name = tensor<string, []>("input_219_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_3_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_3_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(195468288)))];
+            tensor<fp16, [1, 126, 1024]> linear_36_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_3_feed_forward2_linear2_weight_to_fp16, x = input_219_cast_fp16)[name = tensor<string, []>("linear_36_cast_fp16")];
+            tensor<fp16, []> var_850_to_fp16 = const()[name = tensor<string, []>("op_850_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_851_cast_fp16 = mul(x = linear_36_cast_fp16, y = var_850_to_fp16)[name = tensor<string, []>("op_851_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_225_cast_fp16 = add(x = input_213_cast_fp16, y = var_851_cast_fp16)[name = tensor<string, []>("input_225_cast_fp16")];
+            tensor<int32, [1]> input_227_axes_0 = const()[name = tensor<string, []>("input_227_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_3_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_3_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203856960)))];
+            tensor<fp16, [1024]> model_layers_3_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_3_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203859072)))];
+            tensor<fp16, [1, 126, 1024]> input_227_cast_fp16 = layer_norm(axes = input_227_axes_0, beta = model_layers_3_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_3_norm_out_weight_to_fp16, x = input_225_cast_fp16)[name = tensor<string, []>("input_227_cast_fp16")];
+            tensor<int32, [1]> input_229_axes_0 = const()[name = tensor<string, []>("input_229_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_4_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_4_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203861184)))];
+            tensor<fp16, [1024]> model_layers_4_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_4_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203863296)))];
+            tensor<fp16, [1, 126, 1024]> input_229_cast_fp16 = layer_norm(axes = input_229_axes_0, beta = model_layers_4_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_4_norm_feed_forward1_weight_to_fp16, x = input_227_cast_fp16)[name = tensor<string, []>("input_229_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_4_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_4_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(203865408)))];
+            tensor<fp16, [1, 126, 4096]> linear_37_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_4_feed_forward1_linear1_weight_to_fp16, x = input_229_cast_fp16)[name = tensor<string, []>("linear_37_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_233_cast_fp16 = silu(x = linear_37_cast_fp16)[name = tensor<string, []>("input_233_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_4_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_4_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(212254080)))];
+            tensor<fp16, [1, 126, 1024]> linear_38_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_4_feed_forward1_linear2_weight_to_fp16, x = input_233_cast_fp16)[name = tensor<string, []>("linear_38_cast_fp16")];
+            tensor<fp16, []> var_879_to_fp16 = const()[name = tensor<string, []>("op_879_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_880_cast_fp16 = mul(x = linear_38_cast_fp16, y = var_879_to_fp16)[name = tensor<string, []>("op_880_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_239_cast_fp16 = add(x = input_227_cast_fp16, y = var_880_cast_fp16)[name = tensor<string, []>("input_239_cast_fp16")];
+            tensor<int32, [1]> query_9_axes_0 = const()[name = tensor<string, []>("query_9_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_4_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_4_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220642752)))];
+            tensor<fp16, [1024]> model_layers_4_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_4_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220644864)))];
+            tensor<fp16, [1, 126, 1024]> query_9_cast_fp16 = layer_norm(axes = query_9_axes_0, beta = model_layers_4_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_4_norm_self_att_weight_to_fp16, x = input_239_cast_fp16)[name = tensor<string, []>("query_9_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_4_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_4_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220646976)))];
+            tensor<fp16, [1, 126, 1024]> linear_39_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_4_self_attn_linear_q_weight_to_fp16, x = query_9_cast_fp16)[name = tensor<string, []>("linear_39_cast_fp16")];
+            tensor<int32, [4]> var_896 = const()[name = tensor<string, []>("op_896"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_25_cast_fp16 = reshape(shape = var_896, x = linear_39_cast_fp16)[name = tensor<string, []>("q_25_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_4_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_4_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(222744192)))];
+            tensor<fp16, [1, 126, 1024]> linear_40_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_4_self_attn_linear_k_weight_to_fp16, x = query_9_cast_fp16)[name = tensor<string, []>("linear_40_cast_fp16")];
+            tensor<int32, [4]> var_900 = const()[name = tensor<string, []>("op_900"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_17_cast_fp16 = reshape(shape = var_900, x = linear_40_cast_fp16)[name = tensor<string, []>("k_17_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_4_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_4_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(224841408)))];
+            tensor<fp16, [1, 126, 1024]> linear_41_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_4_self_attn_linear_v_weight_to_fp16, x = query_9_cast_fp16)[name = tensor<string, []>("linear_41_cast_fp16")];
+            tensor<int32, [4]> var_904 = const()[name = tensor<string, []>("op_904"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_9_cast_fp16 = reshape(shape = var_904, x = linear_41_cast_fp16)[name = tensor<string, []>("v_9_cast_fp16")];
+            tensor<int32, [4]> value_9_perm_0 = const()[name = tensor<string, []>("value_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_4_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_4_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(226938624)))];
+            tensor<fp16, [1, 126, 8, 128]> var_916_cast_fp16 = add(x = q_25_cast_fp16, y = model_layers_4_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_916_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_4_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_4_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(226940736)))];
+            tensor<fp16, [1, 126, 8, 128]> var_918_cast_fp16 = add(x = q_25_cast_fp16, y = model_layers_4_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_918_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_9_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_9_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_95_transpose_x_0 = const()[name = tensor<string, []>("x_95_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_95_transpose_y_0 = const()[name = tensor<string, []>("x_95_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_920_to_fp16 = const()[name = tensor<string, []>("op_920_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(226942848)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_259 = transpose(perm = q_with_bias_v_9_perm_0, x = var_918_cast_fp16)[name = tensor<string, []>("transpose_259")];
+            tensor<fp16, [1, 8, 126, 251]> x_95_cast_fp16 = matmul(transpose_x = x_95_transpose_x_0, transpose_y = x_95_transpose_y_0, x = transpose_259, y = var_920_to_fp16)[name = tensor<string, []>("x_95_cast_fp16")];
+            tensor<int32, [8]> x_97_pad_0 = const()[name = tensor<string, []>("x_97_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_97_mode_0 = const()[name = tensor<string, []>("x_97_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_53_to_fp16 = const()[name = tensor<string, []>("const_53_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_97_cast_fp16 = pad(constant_val = const_53_to_fp16, mode = x_97_mode_0, pad = x_97_pad_0, x = x_95_cast_fp16)[name = tensor<string, []>("x_97_cast_fp16")];
+            tensor<int32, [4]> var_928 = const()[name = tensor<string, []>("op_928"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_99_cast_fp16 = reshape(shape = var_928, x = x_97_cast_fp16)[name = tensor<string, []>("x_99_cast_fp16")];
+            tensor<int32, [4]> var_932_begin_0 = const()[name = tensor<string, []>("op_932_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_932_end_0 = const()[name = tensor<string, []>("op_932_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_932_end_mask_0 = const()[name = tensor<string, []>("op_932_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_932_cast_fp16 = slice_by_index(begin = var_932_begin_0, end = var_932_end_0, end_mask = var_932_end_mask_0, x = x_99_cast_fp16)[name = tensor<string, []>("op_932_cast_fp16")];
+            tensor<int32, [4]> var_933 = const()[name = tensor<string, []>("op_933"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_17_cast_fp16 = reshape(shape = var_933, x = var_932_cast_fp16)[name = tensor<string, []>("matrix_bd_17_cast_fp16")];
+            tensor<bool, []> matrix_ac_9_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_9_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_9_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_9_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_80_perm_0 = const()[name = tensor<string, []>("transpose_80_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_81_perm_0 = const()[name = tensor<string, []>("transpose_81_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_257 = transpose(perm = transpose_81_perm_0, x = k_17_cast_fp16)[name = tensor<string, []>("transpose_257")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_258 = transpose(perm = transpose_80_perm_0, x = var_916_cast_fp16)[name = tensor<string, []>("transpose_258")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_9_cast_fp16 = matmul(transpose_x = matrix_ac_9_transpose_x_0, transpose_y = matrix_ac_9_transpose_y_0, x = transpose_258, y = transpose_257)[name = tensor<string, []>("matrix_ac_9_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_19_begin_0 = const()[name = tensor<string, []>("matrix_bd_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_19_end_0 = const()[name = tensor<string, []>("matrix_bd_19_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_19_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_19_cast_fp16 = slice_by_index(begin = matrix_bd_19_begin_0, end = matrix_bd_19_end_0, end_mask = matrix_bd_19_end_mask_0, x = matrix_bd_17_cast_fp16)[name = tensor<string, []>("matrix_bd_19_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_942_cast_fp16 = add(x = matrix_ac_9_cast_fp16, y = matrix_bd_19_cast_fp16)[name = tensor<string, []>("op_942_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_17_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_17_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_17_cast_fp16 = mul(x = var_942_cast_fp16, y = _inversed_scores_17_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_17_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_19_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_17_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_19_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_948_cast_fp16 = softmax(axis = var_25, x = scores_19_cast_fp16)[name = tensor<string, []>("op_948_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_241_cast_fp16 = select(a = var_6_to_fp16, b = var_948_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_241_cast_fp16")];
+            tensor<bool, []> x_101_transpose_x_0 = const()[name = tensor<string, []>("x_101_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_101_transpose_y_0 = const()[name = tensor<string, []>("x_101_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_260 = transpose(perm = value_9_perm_0, x = v_9_cast_fp16)[name = tensor<string, []>("transpose_260")];
+            tensor<fp16, [1, 8, 126, 128]> x_101_cast_fp16 = matmul(transpose_x = x_101_transpose_x_0, transpose_y = x_101_transpose_y_0, x = input_241_cast_fp16, y = transpose_260)[name = tensor<string, []>("x_101_cast_fp16")];
+            tensor<int32, [4]> var_952_perm_0 = const()[name = tensor<string, []>("op_952_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_953 = const()[name = tensor<string, []>("op_953"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_256 = transpose(perm = var_952_perm_0, x = x_101_cast_fp16)[name = tensor<string, []>("transpose_256")];
+            tensor<fp16, [1, 126, 1024]> input_243_cast_fp16 = reshape(shape = var_953, x = transpose_256)[name = tensor<string, []>("input_243_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_4_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_4_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(227456960)))];
+            tensor<fp16, [1, 126, 1024]> linear_43_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_4_self_attn_linear_out_weight_to_fp16, x = input_243_cast_fp16)[name = tensor<string, []>("linear_43_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_247_cast_fp16 = add(x = input_239_cast_fp16, y = linear_43_cast_fp16)[name = tensor<string, []>("input_247_cast_fp16")];
+            tensor<int32, [1]> x_105_axes_0 = const()[name = tensor<string, []>("x_105_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_4_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_4_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(229554176)))];
+            tensor<fp16, [1024]> model_layers_4_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_4_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(229556288)))];
+            tensor<fp16, [1, 126, 1024]> x_105_cast_fp16 = layer_norm(axes = x_105_axes_0, beta = model_layers_4_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_4_norm_conv_weight_to_fp16, x = input_247_cast_fp16)[name = tensor<string, []>("x_105_cast_fp16")];
+            tensor<int32, [3]> input_249_perm_0 = const()[name = tensor<string, []>("input_249_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_969 = const()[name = tensor<string, []>("op_969"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_971 = const()[name = tensor<string, []>("op_971"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_251_pad_type_0 = const()[name = tensor<string, []>("input_251_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_251_pad_0 = const()[name = tensor<string, []>("input_251_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_4_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_4_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(229558400)))];
+            tensor<fp16, [1, 1024, 126]> transpose_255 = transpose(perm = input_249_perm_0, x = x_105_cast_fp16)[name = tensor<string, []>("transpose_255")];
+            tensor<fp16, [1, 2048, 126]> input_251_cast_fp16 = conv(dilations = var_971, groups = var_27, pad = input_251_pad_0, pad_type = input_251_pad_type_0, strides = var_969, weight = model_layers_4_conv_pointwise_conv1_weight_to_fp16, x = transpose_255)[name = tensor<string, []>("input_251_cast_fp16")];
+            tensor<int32, []> x_107_split_num_splits_0 = const()[name = tensor<string, []>("x_107_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_107_split_axis_0 = const()[name = tensor<string, []>("x_107_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_107_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_107_split_cast_fp16_1 = split(axis = x_107_split_axis_0, num_splits = x_107_split_num_splits_0, x = input_251_cast_fp16)[name = tensor<string, []>("x_107_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_107_split_1_sigmoid_cast_fp16 = sigmoid(x = x_107_split_cast_fp16_1)[name = tensor<string, []>("x_107_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_107_cast_fp16 = mul(x = x_107_split_cast_fp16_0, y = x_107_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_107_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_253_cast_fp16 = select(a = var_6_to_fp16, b = x_107_cast_fp16, cond = var_323)[name = tensor<string, []>("input_253_cast_fp16")];
+            tensor<int32, [6]> input_255_pad_0 = const()[name = tensor<string, []>("input_255_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_255_mode_0 = const()[name = tensor<string, []>("input_255_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_56_to_fp16 = const()[name = tensor<string, []>("const_56_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_255_cast_fp16 = pad(constant_val = const_56_to_fp16, mode = input_255_mode_0, pad = input_255_pad_0, x = input_253_cast_fp16)[name = tensor<string, []>("input_255_cast_fp16")];
+            tensor<int32, [1]> var_980 = const()[name = tensor<string, []>("op_980"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_982 = const()[name = tensor<string, []>("op_982"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_257_pad_type_0 = const()[name = tensor<string, []>("input_257_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_257_pad_0 = const()[name = tensor<string, []>("input_257_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_259_weight_0_to_fp16 = const()[name = tensor<string, []>("input_259_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(233752768)))];
+            tensor<fp16, [1024]> input_259_bias_0_to_fp16 = const()[name = tensor<string, []>("input_259_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(233771264)))];
+            tensor<fp16, [1, 1024, 126]> input_259_cast_fp16 = conv(bias = input_259_bias_0_to_fp16, dilations = var_982, groups = var_5, pad = input_257_pad_0, pad_type = input_257_pad_type_0, strides = var_980, weight = input_259_weight_0_to_fp16, x = input_255_cast_fp16)[name = tensor<string, []>("input_259_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_261_cast_fp16 = silu(x = input_259_cast_fp16)[name = tensor<string, []>("input_261_cast_fp16")];
+            tensor<int32, [1]> var_992 = const()[name = tensor<string, []>("op_992"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_994 = const()[name = tensor<string, []>("op_994"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_109_pad_type_0 = const()[name = tensor<string, []>("x_109_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_109_pad_0 = const()[name = tensor<string, []>("x_109_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_4_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_4_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(233773376)))];
+            tensor<fp16, [1, 1024, 126]> x_109_cast_fp16 = conv(dilations = var_994, groups = var_27, pad = x_109_pad_0, pad_type = x_109_pad_type_0, strides = var_992, weight = model_layers_4_conv_pointwise_conv2_weight_to_fp16, x = input_261_cast_fp16)[name = tensor<string, []>("x_109_cast_fp16")];
+            tensor<int32, [3]> input_263_perm_0 = const()[name = tensor<string, []>("input_263_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_254 = transpose(perm = input_263_perm_0, x = x_109_cast_fp16)[name = tensor<string, []>("transpose_254")];
+            tensor<fp16, [1, 126, 1024]> input_265_cast_fp16 = add(x = input_247_cast_fp16, y = transpose_254)[name = tensor<string, []>("input_265_cast_fp16")];
+            tensor<int32, [1]> input_267_axes_0 = const()[name = tensor<string, []>("input_267_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_4_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_4_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(235870592)))];
+            tensor<fp16, [1024]> model_layers_4_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_4_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(235872704)))];
+            tensor<fp16, [1, 126, 1024]> input_267_cast_fp16 = layer_norm(axes = input_267_axes_0, beta = model_layers_4_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_4_norm_feed_forward2_weight_to_fp16, x = input_265_cast_fp16)[name = tensor<string, []>("input_267_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_4_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_4_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(235874816)))];
+            tensor<fp16, [1, 126, 4096]> linear_44_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_4_feed_forward2_linear1_weight_to_fp16, x = input_267_cast_fp16)[name = tensor<string, []>("linear_44_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_271_cast_fp16 = silu(x = linear_44_cast_fp16)[name = tensor<string, []>("input_271_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_4_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_4_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(244263488)))];
+            tensor<fp16, [1, 126, 1024]> linear_45_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_4_feed_forward2_linear2_weight_to_fp16, x = input_271_cast_fp16)[name = tensor<string, []>("linear_45_cast_fp16")];
+            tensor<fp16, []> var_1013_to_fp16 = const()[name = tensor<string, []>("op_1013_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1014_cast_fp16 = mul(x = linear_45_cast_fp16, y = var_1013_to_fp16)[name = tensor<string, []>("op_1014_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_277_cast_fp16 = add(x = input_265_cast_fp16, y = var_1014_cast_fp16)[name = tensor<string, []>("input_277_cast_fp16")];
+            tensor<int32, [1]> input_279_axes_0 = const()[name = tensor<string, []>("input_279_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_4_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_4_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(252652160)))];
+            tensor<fp16, [1024]> model_layers_4_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_4_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(252654272)))];
+            tensor<fp16, [1, 126, 1024]> input_279_cast_fp16 = layer_norm(axes = input_279_axes_0, beta = model_layers_4_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_4_norm_out_weight_to_fp16, x = input_277_cast_fp16)[name = tensor<string, []>("input_279_cast_fp16")];
+            tensor<int32, [1]> input_281_axes_0 = const()[name = tensor<string, []>("input_281_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_5_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_5_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(252656384)))];
+            tensor<fp16, [1024]> model_layers_5_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_5_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(252658496)))];
+            tensor<fp16, [1, 126, 1024]> input_281_cast_fp16 = layer_norm(axes = input_281_axes_0, beta = model_layers_5_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_5_norm_feed_forward1_weight_to_fp16, x = input_279_cast_fp16)[name = tensor<string, []>("input_281_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_5_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_5_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(252660608)))];
+            tensor<fp16, [1, 126, 4096]> linear_46_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_5_feed_forward1_linear1_weight_to_fp16, x = input_281_cast_fp16)[name = tensor<string, []>("linear_46_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_285_cast_fp16 = silu(x = linear_46_cast_fp16)[name = tensor<string, []>("input_285_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_5_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_5_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(261049280)))];
+            tensor<fp16, [1, 126, 1024]> linear_47_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_5_feed_forward1_linear2_weight_to_fp16, x = input_285_cast_fp16)[name = tensor<string, []>("linear_47_cast_fp16")];
+            tensor<fp16, []> var_1042_to_fp16 = const()[name = tensor<string, []>("op_1042_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1043_cast_fp16 = mul(x = linear_47_cast_fp16, y = var_1042_to_fp16)[name = tensor<string, []>("op_1043_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_291_cast_fp16 = add(x = input_279_cast_fp16, y = var_1043_cast_fp16)[name = tensor<string, []>("input_291_cast_fp16")];
+            tensor<int32, [1]> query_11_axes_0 = const()[name = tensor<string, []>("query_11_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_5_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_5_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(269437952)))];
+            tensor<fp16, [1024]> model_layers_5_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_5_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(269440064)))];
+            tensor<fp16, [1, 126, 1024]> query_11_cast_fp16 = layer_norm(axes = query_11_axes_0, beta = model_layers_5_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_5_norm_self_att_weight_to_fp16, x = input_291_cast_fp16)[name = tensor<string, []>("query_11_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_5_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_5_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(269442176)))];
+            tensor<fp16, [1, 126, 1024]> linear_48_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_5_self_attn_linear_q_weight_to_fp16, x = query_11_cast_fp16)[name = tensor<string, []>("linear_48_cast_fp16")];
+            tensor<int32, [4]> var_1059 = const()[name = tensor<string, []>("op_1059"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_31_cast_fp16 = reshape(shape = var_1059, x = linear_48_cast_fp16)[name = tensor<string, []>("q_31_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_5_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_5_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(271539392)))];
+            tensor<fp16, [1, 126, 1024]> linear_49_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_5_self_attn_linear_k_weight_to_fp16, x = query_11_cast_fp16)[name = tensor<string, []>("linear_49_cast_fp16")];
+            tensor<int32, [4]> var_1063 = const()[name = tensor<string, []>("op_1063"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_21_cast_fp16 = reshape(shape = var_1063, x = linear_49_cast_fp16)[name = tensor<string, []>("k_21_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_5_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_5_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(273636608)))];
+            tensor<fp16, [1, 126, 1024]> linear_50_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_5_self_attn_linear_v_weight_to_fp16, x = query_11_cast_fp16)[name = tensor<string, []>("linear_50_cast_fp16")];
+            tensor<int32, [4]> var_1067 = const()[name = tensor<string, []>("op_1067"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_11_cast_fp16 = reshape(shape = var_1067, x = linear_50_cast_fp16)[name = tensor<string, []>("v_11_cast_fp16")];
+            tensor<int32, [4]> value_11_perm_0 = const()[name = tensor<string, []>("value_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_5_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_5_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(275733824)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1079_cast_fp16 = add(x = q_31_cast_fp16, y = model_layers_5_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_1079_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_5_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_5_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(275735936)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1081_cast_fp16 = add(x = q_31_cast_fp16, y = model_layers_5_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_1081_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_11_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_11_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_117_transpose_x_0 = const()[name = tensor<string, []>("x_117_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_117_transpose_y_0 = const()[name = tensor<string, []>("x_117_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_1083_to_fp16 = const()[name = tensor<string, []>("op_1083_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(275738048)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_252 = transpose(perm = q_with_bias_v_11_perm_0, x = var_1081_cast_fp16)[name = tensor<string, []>("transpose_252")];
+            tensor<fp16, [1, 8, 126, 251]> x_117_cast_fp16 = matmul(transpose_x = x_117_transpose_x_0, transpose_y = x_117_transpose_y_0, x = transpose_252, y = var_1083_to_fp16)[name = tensor<string, []>("x_117_cast_fp16")];
+            tensor<int32, [8]> x_119_pad_0 = const()[name = tensor<string, []>("x_119_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_119_mode_0 = const()[name = tensor<string, []>("x_119_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_63_to_fp16 = const()[name = tensor<string, []>("const_63_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_119_cast_fp16 = pad(constant_val = const_63_to_fp16, mode = x_119_mode_0, pad = x_119_pad_0, x = x_117_cast_fp16)[name = tensor<string, []>("x_119_cast_fp16")];
+            tensor<int32, [4]> var_1091 = const()[name = tensor<string, []>("op_1091"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_121_cast_fp16 = reshape(shape = var_1091, x = x_119_cast_fp16)[name = tensor<string, []>("x_121_cast_fp16")];
+            tensor<int32, [4]> var_1095_begin_0 = const()[name = tensor<string, []>("op_1095_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1095_end_0 = const()[name = tensor<string, []>("op_1095_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_1095_end_mask_0 = const()[name = tensor<string, []>("op_1095_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_1095_cast_fp16 = slice_by_index(begin = var_1095_begin_0, end = var_1095_end_0, end_mask = var_1095_end_mask_0, x = x_121_cast_fp16)[name = tensor<string, []>("op_1095_cast_fp16")];
+            tensor<int32, [4]> var_1096 = const()[name = tensor<string, []>("op_1096"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_21_cast_fp16 = reshape(shape = var_1096, x = var_1095_cast_fp16)[name = tensor<string, []>("matrix_bd_21_cast_fp16")];
+            tensor<bool, []> matrix_ac_11_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_11_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_11_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_11_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_82_perm_0 = const()[name = tensor<string, []>("transpose_82_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_83_perm_0 = const()[name = tensor<string, []>("transpose_83_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_250 = transpose(perm = transpose_83_perm_0, x = k_21_cast_fp16)[name = tensor<string, []>("transpose_250")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_251 = transpose(perm = transpose_82_perm_0, x = var_1079_cast_fp16)[name = tensor<string, []>("transpose_251")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_11_cast_fp16 = matmul(transpose_x = matrix_ac_11_transpose_x_0, transpose_y = matrix_ac_11_transpose_y_0, x = transpose_251, y = transpose_250)[name = tensor<string, []>("matrix_ac_11_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_23_begin_0 = const()[name = tensor<string, []>("matrix_bd_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_23_end_0 = const()[name = tensor<string, []>("matrix_bd_23_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_23_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_23_cast_fp16 = slice_by_index(begin = matrix_bd_23_begin_0, end = matrix_bd_23_end_0, end_mask = matrix_bd_23_end_mask_0, x = matrix_bd_21_cast_fp16)[name = tensor<string, []>("matrix_bd_23_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1105_cast_fp16 = add(x = matrix_ac_11_cast_fp16, y = matrix_bd_23_cast_fp16)[name = tensor<string, []>("op_1105_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_21_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_21_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_21_cast_fp16 = mul(x = var_1105_cast_fp16, y = _inversed_scores_21_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_21_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_23_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_21_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_23_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1111_cast_fp16 = softmax(axis = var_25, x = scores_23_cast_fp16)[name = tensor<string, []>("op_1111_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_293_cast_fp16 = select(a = var_6_to_fp16, b = var_1111_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_293_cast_fp16")];
+            tensor<bool, []> x_123_transpose_x_0 = const()[name = tensor<string, []>("x_123_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_123_transpose_y_0 = const()[name = tensor<string, []>("x_123_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_253 = transpose(perm = value_11_perm_0, x = v_11_cast_fp16)[name = tensor<string, []>("transpose_253")];
+            tensor<fp16, [1, 8, 126, 128]> x_123_cast_fp16 = matmul(transpose_x = x_123_transpose_x_0, transpose_y = x_123_transpose_y_0, x = input_293_cast_fp16, y = transpose_253)[name = tensor<string, []>("x_123_cast_fp16")];
+            tensor<int32, [4]> var_1115_perm_0 = const()[name = tensor<string, []>("op_1115_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1116 = const()[name = tensor<string, []>("op_1116"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_249 = transpose(perm = var_1115_perm_0, x = x_123_cast_fp16)[name = tensor<string, []>("transpose_249")];
+            tensor<fp16, [1, 126, 1024]> input_295_cast_fp16 = reshape(shape = var_1116, x = transpose_249)[name = tensor<string, []>("input_295_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_5_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_5_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(276252160)))];
+            tensor<fp16, [1, 126, 1024]> linear_52_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_5_self_attn_linear_out_weight_to_fp16, x = input_295_cast_fp16)[name = tensor<string, []>("linear_52_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_299_cast_fp16 = add(x = input_291_cast_fp16, y = linear_52_cast_fp16)[name = tensor<string, []>("input_299_cast_fp16")];
+            tensor<int32, [1]> x_127_axes_0 = const()[name = tensor<string, []>("x_127_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_5_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_5_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(278349376)))];
+            tensor<fp16, [1024]> model_layers_5_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_5_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(278351488)))];
+            tensor<fp16, [1, 126, 1024]> x_127_cast_fp16 = layer_norm(axes = x_127_axes_0, beta = model_layers_5_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_5_norm_conv_weight_to_fp16, x = input_299_cast_fp16)[name = tensor<string, []>("x_127_cast_fp16")];
+            tensor<int32, [3]> input_301_perm_0 = const()[name = tensor<string, []>("input_301_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_1132 = const()[name = tensor<string, []>("op_1132"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_1134 = const()[name = tensor<string, []>("op_1134"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_303_pad_type_0 = const()[name = tensor<string, []>("input_303_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_303_pad_0 = const()[name = tensor<string, []>("input_303_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_5_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_5_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(278353600)))];
+            tensor<fp16, [1, 1024, 126]> transpose_248 = transpose(perm = input_301_perm_0, x = x_127_cast_fp16)[name = tensor<string, []>("transpose_248")];
+            tensor<fp16, [1, 2048, 126]> input_303_cast_fp16 = conv(dilations = var_1134, groups = var_27, pad = input_303_pad_0, pad_type = input_303_pad_type_0, strides = var_1132, weight = model_layers_5_conv_pointwise_conv1_weight_to_fp16, x = transpose_248)[name = tensor<string, []>("input_303_cast_fp16")];
+            tensor<int32, []> x_129_split_num_splits_0 = const()[name = tensor<string, []>("x_129_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_129_split_axis_0 = const()[name = tensor<string, []>("x_129_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_129_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_129_split_cast_fp16_1 = split(axis = x_129_split_axis_0, num_splits = x_129_split_num_splits_0, x = input_303_cast_fp16)[name = tensor<string, []>("x_129_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_129_split_1_sigmoid_cast_fp16 = sigmoid(x = x_129_split_cast_fp16_1)[name = tensor<string, []>("x_129_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_129_cast_fp16 = mul(x = x_129_split_cast_fp16_0, y = x_129_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_129_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_305_cast_fp16 = select(a = var_6_to_fp16, b = x_129_cast_fp16, cond = var_323)[name = tensor<string, []>("input_305_cast_fp16")];
+            tensor<int32, [6]> input_307_pad_0 = const()[name = tensor<string, []>("input_307_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_307_mode_0 = const()[name = tensor<string, []>("input_307_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_66_to_fp16 = const()[name = tensor<string, []>("const_66_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_307_cast_fp16 = pad(constant_val = const_66_to_fp16, mode = input_307_mode_0, pad = input_307_pad_0, x = input_305_cast_fp16)[name = tensor<string, []>("input_307_cast_fp16")];
+            tensor<int32, [1]> var_1143 = const()[name = tensor<string, []>("op_1143"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_1145 = const()[name = tensor<string, []>("op_1145"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_309_pad_type_0 = const()[name = tensor<string, []>("input_309_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_309_pad_0 = const()[name = tensor<string, []>("input_309_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_311_weight_0_to_fp16 = const()[name = tensor<string, []>("input_311_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(282547968)))];
+            tensor<fp16, [1024]> input_311_bias_0_to_fp16 = const()[name = tensor<string, []>("input_311_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(282566464)))];
+            tensor<fp16, [1, 1024, 126]> input_311_cast_fp16 = conv(bias = input_311_bias_0_to_fp16, dilations = var_1145, groups = var_5, pad = input_309_pad_0, pad_type = input_309_pad_type_0, strides = var_1143, weight = input_311_weight_0_to_fp16, x = input_307_cast_fp16)[name = tensor<string, []>("input_311_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_313_cast_fp16 = silu(x = input_311_cast_fp16)[name = tensor<string, []>("input_313_cast_fp16")];
+            tensor<int32, [1]> var_1155 = const()[name = tensor<string, []>("op_1155"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_1157 = const()[name = tensor<string, []>("op_1157"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_131_pad_type_0 = const()[name = tensor<string, []>("x_131_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_131_pad_0 = const()[name = tensor<string, []>("x_131_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_5_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_5_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(282568576)))];
+            tensor<fp16, [1, 1024, 126]> x_131_cast_fp16 = conv(dilations = var_1157, groups = var_27, pad = x_131_pad_0, pad_type = x_131_pad_type_0, strides = var_1155, weight = model_layers_5_conv_pointwise_conv2_weight_to_fp16, x = input_313_cast_fp16)[name = tensor<string, []>("x_131_cast_fp16")];
+            tensor<int32, [3]> input_315_perm_0 = const()[name = tensor<string, []>("input_315_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_247 = transpose(perm = input_315_perm_0, x = x_131_cast_fp16)[name = tensor<string, []>("transpose_247")];
+            tensor<fp16, [1, 126, 1024]> input_317_cast_fp16 = add(x = input_299_cast_fp16, y = transpose_247)[name = tensor<string, []>("input_317_cast_fp16")];
+            tensor<int32, [1]> input_319_axes_0 = const()[name = tensor<string, []>("input_319_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_5_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_5_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(284665792)))];
+            tensor<fp16, [1024]> model_layers_5_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_5_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(284667904)))];
+            tensor<fp16, [1, 126, 1024]> input_319_cast_fp16 = layer_norm(axes = input_319_axes_0, beta = model_layers_5_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_5_norm_feed_forward2_weight_to_fp16, x = input_317_cast_fp16)[name = tensor<string, []>("input_319_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_5_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_5_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(284670016)))];
+            tensor<fp16, [1, 126, 4096]> linear_53_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_5_feed_forward2_linear1_weight_to_fp16, x = input_319_cast_fp16)[name = tensor<string, []>("linear_53_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_323_cast_fp16 = silu(x = linear_53_cast_fp16)[name = tensor<string, []>("input_323_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_5_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_5_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(293058688)))];
+            tensor<fp16, [1, 126, 1024]> linear_54_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_5_feed_forward2_linear2_weight_to_fp16, x = input_323_cast_fp16)[name = tensor<string, []>("linear_54_cast_fp16")];
+            tensor<fp16, []> var_1176_to_fp16 = const()[name = tensor<string, []>("op_1176_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1177_cast_fp16 = mul(x = linear_54_cast_fp16, y = var_1176_to_fp16)[name = tensor<string, []>("op_1177_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_329_cast_fp16 = add(x = input_317_cast_fp16, y = var_1177_cast_fp16)[name = tensor<string, []>("input_329_cast_fp16")];
+            tensor<int32, [1]> input_331_axes_0 = const()[name = tensor<string, []>("input_331_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_5_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_5_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(301447360)))];
+            tensor<fp16, [1024]> model_layers_5_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_5_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(301449472)))];
+            tensor<fp16, [1, 126, 1024]> input_331_cast_fp16 = layer_norm(axes = input_331_axes_0, beta = model_layers_5_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_5_norm_out_weight_to_fp16, x = input_329_cast_fp16)[name = tensor<string, []>("input_331_cast_fp16")];
+            tensor<int32, [1]> input_333_axes_0 = const()[name = tensor<string, []>("input_333_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_6_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_6_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(301451584)))];
+            tensor<fp16, [1024]> model_layers_6_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_6_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(301453696)))];
+            tensor<fp16, [1, 126, 1024]> input_333_cast_fp16 = layer_norm(axes = input_333_axes_0, beta = model_layers_6_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_6_norm_feed_forward1_weight_to_fp16, x = input_331_cast_fp16)[name = tensor<string, []>("input_333_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_6_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_6_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(301455808)))];
+            tensor<fp16, [1, 126, 4096]> linear_55_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_6_feed_forward1_linear1_weight_to_fp16, x = input_333_cast_fp16)[name = tensor<string, []>("linear_55_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_337_cast_fp16 = silu(x = linear_55_cast_fp16)[name = tensor<string, []>("input_337_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_6_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_6_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(309844480)))];
+            tensor<fp16, [1, 126, 1024]> linear_56_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_6_feed_forward1_linear2_weight_to_fp16, x = input_337_cast_fp16)[name = tensor<string, []>("linear_56_cast_fp16")];
+            tensor<fp16, []> var_1205_to_fp16 = const()[name = tensor<string, []>("op_1205_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1206_cast_fp16 = mul(x = linear_56_cast_fp16, y = var_1205_to_fp16)[name = tensor<string, []>("op_1206_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_343_cast_fp16 = add(x = input_331_cast_fp16, y = var_1206_cast_fp16)[name = tensor<string, []>("input_343_cast_fp16")];
+            tensor<int32, [1]> query_13_axes_0 = const()[name = tensor<string, []>("query_13_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_6_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_6_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(318233152)))];
+            tensor<fp16, [1024]> model_layers_6_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_6_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(318235264)))];
+            tensor<fp16, [1, 126, 1024]> query_13_cast_fp16 = layer_norm(axes = query_13_axes_0, beta = model_layers_6_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_6_norm_self_att_weight_to_fp16, x = input_343_cast_fp16)[name = tensor<string, []>("query_13_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_6_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_6_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(318237376)))];
+            tensor<fp16, [1, 126, 1024]> linear_57_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_6_self_attn_linear_q_weight_to_fp16, x = query_13_cast_fp16)[name = tensor<string, []>("linear_57_cast_fp16")];
+            tensor<int32, [4]> var_1222 = const()[name = tensor<string, []>("op_1222"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_37_cast_fp16 = reshape(shape = var_1222, x = linear_57_cast_fp16)[name = tensor<string, []>("q_37_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_6_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_6_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(320334592)))];
+            tensor<fp16, [1, 126, 1024]> linear_58_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_6_self_attn_linear_k_weight_to_fp16, x = query_13_cast_fp16)[name = tensor<string, []>("linear_58_cast_fp16")];
+            tensor<int32, [4]> var_1226 = const()[name = tensor<string, []>("op_1226"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_25_cast_fp16 = reshape(shape = var_1226, x = linear_58_cast_fp16)[name = tensor<string, []>("k_25_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_6_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_6_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(322431808)))];
+            tensor<fp16, [1, 126, 1024]> linear_59_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_6_self_attn_linear_v_weight_to_fp16, x = query_13_cast_fp16)[name = tensor<string, []>("linear_59_cast_fp16")];
+            tensor<int32, [4]> var_1230 = const()[name = tensor<string, []>("op_1230"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_13_cast_fp16 = reshape(shape = var_1230, x = linear_59_cast_fp16)[name = tensor<string, []>("v_13_cast_fp16")];
+            tensor<int32, [4]> value_13_perm_0 = const()[name = tensor<string, []>("value_13_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_6_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_6_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(324529024)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1242_cast_fp16 = add(x = q_37_cast_fp16, y = model_layers_6_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_1242_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_6_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_6_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(324531136)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1244_cast_fp16 = add(x = q_37_cast_fp16, y = model_layers_6_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_1244_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_13_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_13_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_139_transpose_x_0 = const()[name = tensor<string, []>("x_139_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_139_transpose_y_0 = const()[name = tensor<string, []>("x_139_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_1246_to_fp16 = const()[name = tensor<string, []>("op_1246_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(324533248)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_245 = transpose(perm = q_with_bias_v_13_perm_0, x = var_1244_cast_fp16)[name = tensor<string, []>("transpose_245")];
+            tensor<fp16, [1, 8, 126, 251]> x_139_cast_fp16 = matmul(transpose_x = x_139_transpose_x_0, transpose_y = x_139_transpose_y_0, x = transpose_245, y = var_1246_to_fp16)[name = tensor<string, []>("x_139_cast_fp16")];
+            tensor<int32, [8]> x_141_pad_0 = const()[name = tensor<string, []>("x_141_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_141_mode_0 = const()[name = tensor<string, []>("x_141_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_73_to_fp16 = const()[name = tensor<string, []>("const_73_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_141_cast_fp16 = pad(constant_val = const_73_to_fp16, mode = x_141_mode_0, pad = x_141_pad_0, x = x_139_cast_fp16)[name = tensor<string, []>("x_141_cast_fp16")];
+            tensor<int32, [4]> var_1254 = const()[name = tensor<string, []>("op_1254"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_143_cast_fp16 = reshape(shape = var_1254, x = x_141_cast_fp16)[name = tensor<string, []>("x_143_cast_fp16")];
+            tensor<int32, [4]> var_1258_begin_0 = const()[name = tensor<string, []>("op_1258_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1258_end_0 = const()[name = tensor<string, []>("op_1258_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_1258_end_mask_0 = const()[name = tensor<string, []>("op_1258_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_1258_cast_fp16 = slice_by_index(begin = var_1258_begin_0, end = var_1258_end_0, end_mask = var_1258_end_mask_0, x = x_143_cast_fp16)[name = tensor<string, []>("op_1258_cast_fp16")];
+            tensor<int32, [4]> var_1259 = const()[name = tensor<string, []>("op_1259"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_25_cast_fp16 = reshape(shape = var_1259, x = var_1258_cast_fp16)[name = tensor<string, []>("matrix_bd_25_cast_fp16")];
+            tensor<bool, []> matrix_ac_13_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_13_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_13_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_13_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_84_perm_0 = const()[name = tensor<string, []>("transpose_84_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_85_perm_0 = const()[name = tensor<string, []>("transpose_85_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_243 = transpose(perm = transpose_85_perm_0, x = k_25_cast_fp16)[name = tensor<string, []>("transpose_243")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_244 = transpose(perm = transpose_84_perm_0, x = var_1242_cast_fp16)[name = tensor<string, []>("transpose_244")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_13_cast_fp16 = matmul(transpose_x = matrix_ac_13_transpose_x_0, transpose_y = matrix_ac_13_transpose_y_0, x = transpose_244, y = transpose_243)[name = tensor<string, []>("matrix_ac_13_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_27_begin_0 = const()[name = tensor<string, []>("matrix_bd_27_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_27_end_0 = const()[name = tensor<string, []>("matrix_bd_27_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_27_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_27_cast_fp16 = slice_by_index(begin = matrix_bd_27_begin_0, end = matrix_bd_27_end_0, end_mask = matrix_bd_27_end_mask_0, x = matrix_bd_25_cast_fp16)[name = tensor<string, []>("matrix_bd_27_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1268_cast_fp16 = add(x = matrix_ac_13_cast_fp16, y = matrix_bd_27_cast_fp16)[name = tensor<string, []>("op_1268_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_25_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_25_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_25_cast_fp16 = mul(x = var_1268_cast_fp16, y = _inversed_scores_25_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_25_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_27_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_25_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_27_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1274_cast_fp16 = softmax(axis = var_25, x = scores_27_cast_fp16)[name = tensor<string, []>("op_1274_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_345_cast_fp16 = select(a = var_6_to_fp16, b = var_1274_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_345_cast_fp16")];
+            tensor<bool, []> x_145_transpose_x_0 = const()[name = tensor<string, []>("x_145_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_145_transpose_y_0 = const()[name = tensor<string, []>("x_145_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_246 = transpose(perm = value_13_perm_0, x = v_13_cast_fp16)[name = tensor<string, []>("transpose_246")];
+            tensor<fp16, [1, 8, 126, 128]> x_145_cast_fp16 = matmul(transpose_x = x_145_transpose_x_0, transpose_y = x_145_transpose_y_0, x = input_345_cast_fp16, y = transpose_246)[name = tensor<string, []>("x_145_cast_fp16")];
+            tensor<int32, [4]> var_1278_perm_0 = const()[name = tensor<string, []>("op_1278_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1279 = const()[name = tensor<string, []>("op_1279"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_242 = transpose(perm = var_1278_perm_0, x = x_145_cast_fp16)[name = tensor<string, []>("transpose_242")];
+            tensor<fp16, [1, 126, 1024]> input_347_cast_fp16 = reshape(shape = var_1279, x = transpose_242)[name = tensor<string, []>("input_347_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_6_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_6_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(325047360)))];
+            tensor<fp16, [1, 126, 1024]> linear_61_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_6_self_attn_linear_out_weight_to_fp16, x = input_347_cast_fp16)[name = tensor<string, []>("linear_61_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_351_cast_fp16 = add(x = input_343_cast_fp16, y = linear_61_cast_fp16)[name = tensor<string, []>("input_351_cast_fp16")];
+            tensor<int32, [1]> x_149_axes_0 = const()[name = tensor<string, []>("x_149_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_6_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_6_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(327144576)))];
+            tensor<fp16, [1024]> model_layers_6_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_6_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(327146688)))];
+            tensor<fp16, [1, 126, 1024]> x_149_cast_fp16 = layer_norm(axes = x_149_axes_0, beta = model_layers_6_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_6_norm_conv_weight_to_fp16, x = input_351_cast_fp16)[name = tensor<string, []>("x_149_cast_fp16")];
+            tensor<int32, [3]> input_353_perm_0 = const()[name = tensor<string, []>("input_353_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_1295 = const()[name = tensor<string, []>("op_1295"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_1297 = const()[name = tensor<string, []>("op_1297"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_355_pad_type_0 = const()[name = tensor<string, []>("input_355_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_355_pad_0 = const()[name = tensor<string, []>("input_355_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_6_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_6_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(327148800)))];
+            tensor<fp16, [1, 1024, 126]> transpose_241 = transpose(perm = input_353_perm_0, x = x_149_cast_fp16)[name = tensor<string, []>("transpose_241")];
+            tensor<fp16, [1, 2048, 126]> input_355_cast_fp16 = conv(dilations = var_1297, groups = var_27, pad = input_355_pad_0, pad_type = input_355_pad_type_0, strides = var_1295, weight = model_layers_6_conv_pointwise_conv1_weight_to_fp16, x = transpose_241)[name = tensor<string, []>("input_355_cast_fp16")];
+            tensor<int32, []> x_151_split_num_splits_0 = const()[name = tensor<string, []>("x_151_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_151_split_axis_0 = const()[name = tensor<string, []>("x_151_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_151_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_151_split_cast_fp16_1 = split(axis = x_151_split_axis_0, num_splits = x_151_split_num_splits_0, x = input_355_cast_fp16)[name = tensor<string, []>("x_151_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_151_split_1_sigmoid_cast_fp16 = sigmoid(x = x_151_split_cast_fp16_1)[name = tensor<string, []>("x_151_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_151_cast_fp16 = mul(x = x_151_split_cast_fp16_0, y = x_151_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_151_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_357_cast_fp16 = select(a = var_6_to_fp16, b = x_151_cast_fp16, cond = var_323)[name = tensor<string, []>("input_357_cast_fp16")];
+            tensor<int32, [6]> input_359_pad_0 = const()[name = tensor<string, []>("input_359_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_359_mode_0 = const()[name = tensor<string, []>("input_359_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_76_to_fp16 = const()[name = tensor<string, []>("const_76_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_359_cast_fp16 = pad(constant_val = const_76_to_fp16, mode = input_359_mode_0, pad = input_359_pad_0, x = input_357_cast_fp16)[name = tensor<string, []>("input_359_cast_fp16")];
+            tensor<int32, [1]> var_1306 = const()[name = tensor<string, []>("op_1306"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_1308 = const()[name = tensor<string, []>("op_1308"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_361_pad_type_0 = const()[name = tensor<string, []>("input_361_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_361_pad_0 = const()[name = tensor<string, []>("input_361_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_363_weight_0_to_fp16 = const()[name = tensor<string, []>("input_363_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(331343168)))];
+            tensor<fp16, [1024]> input_363_bias_0_to_fp16 = const()[name = tensor<string, []>("input_363_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(331361664)))];
+            tensor<fp16, [1, 1024, 126]> input_363_cast_fp16 = conv(bias = input_363_bias_0_to_fp16, dilations = var_1308, groups = var_5, pad = input_361_pad_0, pad_type = input_361_pad_type_0, strides = var_1306, weight = input_363_weight_0_to_fp16, x = input_359_cast_fp16)[name = tensor<string, []>("input_363_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_365_cast_fp16 = silu(x = input_363_cast_fp16)[name = tensor<string, []>("input_365_cast_fp16")];
+            tensor<int32, [1]> var_1318 = const()[name = tensor<string, []>("op_1318"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_1320 = const()[name = tensor<string, []>("op_1320"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_153_pad_type_0 = const()[name = tensor<string, []>("x_153_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_153_pad_0 = const()[name = tensor<string, []>("x_153_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_6_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_6_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(331363776)))];
+            tensor<fp16, [1, 1024, 126]> x_153_cast_fp16 = conv(dilations = var_1320, groups = var_27, pad = x_153_pad_0, pad_type = x_153_pad_type_0, strides = var_1318, weight = model_layers_6_conv_pointwise_conv2_weight_to_fp16, x = input_365_cast_fp16)[name = tensor<string, []>("x_153_cast_fp16")];
+            tensor<int32, [3]> input_367_perm_0 = const()[name = tensor<string, []>("input_367_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_240 = transpose(perm = input_367_perm_0, x = x_153_cast_fp16)[name = tensor<string, []>("transpose_240")];
+            tensor<fp16, [1, 126, 1024]> input_369_cast_fp16 = add(x = input_351_cast_fp16, y = transpose_240)[name = tensor<string, []>("input_369_cast_fp16")];
+            tensor<int32, [1]> input_371_axes_0 = const()[name = tensor<string, []>("input_371_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_6_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_6_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(333460992)))];
+            tensor<fp16, [1024]> model_layers_6_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_6_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(333463104)))];
+            tensor<fp16, [1, 126, 1024]> input_371_cast_fp16 = layer_norm(axes = input_371_axes_0, beta = model_layers_6_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_6_norm_feed_forward2_weight_to_fp16, x = input_369_cast_fp16)[name = tensor<string, []>("input_371_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_6_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_6_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(333465216)))];
+            tensor<fp16, [1, 126, 4096]> linear_62_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_6_feed_forward2_linear1_weight_to_fp16, x = input_371_cast_fp16)[name = tensor<string, []>("linear_62_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_375_cast_fp16 = silu(x = linear_62_cast_fp16)[name = tensor<string, []>("input_375_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_6_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_6_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(341853888)))];
+            tensor<fp16, [1, 126, 1024]> linear_63_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_6_feed_forward2_linear2_weight_to_fp16, x = input_375_cast_fp16)[name = tensor<string, []>("linear_63_cast_fp16")];
+            tensor<fp16, []> var_1339_to_fp16 = const()[name = tensor<string, []>("op_1339_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1340_cast_fp16 = mul(x = linear_63_cast_fp16, y = var_1339_to_fp16)[name = tensor<string, []>("op_1340_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_381_cast_fp16 = add(x = input_369_cast_fp16, y = var_1340_cast_fp16)[name = tensor<string, []>("input_381_cast_fp16")];
+            tensor<int32, [1]> input_383_axes_0 = const()[name = tensor<string, []>("input_383_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_6_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_6_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(350242560)))];
+            tensor<fp16, [1024]> model_layers_6_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_6_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(350244672)))];
+            tensor<fp16, [1, 126, 1024]> input_383_cast_fp16 = layer_norm(axes = input_383_axes_0, beta = model_layers_6_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_6_norm_out_weight_to_fp16, x = input_381_cast_fp16)[name = tensor<string, []>("input_383_cast_fp16")];
+            tensor<int32, [1]> input_385_axes_0 = const()[name = tensor<string, []>("input_385_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_7_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_7_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(350246784)))];
+            tensor<fp16, [1024]> model_layers_7_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_7_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(350248896)))];
+            tensor<fp16, [1, 126, 1024]> input_385_cast_fp16 = layer_norm(axes = input_385_axes_0, beta = model_layers_7_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_7_norm_feed_forward1_weight_to_fp16, x = input_383_cast_fp16)[name = tensor<string, []>("input_385_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_7_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_7_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(350251008)))];
+            tensor<fp16, [1, 126, 4096]> linear_64_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_7_feed_forward1_linear1_weight_to_fp16, x = input_385_cast_fp16)[name = tensor<string, []>("linear_64_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_389_cast_fp16 = silu(x = linear_64_cast_fp16)[name = tensor<string, []>("input_389_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_7_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_7_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(358639680)))];
+            tensor<fp16, [1, 126, 1024]> linear_65_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_7_feed_forward1_linear2_weight_to_fp16, x = input_389_cast_fp16)[name = tensor<string, []>("linear_65_cast_fp16")];
+            tensor<fp16, []> var_1368_to_fp16 = const()[name = tensor<string, []>("op_1368_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1369_cast_fp16 = mul(x = linear_65_cast_fp16, y = var_1368_to_fp16)[name = tensor<string, []>("op_1369_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_395_cast_fp16 = add(x = input_383_cast_fp16, y = var_1369_cast_fp16)[name = tensor<string, []>("input_395_cast_fp16")];
+            tensor<int32, [1]> query_15_axes_0 = const()[name = tensor<string, []>("query_15_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_7_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_7_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(367028352)))];
+            tensor<fp16, [1024]> model_layers_7_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_7_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(367030464)))];
+            tensor<fp16, [1, 126, 1024]> query_15_cast_fp16 = layer_norm(axes = query_15_axes_0, beta = model_layers_7_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_7_norm_self_att_weight_to_fp16, x = input_395_cast_fp16)[name = tensor<string, []>("query_15_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_7_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_7_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(367032576)))];
+            tensor<fp16, [1, 126, 1024]> linear_66_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_7_self_attn_linear_q_weight_to_fp16, x = query_15_cast_fp16)[name = tensor<string, []>("linear_66_cast_fp16")];
+            tensor<int32, [4]> var_1385 = const()[name = tensor<string, []>("op_1385"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_43_cast_fp16 = reshape(shape = var_1385, x = linear_66_cast_fp16)[name = tensor<string, []>("q_43_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_7_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_7_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(369129792)))];
+            tensor<fp16, [1, 126, 1024]> linear_67_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_7_self_attn_linear_k_weight_to_fp16, x = query_15_cast_fp16)[name = tensor<string, []>("linear_67_cast_fp16")];
+            tensor<int32, [4]> var_1389 = const()[name = tensor<string, []>("op_1389"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_29_cast_fp16 = reshape(shape = var_1389, x = linear_67_cast_fp16)[name = tensor<string, []>("k_29_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_7_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_7_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(371227008)))];
+            tensor<fp16, [1, 126, 1024]> linear_68_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_7_self_attn_linear_v_weight_to_fp16, x = query_15_cast_fp16)[name = tensor<string, []>("linear_68_cast_fp16")];
+            tensor<int32, [4]> var_1393 = const()[name = tensor<string, []>("op_1393"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_15_cast_fp16 = reshape(shape = var_1393, x = linear_68_cast_fp16)[name = tensor<string, []>("v_15_cast_fp16")];
+            tensor<int32, [4]> value_15_perm_0 = const()[name = tensor<string, []>("value_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_7_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_7_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(373324224)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1405_cast_fp16 = add(x = q_43_cast_fp16, y = model_layers_7_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_1405_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_7_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_7_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(373326336)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1407_cast_fp16 = add(x = q_43_cast_fp16, y = model_layers_7_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_1407_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_15_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_15_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_161_transpose_x_0 = const()[name = tensor<string, []>("x_161_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_161_transpose_y_0 = const()[name = tensor<string, []>("x_161_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_1409_to_fp16 = const()[name = tensor<string, []>("op_1409_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(373328448)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_238 = transpose(perm = q_with_bias_v_15_perm_0, x = var_1407_cast_fp16)[name = tensor<string, []>("transpose_238")];
+            tensor<fp16, [1, 8, 126, 251]> x_161_cast_fp16 = matmul(transpose_x = x_161_transpose_x_0, transpose_y = x_161_transpose_y_0, x = transpose_238, y = var_1409_to_fp16)[name = tensor<string, []>("x_161_cast_fp16")];
+            tensor<int32, [8]> x_163_pad_0 = const()[name = tensor<string, []>("x_163_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_163_mode_0 = const()[name = tensor<string, []>("x_163_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_83_to_fp16 = const()[name = tensor<string, []>("const_83_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_163_cast_fp16 = pad(constant_val = const_83_to_fp16, mode = x_163_mode_0, pad = x_163_pad_0, x = x_161_cast_fp16)[name = tensor<string, []>("x_163_cast_fp16")];
+            tensor<int32, [4]> var_1417 = const()[name = tensor<string, []>("op_1417"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_165_cast_fp16 = reshape(shape = var_1417, x = x_163_cast_fp16)[name = tensor<string, []>("x_165_cast_fp16")];
+            tensor<int32, [4]> var_1421_begin_0 = const()[name = tensor<string, []>("op_1421_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1421_end_0 = const()[name = tensor<string, []>("op_1421_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_1421_end_mask_0 = const()[name = tensor<string, []>("op_1421_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_1421_cast_fp16 = slice_by_index(begin = var_1421_begin_0, end = var_1421_end_0, end_mask = var_1421_end_mask_0, x = x_165_cast_fp16)[name = tensor<string, []>("op_1421_cast_fp16")];
+            tensor<int32, [4]> var_1422 = const()[name = tensor<string, []>("op_1422"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_29_cast_fp16 = reshape(shape = var_1422, x = var_1421_cast_fp16)[name = tensor<string, []>("matrix_bd_29_cast_fp16")];
+            tensor<bool, []> matrix_ac_15_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_15_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_15_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_15_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_86_perm_0 = const()[name = tensor<string, []>("transpose_86_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_87_perm_0 = const()[name = tensor<string, []>("transpose_87_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_236 = transpose(perm = transpose_87_perm_0, x = k_29_cast_fp16)[name = tensor<string, []>("transpose_236")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_237 = transpose(perm = transpose_86_perm_0, x = var_1405_cast_fp16)[name = tensor<string, []>("transpose_237")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_15_cast_fp16 = matmul(transpose_x = matrix_ac_15_transpose_x_0, transpose_y = matrix_ac_15_transpose_y_0, x = transpose_237, y = transpose_236)[name = tensor<string, []>("matrix_ac_15_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_31_begin_0 = const()[name = tensor<string, []>("matrix_bd_31_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_31_end_0 = const()[name = tensor<string, []>("matrix_bd_31_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_31_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_31_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_31_cast_fp16 = slice_by_index(begin = matrix_bd_31_begin_0, end = matrix_bd_31_end_0, end_mask = matrix_bd_31_end_mask_0, x = matrix_bd_29_cast_fp16)[name = tensor<string, []>("matrix_bd_31_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1431_cast_fp16 = add(x = matrix_ac_15_cast_fp16, y = matrix_bd_31_cast_fp16)[name = tensor<string, []>("op_1431_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_29_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_29_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_29_cast_fp16 = mul(x = var_1431_cast_fp16, y = _inversed_scores_29_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_29_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_31_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_29_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_31_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1437_cast_fp16 = softmax(axis = var_25, x = scores_31_cast_fp16)[name = tensor<string, []>("op_1437_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_397_cast_fp16 = select(a = var_6_to_fp16, b = var_1437_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_397_cast_fp16")];
+            tensor<bool, []> x_167_transpose_x_0 = const()[name = tensor<string, []>("x_167_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_167_transpose_y_0 = const()[name = tensor<string, []>("x_167_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_239 = transpose(perm = value_15_perm_0, x = v_15_cast_fp16)[name = tensor<string, []>("transpose_239")];
+            tensor<fp16, [1, 8, 126, 128]> x_167_cast_fp16 = matmul(transpose_x = x_167_transpose_x_0, transpose_y = x_167_transpose_y_0, x = input_397_cast_fp16, y = transpose_239)[name = tensor<string, []>("x_167_cast_fp16")];
+            tensor<int32, [4]> var_1441_perm_0 = const()[name = tensor<string, []>("op_1441_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1442 = const()[name = tensor<string, []>("op_1442"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_235 = transpose(perm = var_1441_perm_0, x = x_167_cast_fp16)[name = tensor<string, []>("transpose_235")];
+            tensor<fp16, [1, 126, 1024]> input_399_cast_fp16 = reshape(shape = var_1442, x = transpose_235)[name = tensor<string, []>("input_399_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_7_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_7_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(373842560)))];
+            tensor<fp16, [1, 126, 1024]> linear_70_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_7_self_attn_linear_out_weight_to_fp16, x = input_399_cast_fp16)[name = tensor<string, []>("linear_70_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_403_cast_fp16 = add(x = input_395_cast_fp16, y = linear_70_cast_fp16)[name = tensor<string, []>("input_403_cast_fp16")];
+            tensor<int32, [1]> x_171_axes_0 = const()[name = tensor<string, []>("x_171_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_7_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_7_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(375939776)))];
+            tensor<fp16, [1024]> model_layers_7_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_7_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(375941888)))];
+            tensor<fp16, [1, 126, 1024]> x_171_cast_fp16 = layer_norm(axes = x_171_axes_0, beta = model_layers_7_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_7_norm_conv_weight_to_fp16, x = input_403_cast_fp16)[name = tensor<string, []>("x_171_cast_fp16")];
+            tensor<int32, [3]> input_405_perm_0 = const()[name = tensor<string, []>("input_405_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_1458 = const()[name = tensor<string, []>("op_1458"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_1460 = const()[name = tensor<string, []>("op_1460"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_407_pad_type_0 = const()[name = tensor<string, []>("input_407_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_407_pad_0 = const()[name = tensor<string, []>("input_407_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_7_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_7_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(375944000)))];
+            tensor<fp16, [1, 1024, 126]> transpose_234 = transpose(perm = input_405_perm_0, x = x_171_cast_fp16)[name = tensor<string, []>("transpose_234")];
+            tensor<fp16, [1, 2048, 126]> input_407_cast_fp16 = conv(dilations = var_1460, groups = var_27, pad = input_407_pad_0, pad_type = input_407_pad_type_0, strides = var_1458, weight = model_layers_7_conv_pointwise_conv1_weight_to_fp16, x = transpose_234)[name = tensor<string, []>("input_407_cast_fp16")];
+            tensor<int32, []> x_173_split_num_splits_0 = const()[name = tensor<string, []>("x_173_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_173_split_axis_0 = const()[name = tensor<string, []>("x_173_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_173_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_173_split_cast_fp16_1 = split(axis = x_173_split_axis_0, num_splits = x_173_split_num_splits_0, x = input_407_cast_fp16)[name = tensor<string, []>("x_173_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_173_split_1_sigmoid_cast_fp16 = sigmoid(x = x_173_split_cast_fp16_1)[name = tensor<string, []>("x_173_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_173_cast_fp16 = mul(x = x_173_split_cast_fp16_0, y = x_173_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_173_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_409_cast_fp16 = select(a = var_6_to_fp16, b = x_173_cast_fp16, cond = var_323)[name = tensor<string, []>("input_409_cast_fp16")];
+            tensor<int32, [6]> input_411_pad_0 = const()[name = tensor<string, []>("input_411_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_411_mode_0 = const()[name = tensor<string, []>("input_411_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_86_to_fp16 = const()[name = tensor<string, []>("const_86_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_411_cast_fp16 = pad(constant_val = const_86_to_fp16, mode = input_411_mode_0, pad = input_411_pad_0, x = input_409_cast_fp16)[name = tensor<string, []>("input_411_cast_fp16")];
+            tensor<int32, [1]> var_1469 = const()[name = tensor<string, []>("op_1469"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_1471 = const()[name = tensor<string, []>("op_1471"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_413_pad_type_0 = const()[name = tensor<string, []>("input_413_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_413_pad_0 = const()[name = tensor<string, []>("input_413_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_415_weight_0_to_fp16 = const()[name = tensor<string, []>("input_415_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(380138368)))];
+            tensor<fp16, [1024]> input_415_bias_0_to_fp16 = const()[name = tensor<string, []>("input_415_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(380156864)))];
+            tensor<fp16, [1, 1024, 126]> input_415_cast_fp16 = conv(bias = input_415_bias_0_to_fp16, dilations = var_1471, groups = var_5, pad = input_413_pad_0, pad_type = input_413_pad_type_0, strides = var_1469, weight = input_415_weight_0_to_fp16, x = input_411_cast_fp16)[name = tensor<string, []>("input_415_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_417_cast_fp16 = silu(x = input_415_cast_fp16)[name = tensor<string, []>("input_417_cast_fp16")];
+            tensor<int32, [1]> var_1481 = const()[name = tensor<string, []>("op_1481"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_1483 = const()[name = tensor<string, []>("op_1483"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_175_pad_type_0 = const()[name = tensor<string, []>("x_175_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_175_pad_0 = const()[name = tensor<string, []>("x_175_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_7_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_7_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(380158976)))];
+            tensor<fp16, [1, 1024, 126]> x_175_cast_fp16 = conv(dilations = var_1483, groups = var_27, pad = x_175_pad_0, pad_type = x_175_pad_type_0, strides = var_1481, weight = model_layers_7_conv_pointwise_conv2_weight_to_fp16, x = input_417_cast_fp16)[name = tensor<string, []>("x_175_cast_fp16")];
+            tensor<int32, [3]> input_419_perm_0 = const()[name = tensor<string, []>("input_419_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_233 = transpose(perm = input_419_perm_0, x = x_175_cast_fp16)[name = tensor<string, []>("transpose_233")];
+            tensor<fp16, [1, 126, 1024]> input_421_cast_fp16 = add(x = input_403_cast_fp16, y = transpose_233)[name = tensor<string, []>("input_421_cast_fp16")];
+            tensor<int32, [1]> input_423_axes_0 = const()[name = tensor<string, []>("input_423_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_7_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_7_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(382256192)))];
+            tensor<fp16, [1024]> model_layers_7_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_7_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(382258304)))];
+            tensor<fp16, [1, 126, 1024]> input_423_cast_fp16 = layer_norm(axes = input_423_axes_0, beta = model_layers_7_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_7_norm_feed_forward2_weight_to_fp16, x = input_421_cast_fp16)[name = tensor<string, []>("input_423_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_7_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_7_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(382260416)))];
+            tensor<fp16, [1, 126, 4096]> linear_71_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_7_feed_forward2_linear1_weight_to_fp16, x = input_423_cast_fp16)[name = tensor<string, []>("linear_71_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_427_cast_fp16 = silu(x = linear_71_cast_fp16)[name = tensor<string, []>("input_427_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_7_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_7_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(390649088)))];
+            tensor<fp16, [1, 126, 1024]> linear_72_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_7_feed_forward2_linear2_weight_to_fp16, x = input_427_cast_fp16)[name = tensor<string, []>("linear_72_cast_fp16")];
+            tensor<fp16, []> var_1502_to_fp16 = const()[name = tensor<string, []>("op_1502_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1503_cast_fp16 = mul(x = linear_72_cast_fp16, y = var_1502_to_fp16)[name = tensor<string, []>("op_1503_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_433_cast_fp16 = add(x = input_421_cast_fp16, y = var_1503_cast_fp16)[name = tensor<string, []>("input_433_cast_fp16")];
+            tensor<int32, [1]> input_435_axes_0 = const()[name = tensor<string, []>("input_435_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_7_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_7_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(399037760)))];
+            tensor<fp16, [1024]> model_layers_7_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_7_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(399039872)))];
+            tensor<fp16, [1, 126, 1024]> input_435_cast_fp16 = layer_norm(axes = input_435_axes_0, beta = model_layers_7_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_7_norm_out_weight_to_fp16, x = input_433_cast_fp16)[name = tensor<string, []>("input_435_cast_fp16")];
+            tensor<int32, [1]> input_437_axes_0 = const()[name = tensor<string, []>("input_437_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_8_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_8_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(399041984)))];
+            tensor<fp16, [1024]> model_layers_8_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_8_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(399044096)))];
+            tensor<fp16, [1, 126, 1024]> input_437_cast_fp16 = layer_norm(axes = input_437_axes_0, beta = model_layers_8_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_8_norm_feed_forward1_weight_to_fp16, x = input_435_cast_fp16)[name = tensor<string, []>("input_437_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_8_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_8_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(399046208)))];
+            tensor<fp16, [1, 126, 4096]> linear_73_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_8_feed_forward1_linear1_weight_to_fp16, x = input_437_cast_fp16)[name = tensor<string, []>("linear_73_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_441_cast_fp16 = silu(x = linear_73_cast_fp16)[name = tensor<string, []>("input_441_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_8_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_8_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(407434880)))];
+            tensor<fp16, [1, 126, 1024]> linear_74_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_8_feed_forward1_linear2_weight_to_fp16, x = input_441_cast_fp16)[name = tensor<string, []>("linear_74_cast_fp16")];
+            tensor<fp16, []> var_1531_to_fp16 = const()[name = tensor<string, []>("op_1531_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1532_cast_fp16 = mul(x = linear_74_cast_fp16, y = var_1531_to_fp16)[name = tensor<string, []>("op_1532_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_447_cast_fp16 = add(x = input_435_cast_fp16, y = var_1532_cast_fp16)[name = tensor<string, []>("input_447_cast_fp16")];
+            tensor<int32, [1]> query_17_axes_0 = const()[name = tensor<string, []>("query_17_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_8_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_8_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(415823552)))];
+            tensor<fp16, [1024]> model_layers_8_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_8_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(415825664)))];
+            tensor<fp16, [1, 126, 1024]> query_17_cast_fp16 = layer_norm(axes = query_17_axes_0, beta = model_layers_8_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_8_norm_self_att_weight_to_fp16, x = input_447_cast_fp16)[name = tensor<string, []>("query_17_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_8_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_8_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(415827776)))];
+            tensor<fp16, [1, 126, 1024]> linear_75_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_8_self_attn_linear_q_weight_to_fp16, x = query_17_cast_fp16)[name = tensor<string, []>("linear_75_cast_fp16")];
+            tensor<int32, [4]> var_1548 = const()[name = tensor<string, []>("op_1548"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_49_cast_fp16 = reshape(shape = var_1548, x = linear_75_cast_fp16)[name = tensor<string, []>("q_49_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_8_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_8_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(417924992)))];
+            tensor<fp16, [1, 126, 1024]> linear_76_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_8_self_attn_linear_k_weight_to_fp16, x = query_17_cast_fp16)[name = tensor<string, []>("linear_76_cast_fp16")];
+            tensor<int32, [4]> var_1552 = const()[name = tensor<string, []>("op_1552"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_33_cast_fp16 = reshape(shape = var_1552, x = linear_76_cast_fp16)[name = tensor<string, []>("k_33_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_8_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_8_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(420022208)))];
+            tensor<fp16, [1, 126, 1024]> linear_77_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_8_self_attn_linear_v_weight_to_fp16, x = query_17_cast_fp16)[name = tensor<string, []>("linear_77_cast_fp16")];
+            tensor<int32, [4]> var_1556 = const()[name = tensor<string, []>("op_1556"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_17_cast_fp16 = reshape(shape = var_1556, x = linear_77_cast_fp16)[name = tensor<string, []>("v_17_cast_fp16")];
+            tensor<int32, [4]> value_17_perm_0 = const()[name = tensor<string, []>("value_17_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_8_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_8_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(422119424)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1568_cast_fp16 = add(x = q_49_cast_fp16, y = model_layers_8_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_1568_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_8_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_8_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(422121536)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1570_cast_fp16 = add(x = q_49_cast_fp16, y = model_layers_8_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_1570_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_17_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_17_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_183_transpose_x_0 = const()[name = tensor<string, []>("x_183_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_183_transpose_y_0 = const()[name = tensor<string, []>("x_183_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_1572_to_fp16 = const()[name = tensor<string, []>("op_1572_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(422123648)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_231 = transpose(perm = q_with_bias_v_17_perm_0, x = var_1570_cast_fp16)[name = tensor<string, []>("transpose_231")];
+            tensor<fp16, [1, 8, 126, 251]> x_183_cast_fp16 = matmul(transpose_x = x_183_transpose_x_0, transpose_y = x_183_transpose_y_0, x = transpose_231, y = var_1572_to_fp16)[name = tensor<string, []>("x_183_cast_fp16")];
+            tensor<int32, [8]> x_185_pad_0 = const()[name = tensor<string, []>("x_185_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_185_mode_0 = const()[name = tensor<string, []>("x_185_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_93_to_fp16 = const()[name = tensor<string, []>("const_93_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_185_cast_fp16 = pad(constant_val = const_93_to_fp16, mode = x_185_mode_0, pad = x_185_pad_0, x = x_183_cast_fp16)[name = tensor<string, []>("x_185_cast_fp16")];
+            tensor<int32, [4]> var_1580 = const()[name = tensor<string, []>("op_1580"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_187_cast_fp16 = reshape(shape = var_1580, x = x_185_cast_fp16)[name = tensor<string, []>("x_187_cast_fp16")];
+            tensor<int32, [4]> var_1584_begin_0 = const()[name = tensor<string, []>("op_1584_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1584_end_0 = const()[name = tensor<string, []>("op_1584_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_1584_end_mask_0 = const()[name = tensor<string, []>("op_1584_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_1584_cast_fp16 = slice_by_index(begin = var_1584_begin_0, end = var_1584_end_0, end_mask = var_1584_end_mask_0, x = x_187_cast_fp16)[name = tensor<string, []>("op_1584_cast_fp16")];
+            tensor<int32, [4]> var_1585 = const()[name = tensor<string, []>("op_1585"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_33_cast_fp16 = reshape(shape = var_1585, x = var_1584_cast_fp16)[name = tensor<string, []>("matrix_bd_33_cast_fp16")];
+            tensor<bool, []> matrix_ac_17_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_17_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_17_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_17_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_88_perm_0 = const()[name = tensor<string, []>("transpose_88_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_89_perm_0 = const()[name = tensor<string, []>("transpose_89_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_229 = transpose(perm = transpose_89_perm_0, x = k_33_cast_fp16)[name = tensor<string, []>("transpose_229")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_230 = transpose(perm = transpose_88_perm_0, x = var_1568_cast_fp16)[name = tensor<string, []>("transpose_230")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_17_cast_fp16 = matmul(transpose_x = matrix_ac_17_transpose_x_0, transpose_y = matrix_ac_17_transpose_y_0, x = transpose_230, y = transpose_229)[name = tensor<string, []>("matrix_ac_17_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_35_begin_0 = const()[name = tensor<string, []>("matrix_bd_35_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_35_end_0 = const()[name = tensor<string, []>("matrix_bd_35_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_35_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_35_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_35_cast_fp16 = slice_by_index(begin = matrix_bd_35_begin_0, end = matrix_bd_35_end_0, end_mask = matrix_bd_35_end_mask_0, x = matrix_bd_33_cast_fp16)[name = tensor<string, []>("matrix_bd_35_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1594_cast_fp16 = add(x = matrix_ac_17_cast_fp16, y = matrix_bd_35_cast_fp16)[name = tensor<string, []>("op_1594_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_33_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_33_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_33_cast_fp16 = mul(x = var_1594_cast_fp16, y = _inversed_scores_33_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_33_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_35_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_33_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_35_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1600_cast_fp16 = softmax(axis = var_25, x = scores_35_cast_fp16)[name = tensor<string, []>("op_1600_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_449_cast_fp16 = select(a = var_6_to_fp16, b = var_1600_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_449_cast_fp16")];
+            tensor<bool, []> x_189_transpose_x_0 = const()[name = tensor<string, []>("x_189_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_189_transpose_y_0 = const()[name = tensor<string, []>("x_189_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_232 = transpose(perm = value_17_perm_0, x = v_17_cast_fp16)[name = tensor<string, []>("transpose_232")];
+            tensor<fp16, [1, 8, 126, 128]> x_189_cast_fp16 = matmul(transpose_x = x_189_transpose_x_0, transpose_y = x_189_transpose_y_0, x = input_449_cast_fp16, y = transpose_232)[name = tensor<string, []>("x_189_cast_fp16")];
+            tensor<int32, [4]> var_1604_perm_0 = const()[name = tensor<string, []>("op_1604_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1605 = const()[name = tensor<string, []>("op_1605"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_228 = transpose(perm = var_1604_perm_0, x = x_189_cast_fp16)[name = tensor<string, []>("transpose_228")];
+            tensor<fp16, [1, 126, 1024]> input_451_cast_fp16 = reshape(shape = var_1605, x = transpose_228)[name = tensor<string, []>("input_451_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_8_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_8_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(422637760)))];
+            tensor<fp16, [1, 126, 1024]> linear_79_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_8_self_attn_linear_out_weight_to_fp16, x = input_451_cast_fp16)[name = tensor<string, []>("linear_79_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_455_cast_fp16 = add(x = input_447_cast_fp16, y = linear_79_cast_fp16)[name = tensor<string, []>("input_455_cast_fp16")];
+            tensor<int32, [1]> x_193_axes_0 = const()[name = tensor<string, []>("x_193_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_8_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_8_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(424734976)))];
+            tensor<fp16, [1024]> model_layers_8_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_8_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(424737088)))];
+            tensor<fp16, [1, 126, 1024]> x_193_cast_fp16 = layer_norm(axes = x_193_axes_0, beta = model_layers_8_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_8_norm_conv_weight_to_fp16, x = input_455_cast_fp16)[name = tensor<string, []>("x_193_cast_fp16")];
+            tensor<int32, [3]> input_457_perm_0 = const()[name = tensor<string, []>("input_457_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_1621 = const()[name = tensor<string, []>("op_1621"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_1623 = const()[name = tensor<string, []>("op_1623"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_459_pad_type_0 = const()[name = tensor<string, []>("input_459_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_459_pad_0 = const()[name = tensor<string, []>("input_459_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_8_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_8_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(424739200)))];
+            tensor<fp16, [1, 1024, 126]> transpose_227 = transpose(perm = input_457_perm_0, x = x_193_cast_fp16)[name = tensor<string, []>("transpose_227")];
+            tensor<fp16, [1, 2048, 126]> input_459_cast_fp16 = conv(dilations = var_1623, groups = var_27, pad = input_459_pad_0, pad_type = input_459_pad_type_0, strides = var_1621, weight = model_layers_8_conv_pointwise_conv1_weight_to_fp16, x = transpose_227)[name = tensor<string, []>("input_459_cast_fp16")];
+            tensor<int32, []> x_195_split_num_splits_0 = const()[name = tensor<string, []>("x_195_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_195_split_axis_0 = const()[name = tensor<string, []>("x_195_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_195_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_195_split_cast_fp16_1 = split(axis = x_195_split_axis_0, num_splits = x_195_split_num_splits_0, x = input_459_cast_fp16)[name = tensor<string, []>("x_195_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_195_split_1_sigmoid_cast_fp16 = sigmoid(x = x_195_split_cast_fp16_1)[name = tensor<string, []>("x_195_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_195_cast_fp16 = mul(x = x_195_split_cast_fp16_0, y = x_195_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_195_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_461_cast_fp16 = select(a = var_6_to_fp16, b = x_195_cast_fp16, cond = var_323)[name = tensor<string, []>("input_461_cast_fp16")];
+            tensor<int32, [6]> input_463_pad_0 = const()[name = tensor<string, []>("input_463_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_463_mode_0 = const()[name = tensor<string, []>("input_463_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_96_to_fp16 = const()[name = tensor<string, []>("const_96_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_463_cast_fp16 = pad(constant_val = const_96_to_fp16, mode = input_463_mode_0, pad = input_463_pad_0, x = input_461_cast_fp16)[name = tensor<string, []>("input_463_cast_fp16")];
+            tensor<int32, [1]> var_1632 = const()[name = tensor<string, []>("op_1632"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_1634 = const()[name = tensor<string, []>("op_1634"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_465_pad_type_0 = const()[name = tensor<string, []>("input_465_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_465_pad_0 = const()[name = tensor<string, []>("input_465_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_467_weight_0_to_fp16 = const()[name = tensor<string, []>("input_467_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(428933568)))];
+            tensor<fp16, [1024]> input_467_bias_0_to_fp16 = const()[name = tensor<string, []>("input_467_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(428952064)))];
+            tensor<fp16, [1, 1024, 126]> input_467_cast_fp16 = conv(bias = input_467_bias_0_to_fp16, dilations = var_1634, groups = var_5, pad = input_465_pad_0, pad_type = input_465_pad_type_0, strides = var_1632, weight = input_467_weight_0_to_fp16, x = input_463_cast_fp16)[name = tensor<string, []>("input_467_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_469_cast_fp16 = silu(x = input_467_cast_fp16)[name = tensor<string, []>("input_469_cast_fp16")];
+            tensor<int32, [1]> var_1644 = const()[name = tensor<string, []>("op_1644"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_1646 = const()[name = tensor<string, []>("op_1646"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_197_pad_type_0 = const()[name = tensor<string, []>("x_197_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_197_pad_0 = const()[name = tensor<string, []>("x_197_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_8_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_8_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(428954176)))];
+            tensor<fp16, [1, 1024, 126]> x_197_cast_fp16 = conv(dilations = var_1646, groups = var_27, pad = x_197_pad_0, pad_type = x_197_pad_type_0, strides = var_1644, weight = model_layers_8_conv_pointwise_conv2_weight_to_fp16, x = input_469_cast_fp16)[name = tensor<string, []>("x_197_cast_fp16")];
+            tensor<int32, [3]> input_471_perm_0 = const()[name = tensor<string, []>("input_471_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_226 = transpose(perm = input_471_perm_0, x = x_197_cast_fp16)[name = tensor<string, []>("transpose_226")];
+            tensor<fp16, [1, 126, 1024]> input_473_cast_fp16 = add(x = input_455_cast_fp16, y = transpose_226)[name = tensor<string, []>("input_473_cast_fp16")];
+            tensor<int32, [1]> input_475_axes_0 = const()[name = tensor<string, []>("input_475_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_8_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_8_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(431051392)))];
+            tensor<fp16, [1024]> model_layers_8_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_8_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(431053504)))];
+            tensor<fp16, [1, 126, 1024]> input_475_cast_fp16 = layer_norm(axes = input_475_axes_0, beta = model_layers_8_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_8_norm_feed_forward2_weight_to_fp16, x = input_473_cast_fp16)[name = tensor<string, []>("input_475_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_8_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_8_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(431055616)))];
+            tensor<fp16, [1, 126, 4096]> linear_80_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_8_feed_forward2_linear1_weight_to_fp16, x = input_475_cast_fp16)[name = tensor<string, []>("linear_80_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_479_cast_fp16 = silu(x = linear_80_cast_fp16)[name = tensor<string, []>("input_479_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_8_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_8_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(439444288)))];
+            tensor<fp16, [1, 126, 1024]> linear_81_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_8_feed_forward2_linear2_weight_to_fp16, x = input_479_cast_fp16)[name = tensor<string, []>("linear_81_cast_fp16")];
+            tensor<fp16, []> var_1665_to_fp16 = const()[name = tensor<string, []>("op_1665_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1666_cast_fp16 = mul(x = linear_81_cast_fp16, y = var_1665_to_fp16)[name = tensor<string, []>("op_1666_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_485_cast_fp16 = add(x = input_473_cast_fp16, y = var_1666_cast_fp16)[name = tensor<string, []>("input_485_cast_fp16")];
+            tensor<int32, [1]> input_487_axes_0 = const()[name = tensor<string, []>("input_487_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_8_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_8_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(447832960)))];
+            tensor<fp16, [1024]> model_layers_8_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_8_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(447835072)))];
+            tensor<fp16, [1, 126, 1024]> input_487_cast_fp16 = layer_norm(axes = input_487_axes_0, beta = model_layers_8_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_8_norm_out_weight_to_fp16, x = input_485_cast_fp16)[name = tensor<string, []>("input_487_cast_fp16")];
+            tensor<int32, [1]> input_489_axes_0 = const()[name = tensor<string, []>("input_489_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_9_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_9_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(447837184)))];
+            tensor<fp16, [1024]> model_layers_9_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_9_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(447839296)))];
+            tensor<fp16, [1, 126, 1024]> input_489_cast_fp16 = layer_norm(axes = input_489_axes_0, beta = model_layers_9_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_9_norm_feed_forward1_weight_to_fp16, x = input_487_cast_fp16)[name = tensor<string, []>("input_489_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_9_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_9_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(447841408)))];
+            tensor<fp16, [1, 126, 4096]> linear_82_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_9_feed_forward1_linear1_weight_to_fp16, x = input_489_cast_fp16)[name = tensor<string, []>("linear_82_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_493_cast_fp16 = silu(x = linear_82_cast_fp16)[name = tensor<string, []>("input_493_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_9_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_9_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(456230080)))];
+            tensor<fp16, [1, 126, 1024]> linear_83_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_9_feed_forward1_linear2_weight_to_fp16, x = input_493_cast_fp16)[name = tensor<string, []>("linear_83_cast_fp16")];
+            tensor<fp16, []> var_1694_to_fp16 = const()[name = tensor<string, []>("op_1694_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1695_cast_fp16 = mul(x = linear_83_cast_fp16, y = var_1694_to_fp16)[name = tensor<string, []>("op_1695_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_499_cast_fp16 = add(x = input_487_cast_fp16, y = var_1695_cast_fp16)[name = tensor<string, []>("input_499_cast_fp16")];
+            tensor<int32, [1]> query_19_axes_0 = const()[name = tensor<string, []>("query_19_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_9_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_9_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(464618752)))];
+            tensor<fp16, [1024]> model_layers_9_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_9_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(464620864)))];
+            tensor<fp16, [1, 126, 1024]> query_19_cast_fp16 = layer_norm(axes = query_19_axes_0, beta = model_layers_9_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_9_norm_self_att_weight_to_fp16, x = input_499_cast_fp16)[name = tensor<string, []>("query_19_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_9_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_9_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(464622976)))];
+            tensor<fp16, [1, 126, 1024]> linear_84_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_9_self_attn_linear_q_weight_to_fp16, x = query_19_cast_fp16)[name = tensor<string, []>("linear_84_cast_fp16")];
+            tensor<int32, [4]> var_1711 = const()[name = tensor<string, []>("op_1711"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_55_cast_fp16 = reshape(shape = var_1711, x = linear_84_cast_fp16)[name = tensor<string, []>("q_55_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_9_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_9_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(466720192)))];
+            tensor<fp16, [1, 126, 1024]> linear_85_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_9_self_attn_linear_k_weight_to_fp16, x = query_19_cast_fp16)[name = tensor<string, []>("linear_85_cast_fp16")];
+            tensor<int32, [4]> var_1715 = const()[name = tensor<string, []>("op_1715"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_37_cast_fp16 = reshape(shape = var_1715, x = linear_85_cast_fp16)[name = tensor<string, []>("k_37_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_9_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_9_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(468817408)))];
+            tensor<fp16, [1, 126, 1024]> linear_86_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_9_self_attn_linear_v_weight_to_fp16, x = query_19_cast_fp16)[name = tensor<string, []>("linear_86_cast_fp16")];
+            tensor<int32, [4]> var_1719 = const()[name = tensor<string, []>("op_1719"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_19_cast_fp16 = reshape(shape = var_1719, x = linear_86_cast_fp16)[name = tensor<string, []>("v_19_cast_fp16")];
+            tensor<int32, [4]> value_19_perm_0 = const()[name = tensor<string, []>("value_19_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_9_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_9_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(470914624)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1731_cast_fp16 = add(x = q_55_cast_fp16, y = model_layers_9_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_1731_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_9_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_9_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(470916736)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1733_cast_fp16 = add(x = q_55_cast_fp16, y = model_layers_9_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_1733_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_19_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_19_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_205_transpose_x_0 = const()[name = tensor<string, []>("x_205_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_205_transpose_y_0 = const()[name = tensor<string, []>("x_205_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_1735_to_fp16 = const()[name = tensor<string, []>("op_1735_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(470918848)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_224 = transpose(perm = q_with_bias_v_19_perm_0, x = var_1733_cast_fp16)[name = tensor<string, []>("transpose_224")];
+            tensor<fp16, [1, 8, 126, 251]> x_205_cast_fp16 = matmul(transpose_x = x_205_transpose_x_0, transpose_y = x_205_transpose_y_0, x = transpose_224, y = var_1735_to_fp16)[name = tensor<string, []>("x_205_cast_fp16")];
+            tensor<int32, [8]> x_207_pad_0 = const()[name = tensor<string, []>("x_207_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_207_mode_0 = const()[name = tensor<string, []>("x_207_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_103_to_fp16 = const()[name = tensor<string, []>("const_103_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_207_cast_fp16 = pad(constant_val = const_103_to_fp16, mode = x_207_mode_0, pad = x_207_pad_0, x = x_205_cast_fp16)[name = tensor<string, []>("x_207_cast_fp16")];
+            tensor<int32, [4]> var_1743 = const()[name = tensor<string, []>("op_1743"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_209_cast_fp16 = reshape(shape = var_1743, x = x_207_cast_fp16)[name = tensor<string, []>("x_209_cast_fp16")];
+            tensor<int32, [4]> var_1747_begin_0 = const()[name = tensor<string, []>("op_1747_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1747_end_0 = const()[name = tensor<string, []>("op_1747_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_1747_end_mask_0 = const()[name = tensor<string, []>("op_1747_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_1747_cast_fp16 = slice_by_index(begin = var_1747_begin_0, end = var_1747_end_0, end_mask = var_1747_end_mask_0, x = x_209_cast_fp16)[name = tensor<string, []>("op_1747_cast_fp16")];
+            tensor<int32, [4]> var_1748 = const()[name = tensor<string, []>("op_1748"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_37_cast_fp16 = reshape(shape = var_1748, x = var_1747_cast_fp16)[name = tensor<string, []>("matrix_bd_37_cast_fp16")];
+            tensor<bool, []> matrix_ac_19_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_19_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_19_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_19_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_90_perm_0 = const()[name = tensor<string, []>("transpose_90_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_91_perm_0 = const()[name = tensor<string, []>("transpose_91_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_222 = transpose(perm = transpose_91_perm_0, x = k_37_cast_fp16)[name = tensor<string, []>("transpose_222")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_223 = transpose(perm = transpose_90_perm_0, x = var_1731_cast_fp16)[name = tensor<string, []>("transpose_223")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_19_cast_fp16 = matmul(transpose_x = matrix_ac_19_transpose_x_0, transpose_y = matrix_ac_19_transpose_y_0, x = transpose_223, y = transpose_222)[name = tensor<string, []>("matrix_ac_19_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_39_begin_0 = const()[name = tensor<string, []>("matrix_bd_39_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_39_end_0 = const()[name = tensor<string, []>("matrix_bd_39_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_39_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_39_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_39_cast_fp16 = slice_by_index(begin = matrix_bd_39_begin_0, end = matrix_bd_39_end_0, end_mask = matrix_bd_39_end_mask_0, x = matrix_bd_37_cast_fp16)[name = tensor<string, []>("matrix_bd_39_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1757_cast_fp16 = add(x = matrix_ac_19_cast_fp16, y = matrix_bd_39_cast_fp16)[name = tensor<string, []>("op_1757_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_37_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_37_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_37_cast_fp16 = mul(x = var_1757_cast_fp16, y = _inversed_scores_37_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_37_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_39_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_37_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_39_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1763_cast_fp16 = softmax(axis = var_25, x = scores_39_cast_fp16)[name = tensor<string, []>("op_1763_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_501_cast_fp16 = select(a = var_6_to_fp16, b = var_1763_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_501_cast_fp16")];
+            tensor<bool, []> x_211_transpose_x_0 = const()[name = tensor<string, []>("x_211_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_211_transpose_y_0 = const()[name = tensor<string, []>("x_211_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_225 = transpose(perm = value_19_perm_0, x = v_19_cast_fp16)[name = tensor<string, []>("transpose_225")];
+            tensor<fp16, [1, 8, 126, 128]> x_211_cast_fp16 = matmul(transpose_x = x_211_transpose_x_0, transpose_y = x_211_transpose_y_0, x = input_501_cast_fp16, y = transpose_225)[name = tensor<string, []>("x_211_cast_fp16")];
+            tensor<int32, [4]> var_1767_perm_0 = const()[name = tensor<string, []>("op_1767_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1768 = const()[name = tensor<string, []>("op_1768"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_221 = transpose(perm = var_1767_perm_0, x = x_211_cast_fp16)[name = tensor<string, []>("transpose_221")];
+            tensor<fp16, [1, 126, 1024]> input_503_cast_fp16 = reshape(shape = var_1768, x = transpose_221)[name = tensor<string, []>("input_503_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_9_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_9_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(471432960)))];
+            tensor<fp16, [1, 126, 1024]> linear_88_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_9_self_attn_linear_out_weight_to_fp16, x = input_503_cast_fp16)[name = tensor<string, []>("linear_88_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_507_cast_fp16 = add(x = input_499_cast_fp16, y = linear_88_cast_fp16)[name = tensor<string, []>("input_507_cast_fp16")];
+            tensor<int32, [1]> x_215_axes_0 = const()[name = tensor<string, []>("x_215_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_9_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_9_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(473530176)))];
+            tensor<fp16, [1024]> model_layers_9_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_9_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(473532288)))];
+            tensor<fp16, [1, 126, 1024]> x_215_cast_fp16 = layer_norm(axes = x_215_axes_0, beta = model_layers_9_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_9_norm_conv_weight_to_fp16, x = input_507_cast_fp16)[name = tensor<string, []>("x_215_cast_fp16")];
+            tensor<int32, [3]> input_509_perm_0 = const()[name = tensor<string, []>("input_509_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_1784 = const()[name = tensor<string, []>("op_1784"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_1786 = const()[name = tensor<string, []>("op_1786"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_511_pad_type_0 = const()[name = tensor<string, []>("input_511_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_511_pad_0 = const()[name = tensor<string, []>("input_511_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_9_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_9_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(473534400)))];
+            tensor<fp16, [1, 1024, 126]> transpose_220 = transpose(perm = input_509_perm_0, x = x_215_cast_fp16)[name = tensor<string, []>("transpose_220")];
+            tensor<fp16, [1, 2048, 126]> input_511_cast_fp16 = conv(dilations = var_1786, groups = var_27, pad = input_511_pad_0, pad_type = input_511_pad_type_0, strides = var_1784, weight = model_layers_9_conv_pointwise_conv1_weight_to_fp16, x = transpose_220)[name = tensor<string, []>("input_511_cast_fp16")];
+            tensor<int32, []> x_217_split_num_splits_0 = const()[name = tensor<string, []>("x_217_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_217_split_axis_0 = const()[name = tensor<string, []>("x_217_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_217_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_217_split_cast_fp16_1 = split(axis = x_217_split_axis_0, num_splits = x_217_split_num_splits_0, x = input_511_cast_fp16)[name = tensor<string, []>("x_217_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_217_split_1_sigmoid_cast_fp16 = sigmoid(x = x_217_split_cast_fp16_1)[name = tensor<string, []>("x_217_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_217_cast_fp16 = mul(x = x_217_split_cast_fp16_0, y = x_217_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_217_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_513_cast_fp16 = select(a = var_6_to_fp16, b = x_217_cast_fp16, cond = var_323)[name = tensor<string, []>("input_513_cast_fp16")];
+            tensor<int32, [6]> input_515_pad_0 = const()[name = tensor<string, []>("input_515_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_515_mode_0 = const()[name = tensor<string, []>("input_515_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_106_to_fp16 = const()[name = tensor<string, []>("const_106_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_515_cast_fp16 = pad(constant_val = const_106_to_fp16, mode = input_515_mode_0, pad = input_515_pad_0, x = input_513_cast_fp16)[name = tensor<string, []>("input_515_cast_fp16")];
+            tensor<int32, [1]> var_1795 = const()[name = tensor<string, []>("op_1795"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_1797 = const()[name = tensor<string, []>("op_1797"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_517_pad_type_0 = const()[name = tensor<string, []>("input_517_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_517_pad_0 = const()[name = tensor<string, []>("input_517_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_519_weight_0_to_fp16 = const()[name = tensor<string, []>("input_519_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(477728768)))];
+            tensor<fp16, [1024]> input_519_bias_0_to_fp16 = const()[name = tensor<string, []>("input_519_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(477747264)))];
+            tensor<fp16, [1, 1024, 126]> input_519_cast_fp16 = conv(bias = input_519_bias_0_to_fp16, dilations = var_1797, groups = var_5, pad = input_517_pad_0, pad_type = input_517_pad_type_0, strides = var_1795, weight = input_519_weight_0_to_fp16, x = input_515_cast_fp16)[name = tensor<string, []>("input_519_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_521_cast_fp16 = silu(x = input_519_cast_fp16)[name = tensor<string, []>("input_521_cast_fp16")];
+            tensor<int32, [1]> var_1807 = const()[name = tensor<string, []>("op_1807"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_1809 = const()[name = tensor<string, []>("op_1809"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_219_pad_type_0 = const()[name = tensor<string, []>("x_219_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_219_pad_0 = const()[name = tensor<string, []>("x_219_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_9_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_9_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(477749376)))];
+            tensor<fp16, [1, 1024, 126]> x_219_cast_fp16 = conv(dilations = var_1809, groups = var_27, pad = x_219_pad_0, pad_type = x_219_pad_type_0, strides = var_1807, weight = model_layers_9_conv_pointwise_conv2_weight_to_fp16, x = input_521_cast_fp16)[name = tensor<string, []>("x_219_cast_fp16")];
+            tensor<int32, [3]> input_523_perm_0 = const()[name = tensor<string, []>("input_523_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_219 = transpose(perm = input_523_perm_0, x = x_219_cast_fp16)[name = tensor<string, []>("transpose_219")];
+            tensor<fp16, [1, 126, 1024]> input_525_cast_fp16 = add(x = input_507_cast_fp16, y = transpose_219)[name = tensor<string, []>("input_525_cast_fp16")];
+            tensor<int32, [1]> input_527_axes_0 = const()[name = tensor<string, []>("input_527_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_9_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_9_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(479846592)))];
+            tensor<fp16, [1024]> model_layers_9_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_9_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(479848704)))];
+            tensor<fp16, [1, 126, 1024]> input_527_cast_fp16 = layer_norm(axes = input_527_axes_0, beta = model_layers_9_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_9_norm_feed_forward2_weight_to_fp16, x = input_525_cast_fp16)[name = tensor<string, []>("input_527_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_9_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_9_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(479850816)))];
+            tensor<fp16, [1, 126, 4096]> linear_89_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_9_feed_forward2_linear1_weight_to_fp16, x = input_527_cast_fp16)[name = tensor<string, []>("linear_89_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_531_cast_fp16 = silu(x = linear_89_cast_fp16)[name = tensor<string, []>("input_531_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_9_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_9_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(488239488)))];
+            tensor<fp16, [1, 126, 1024]> linear_90_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_9_feed_forward2_linear2_weight_to_fp16, x = input_531_cast_fp16)[name = tensor<string, []>("linear_90_cast_fp16")];
+            tensor<fp16, []> var_1828_to_fp16 = const()[name = tensor<string, []>("op_1828_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1829_cast_fp16 = mul(x = linear_90_cast_fp16, y = var_1828_to_fp16)[name = tensor<string, []>("op_1829_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_537_cast_fp16 = add(x = input_525_cast_fp16, y = var_1829_cast_fp16)[name = tensor<string, []>("input_537_cast_fp16")];
+            tensor<int32, [1]> input_539_axes_0 = const()[name = tensor<string, []>("input_539_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_9_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_9_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496628160)))];
+            tensor<fp16, [1024]> model_layers_9_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_9_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496630272)))];
+            tensor<fp16, [1, 126, 1024]> input_539_cast_fp16 = layer_norm(axes = input_539_axes_0, beta = model_layers_9_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_9_norm_out_weight_to_fp16, x = input_537_cast_fp16)[name = tensor<string, []>("input_539_cast_fp16")];
+            tensor<int32, [1]> input_541_axes_0 = const()[name = tensor<string, []>("input_541_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_10_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_10_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496632384)))];
+            tensor<fp16, [1024]> model_layers_10_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_10_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496634496)))];
+            tensor<fp16, [1, 126, 1024]> input_541_cast_fp16 = layer_norm(axes = input_541_axes_0, beta = model_layers_10_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_10_norm_feed_forward1_weight_to_fp16, x = input_539_cast_fp16)[name = tensor<string, []>("input_541_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_10_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_10_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496636608)))];
+            tensor<fp16, [1, 126, 4096]> linear_91_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_10_feed_forward1_linear1_weight_to_fp16, x = input_541_cast_fp16)[name = tensor<string, []>("linear_91_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_545_cast_fp16 = silu(x = linear_91_cast_fp16)[name = tensor<string, []>("input_545_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_10_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_10_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(505025280)))];
+            tensor<fp16, [1, 126, 1024]> linear_92_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_10_feed_forward1_linear2_weight_to_fp16, x = input_545_cast_fp16)[name = tensor<string, []>("linear_92_cast_fp16")];
+            tensor<fp16, []> var_1857_to_fp16 = const()[name = tensor<string, []>("op_1857_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1858_cast_fp16 = mul(x = linear_92_cast_fp16, y = var_1857_to_fp16)[name = tensor<string, []>("op_1858_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_551_cast_fp16 = add(x = input_539_cast_fp16, y = var_1858_cast_fp16)[name = tensor<string, []>("input_551_cast_fp16")];
+            tensor<int32, [1]> query_21_axes_0 = const()[name = tensor<string, []>("query_21_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_10_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_10_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(513413952)))];
+            tensor<fp16, [1024]> model_layers_10_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_10_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(513416064)))];
+            tensor<fp16, [1, 126, 1024]> query_21_cast_fp16 = layer_norm(axes = query_21_axes_0, beta = model_layers_10_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_10_norm_self_att_weight_to_fp16, x = input_551_cast_fp16)[name = tensor<string, []>("query_21_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_10_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_10_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(513418176)))];
+            tensor<fp16, [1, 126, 1024]> linear_93_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_10_self_attn_linear_q_weight_to_fp16, x = query_21_cast_fp16)[name = tensor<string, []>("linear_93_cast_fp16")];
+            tensor<int32, [4]> var_1874 = const()[name = tensor<string, []>("op_1874"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_61_cast_fp16 = reshape(shape = var_1874, x = linear_93_cast_fp16)[name = tensor<string, []>("q_61_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_10_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_10_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(515515392)))];
+            tensor<fp16, [1, 126, 1024]> linear_94_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_10_self_attn_linear_k_weight_to_fp16, x = query_21_cast_fp16)[name = tensor<string, []>("linear_94_cast_fp16")];
+            tensor<int32, [4]> var_1878 = const()[name = tensor<string, []>("op_1878"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_41_cast_fp16 = reshape(shape = var_1878, x = linear_94_cast_fp16)[name = tensor<string, []>("k_41_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_10_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_10_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(517612608)))];
+            tensor<fp16, [1, 126, 1024]> linear_95_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_10_self_attn_linear_v_weight_to_fp16, x = query_21_cast_fp16)[name = tensor<string, []>("linear_95_cast_fp16")];
+            tensor<int32, [4]> var_1882 = const()[name = tensor<string, []>("op_1882"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_21_cast_fp16 = reshape(shape = var_1882, x = linear_95_cast_fp16)[name = tensor<string, []>("v_21_cast_fp16")];
+            tensor<int32, [4]> value_21_perm_0 = const()[name = tensor<string, []>("value_21_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_10_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_10_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(519709824)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1894_cast_fp16 = add(x = q_61_cast_fp16, y = model_layers_10_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_1894_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_10_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_10_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(519711936)))];
+            tensor<fp16, [1, 126, 8, 128]> var_1896_cast_fp16 = add(x = q_61_cast_fp16, y = model_layers_10_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_1896_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_21_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_21_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_227_transpose_x_0 = const()[name = tensor<string, []>("x_227_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_227_transpose_y_0 = const()[name = tensor<string, []>("x_227_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_1898_to_fp16 = const()[name = tensor<string, []>("op_1898_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(519714048)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_217 = transpose(perm = q_with_bias_v_21_perm_0, x = var_1896_cast_fp16)[name = tensor<string, []>("transpose_217")];
+            tensor<fp16, [1, 8, 126, 251]> x_227_cast_fp16 = matmul(transpose_x = x_227_transpose_x_0, transpose_y = x_227_transpose_y_0, x = transpose_217, y = var_1898_to_fp16)[name = tensor<string, []>("x_227_cast_fp16")];
+            tensor<int32, [8]> x_229_pad_0 = const()[name = tensor<string, []>("x_229_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_229_mode_0 = const()[name = tensor<string, []>("x_229_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_113_to_fp16 = const()[name = tensor<string, []>("const_113_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_229_cast_fp16 = pad(constant_val = const_113_to_fp16, mode = x_229_mode_0, pad = x_229_pad_0, x = x_227_cast_fp16)[name = tensor<string, []>("x_229_cast_fp16")];
+            tensor<int32, [4]> var_1906 = const()[name = tensor<string, []>("op_1906"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_231_cast_fp16 = reshape(shape = var_1906, x = x_229_cast_fp16)[name = tensor<string, []>("x_231_cast_fp16")];
+            tensor<int32, [4]> var_1910_begin_0 = const()[name = tensor<string, []>("op_1910_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_1910_end_0 = const()[name = tensor<string, []>("op_1910_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_1910_end_mask_0 = const()[name = tensor<string, []>("op_1910_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_1910_cast_fp16 = slice_by_index(begin = var_1910_begin_0, end = var_1910_end_0, end_mask = var_1910_end_mask_0, x = x_231_cast_fp16)[name = tensor<string, []>("op_1910_cast_fp16")];
+            tensor<int32, [4]> var_1911 = const()[name = tensor<string, []>("op_1911"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_41_cast_fp16 = reshape(shape = var_1911, x = var_1910_cast_fp16)[name = tensor<string, []>("matrix_bd_41_cast_fp16")];
+            tensor<bool, []> matrix_ac_21_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_21_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_21_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_21_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_92_perm_0 = const()[name = tensor<string, []>("transpose_92_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_93_perm_0 = const()[name = tensor<string, []>("transpose_93_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_215 = transpose(perm = transpose_93_perm_0, x = k_41_cast_fp16)[name = tensor<string, []>("transpose_215")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_216 = transpose(perm = transpose_92_perm_0, x = var_1894_cast_fp16)[name = tensor<string, []>("transpose_216")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_21_cast_fp16 = matmul(transpose_x = matrix_ac_21_transpose_x_0, transpose_y = matrix_ac_21_transpose_y_0, x = transpose_216, y = transpose_215)[name = tensor<string, []>("matrix_ac_21_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_43_begin_0 = const()[name = tensor<string, []>("matrix_bd_43_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_43_end_0 = const()[name = tensor<string, []>("matrix_bd_43_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_43_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_43_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_43_cast_fp16 = slice_by_index(begin = matrix_bd_43_begin_0, end = matrix_bd_43_end_0, end_mask = matrix_bd_43_end_mask_0, x = matrix_bd_41_cast_fp16)[name = tensor<string, []>("matrix_bd_43_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1920_cast_fp16 = add(x = matrix_ac_21_cast_fp16, y = matrix_bd_43_cast_fp16)[name = tensor<string, []>("op_1920_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_41_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_41_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_41_cast_fp16 = mul(x = var_1920_cast_fp16, y = _inversed_scores_41_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_41_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_43_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_41_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_43_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_1926_cast_fp16 = softmax(axis = var_25, x = scores_43_cast_fp16)[name = tensor<string, []>("op_1926_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_553_cast_fp16 = select(a = var_6_to_fp16, b = var_1926_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_553_cast_fp16")];
+            tensor<bool, []> x_233_transpose_x_0 = const()[name = tensor<string, []>("x_233_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_233_transpose_y_0 = const()[name = tensor<string, []>("x_233_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_218 = transpose(perm = value_21_perm_0, x = v_21_cast_fp16)[name = tensor<string, []>("transpose_218")];
+            tensor<fp16, [1, 8, 126, 128]> x_233_cast_fp16 = matmul(transpose_x = x_233_transpose_x_0, transpose_y = x_233_transpose_y_0, x = input_553_cast_fp16, y = transpose_218)[name = tensor<string, []>("x_233_cast_fp16")];
+            tensor<int32, [4]> var_1930_perm_0 = const()[name = tensor<string, []>("op_1930_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_1931 = const()[name = tensor<string, []>("op_1931"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_214 = transpose(perm = var_1930_perm_0, x = x_233_cast_fp16)[name = tensor<string, []>("transpose_214")];
+            tensor<fp16, [1, 126, 1024]> input_555_cast_fp16 = reshape(shape = var_1931, x = transpose_214)[name = tensor<string, []>("input_555_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_10_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_10_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(520228160)))];
+            tensor<fp16, [1, 126, 1024]> linear_97_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_10_self_attn_linear_out_weight_to_fp16, x = input_555_cast_fp16)[name = tensor<string, []>("linear_97_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_559_cast_fp16 = add(x = input_551_cast_fp16, y = linear_97_cast_fp16)[name = tensor<string, []>("input_559_cast_fp16")];
+            tensor<int32, [1]> x_237_axes_0 = const()[name = tensor<string, []>("x_237_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_10_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_10_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(522325376)))];
+            tensor<fp16, [1024]> model_layers_10_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_10_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(522327488)))];
+            tensor<fp16, [1, 126, 1024]> x_237_cast_fp16 = layer_norm(axes = x_237_axes_0, beta = model_layers_10_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_10_norm_conv_weight_to_fp16, x = input_559_cast_fp16)[name = tensor<string, []>("x_237_cast_fp16")];
+            tensor<int32, [3]> input_561_perm_0 = const()[name = tensor<string, []>("input_561_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_1947 = const()[name = tensor<string, []>("op_1947"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_1949 = const()[name = tensor<string, []>("op_1949"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_563_pad_type_0 = const()[name = tensor<string, []>("input_563_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_563_pad_0 = const()[name = tensor<string, []>("input_563_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_10_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_10_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(522329600)))];
+            tensor<fp16, [1, 1024, 126]> transpose_213 = transpose(perm = input_561_perm_0, x = x_237_cast_fp16)[name = tensor<string, []>("transpose_213")];
+            tensor<fp16, [1, 2048, 126]> input_563_cast_fp16 = conv(dilations = var_1949, groups = var_27, pad = input_563_pad_0, pad_type = input_563_pad_type_0, strides = var_1947, weight = model_layers_10_conv_pointwise_conv1_weight_to_fp16, x = transpose_213)[name = tensor<string, []>("input_563_cast_fp16")];
+            tensor<int32, []> x_239_split_num_splits_0 = const()[name = tensor<string, []>("x_239_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_239_split_axis_0 = const()[name = tensor<string, []>("x_239_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_239_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_239_split_cast_fp16_1 = split(axis = x_239_split_axis_0, num_splits = x_239_split_num_splits_0, x = input_563_cast_fp16)[name = tensor<string, []>("x_239_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_239_split_1_sigmoid_cast_fp16 = sigmoid(x = x_239_split_cast_fp16_1)[name = tensor<string, []>("x_239_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_239_cast_fp16 = mul(x = x_239_split_cast_fp16_0, y = x_239_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_239_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_565_cast_fp16 = select(a = var_6_to_fp16, b = x_239_cast_fp16, cond = var_323)[name = tensor<string, []>("input_565_cast_fp16")];
+            tensor<int32, [6]> input_567_pad_0 = const()[name = tensor<string, []>("input_567_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_567_mode_0 = const()[name = tensor<string, []>("input_567_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_116_to_fp16 = const()[name = tensor<string, []>("const_116_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_567_cast_fp16 = pad(constant_val = const_116_to_fp16, mode = input_567_mode_0, pad = input_567_pad_0, x = input_565_cast_fp16)[name = tensor<string, []>("input_567_cast_fp16")];
+            tensor<int32, [1]> var_1958 = const()[name = tensor<string, []>("op_1958"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_1960 = const()[name = tensor<string, []>("op_1960"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_569_pad_type_0 = const()[name = tensor<string, []>("input_569_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_569_pad_0 = const()[name = tensor<string, []>("input_569_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_571_weight_0_to_fp16 = const()[name = tensor<string, []>("input_571_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526523968)))];
+            tensor<fp16, [1024]> input_571_bias_0_to_fp16 = const()[name = tensor<string, []>("input_571_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526542464)))];
+            tensor<fp16, [1, 1024, 126]> input_571_cast_fp16 = conv(bias = input_571_bias_0_to_fp16, dilations = var_1960, groups = var_5, pad = input_569_pad_0, pad_type = input_569_pad_type_0, strides = var_1958, weight = input_571_weight_0_to_fp16, x = input_567_cast_fp16)[name = tensor<string, []>("input_571_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_573_cast_fp16 = silu(x = input_571_cast_fp16)[name = tensor<string, []>("input_573_cast_fp16")];
+            tensor<int32, [1]> var_1970 = const()[name = tensor<string, []>("op_1970"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_1972 = const()[name = tensor<string, []>("op_1972"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_241_pad_type_0 = const()[name = tensor<string, []>("x_241_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_241_pad_0 = const()[name = tensor<string, []>("x_241_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_10_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_10_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(526544576)))];
+            tensor<fp16, [1, 1024, 126]> x_241_cast_fp16 = conv(dilations = var_1972, groups = var_27, pad = x_241_pad_0, pad_type = x_241_pad_type_0, strides = var_1970, weight = model_layers_10_conv_pointwise_conv2_weight_to_fp16, x = input_573_cast_fp16)[name = tensor<string, []>("x_241_cast_fp16")];
+            tensor<int32, [3]> input_575_perm_0 = const()[name = tensor<string, []>("input_575_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_212 = transpose(perm = input_575_perm_0, x = x_241_cast_fp16)[name = tensor<string, []>("transpose_212")];
+            tensor<fp16, [1, 126, 1024]> input_577_cast_fp16 = add(x = input_559_cast_fp16, y = transpose_212)[name = tensor<string, []>("input_577_cast_fp16")];
+            tensor<int32, [1]> input_579_axes_0 = const()[name = tensor<string, []>("input_579_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_10_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_10_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(528641792)))];
+            tensor<fp16, [1024]> model_layers_10_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_10_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(528643904)))];
+            tensor<fp16, [1, 126, 1024]> input_579_cast_fp16 = layer_norm(axes = input_579_axes_0, beta = model_layers_10_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_10_norm_feed_forward2_weight_to_fp16, x = input_577_cast_fp16)[name = tensor<string, []>("input_579_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_10_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_10_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(528646016)))];
+            tensor<fp16, [1, 126, 4096]> linear_98_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_10_feed_forward2_linear1_weight_to_fp16, x = input_579_cast_fp16)[name = tensor<string, []>("linear_98_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_583_cast_fp16 = silu(x = linear_98_cast_fp16)[name = tensor<string, []>("input_583_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_10_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_10_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(537034688)))];
+            tensor<fp16, [1, 126, 1024]> linear_99_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_10_feed_forward2_linear2_weight_to_fp16, x = input_583_cast_fp16)[name = tensor<string, []>("linear_99_cast_fp16")];
+            tensor<fp16, []> var_1991_to_fp16 = const()[name = tensor<string, []>("op_1991_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_1992_cast_fp16 = mul(x = linear_99_cast_fp16, y = var_1991_to_fp16)[name = tensor<string, []>("op_1992_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_589_cast_fp16 = add(x = input_577_cast_fp16, y = var_1992_cast_fp16)[name = tensor<string, []>("input_589_cast_fp16")];
+            tensor<int32, [1]> input_591_axes_0 = const()[name = tensor<string, []>("input_591_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_10_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_10_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(545423360)))];
+            tensor<fp16, [1024]> model_layers_10_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_10_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(545425472)))];
+            tensor<fp16, [1, 126, 1024]> input_591_cast_fp16 = layer_norm(axes = input_591_axes_0, beta = model_layers_10_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_10_norm_out_weight_to_fp16, x = input_589_cast_fp16)[name = tensor<string, []>("input_591_cast_fp16")];
+            tensor<int32, [1]> input_593_axes_0 = const()[name = tensor<string, []>("input_593_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_11_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_11_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(545427584)))];
+            tensor<fp16, [1024]> model_layers_11_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_11_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(545429696)))];
+            tensor<fp16, [1, 126, 1024]> input_593_cast_fp16 = layer_norm(axes = input_593_axes_0, beta = model_layers_11_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_11_norm_feed_forward1_weight_to_fp16, x = input_591_cast_fp16)[name = tensor<string, []>("input_593_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_11_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_11_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(545431808)))];
+            tensor<fp16, [1, 126, 4096]> linear_100_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_11_feed_forward1_linear1_weight_to_fp16, x = input_593_cast_fp16)[name = tensor<string, []>("linear_100_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_597_cast_fp16 = silu(x = linear_100_cast_fp16)[name = tensor<string, []>("input_597_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_11_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_11_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(553820480)))];
+            tensor<fp16, [1, 126, 1024]> linear_101_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_11_feed_forward1_linear2_weight_to_fp16, x = input_597_cast_fp16)[name = tensor<string, []>("linear_101_cast_fp16")];
+            tensor<fp16, []> var_2020_to_fp16 = const()[name = tensor<string, []>("op_2020_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2021_cast_fp16 = mul(x = linear_101_cast_fp16, y = var_2020_to_fp16)[name = tensor<string, []>("op_2021_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_603_cast_fp16 = add(x = input_591_cast_fp16, y = var_2021_cast_fp16)[name = tensor<string, []>("input_603_cast_fp16")];
+            tensor<int32, [1]> query_23_axes_0 = const()[name = tensor<string, []>("query_23_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_11_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_11_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(562209152)))];
+            tensor<fp16, [1024]> model_layers_11_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_11_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(562211264)))];
+            tensor<fp16, [1, 126, 1024]> query_23_cast_fp16 = layer_norm(axes = query_23_axes_0, beta = model_layers_11_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_11_norm_self_att_weight_to_fp16, x = input_603_cast_fp16)[name = tensor<string, []>("query_23_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_11_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_11_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(562213376)))];
+            tensor<fp16, [1, 126, 1024]> linear_102_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_11_self_attn_linear_q_weight_to_fp16, x = query_23_cast_fp16)[name = tensor<string, []>("linear_102_cast_fp16")];
+            tensor<int32, [4]> var_2037 = const()[name = tensor<string, []>("op_2037"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_67_cast_fp16 = reshape(shape = var_2037, x = linear_102_cast_fp16)[name = tensor<string, []>("q_67_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_11_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_11_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(564310592)))];
+            tensor<fp16, [1, 126, 1024]> linear_103_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_11_self_attn_linear_k_weight_to_fp16, x = query_23_cast_fp16)[name = tensor<string, []>("linear_103_cast_fp16")];
+            tensor<int32, [4]> var_2041 = const()[name = tensor<string, []>("op_2041"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_45_cast_fp16 = reshape(shape = var_2041, x = linear_103_cast_fp16)[name = tensor<string, []>("k_45_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_11_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_11_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(566407808)))];
+            tensor<fp16, [1, 126, 1024]> linear_104_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_11_self_attn_linear_v_weight_to_fp16, x = query_23_cast_fp16)[name = tensor<string, []>("linear_104_cast_fp16")];
+            tensor<int32, [4]> var_2045 = const()[name = tensor<string, []>("op_2045"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_23_cast_fp16 = reshape(shape = var_2045, x = linear_104_cast_fp16)[name = tensor<string, []>("v_23_cast_fp16")];
+            tensor<int32, [4]> value_23_perm_0 = const()[name = tensor<string, []>("value_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_11_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_11_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(568505024)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2057_cast_fp16 = add(x = q_67_cast_fp16, y = model_layers_11_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_2057_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_11_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_11_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(568507136)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2059_cast_fp16 = add(x = q_67_cast_fp16, y = model_layers_11_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_2059_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_23_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_23_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_249_transpose_x_0 = const()[name = tensor<string, []>("x_249_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_249_transpose_y_0 = const()[name = tensor<string, []>("x_249_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_2061_to_fp16 = const()[name = tensor<string, []>("op_2061_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(568509248)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_210 = transpose(perm = q_with_bias_v_23_perm_0, x = var_2059_cast_fp16)[name = tensor<string, []>("transpose_210")];
+            tensor<fp16, [1, 8, 126, 251]> x_249_cast_fp16 = matmul(transpose_x = x_249_transpose_x_0, transpose_y = x_249_transpose_y_0, x = transpose_210, y = var_2061_to_fp16)[name = tensor<string, []>("x_249_cast_fp16")];
+            tensor<int32, [8]> x_251_pad_0 = const()[name = tensor<string, []>("x_251_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_251_mode_0 = const()[name = tensor<string, []>("x_251_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_123_to_fp16 = const()[name = tensor<string, []>("const_123_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_251_cast_fp16 = pad(constant_val = const_123_to_fp16, mode = x_251_mode_0, pad = x_251_pad_0, x = x_249_cast_fp16)[name = tensor<string, []>("x_251_cast_fp16")];
+            tensor<int32, [4]> var_2069 = const()[name = tensor<string, []>("op_2069"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_253_cast_fp16 = reshape(shape = var_2069, x = x_251_cast_fp16)[name = tensor<string, []>("x_253_cast_fp16")];
+            tensor<int32, [4]> var_2073_begin_0 = const()[name = tensor<string, []>("op_2073_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2073_end_0 = const()[name = tensor<string, []>("op_2073_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_2073_end_mask_0 = const()[name = tensor<string, []>("op_2073_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_2073_cast_fp16 = slice_by_index(begin = var_2073_begin_0, end = var_2073_end_0, end_mask = var_2073_end_mask_0, x = x_253_cast_fp16)[name = tensor<string, []>("op_2073_cast_fp16")];
+            tensor<int32, [4]> var_2074 = const()[name = tensor<string, []>("op_2074"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_45_cast_fp16 = reshape(shape = var_2074, x = var_2073_cast_fp16)[name = tensor<string, []>("matrix_bd_45_cast_fp16")];
+            tensor<bool, []> matrix_ac_23_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_23_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_23_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_23_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_94_perm_0 = const()[name = tensor<string, []>("transpose_94_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_95_perm_0 = const()[name = tensor<string, []>("transpose_95_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_208 = transpose(perm = transpose_95_perm_0, x = k_45_cast_fp16)[name = tensor<string, []>("transpose_208")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_209 = transpose(perm = transpose_94_perm_0, x = var_2057_cast_fp16)[name = tensor<string, []>("transpose_209")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_23_cast_fp16 = matmul(transpose_x = matrix_ac_23_transpose_x_0, transpose_y = matrix_ac_23_transpose_y_0, x = transpose_209, y = transpose_208)[name = tensor<string, []>("matrix_ac_23_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_47_begin_0 = const()[name = tensor<string, []>("matrix_bd_47_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_47_end_0 = const()[name = tensor<string, []>("matrix_bd_47_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_47_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_47_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_47_cast_fp16 = slice_by_index(begin = matrix_bd_47_begin_0, end = matrix_bd_47_end_0, end_mask = matrix_bd_47_end_mask_0, x = matrix_bd_45_cast_fp16)[name = tensor<string, []>("matrix_bd_47_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2083_cast_fp16 = add(x = matrix_ac_23_cast_fp16, y = matrix_bd_47_cast_fp16)[name = tensor<string, []>("op_2083_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_45_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_45_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_45_cast_fp16 = mul(x = var_2083_cast_fp16, y = _inversed_scores_45_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_45_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_47_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_45_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_47_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2089_cast_fp16 = softmax(axis = var_25, x = scores_47_cast_fp16)[name = tensor<string, []>("op_2089_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_605_cast_fp16 = select(a = var_6_to_fp16, b = var_2089_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_605_cast_fp16")];
+            tensor<bool, []> x_255_transpose_x_0 = const()[name = tensor<string, []>("x_255_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_255_transpose_y_0 = const()[name = tensor<string, []>("x_255_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_211 = transpose(perm = value_23_perm_0, x = v_23_cast_fp16)[name = tensor<string, []>("transpose_211")];
+            tensor<fp16, [1, 8, 126, 128]> x_255_cast_fp16 = matmul(transpose_x = x_255_transpose_x_0, transpose_y = x_255_transpose_y_0, x = input_605_cast_fp16, y = transpose_211)[name = tensor<string, []>("x_255_cast_fp16")];
+            tensor<int32, [4]> var_2093_perm_0 = const()[name = tensor<string, []>("op_2093_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2094 = const()[name = tensor<string, []>("op_2094"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_207 = transpose(perm = var_2093_perm_0, x = x_255_cast_fp16)[name = tensor<string, []>("transpose_207")];
+            tensor<fp16, [1, 126, 1024]> input_607_cast_fp16 = reshape(shape = var_2094, x = transpose_207)[name = tensor<string, []>("input_607_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_11_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_11_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(569023360)))];
+            tensor<fp16, [1, 126, 1024]> linear_106_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_11_self_attn_linear_out_weight_to_fp16, x = input_607_cast_fp16)[name = tensor<string, []>("linear_106_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_611_cast_fp16 = add(x = input_603_cast_fp16, y = linear_106_cast_fp16)[name = tensor<string, []>("input_611_cast_fp16")];
+            tensor<int32, [1]> x_259_axes_0 = const()[name = tensor<string, []>("x_259_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_11_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_11_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(571120576)))];
+            tensor<fp16, [1024]> model_layers_11_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_11_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(571122688)))];
+            tensor<fp16, [1, 126, 1024]> x_259_cast_fp16 = layer_norm(axes = x_259_axes_0, beta = model_layers_11_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_11_norm_conv_weight_to_fp16, x = input_611_cast_fp16)[name = tensor<string, []>("x_259_cast_fp16")];
+            tensor<int32, [3]> input_613_perm_0 = const()[name = tensor<string, []>("input_613_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_2110 = const()[name = tensor<string, []>("op_2110"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_2112 = const()[name = tensor<string, []>("op_2112"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_615_pad_type_0 = const()[name = tensor<string, []>("input_615_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_615_pad_0 = const()[name = tensor<string, []>("input_615_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_11_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_11_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(571124800)))];
+            tensor<fp16, [1, 1024, 126]> transpose_206 = transpose(perm = input_613_perm_0, x = x_259_cast_fp16)[name = tensor<string, []>("transpose_206")];
+            tensor<fp16, [1, 2048, 126]> input_615_cast_fp16 = conv(dilations = var_2112, groups = var_27, pad = input_615_pad_0, pad_type = input_615_pad_type_0, strides = var_2110, weight = model_layers_11_conv_pointwise_conv1_weight_to_fp16, x = transpose_206)[name = tensor<string, []>("input_615_cast_fp16")];
+            tensor<int32, []> x_261_split_num_splits_0 = const()[name = tensor<string, []>("x_261_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_261_split_axis_0 = const()[name = tensor<string, []>("x_261_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_261_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_261_split_cast_fp16_1 = split(axis = x_261_split_axis_0, num_splits = x_261_split_num_splits_0, x = input_615_cast_fp16)[name = tensor<string, []>("x_261_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_261_split_1_sigmoid_cast_fp16 = sigmoid(x = x_261_split_cast_fp16_1)[name = tensor<string, []>("x_261_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_261_cast_fp16 = mul(x = x_261_split_cast_fp16_0, y = x_261_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_261_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_617_cast_fp16 = select(a = var_6_to_fp16, b = x_261_cast_fp16, cond = var_323)[name = tensor<string, []>("input_617_cast_fp16")];
+            tensor<int32, [6]> input_619_pad_0 = const()[name = tensor<string, []>("input_619_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_619_mode_0 = const()[name = tensor<string, []>("input_619_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_126_to_fp16 = const()[name = tensor<string, []>("const_126_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_619_cast_fp16 = pad(constant_val = const_126_to_fp16, mode = input_619_mode_0, pad = input_619_pad_0, x = input_617_cast_fp16)[name = tensor<string, []>("input_619_cast_fp16")];
+            tensor<int32, [1]> var_2121 = const()[name = tensor<string, []>("op_2121"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_2123 = const()[name = tensor<string, []>("op_2123"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_621_pad_type_0 = const()[name = tensor<string, []>("input_621_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_621_pad_0 = const()[name = tensor<string, []>("input_621_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_623_weight_0_to_fp16 = const()[name = tensor<string, []>("input_623_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(575319168)))];
+            tensor<fp16, [1024]> input_623_bias_0_to_fp16 = const()[name = tensor<string, []>("input_623_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(575337664)))];
+            tensor<fp16, [1, 1024, 126]> input_623_cast_fp16 = conv(bias = input_623_bias_0_to_fp16, dilations = var_2123, groups = var_5, pad = input_621_pad_0, pad_type = input_621_pad_type_0, strides = var_2121, weight = input_623_weight_0_to_fp16, x = input_619_cast_fp16)[name = tensor<string, []>("input_623_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_625_cast_fp16 = silu(x = input_623_cast_fp16)[name = tensor<string, []>("input_625_cast_fp16")];
+            tensor<int32, [1]> var_2133 = const()[name = tensor<string, []>("op_2133"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_2135 = const()[name = tensor<string, []>("op_2135"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_263_pad_type_0 = const()[name = tensor<string, []>("x_263_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_263_pad_0 = const()[name = tensor<string, []>("x_263_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_11_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_11_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(575339776)))];
+            tensor<fp16, [1, 1024, 126]> x_263_cast_fp16 = conv(dilations = var_2135, groups = var_27, pad = x_263_pad_0, pad_type = x_263_pad_type_0, strides = var_2133, weight = model_layers_11_conv_pointwise_conv2_weight_to_fp16, x = input_625_cast_fp16)[name = tensor<string, []>("x_263_cast_fp16")];
+            tensor<int32, [3]> input_627_perm_0 = const()[name = tensor<string, []>("input_627_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_205 = transpose(perm = input_627_perm_0, x = x_263_cast_fp16)[name = tensor<string, []>("transpose_205")];
+            tensor<fp16, [1, 126, 1024]> input_629_cast_fp16 = add(x = input_611_cast_fp16, y = transpose_205)[name = tensor<string, []>("input_629_cast_fp16")];
+            tensor<int32, [1]> input_631_axes_0 = const()[name = tensor<string, []>("input_631_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_11_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_11_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(577436992)))];
+            tensor<fp16, [1024]> model_layers_11_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_11_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(577439104)))];
+            tensor<fp16, [1, 126, 1024]> input_631_cast_fp16 = layer_norm(axes = input_631_axes_0, beta = model_layers_11_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_11_norm_feed_forward2_weight_to_fp16, x = input_629_cast_fp16)[name = tensor<string, []>("input_631_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_11_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_11_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(577441216)))];
+            tensor<fp16, [1, 126, 4096]> linear_107_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_11_feed_forward2_linear1_weight_to_fp16, x = input_631_cast_fp16)[name = tensor<string, []>("linear_107_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_635_cast_fp16 = silu(x = linear_107_cast_fp16)[name = tensor<string, []>("input_635_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_11_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_11_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(585829888)))];
+            tensor<fp16, [1, 126, 1024]> linear_108_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_11_feed_forward2_linear2_weight_to_fp16, x = input_635_cast_fp16)[name = tensor<string, []>("linear_108_cast_fp16")];
+            tensor<fp16, []> var_2154_to_fp16 = const()[name = tensor<string, []>("op_2154_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2155_cast_fp16 = mul(x = linear_108_cast_fp16, y = var_2154_to_fp16)[name = tensor<string, []>("op_2155_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_641_cast_fp16 = add(x = input_629_cast_fp16, y = var_2155_cast_fp16)[name = tensor<string, []>("input_641_cast_fp16")];
+            tensor<int32, [1]> input_643_axes_0 = const()[name = tensor<string, []>("input_643_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_11_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_11_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(594218560)))];
+            tensor<fp16, [1024]> model_layers_11_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_11_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(594220672)))];
+            tensor<fp16, [1, 126, 1024]> input_643_cast_fp16 = layer_norm(axes = input_643_axes_0, beta = model_layers_11_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_11_norm_out_weight_to_fp16, x = input_641_cast_fp16)[name = tensor<string, []>("input_643_cast_fp16")];
+            tensor<int32, [1]> input_645_axes_0 = const()[name = tensor<string, []>("input_645_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_12_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_12_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(594222784)))];
+            tensor<fp16, [1024]> model_layers_12_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_12_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(594224896)))];
+            tensor<fp16, [1, 126, 1024]> input_645_cast_fp16 = layer_norm(axes = input_645_axes_0, beta = model_layers_12_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_12_norm_feed_forward1_weight_to_fp16, x = input_643_cast_fp16)[name = tensor<string, []>("input_645_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_12_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_12_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(594227008)))];
+            tensor<fp16, [1, 126, 4096]> linear_109_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_12_feed_forward1_linear1_weight_to_fp16, x = input_645_cast_fp16)[name = tensor<string, []>("linear_109_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_649_cast_fp16 = silu(x = linear_109_cast_fp16)[name = tensor<string, []>("input_649_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_12_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_12_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(602615680)))];
+            tensor<fp16, [1, 126, 1024]> linear_110_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_12_feed_forward1_linear2_weight_to_fp16, x = input_649_cast_fp16)[name = tensor<string, []>("linear_110_cast_fp16")];
+            tensor<fp16, []> var_2183_to_fp16 = const()[name = tensor<string, []>("op_2183_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2184_cast_fp16 = mul(x = linear_110_cast_fp16, y = var_2183_to_fp16)[name = tensor<string, []>("op_2184_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_655_cast_fp16 = add(x = input_643_cast_fp16, y = var_2184_cast_fp16)[name = tensor<string, []>("input_655_cast_fp16")];
+            tensor<int32, [1]> query_25_axes_0 = const()[name = tensor<string, []>("query_25_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_12_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_12_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(611004352)))];
+            tensor<fp16, [1024]> model_layers_12_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_12_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(611006464)))];
+            tensor<fp16, [1, 126, 1024]> query_25_cast_fp16 = layer_norm(axes = query_25_axes_0, beta = model_layers_12_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_12_norm_self_att_weight_to_fp16, x = input_655_cast_fp16)[name = tensor<string, []>("query_25_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_12_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_12_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(611008576)))];
+            tensor<fp16, [1, 126, 1024]> linear_111_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_12_self_attn_linear_q_weight_to_fp16, x = query_25_cast_fp16)[name = tensor<string, []>("linear_111_cast_fp16")];
+            tensor<int32, [4]> var_2200 = const()[name = tensor<string, []>("op_2200"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_73_cast_fp16 = reshape(shape = var_2200, x = linear_111_cast_fp16)[name = tensor<string, []>("q_73_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_12_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_12_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(613105792)))];
+            tensor<fp16, [1, 126, 1024]> linear_112_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_12_self_attn_linear_k_weight_to_fp16, x = query_25_cast_fp16)[name = tensor<string, []>("linear_112_cast_fp16")];
+            tensor<int32, [4]> var_2204 = const()[name = tensor<string, []>("op_2204"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_49_cast_fp16 = reshape(shape = var_2204, x = linear_112_cast_fp16)[name = tensor<string, []>("k_49_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_12_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_12_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(615203008)))];
+            tensor<fp16, [1, 126, 1024]> linear_113_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_12_self_attn_linear_v_weight_to_fp16, x = query_25_cast_fp16)[name = tensor<string, []>("linear_113_cast_fp16")];
+            tensor<int32, [4]> var_2208 = const()[name = tensor<string, []>("op_2208"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_25_cast_fp16 = reshape(shape = var_2208, x = linear_113_cast_fp16)[name = tensor<string, []>("v_25_cast_fp16")];
+            tensor<int32, [4]> value_25_perm_0 = const()[name = tensor<string, []>("value_25_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_12_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_12_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(617300224)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2220_cast_fp16 = add(x = q_73_cast_fp16, y = model_layers_12_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_2220_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_12_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_12_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(617302336)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2222_cast_fp16 = add(x = q_73_cast_fp16, y = model_layers_12_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_2222_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_25_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_25_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_271_transpose_x_0 = const()[name = tensor<string, []>("x_271_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_271_transpose_y_0 = const()[name = tensor<string, []>("x_271_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_2224_to_fp16 = const()[name = tensor<string, []>("op_2224_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(617304448)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_203 = transpose(perm = q_with_bias_v_25_perm_0, x = var_2222_cast_fp16)[name = tensor<string, []>("transpose_203")];
+            tensor<fp16, [1, 8, 126, 251]> x_271_cast_fp16 = matmul(transpose_x = x_271_transpose_x_0, transpose_y = x_271_transpose_y_0, x = transpose_203, y = var_2224_to_fp16)[name = tensor<string, []>("x_271_cast_fp16")];
+            tensor<int32, [8]> x_273_pad_0 = const()[name = tensor<string, []>("x_273_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_273_mode_0 = const()[name = tensor<string, []>("x_273_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_133_to_fp16 = const()[name = tensor<string, []>("const_133_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_273_cast_fp16 = pad(constant_val = const_133_to_fp16, mode = x_273_mode_0, pad = x_273_pad_0, x = x_271_cast_fp16)[name = tensor<string, []>("x_273_cast_fp16")];
+            tensor<int32, [4]> var_2232 = const()[name = tensor<string, []>("op_2232"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_275_cast_fp16 = reshape(shape = var_2232, x = x_273_cast_fp16)[name = tensor<string, []>("x_275_cast_fp16")];
+            tensor<int32, [4]> var_2236_begin_0 = const()[name = tensor<string, []>("op_2236_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2236_end_0 = const()[name = tensor<string, []>("op_2236_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_2236_end_mask_0 = const()[name = tensor<string, []>("op_2236_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_2236_cast_fp16 = slice_by_index(begin = var_2236_begin_0, end = var_2236_end_0, end_mask = var_2236_end_mask_0, x = x_275_cast_fp16)[name = tensor<string, []>("op_2236_cast_fp16")];
+            tensor<int32, [4]> var_2237 = const()[name = tensor<string, []>("op_2237"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_49_cast_fp16 = reshape(shape = var_2237, x = var_2236_cast_fp16)[name = tensor<string, []>("matrix_bd_49_cast_fp16")];
+            tensor<bool, []> matrix_ac_25_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_25_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_25_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_25_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_96_perm_0 = const()[name = tensor<string, []>("transpose_96_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_97_perm_0 = const()[name = tensor<string, []>("transpose_97_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_201 = transpose(perm = transpose_97_perm_0, x = k_49_cast_fp16)[name = tensor<string, []>("transpose_201")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_202 = transpose(perm = transpose_96_perm_0, x = var_2220_cast_fp16)[name = tensor<string, []>("transpose_202")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_25_cast_fp16 = matmul(transpose_x = matrix_ac_25_transpose_x_0, transpose_y = matrix_ac_25_transpose_y_0, x = transpose_202, y = transpose_201)[name = tensor<string, []>("matrix_ac_25_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_51_begin_0 = const()[name = tensor<string, []>("matrix_bd_51_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_51_end_0 = const()[name = tensor<string, []>("matrix_bd_51_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_51_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_51_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_51_cast_fp16 = slice_by_index(begin = matrix_bd_51_begin_0, end = matrix_bd_51_end_0, end_mask = matrix_bd_51_end_mask_0, x = matrix_bd_49_cast_fp16)[name = tensor<string, []>("matrix_bd_51_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2246_cast_fp16 = add(x = matrix_ac_25_cast_fp16, y = matrix_bd_51_cast_fp16)[name = tensor<string, []>("op_2246_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_49_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_49_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_49_cast_fp16 = mul(x = var_2246_cast_fp16, y = _inversed_scores_49_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_49_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_51_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_49_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_51_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2252_cast_fp16 = softmax(axis = var_25, x = scores_51_cast_fp16)[name = tensor<string, []>("op_2252_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_657_cast_fp16 = select(a = var_6_to_fp16, b = var_2252_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_657_cast_fp16")];
+            tensor<bool, []> x_277_transpose_x_0 = const()[name = tensor<string, []>("x_277_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_277_transpose_y_0 = const()[name = tensor<string, []>("x_277_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_204 = transpose(perm = value_25_perm_0, x = v_25_cast_fp16)[name = tensor<string, []>("transpose_204")];
+            tensor<fp16, [1, 8, 126, 128]> x_277_cast_fp16 = matmul(transpose_x = x_277_transpose_x_0, transpose_y = x_277_transpose_y_0, x = input_657_cast_fp16, y = transpose_204)[name = tensor<string, []>("x_277_cast_fp16")];
+            tensor<int32, [4]> var_2256_perm_0 = const()[name = tensor<string, []>("op_2256_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2257 = const()[name = tensor<string, []>("op_2257"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_200 = transpose(perm = var_2256_perm_0, x = x_277_cast_fp16)[name = tensor<string, []>("transpose_200")];
+            tensor<fp16, [1, 126, 1024]> input_659_cast_fp16 = reshape(shape = var_2257, x = transpose_200)[name = tensor<string, []>("input_659_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_12_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_12_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(617818560)))];
+            tensor<fp16, [1, 126, 1024]> linear_115_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_12_self_attn_linear_out_weight_to_fp16, x = input_659_cast_fp16)[name = tensor<string, []>("linear_115_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_663_cast_fp16 = add(x = input_655_cast_fp16, y = linear_115_cast_fp16)[name = tensor<string, []>("input_663_cast_fp16")];
+            tensor<int32, [1]> x_281_axes_0 = const()[name = tensor<string, []>("x_281_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_12_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_12_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(619915776)))];
+            tensor<fp16, [1024]> model_layers_12_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_12_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(619917888)))];
+            tensor<fp16, [1, 126, 1024]> x_281_cast_fp16 = layer_norm(axes = x_281_axes_0, beta = model_layers_12_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_12_norm_conv_weight_to_fp16, x = input_663_cast_fp16)[name = tensor<string, []>("x_281_cast_fp16")];
+            tensor<int32, [3]> input_665_perm_0 = const()[name = tensor<string, []>("input_665_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_2273 = const()[name = tensor<string, []>("op_2273"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_2275 = const()[name = tensor<string, []>("op_2275"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_667_pad_type_0 = const()[name = tensor<string, []>("input_667_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_667_pad_0 = const()[name = tensor<string, []>("input_667_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_12_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_12_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(619920000)))];
+            tensor<fp16, [1, 1024, 126]> transpose_199 = transpose(perm = input_665_perm_0, x = x_281_cast_fp16)[name = tensor<string, []>("transpose_199")];
+            tensor<fp16, [1, 2048, 126]> input_667_cast_fp16 = conv(dilations = var_2275, groups = var_27, pad = input_667_pad_0, pad_type = input_667_pad_type_0, strides = var_2273, weight = model_layers_12_conv_pointwise_conv1_weight_to_fp16, x = transpose_199)[name = tensor<string, []>("input_667_cast_fp16")];
+            tensor<int32, []> x_283_split_num_splits_0 = const()[name = tensor<string, []>("x_283_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_283_split_axis_0 = const()[name = tensor<string, []>("x_283_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_283_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_283_split_cast_fp16_1 = split(axis = x_283_split_axis_0, num_splits = x_283_split_num_splits_0, x = input_667_cast_fp16)[name = tensor<string, []>("x_283_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_283_split_1_sigmoid_cast_fp16 = sigmoid(x = x_283_split_cast_fp16_1)[name = tensor<string, []>("x_283_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_283_cast_fp16 = mul(x = x_283_split_cast_fp16_0, y = x_283_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_283_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_669_cast_fp16 = select(a = var_6_to_fp16, b = x_283_cast_fp16, cond = var_323)[name = tensor<string, []>("input_669_cast_fp16")];
+            tensor<int32, [6]> input_671_pad_0 = const()[name = tensor<string, []>("input_671_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_671_mode_0 = const()[name = tensor<string, []>("input_671_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_136_to_fp16 = const()[name = tensor<string, []>("const_136_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_671_cast_fp16 = pad(constant_val = const_136_to_fp16, mode = input_671_mode_0, pad = input_671_pad_0, x = input_669_cast_fp16)[name = tensor<string, []>("input_671_cast_fp16")];
+            tensor<int32, [1]> var_2284 = const()[name = tensor<string, []>("op_2284"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_2286 = const()[name = tensor<string, []>("op_2286"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_673_pad_type_0 = const()[name = tensor<string, []>("input_673_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_673_pad_0 = const()[name = tensor<string, []>("input_673_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_675_weight_0_to_fp16 = const()[name = tensor<string, []>("input_675_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(624114368)))];
+            tensor<fp16, [1024]> input_675_bias_0_to_fp16 = const()[name = tensor<string, []>("input_675_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(624132864)))];
+            tensor<fp16, [1, 1024, 126]> input_675_cast_fp16 = conv(bias = input_675_bias_0_to_fp16, dilations = var_2286, groups = var_5, pad = input_673_pad_0, pad_type = input_673_pad_type_0, strides = var_2284, weight = input_675_weight_0_to_fp16, x = input_671_cast_fp16)[name = tensor<string, []>("input_675_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_677_cast_fp16 = silu(x = input_675_cast_fp16)[name = tensor<string, []>("input_677_cast_fp16")];
+            tensor<int32, [1]> var_2296 = const()[name = tensor<string, []>("op_2296"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_2298 = const()[name = tensor<string, []>("op_2298"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_285_pad_type_0 = const()[name = tensor<string, []>("x_285_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_285_pad_0 = const()[name = tensor<string, []>("x_285_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_12_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_12_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(624134976)))];
+            tensor<fp16, [1, 1024, 126]> x_285_cast_fp16 = conv(dilations = var_2298, groups = var_27, pad = x_285_pad_0, pad_type = x_285_pad_type_0, strides = var_2296, weight = model_layers_12_conv_pointwise_conv2_weight_to_fp16, x = input_677_cast_fp16)[name = tensor<string, []>("x_285_cast_fp16")];
+            tensor<int32, [3]> input_679_perm_0 = const()[name = tensor<string, []>("input_679_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_198 = transpose(perm = input_679_perm_0, x = x_285_cast_fp16)[name = tensor<string, []>("transpose_198")];
+            tensor<fp16, [1, 126, 1024]> input_681_cast_fp16 = add(x = input_663_cast_fp16, y = transpose_198)[name = tensor<string, []>("input_681_cast_fp16")];
+            tensor<int32, [1]> input_683_axes_0 = const()[name = tensor<string, []>("input_683_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_12_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_12_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(626232192)))];
+            tensor<fp16, [1024]> model_layers_12_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_12_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(626234304)))];
+            tensor<fp16, [1, 126, 1024]> input_683_cast_fp16 = layer_norm(axes = input_683_axes_0, beta = model_layers_12_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_12_norm_feed_forward2_weight_to_fp16, x = input_681_cast_fp16)[name = tensor<string, []>("input_683_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_12_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_12_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(626236416)))];
+            tensor<fp16, [1, 126, 4096]> linear_116_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_12_feed_forward2_linear1_weight_to_fp16, x = input_683_cast_fp16)[name = tensor<string, []>("linear_116_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_687_cast_fp16 = silu(x = linear_116_cast_fp16)[name = tensor<string, []>("input_687_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_12_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_12_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(634625088)))];
+            tensor<fp16, [1, 126, 1024]> linear_117_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_12_feed_forward2_linear2_weight_to_fp16, x = input_687_cast_fp16)[name = tensor<string, []>("linear_117_cast_fp16")];
+            tensor<fp16, []> var_2317_to_fp16 = const()[name = tensor<string, []>("op_2317_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2318_cast_fp16 = mul(x = linear_117_cast_fp16, y = var_2317_to_fp16)[name = tensor<string, []>("op_2318_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_693_cast_fp16 = add(x = input_681_cast_fp16, y = var_2318_cast_fp16)[name = tensor<string, []>("input_693_cast_fp16")];
+            tensor<int32, [1]> input_695_axes_0 = const()[name = tensor<string, []>("input_695_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_12_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_12_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643013760)))];
+            tensor<fp16, [1024]> model_layers_12_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_12_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643015872)))];
+            tensor<fp16, [1, 126, 1024]> input_695_cast_fp16 = layer_norm(axes = input_695_axes_0, beta = model_layers_12_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_12_norm_out_weight_to_fp16, x = input_693_cast_fp16)[name = tensor<string, []>("input_695_cast_fp16")];
+            tensor<int32, [1]> input_697_axes_0 = const()[name = tensor<string, []>("input_697_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_13_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_13_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643017984)))];
+            tensor<fp16, [1024]> model_layers_13_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_13_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643020096)))];
+            tensor<fp16, [1, 126, 1024]> input_697_cast_fp16 = layer_norm(axes = input_697_axes_0, beta = model_layers_13_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_13_norm_feed_forward1_weight_to_fp16, x = input_695_cast_fp16)[name = tensor<string, []>("input_697_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_13_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_13_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643022208)))];
+            tensor<fp16, [1, 126, 4096]> linear_118_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_13_feed_forward1_linear1_weight_to_fp16, x = input_697_cast_fp16)[name = tensor<string, []>("linear_118_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_701_cast_fp16 = silu(x = linear_118_cast_fp16)[name = tensor<string, []>("input_701_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_13_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_13_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(651410880)))];
+            tensor<fp16, [1, 126, 1024]> linear_119_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_13_feed_forward1_linear2_weight_to_fp16, x = input_701_cast_fp16)[name = tensor<string, []>("linear_119_cast_fp16")];
+            tensor<fp16, []> var_2346_to_fp16 = const()[name = tensor<string, []>("op_2346_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2347_cast_fp16 = mul(x = linear_119_cast_fp16, y = var_2346_to_fp16)[name = tensor<string, []>("op_2347_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_707_cast_fp16 = add(x = input_695_cast_fp16, y = var_2347_cast_fp16)[name = tensor<string, []>("input_707_cast_fp16")];
+            tensor<int32, [1]> query_27_axes_0 = const()[name = tensor<string, []>("query_27_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_13_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_13_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(659799552)))];
+            tensor<fp16, [1024]> model_layers_13_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_13_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(659801664)))];
+            tensor<fp16, [1, 126, 1024]> query_27_cast_fp16 = layer_norm(axes = query_27_axes_0, beta = model_layers_13_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_13_norm_self_att_weight_to_fp16, x = input_707_cast_fp16)[name = tensor<string, []>("query_27_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_13_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_13_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(659803776)))];
+            tensor<fp16, [1, 126, 1024]> linear_120_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_13_self_attn_linear_q_weight_to_fp16, x = query_27_cast_fp16)[name = tensor<string, []>("linear_120_cast_fp16")];
+            tensor<int32, [4]> var_2363 = const()[name = tensor<string, []>("op_2363"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_79_cast_fp16 = reshape(shape = var_2363, x = linear_120_cast_fp16)[name = tensor<string, []>("q_79_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_13_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_13_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(661900992)))];
+            tensor<fp16, [1, 126, 1024]> linear_121_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_13_self_attn_linear_k_weight_to_fp16, x = query_27_cast_fp16)[name = tensor<string, []>("linear_121_cast_fp16")];
+            tensor<int32, [4]> var_2367 = const()[name = tensor<string, []>("op_2367"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_53_cast_fp16 = reshape(shape = var_2367, x = linear_121_cast_fp16)[name = tensor<string, []>("k_53_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_13_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_13_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(663998208)))];
+            tensor<fp16, [1, 126, 1024]> linear_122_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_13_self_attn_linear_v_weight_to_fp16, x = query_27_cast_fp16)[name = tensor<string, []>("linear_122_cast_fp16")];
+            tensor<int32, [4]> var_2371 = const()[name = tensor<string, []>("op_2371"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_27_cast_fp16 = reshape(shape = var_2371, x = linear_122_cast_fp16)[name = tensor<string, []>("v_27_cast_fp16")];
+            tensor<int32, [4]> value_27_perm_0 = const()[name = tensor<string, []>("value_27_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_13_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_13_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(666095424)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2383_cast_fp16 = add(x = q_79_cast_fp16, y = model_layers_13_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_2383_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_13_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_13_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(666097536)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2385_cast_fp16 = add(x = q_79_cast_fp16, y = model_layers_13_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_2385_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_27_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_27_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_293_transpose_x_0 = const()[name = tensor<string, []>("x_293_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_293_transpose_y_0 = const()[name = tensor<string, []>("x_293_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_2387_to_fp16 = const()[name = tensor<string, []>("op_2387_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(666099648)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_196 = transpose(perm = q_with_bias_v_27_perm_0, x = var_2385_cast_fp16)[name = tensor<string, []>("transpose_196")];
+            tensor<fp16, [1, 8, 126, 251]> x_293_cast_fp16 = matmul(transpose_x = x_293_transpose_x_0, transpose_y = x_293_transpose_y_0, x = transpose_196, y = var_2387_to_fp16)[name = tensor<string, []>("x_293_cast_fp16")];
+            tensor<int32, [8]> x_295_pad_0 = const()[name = tensor<string, []>("x_295_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_295_mode_0 = const()[name = tensor<string, []>("x_295_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_143_to_fp16 = const()[name = tensor<string, []>("const_143_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_295_cast_fp16 = pad(constant_val = const_143_to_fp16, mode = x_295_mode_0, pad = x_295_pad_0, x = x_293_cast_fp16)[name = tensor<string, []>("x_295_cast_fp16")];
+            tensor<int32, [4]> var_2395 = const()[name = tensor<string, []>("op_2395"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_297_cast_fp16 = reshape(shape = var_2395, x = x_295_cast_fp16)[name = tensor<string, []>("x_297_cast_fp16")];
+            tensor<int32, [4]> var_2399_begin_0 = const()[name = tensor<string, []>("op_2399_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2399_end_0 = const()[name = tensor<string, []>("op_2399_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_2399_end_mask_0 = const()[name = tensor<string, []>("op_2399_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_2399_cast_fp16 = slice_by_index(begin = var_2399_begin_0, end = var_2399_end_0, end_mask = var_2399_end_mask_0, x = x_297_cast_fp16)[name = tensor<string, []>("op_2399_cast_fp16")];
+            tensor<int32, [4]> var_2400 = const()[name = tensor<string, []>("op_2400"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_53_cast_fp16 = reshape(shape = var_2400, x = var_2399_cast_fp16)[name = tensor<string, []>("matrix_bd_53_cast_fp16")];
+            tensor<bool, []> matrix_ac_27_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_27_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_27_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_27_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_98_perm_0 = const()[name = tensor<string, []>("transpose_98_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_99_perm_0 = const()[name = tensor<string, []>("transpose_99_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_194 = transpose(perm = transpose_99_perm_0, x = k_53_cast_fp16)[name = tensor<string, []>("transpose_194")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_195 = transpose(perm = transpose_98_perm_0, x = var_2383_cast_fp16)[name = tensor<string, []>("transpose_195")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_27_cast_fp16 = matmul(transpose_x = matrix_ac_27_transpose_x_0, transpose_y = matrix_ac_27_transpose_y_0, x = transpose_195, y = transpose_194)[name = tensor<string, []>("matrix_ac_27_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_55_begin_0 = const()[name = tensor<string, []>("matrix_bd_55_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_55_end_0 = const()[name = tensor<string, []>("matrix_bd_55_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_55_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_55_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_55_cast_fp16 = slice_by_index(begin = matrix_bd_55_begin_0, end = matrix_bd_55_end_0, end_mask = matrix_bd_55_end_mask_0, x = matrix_bd_53_cast_fp16)[name = tensor<string, []>("matrix_bd_55_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2409_cast_fp16 = add(x = matrix_ac_27_cast_fp16, y = matrix_bd_55_cast_fp16)[name = tensor<string, []>("op_2409_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_53_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_53_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_53_cast_fp16 = mul(x = var_2409_cast_fp16, y = _inversed_scores_53_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_53_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_55_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_53_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_55_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2415_cast_fp16 = softmax(axis = var_25, x = scores_55_cast_fp16)[name = tensor<string, []>("op_2415_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_709_cast_fp16 = select(a = var_6_to_fp16, b = var_2415_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_709_cast_fp16")];
+            tensor<bool, []> x_299_transpose_x_0 = const()[name = tensor<string, []>("x_299_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_299_transpose_y_0 = const()[name = tensor<string, []>("x_299_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_197 = transpose(perm = value_27_perm_0, x = v_27_cast_fp16)[name = tensor<string, []>("transpose_197")];
+            tensor<fp16, [1, 8, 126, 128]> x_299_cast_fp16 = matmul(transpose_x = x_299_transpose_x_0, transpose_y = x_299_transpose_y_0, x = input_709_cast_fp16, y = transpose_197)[name = tensor<string, []>("x_299_cast_fp16")];
+            tensor<int32, [4]> var_2419_perm_0 = const()[name = tensor<string, []>("op_2419_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2420 = const()[name = tensor<string, []>("op_2420"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_193 = transpose(perm = var_2419_perm_0, x = x_299_cast_fp16)[name = tensor<string, []>("transpose_193")];
+            tensor<fp16, [1, 126, 1024]> input_711_cast_fp16 = reshape(shape = var_2420, x = transpose_193)[name = tensor<string, []>("input_711_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_13_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_13_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(666613760)))];
+            tensor<fp16, [1, 126, 1024]> linear_124_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_13_self_attn_linear_out_weight_to_fp16, x = input_711_cast_fp16)[name = tensor<string, []>("linear_124_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_715_cast_fp16 = add(x = input_707_cast_fp16, y = linear_124_cast_fp16)[name = tensor<string, []>("input_715_cast_fp16")];
+            tensor<int32, [1]> x_303_axes_0 = const()[name = tensor<string, []>("x_303_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_13_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_13_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(668710976)))];
+            tensor<fp16, [1024]> model_layers_13_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_13_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(668713088)))];
+            tensor<fp16, [1, 126, 1024]> x_303_cast_fp16 = layer_norm(axes = x_303_axes_0, beta = model_layers_13_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_13_norm_conv_weight_to_fp16, x = input_715_cast_fp16)[name = tensor<string, []>("x_303_cast_fp16")];
+            tensor<int32, [3]> input_717_perm_0 = const()[name = tensor<string, []>("input_717_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_2436 = const()[name = tensor<string, []>("op_2436"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_2438 = const()[name = tensor<string, []>("op_2438"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_719_pad_type_0 = const()[name = tensor<string, []>("input_719_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_719_pad_0 = const()[name = tensor<string, []>("input_719_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_13_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_13_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(668715200)))];
+            tensor<fp16, [1, 1024, 126]> transpose_192 = transpose(perm = input_717_perm_0, x = x_303_cast_fp16)[name = tensor<string, []>("transpose_192")];
+            tensor<fp16, [1, 2048, 126]> input_719_cast_fp16 = conv(dilations = var_2438, groups = var_27, pad = input_719_pad_0, pad_type = input_719_pad_type_0, strides = var_2436, weight = model_layers_13_conv_pointwise_conv1_weight_to_fp16, x = transpose_192)[name = tensor<string, []>("input_719_cast_fp16")];
+            tensor<int32, []> x_305_split_num_splits_0 = const()[name = tensor<string, []>("x_305_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_305_split_axis_0 = const()[name = tensor<string, []>("x_305_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_305_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_305_split_cast_fp16_1 = split(axis = x_305_split_axis_0, num_splits = x_305_split_num_splits_0, x = input_719_cast_fp16)[name = tensor<string, []>("x_305_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_305_split_1_sigmoid_cast_fp16 = sigmoid(x = x_305_split_cast_fp16_1)[name = tensor<string, []>("x_305_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_305_cast_fp16 = mul(x = x_305_split_cast_fp16_0, y = x_305_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_305_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_721_cast_fp16 = select(a = var_6_to_fp16, b = x_305_cast_fp16, cond = var_323)[name = tensor<string, []>("input_721_cast_fp16")];
+            tensor<int32, [6]> input_723_pad_0 = const()[name = tensor<string, []>("input_723_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_723_mode_0 = const()[name = tensor<string, []>("input_723_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_146_to_fp16 = const()[name = tensor<string, []>("const_146_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_723_cast_fp16 = pad(constant_val = const_146_to_fp16, mode = input_723_mode_0, pad = input_723_pad_0, x = input_721_cast_fp16)[name = tensor<string, []>("input_723_cast_fp16")];
+            tensor<int32, [1]> var_2447 = const()[name = tensor<string, []>("op_2447"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_2449 = const()[name = tensor<string, []>("op_2449"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_725_pad_type_0 = const()[name = tensor<string, []>("input_725_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_725_pad_0 = const()[name = tensor<string, []>("input_725_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_727_weight_0_to_fp16 = const()[name = tensor<string, []>("input_727_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(672909568)))];
+            tensor<fp16, [1024]> input_727_bias_0_to_fp16 = const()[name = tensor<string, []>("input_727_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(672928064)))];
+            tensor<fp16, [1, 1024, 126]> input_727_cast_fp16 = conv(bias = input_727_bias_0_to_fp16, dilations = var_2449, groups = var_5, pad = input_725_pad_0, pad_type = input_725_pad_type_0, strides = var_2447, weight = input_727_weight_0_to_fp16, x = input_723_cast_fp16)[name = tensor<string, []>("input_727_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_729_cast_fp16 = silu(x = input_727_cast_fp16)[name = tensor<string, []>("input_729_cast_fp16")];
+            tensor<int32, [1]> var_2459 = const()[name = tensor<string, []>("op_2459"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_2461 = const()[name = tensor<string, []>("op_2461"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_307_pad_type_0 = const()[name = tensor<string, []>("x_307_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_307_pad_0 = const()[name = tensor<string, []>("x_307_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_13_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_13_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(672930176)))];
+            tensor<fp16, [1, 1024, 126]> x_307_cast_fp16 = conv(dilations = var_2461, groups = var_27, pad = x_307_pad_0, pad_type = x_307_pad_type_0, strides = var_2459, weight = model_layers_13_conv_pointwise_conv2_weight_to_fp16, x = input_729_cast_fp16)[name = tensor<string, []>("x_307_cast_fp16")];
+            tensor<int32, [3]> input_731_perm_0 = const()[name = tensor<string, []>("input_731_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_191 = transpose(perm = input_731_perm_0, x = x_307_cast_fp16)[name = tensor<string, []>("transpose_191")];
+            tensor<fp16, [1, 126, 1024]> input_733_cast_fp16 = add(x = input_715_cast_fp16, y = transpose_191)[name = tensor<string, []>("input_733_cast_fp16")];
+            tensor<int32, [1]> input_735_axes_0 = const()[name = tensor<string, []>("input_735_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_13_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_13_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(675027392)))];
+            tensor<fp16, [1024]> model_layers_13_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_13_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(675029504)))];
+            tensor<fp16, [1, 126, 1024]> input_735_cast_fp16 = layer_norm(axes = input_735_axes_0, beta = model_layers_13_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_13_norm_feed_forward2_weight_to_fp16, x = input_733_cast_fp16)[name = tensor<string, []>("input_735_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_13_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_13_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(675031616)))];
+            tensor<fp16, [1, 126, 4096]> linear_125_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_13_feed_forward2_linear1_weight_to_fp16, x = input_735_cast_fp16)[name = tensor<string, []>("linear_125_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_739_cast_fp16 = silu(x = linear_125_cast_fp16)[name = tensor<string, []>("input_739_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_13_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_13_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(683420288)))];
+            tensor<fp16, [1, 126, 1024]> linear_126_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_13_feed_forward2_linear2_weight_to_fp16, x = input_739_cast_fp16)[name = tensor<string, []>("linear_126_cast_fp16")];
+            tensor<fp16, []> var_2480_to_fp16 = const()[name = tensor<string, []>("op_2480_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2481_cast_fp16 = mul(x = linear_126_cast_fp16, y = var_2480_to_fp16)[name = tensor<string, []>("op_2481_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_745_cast_fp16 = add(x = input_733_cast_fp16, y = var_2481_cast_fp16)[name = tensor<string, []>("input_745_cast_fp16")];
+            tensor<int32, [1]> input_747_axes_0 = const()[name = tensor<string, []>("input_747_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_13_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_13_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(691808960)))];
+            tensor<fp16, [1024]> model_layers_13_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_13_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(691811072)))];
+            tensor<fp16, [1, 126, 1024]> input_747_cast_fp16 = layer_norm(axes = input_747_axes_0, beta = model_layers_13_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_13_norm_out_weight_to_fp16, x = input_745_cast_fp16)[name = tensor<string, []>("input_747_cast_fp16")];
+            tensor<int32, [1]> input_749_axes_0 = const()[name = tensor<string, []>("input_749_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_14_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_14_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(691813184)))];
+            tensor<fp16, [1024]> model_layers_14_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_14_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(691815296)))];
+            tensor<fp16, [1, 126, 1024]> input_749_cast_fp16 = layer_norm(axes = input_749_axes_0, beta = model_layers_14_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_14_norm_feed_forward1_weight_to_fp16, x = input_747_cast_fp16)[name = tensor<string, []>("input_749_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_14_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_14_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(691817408)))];
+            tensor<fp16, [1, 126, 4096]> linear_127_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_14_feed_forward1_linear1_weight_to_fp16, x = input_749_cast_fp16)[name = tensor<string, []>("linear_127_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_753_cast_fp16 = silu(x = linear_127_cast_fp16)[name = tensor<string, []>("input_753_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_14_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_14_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(700206080)))];
+            tensor<fp16, [1, 126, 1024]> linear_128_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_14_feed_forward1_linear2_weight_to_fp16, x = input_753_cast_fp16)[name = tensor<string, []>("linear_128_cast_fp16")];
+            tensor<fp16, []> var_2509_to_fp16 = const()[name = tensor<string, []>("op_2509_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2510_cast_fp16 = mul(x = linear_128_cast_fp16, y = var_2509_to_fp16)[name = tensor<string, []>("op_2510_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_759_cast_fp16 = add(x = input_747_cast_fp16, y = var_2510_cast_fp16)[name = tensor<string, []>("input_759_cast_fp16")];
+            tensor<int32, [1]> query_29_axes_0 = const()[name = tensor<string, []>("query_29_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_14_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_14_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(708594752)))];
+            tensor<fp16, [1024]> model_layers_14_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_14_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(708596864)))];
+            tensor<fp16, [1, 126, 1024]> query_29_cast_fp16 = layer_norm(axes = query_29_axes_0, beta = model_layers_14_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_14_norm_self_att_weight_to_fp16, x = input_759_cast_fp16)[name = tensor<string, []>("query_29_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_14_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_14_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(708598976)))];
+            tensor<fp16, [1, 126, 1024]> linear_129_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_14_self_attn_linear_q_weight_to_fp16, x = query_29_cast_fp16)[name = tensor<string, []>("linear_129_cast_fp16")];
+            tensor<int32, [4]> var_2526 = const()[name = tensor<string, []>("op_2526"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_85_cast_fp16 = reshape(shape = var_2526, x = linear_129_cast_fp16)[name = tensor<string, []>("q_85_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_14_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_14_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(710696192)))];
+            tensor<fp16, [1, 126, 1024]> linear_130_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_14_self_attn_linear_k_weight_to_fp16, x = query_29_cast_fp16)[name = tensor<string, []>("linear_130_cast_fp16")];
+            tensor<int32, [4]> var_2530 = const()[name = tensor<string, []>("op_2530"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_57_cast_fp16 = reshape(shape = var_2530, x = linear_130_cast_fp16)[name = tensor<string, []>("k_57_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_14_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_14_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(712793408)))];
+            tensor<fp16, [1, 126, 1024]> linear_131_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_14_self_attn_linear_v_weight_to_fp16, x = query_29_cast_fp16)[name = tensor<string, []>("linear_131_cast_fp16")];
+            tensor<int32, [4]> var_2534 = const()[name = tensor<string, []>("op_2534"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_29_cast_fp16 = reshape(shape = var_2534, x = linear_131_cast_fp16)[name = tensor<string, []>("v_29_cast_fp16")];
+            tensor<int32, [4]> value_29_perm_0 = const()[name = tensor<string, []>("value_29_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_14_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_14_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(714890624)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2546_cast_fp16 = add(x = q_85_cast_fp16, y = model_layers_14_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_2546_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_14_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_14_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(714892736)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2548_cast_fp16 = add(x = q_85_cast_fp16, y = model_layers_14_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_2548_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_29_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_29_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_315_transpose_x_0 = const()[name = tensor<string, []>("x_315_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_315_transpose_y_0 = const()[name = tensor<string, []>("x_315_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_2550_to_fp16 = const()[name = tensor<string, []>("op_2550_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(714894848)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_189 = transpose(perm = q_with_bias_v_29_perm_0, x = var_2548_cast_fp16)[name = tensor<string, []>("transpose_189")];
+            tensor<fp16, [1, 8, 126, 251]> x_315_cast_fp16 = matmul(transpose_x = x_315_transpose_x_0, transpose_y = x_315_transpose_y_0, x = transpose_189, y = var_2550_to_fp16)[name = tensor<string, []>("x_315_cast_fp16")];
+            tensor<int32, [8]> x_317_pad_0 = const()[name = tensor<string, []>("x_317_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_317_mode_0 = const()[name = tensor<string, []>("x_317_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_153_to_fp16 = const()[name = tensor<string, []>("const_153_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_317_cast_fp16 = pad(constant_val = const_153_to_fp16, mode = x_317_mode_0, pad = x_317_pad_0, x = x_315_cast_fp16)[name = tensor<string, []>("x_317_cast_fp16")];
+            tensor<int32, [4]> var_2558 = const()[name = tensor<string, []>("op_2558"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_319_cast_fp16 = reshape(shape = var_2558, x = x_317_cast_fp16)[name = tensor<string, []>("x_319_cast_fp16")];
+            tensor<int32, [4]> var_2562_begin_0 = const()[name = tensor<string, []>("op_2562_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2562_end_0 = const()[name = tensor<string, []>("op_2562_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_2562_end_mask_0 = const()[name = tensor<string, []>("op_2562_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_2562_cast_fp16 = slice_by_index(begin = var_2562_begin_0, end = var_2562_end_0, end_mask = var_2562_end_mask_0, x = x_319_cast_fp16)[name = tensor<string, []>("op_2562_cast_fp16")];
+            tensor<int32, [4]> var_2563 = const()[name = tensor<string, []>("op_2563"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_57_cast_fp16 = reshape(shape = var_2563, x = var_2562_cast_fp16)[name = tensor<string, []>("matrix_bd_57_cast_fp16")];
+            tensor<bool, []> matrix_ac_29_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_29_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_29_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_29_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_100_perm_0 = const()[name = tensor<string, []>("transpose_100_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_101_perm_0 = const()[name = tensor<string, []>("transpose_101_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_187 = transpose(perm = transpose_101_perm_0, x = k_57_cast_fp16)[name = tensor<string, []>("transpose_187")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_188 = transpose(perm = transpose_100_perm_0, x = var_2546_cast_fp16)[name = tensor<string, []>("transpose_188")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_29_cast_fp16 = matmul(transpose_x = matrix_ac_29_transpose_x_0, transpose_y = matrix_ac_29_transpose_y_0, x = transpose_188, y = transpose_187)[name = tensor<string, []>("matrix_ac_29_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_59_begin_0 = const()[name = tensor<string, []>("matrix_bd_59_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_59_end_0 = const()[name = tensor<string, []>("matrix_bd_59_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_59_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_59_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_59_cast_fp16 = slice_by_index(begin = matrix_bd_59_begin_0, end = matrix_bd_59_end_0, end_mask = matrix_bd_59_end_mask_0, x = matrix_bd_57_cast_fp16)[name = tensor<string, []>("matrix_bd_59_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2572_cast_fp16 = add(x = matrix_ac_29_cast_fp16, y = matrix_bd_59_cast_fp16)[name = tensor<string, []>("op_2572_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_57_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_57_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_57_cast_fp16 = mul(x = var_2572_cast_fp16, y = _inversed_scores_57_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_57_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_59_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_57_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_59_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2578_cast_fp16 = softmax(axis = var_25, x = scores_59_cast_fp16)[name = tensor<string, []>("op_2578_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_761_cast_fp16 = select(a = var_6_to_fp16, b = var_2578_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_761_cast_fp16")];
+            tensor<bool, []> x_321_transpose_x_0 = const()[name = tensor<string, []>("x_321_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_321_transpose_y_0 = const()[name = tensor<string, []>("x_321_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_190 = transpose(perm = value_29_perm_0, x = v_29_cast_fp16)[name = tensor<string, []>("transpose_190")];
+            tensor<fp16, [1, 8, 126, 128]> x_321_cast_fp16 = matmul(transpose_x = x_321_transpose_x_0, transpose_y = x_321_transpose_y_0, x = input_761_cast_fp16, y = transpose_190)[name = tensor<string, []>("x_321_cast_fp16")];
+            tensor<int32, [4]> var_2582_perm_0 = const()[name = tensor<string, []>("op_2582_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2583 = const()[name = tensor<string, []>("op_2583"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_186 = transpose(perm = var_2582_perm_0, x = x_321_cast_fp16)[name = tensor<string, []>("transpose_186")];
+            tensor<fp16, [1, 126, 1024]> input_763_cast_fp16 = reshape(shape = var_2583, x = transpose_186)[name = tensor<string, []>("input_763_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_14_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_14_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(715408960)))];
+            tensor<fp16, [1, 126, 1024]> linear_133_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_14_self_attn_linear_out_weight_to_fp16, x = input_763_cast_fp16)[name = tensor<string, []>("linear_133_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_767_cast_fp16 = add(x = input_759_cast_fp16, y = linear_133_cast_fp16)[name = tensor<string, []>("input_767_cast_fp16")];
+            tensor<int32, [1]> x_325_axes_0 = const()[name = tensor<string, []>("x_325_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_14_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_14_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(717506176)))];
+            tensor<fp16, [1024]> model_layers_14_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_14_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(717508288)))];
+            tensor<fp16, [1, 126, 1024]> x_325_cast_fp16 = layer_norm(axes = x_325_axes_0, beta = model_layers_14_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_14_norm_conv_weight_to_fp16, x = input_767_cast_fp16)[name = tensor<string, []>("x_325_cast_fp16")];
+            tensor<int32, [3]> input_769_perm_0 = const()[name = tensor<string, []>("input_769_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_2599 = const()[name = tensor<string, []>("op_2599"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_2601 = const()[name = tensor<string, []>("op_2601"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_771_pad_type_0 = const()[name = tensor<string, []>("input_771_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_771_pad_0 = const()[name = tensor<string, []>("input_771_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_14_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_14_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(717510400)))];
+            tensor<fp16, [1, 1024, 126]> transpose_185 = transpose(perm = input_769_perm_0, x = x_325_cast_fp16)[name = tensor<string, []>("transpose_185")];
+            tensor<fp16, [1, 2048, 126]> input_771_cast_fp16 = conv(dilations = var_2601, groups = var_27, pad = input_771_pad_0, pad_type = input_771_pad_type_0, strides = var_2599, weight = model_layers_14_conv_pointwise_conv1_weight_to_fp16, x = transpose_185)[name = tensor<string, []>("input_771_cast_fp16")];
+            tensor<int32, []> x_327_split_num_splits_0 = const()[name = tensor<string, []>("x_327_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_327_split_axis_0 = const()[name = tensor<string, []>("x_327_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_327_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_327_split_cast_fp16_1 = split(axis = x_327_split_axis_0, num_splits = x_327_split_num_splits_0, x = input_771_cast_fp16)[name = tensor<string, []>("x_327_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_327_split_1_sigmoid_cast_fp16 = sigmoid(x = x_327_split_cast_fp16_1)[name = tensor<string, []>("x_327_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_327_cast_fp16 = mul(x = x_327_split_cast_fp16_0, y = x_327_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_327_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_773_cast_fp16 = select(a = var_6_to_fp16, b = x_327_cast_fp16, cond = var_323)[name = tensor<string, []>("input_773_cast_fp16")];
+            tensor<int32, [6]> input_775_pad_0 = const()[name = tensor<string, []>("input_775_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_775_mode_0 = const()[name = tensor<string, []>("input_775_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_156_to_fp16 = const()[name = tensor<string, []>("const_156_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_775_cast_fp16 = pad(constant_val = const_156_to_fp16, mode = input_775_mode_0, pad = input_775_pad_0, x = input_773_cast_fp16)[name = tensor<string, []>("input_775_cast_fp16")];
+            tensor<int32, [1]> var_2610 = const()[name = tensor<string, []>("op_2610"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_2612 = const()[name = tensor<string, []>("op_2612"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_777_pad_type_0 = const()[name = tensor<string, []>("input_777_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_777_pad_0 = const()[name = tensor<string, []>("input_777_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_779_weight_0_to_fp16 = const()[name = tensor<string, []>("input_779_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(721704768)))];
+            tensor<fp16, [1024]> input_779_bias_0_to_fp16 = const()[name = tensor<string, []>("input_779_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(721723264)))];
+            tensor<fp16, [1, 1024, 126]> input_779_cast_fp16 = conv(bias = input_779_bias_0_to_fp16, dilations = var_2612, groups = var_5, pad = input_777_pad_0, pad_type = input_777_pad_type_0, strides = var_2610, weight = input_779_weight_0_to_fp16, x = input_775_cast_fp16)[name = tensor<string, []>("input_779_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_781_cast_fp16 = silu(x = input_779_cast_fp16)[name = tensor<string, []>("input_781_cast_fp16")];
+            tensor<int32, [1]> var_2622 = const()[name = tensor<string, []>("op_2622"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_2624 = const()[name = tensor<string, []>("op_2624"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_329_pad_type_0 = const()[name = tensor<string, []>("x_329_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_329_pad_0 = const()[name = tensor<string, []>("x_329_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_14_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_14_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(721725376)))];
+            tensor<fp16, [1, 1024, 126]> x_329_cast_fp16 = conv(dilations = var_2624, groups = var_27, pad = x_329_pad_0, pad_type = x_329_pad_type_0, strides = var_2622, weight = model_layers_14_conv_pointwise_conv2_weight_to_fp16, x = input_781_cast_fp16)[name = tensor<string, []>("x_329_cast_fp16")];
+            tensor<int32, [3]> input_783_perm_0 = const()[name = tensor<string, []>("input_783_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_184 = transpose(perm = input_783_perm_0, x = x_329_cast_fp16)[name = tensor<string, []>("transpose_184")];
+            tensor<fp16, [1, 126, 1024]> input_785_cast_fp16 = add(x = input_767_cast_fp16, y = transpose_184)[name = tensor<string, []>("input_785_cast_fp16")];
+            tensor<int32, [1]> input_787_axes_0 = const()[name = tensor<string, []>("input_787_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_14_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_14_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(723822592)))];
+            tensor<fp16, [1024]> model_layers_14_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_14_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(723824704)))];
+            tensor<fp16, [1, 126, 1024]> input_787_cast_fp16 = layer_norm(axes = input_787_axes_0, beta = model_layers_14_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_14_norm_feed_forward2_weight_to_fp16, x = input_785_cast_fp16)[name = tensor<string, []>("input_787_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_14_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_14_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(723826816)))];
+            tensor<fp16, [1, 126, 4096]> linear_134_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_14_feed_forward2_linear1_weight_to_fp16, x = input_787_cast_fp16)[name = tensor<string, []>("linear_134_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_791_cast_fp16 = silu(x = linear_134_cast_fp16)[name = tensor<string, []>("input_791_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_14_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_14_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(732215488)))];
+            tensor<fp16, [1, 126, 1024]> linear_135_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_14_feed_forward2_linear2_weight_to_fp16, x = input_791_cast_fp16)[name = tensor<string, []>("linear_135_cast_fp16")];
+            tensor<fp16, []> var_2643_to_fp16 = const()[name = tensor<string, []>("op_2643_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2644_cast_fp16 = mul(x = linear_135_cast_fp16, y = var_2643_to_fp16)[name = tensor<string, []>("op_2644_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_797_cast_fp16 = add(x = input_785_cast_fp16, y = var_2644_cast_fp16)[name = tensor<string, []>("input_797_cast_fp16")];
+            tensor<int32, [1]> input_799_axes_0 = const()[name = tensor<string, []>("input_799_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_14_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_14_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(740604160)))];
+            tensor<fp16, [1024]> model_layers_14_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_14_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(740606272)))];
+            tensor<fp16, [1, 126, 1024]> input_799_cast_fp16 = layer_norm(axes = input_799_axes_0, beta = model_layers_14_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_14_norm_out_weight_to_fp16, x = input_797_cast_fp16)[name = tensor<string, []>("input_799_cast_fp16")];
+            tensor<int32, [1]> input_801_axes_0 = const()[name = tensor<string, []>("input_801_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_15_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_15_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(740608384)))];
+            tensor<fp16, [1024]> model_layers_15_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_15_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(740610496)))];
+            tensor<fp16, [1, 126, 1024]> input_801_cast_fp16 = layer_norm(axes = input_801_axes_0, beta = model_layers_15_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_15_norm_feed_forward1_weight_to_fp16, x = input_799_cast_fp16)[name = tensor<string, []>("input_801_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_15_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_15_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(740612608)))];
+            tensor<fp16, [1, 126, 4096]> linear_136_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_15_feed_forward1_linear1_weight_to_fp16, x = input_801_cast_fp16)[name = tensor<string, []>("linear_136_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_805_cast_fp16 = silu(x = linear_136_cast_fp16)[name = tensor<string, []>("input_805_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_15_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_15_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(749001280)))];
+            tensor<fp16, [1, 126, 1024]> linear_137_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_15_feed_forward1_linear2_weight_to_fp16, x = input_805_cast_fp16)[name = tensor<string, []>("linear_137_cast_fp16")];
+            tensor<fp16, []> var_2672_to_fp16 = const()[name = tensor<string, []>("op_2672_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2673_cast_fp16 = mul(x = linear_137_cast_fp16, y = var_2672_to_fp16)[name = tensor<string, []>("op_2673_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_811_cast_fp16 = add(x = input_799_cast_fp16, y = var_2673_cast_fp16)[name = tensor<string, []>("input_811_cast_fp16")];
+            tensor<int32, [1]> query_31_axes_0 = const()[name = tensor<string, []>("query_31_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_15_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_15_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(757389952)))];
+            tensor<fp16, [1024]> model_layers_15_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_15_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(757392064)))];
+            tensor<fp16, [1, 126, 1024]> query_31_cast_fp16 = layer_norm(axes = query_31_axes_0, beta = model_layers_15_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_15_norm_self_att_weight_to_fp16, x = input_811_cast_fp16)[name = tensor<string, []>("query_31_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_15_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_15_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(757394176)))];
+            tensor<fp16, [1, 126, 1024]> linear_138_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_15_self_attn_linear_q_weight_to_fp16, x = query_31_cast_fp16)[name = tensor<string, []>("linear_138_cast_fp16")];
+            tensor<int32, [4]> var_2689 = const()[name = tensor<string, []>("op_2689"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_91_cast_fp16 = reshape(shape = var_2689, x = linear_138_cast_fp16)[name = tensor<string, []>("q_91_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_15_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_15_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(759491392)))];
+            tensor<fp16, [1, 126, 1024]> linear_139_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_15_self_attn_linear_k_weight_to_fp16, x = query_31_cast_fp16)[name = tensor<string, []>("linear_139_cast_fp16")];
+            tensor<int32, [4]> var_2693 = const()[name = tensor<string, []>("op_2693"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_61_cast_fp16 = reshape(shape = var_2693, x = linear_139_cast_fp16)[name = tensor<string, []>("k_61_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_15_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_15_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(761588608)))];
+            tensor<fp16, [1, 126, 1024]> linear_140_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_15_self_attn_linear_v_weight_to_fp16, x = query_31_cast_fp16)[name = tensor<string, []>("linear_140_cast_fp16")];
+            tensor<int32, [4]> var_2697 = const()[name = tensor<string, []>("op_2697"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_31_cast_fp16 = reshape(shape = var_2697, x = linear_140_cast_fp16)[name = tensor<string, []>("v_31_cast_fp16")];
+            tensor<int32, [4]> value_31_perm_0 = const()[name = tensor<string, []>("value_31_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_15_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_15_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(763685824)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2709_cast_fp16 = add(x = q_91_cast_fp16, y = model_layers_15_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_2709_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_15_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_15_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(763687936)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2711_cast_fp16 = add(x = q_91_cast_fp16, y = model_layers_15_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_2711_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_31_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_31_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_337_transpose_x_0 = const()[name = tensor<string, []>("x_337_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_337_transpose_y_0 = const()[name = tensor<string, []>("x_337_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_2713_to_fp16 = const()[name = tensor<string, []>("op_2713_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(763690048)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_182 = transpose(perm = q_with_bias_v_31_perm_0, x = var_2711_cast_fp16)[name = tensor<string, []>("transpose_182")];
+            tensor<fp16, [1, 8, 126, 251]> x_337_cast_fp16 = matmul(transpose_x = x_337_transpose_x_0, transpose_y = x_337_transpose_y_0, x = transpose_182, y = var_2713_to_fp16)[name = tensor<string, []>("x_337_cast_fp16")];
+            tensor<int32, [8]> x_339_pad_0 = const()[name = tensor<string, []>("x_339_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_339_mode_0 = const()[name = tensor<string, []>("x_339_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_163_to_fp16 = const()[name = tensor<string, []>("const_163_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_339_cast_fp16 = pad(constant_val = const_163_to_fp16, mode = x_339_mode_0, pad = x_339_pad_0, x = x_337_cast_fp16)[name = tensor<string, []>("x_339_cast_fp16")];
+            tensor<int32, [4]> var_2721 = const()[name = tensor<string, []>("op_2721"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_341_cast_fp16 = reshape(shape = var_2721, x = x_339_cast_fp16)[name = tensor<string, []>("x_341_cast_fp16")];
+            tensor<int32, [4]> var_2725_begin_0 = const()[name = tensor<string, []>("op_2725_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2725_end_0 = const()[name = tensor<string, []>("op_2725_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_2725_end_mask_0 = const()[name = tensor<string, []>("op_2725_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_2725_cast_fp16 = slice_by_index(begin = var_2725_begin_0, end = var_2725_end_0, end_mask = var_2725_end_mask_0, x = x_341_cast_fp16)[name = tensor<string, []>("op_2725_cast_fp16")];
+            tensor<int32, [4]> var_2726 = const()[name = tensor<string, []>("op_2726"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_61_cast_fp16 = reshape(shape = var_2726, x = var_2725_cast_fp16)[name = tensor<string, []>("matrix_bd_61_cast_fp16")];
+            tensor<bool, []> matrix_ac_31_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_31_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_31_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_31_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_102_perm_0 = const()[name = tensor<string, []>("transpose_102_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_103_perm_0 = const()[name = tensor<string, []>("transpose_103_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_180 = transpose(perm = transpose_103_perm_0, x = k_61_cast_fp16)[name = tensor<string, []>("transpose_180")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_181 = transpose(perm = transpose_102_perm_0, x = var_2709_cast_fp16)[name = tensor<string, []>("transpose_181")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_31_cast_fp16 = matmul(transpose_x = matrix_ac_31_transpose_x_0, transpose_y = matrix_ac_31_transpose_y_0, x = transpose_181, y = transpose_180)[name = tensor<string, []>("matrix_ac_31_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_63_begin_0 = const()[name = tensor<string, []>("matrix_bd_63_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_63_end_0 = const()[name = tensor<string, []>("matrix_bd_63_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_63_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_63_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_63_cast_fp16 = slice_by_index(begin = matrix_bd_63_begin_0, end = matrix_bd_63_end_0, end_mask = matrix_bd_63_end_mask_0, x = matrix_bd_61_cast_fp16)[name = tensor<string, []>("matrix_bd_63_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2735_cast_fp16 = add(x = matrix_ac_31_cast_fp16, y = matrix_bd_63_cast_fp16)[name = tensor<string, []>("op_2735_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_61_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_61_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_61_cast_fp16 = mul(x = var_2735_cast_fp16, y = _inversed_scores_61_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_61_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_63_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_61_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_63_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2741_cast_fp16 = softmax(axis = var_25, x = scores_63_cast_fp16)[name = tensor<string, []>("op_2741_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_813_cast_fp16 = select(a = var_6_to_fp16, b = var_2741_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_813_cast_fp16")];
+            tensor<bool, []> x_343_transpose_x_0 = const()[name = tensor<string, []>("x_343_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_343_transpose_y_0 = const()[name = tensor<string, []>("x_343_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_183 = transpose(perm = value_31_perm_0, x = v_31_cast_fp16)[name = tensor<string, []>("transpose_183")];
+            tensor<fp16, [1, 8, 126, 128]> x_343_cast_fp16 = matmul(transpose_x = x_343_transpose_x_0, transpose_y = x_343_transpose_y_0, x = input_813_cast_fp16, y = transpose_183)[name = tensor<string, []>("x_343_cast_fp16")];
+            tensor<int32, [4]> var_2745_perm_0 = const()[name = tensor<string, []>("op_2745_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2746 = const()[name = tensor<string, []>("op_2746"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_179 = transpose(perm = var_2745_perm_0, x = x_343_cast_fp16)[name = tensor<string, []>("transpose_179")];
+            tensor<fp16, [1, 126, 1024]> input_815_cast_fp16 = reshape(shape = var_2746, x = transpose_179)[name = tensor<string, []>("input_815_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_15_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_15_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(764204160)))];
+            tensor<fp16, [1, 126, 1024]> linear_142_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_15_self_attn_linear_out_weight_to_fp16, x = input_815_cast_fp16)[name = tensor<string, []>("linear_142_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_819_cast_fp16 = add(x = input_811_cast_fp16, y = linear_142_cast_fp16)[name = tensor<string, []>("input_819_cast_fp16")];
+            tensor<int32, [1]> x_347_axes_0 = const()[name = tensor<string, []>("x_347_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_15_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_15_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(766301376)))];
+            tensor<fp16, [1024]> model_layers_15_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_15_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(766303488)))];
+            tensor<fp16, [1, 126, 1024]> x_347_cast_fp16 = layer_norm(axes = x_347_axes_0, beta = model_layers_15_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_15_norm_conv_weight_to_fp16, x = input_819_cast_fp16)[name = tensor<string, []>("x_347_cast_fp16")];
+            tensor<int32, [3]> input_821_perm_0 = const()[name = tensor<string, []>("input_821_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_2762 = const()[name = tensor<string, []>("op_2762"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_2764 = const()[name = tensor<string, []>("op_2764"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_823_pad_type_0 = const()[name = tensor<string, []>("input_823_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_823_pad_0 = const()[name = tensor<string, []>("input_823_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_15_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_15_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(766305600)))];
+            tensor<fp16, [1, 1024, 126]> transpose_178 = transpose(perm = input_821_perm_0, x = x_347_cast_fp16)[name = tensor<string, []>("transpose_178")];
+            tensor<fp16, [1, 2048, 126]> input_823_cast_fp16 = conv(dilations = var_2764, groups = var_27, pad = input_823_pad_0, pad_type = input_823_pad_type_0, strides = var_2762, weight = model_layers_15_conv_pointwise_conv1_weight_to_fp16, x = transpose_178)[name = tensor<string, []>("input_823_cast_fp16")];
+            tensor<int32, []> x_349_split_num_splits_0 = const()[name = tensor<string, []>("x_349_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_349_split_axis_0 = const()[name = tensor<string, []>("x_349_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_349_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_349_split_cast_fp16_1 = split(axis = x_349_split_axis_0, num_splits = x_349_split_num_splits_0, x = input_823_cast_fp16)[name = tensor<string, []>("x_349_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_349_split_1_sigmoid_cast_fp16 = sigmoid(x = x_349_split_cast_fp16_1)[name = tensor<string, []>("x_349_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_349_cast_fp16 = mul(x = x_349_split_cast_fp16_0, y = x_349_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_349_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_825_cast_fp16 = select(a = var_6_to_fp16, b = x_349_cast_fp16, cond = var_323)[name = tensor<string, []>("input_825_cast_fp16")];
+            tensor<int32, [6]> input_827_pad_0 = const()[name = tensor<string, []>("input_827_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_827_mode_0 = const()[name = tensor<string, []>("input_827_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_166_to_fp16 = const()[name = tensor<string, []>("const_166_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_827_cast_fp16 = pad(constant_val = const_166_to_fp16, mode = input_827_mode_0, pad = input_827_pad_0, x = input_825_cast_fp16)[name = tensor<string, []>("input_827_cast_fp16")];
+            tensor<int32, [1]> var_2773 = const()[name = tensor<string, []>("op_2773"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_2775 = const()[name = tensor<string, []>("op_2775"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_829_pad_type_0 = const()[name = tensor<string, []>("input_829_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_829_pad_0 = const()[name = tensor<string, []>("input_829_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_831_weight_0_to_fp16 = const()[name = tensor<string, []>("input_831_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(770499968)))];
+            tensor<fp16, [1024]> input_831_bias_0_to_fp16 = const()[name = tensor<string, []>("input_831_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(770518464)))];
+            tensor<fp16, [1, 1024, 126]> input_831_cast_fp16 = conv(bias = input_831_bias_0_to_fp16, dilations = var_2775, groups = var_5, pad = input_829_pad_0, pad_type = input_829_pad_type_0, strides = var_2773, weight = input_831_weight_0_to_fp16, x = input_827_cast_fp16)[name = tensor<string, []>("input_831_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_833_cast_fp16 = silu(x = input_831_cast_fp16)[name = tensor<string, []>("input_833_cast_fp16")];
+            tensor<int32, [1]> var_2785 = const()[name = tensor<string, []>("op_2785"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_2787 = const()[name = tensor<string, []>("op_2787"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_351_pad_type_0 = const()[name = tensor<string, []>("x_351_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_351_pad_0 = const()[name = tensor<string, []>("x_351_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_15_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_15_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(770520576)))];
+            tensor<fp16, [1, 1024, 126]> x_351_cast_fp16 = conv(dilations = var_2787, groups = var_27, pad = x_351_pad_0, pad_type = x_351_pad_type_0, strides = var_2785, weight = model_layers_15_conv_pointwise_conv2_weight_to_fp16, x = input_833_cast_fp16)[name = tensor<string, []>("x_351_cast_fp16")];
+            tensor<int32, [3]> input_835_perm_0 = const()[name = tensor<string, []>("input_835_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_177 = transpose(perm = input_835_perm_0, x = x_351_cast_fp16)[name = tensor<string, []>("transpose_177")];
+            tensor<fp16, [1, 126, 1024]> input_837_cast_fp16 = add(x = input_819_cast_fp16, y = transpose_177)[name = tensor<string, []>("input_837_cast_fp16")];
+            tensor<int32, [1]> input_839_axes_0 = const()[name = tensor<string, []>("input_839_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_15_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_15_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(772617792)))];
+            tensor<fp16, [1024]> model_layers_15_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_15_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(772619904)))];
+            tensor<fp16, [1, 126, 1024]> input_839_cast_fp16 = layer_norm(axes = input_839_axes_0, beta = model_layers_15_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_15_norm_feed_forward2_weight_to_fp16, x = input_837_cast_fp16)[name = tensor<string, []>("input_839_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_15_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_15_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(772622016)))];
+            tensor<fp16, [1, 126, 4096]> linear_143_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_15_feed_forward2_linear1_weight_to_fp16, x = input_839_cast_fp16)[name = tensor<string, []>("linear_143_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_843_cast_fp16 = silu(x = linear_143_cast_fp16)[name = tensor<string, []>("input_843_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_15_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_15_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(781010688)))];
+            tensor<fp16, [1, 126, 1024]> linear_144_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_15_feed_forward2_linear2_weight_to_fp16, x = input_843_cast_fp16)[name = tensor<string, []>("linear_144_cast_fp16")];
+            tensor<fp16, []> var_2806_to_fp16 = const()[name = tensor<string, []>("op_2806_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2807_cast_fp16 = mul(x = linear_144_cast_fp16, y = var_2806_to_fp16)[name = tensor<string, []>("op_2807_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_849_cast_fp16 = add(x = input_837_cast_fp16, y = var_2807_cast_fp16)[name = tensor<string, []>("input_849_cast_fp16")];
+            tensor<int32, [1]> input_851_axes_0 = const()[name = tensor<string, []>("input_851_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_15_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_15_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(789399360)))];
+            tensor<fp16, [1024]> model_layers_15_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_15_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(789401472)))];
+            tensor<fp16, [1, 126, 1024]> input_851_cast_fp16 = layer_norm(axes = input_851_axes_0, beta = model_layers_15_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_15_norm_out_weight_to_fp16, x = input_849_cast_fp16)[name = tensor<string, []>("input_851_cast_fp16")];
+            tensor<int32, [1]> input_853_axes_0 = const()[name = tensor<string, []>("input_853_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_16_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_16_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(789403584)))];
+            tensor<fp16, [1024]> model_layers_16_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_16_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(789405696)))];
+            tensor<fp16, [1, 126, 1024]> input_853_cast_fp16 = layer_norm(axes = input_853_axes_0, beta = model_layers_16_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_16_norm_feed_forward1_weight_to_fp16, x = input_851_cast_fp16)[name = tensor<string, []>("input_853_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_16_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_16_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(789407808)))];
+            tensor<fp16, [1, 126, 4096]> linear_145_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_16_feed_forward1_linear1_weight_to_fp16, x = input_853_cast_fp16)[name = tensor<string, []>("linear_145_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_857_cast_fp16 = silu(x = linear_145_cast_fp16)[name = tensor<string, []>("input_857_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_16_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_16_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(797796480)))];
+            tensor<fp16, [1, 126, 1024]> linear_146_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_16_feed_forward1_linear2_weight_to_fp16, x = input_857_cast_fp16)[name = tensor<string, []>("linear_146_cast_fp16")];
+            tensor<fp16, []> var_2835_to_fp16 = const()[name = tensor<string, []>("op_2835_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2836_cast_fp16 = mul(x = linear_146_cast_fp16, y = var_2835_to_fp16)[name = tensor<string, []>("op_2836_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_863_cast_fp16 = add(x = input_851_cast_fp16, y = var_2836_cast_fp16)[name = tensor<string, []>("input_863_cast_fp16")];
+            tensor<int32, [1]> query_33_axes_0 = const()[name = tensor<string, []>("query_33_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_16_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_16_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(806185152)))];
+            tensor<fp16, [1024]> model_layers_16_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_16_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(806187264)))];
+            tensor<fp16, [1, 126, 1024]> query_33_cast_fp16 = layer_norm(axes = query_33_axes_0, beta = model_layers_16_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_16_norm_self_att_weight_to_fp16, x = input_863_cast_fp16)[name = tensor<string, []>("query_33_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_16_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_16_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(806189376)))];
+            tensor<fp16, [1, 126, 1024]> linear_147_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_16_self_attn_linear_q_weight_to_fp16, x = query_33_cast_fp16)[name = tensor<string, []>("linear_147_cast_fp16")];
+            tensor<int32, [4]> var_2852 = const()[name = tensor<string, []>("op_2852"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_97_cast_fp16 = reshape(shape = var_2852, x = linear_147_cast_fp16)[name = tensor<string, []>("q_97_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_16_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_16_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(808286592)))];
+            tensor<fp16, [1, 126, 1024]> linear_148_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_16_self_attn_linear_k_weight_to_fp16, x = query_33_cast_fp16)[name = tensor<string, []>("linear_148_cast_fp16")];
+            tensor<int32, [4]> var_2856 = const()[name = tensor<string, []>("op_2856"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_65_cast_fp16 = reshape(shape = var_2856, x = linear_148_cast_fp16)[name = tensor<string, []>("k_65_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_16_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_16_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(810383808)))];
+            tensor<fp16, [1, 126, 1024]> linear_149_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_16_self_attn_linear_v_weight_to_fp16, x = query_33_cast_fp16)[name = tensor<string, []>("linear_149_cast_fp16")];
+            tensor<int32, [4]> var_2860 = const()[name = tensor<string, []>("op_2860"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_33_cast_fp16 = reshape(shape = var_2860, x = linear_149_cast_fp16)[name = tensor<string, []>("v_33_cast_fp16")];
+            tensor<int32, [4]> value_33_perm_0 = const()[name = tensor<string, []>("value_33_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_16_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_16_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(812481024)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2872_cast_fp16 = add(x = q_97_cast_fp16, y = model_layers_16_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_2872_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_16_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_16_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(812483136)))];
+            tensor<fp16, [1, 126, 8, 128]> var_2874_cast_fp16 = add(x = q_97_cast_fp16, y = model_layers_16_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_2874_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_33_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_33_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_359_transpose_x_0 = const()[name = tensor<string, []>("x_359_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_359_transpose_y_0 = const()[name = tensor<string, []>("x_359_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_2876_to_fp16 = const()[name = tensor<string, []>("op_2876_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(812485248)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_175 = transpose(perm = q_with_bias_v_33_perm_0, x = var_2874_cast_fp16)[name = tensor<string, []>("transpose_175")];
+            tensor<fp16, [1, 8, 126, 251]> x_359_cast_fp16 = matmul(transpose_x = x_359_transpose_x_0, transpose_y = x_359_transpose_y_0, x = transpose_175, y = var_2876_to_fp16)[name = tensor<string, []>("x_359_cast_fp16")];
+            tensor<int32, [8]> x_361_pad_0 = const()[name = tensor<string, []>("x_361_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_361_mode_0 = const()[name = tensor<string, []>("x_361_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_173_to_fp16 = const()[name = tensor<string, []>("const_173_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_361_cast_fp16 = pad(constant_val = const_173_to_fp16, mode = x_361_mode_0, pad = x_361_pad_0, x = x_359_cast_fp16)[name = tensor<string, []>("x_361_cast_fp16")];
+            tensor<int32, [4]> var_2884 = const()[name = tensor<string, []>("op_2884"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_363_cast_fp16 = reshape(shape = var_2884, x = x_361_cast_fp16)[name = tensor<string, []>("x_363_cast_fp16")];
+            tensor<int32, [4]> var_2888_begin_0 = const()[name = tensor<string, []>("op_2888_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_2888_end_0 = const()[name = tensor<string, []>("op_2888_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_2888_end_mask_0 = const()[name = tensor<string, []>("op_2888_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_2888_cast_fp16 = slice_by_index(begin = var_2888_begin_0, end = var_2888_end_0, end_mask = var_2888_end_mask_0, x = x_363_cast_fp16)[name = tensor<string, []>("op_2888_cast_fp16")];
+            tensor<int32, [4]> var_2889 = const()[name = tensor<string, []>("op_2889"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_65_cast_fp16 = reshape(shape = var_2889, x = var_2888_cast_fp16)[name = tensor<string, []>("matrix_bd_65_cast_fp16")];
+            tensor<bool, []> matrix_ac_33_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_33_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_33_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_33_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_104_perm_0 = const()[name = tensor<string, []>("transpose_104_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_105_perm_0 = const()[name = tensor<string, []>("transpose_105_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_173 = transpose(perm = transpose_105_perm_0, x = k_65_cast_fp16)[name = tensor<string, []>("transpose_173")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_174 = transpose(perm = transpose_104_perm_0, x = var_2872_cast_fp16)[name = tensor<string, []>("transpose_174")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_33_cast_fp16 = matmul(transpose_x = matrix_ac_33_transpose_x_0, transpose_y = matrix_ac_33_transpose_y_0, x = transpose_174, y = transpose_173)[name = tensor<string, []>("matrix_ac_33_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_67_begin_0 = const()[name = tensor<string, []>("matrix_bd_67_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_67_end_0 = const()[name = tensor<string, []>("matrix_bd_67_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_67_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_67_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_67_cast_fp16 = slice_by_index(begin = matrix_bd_67_begin_0, end = matrix_bd_67_end_0, end_mask = matrix_bd_67_end_mask_0, x = matrix_bd_65_cast_fp16)[name = tensor<string, []>("matrix_bd_67_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2898_cast_fp16 = add(x = matrix_ac_33_cast_fp16, y = matrix_bd_67_cast_fp16)[name = tensor<string, []>("op_2898_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_65_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_65_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_65_cast_fp16 = mul(x = var_2898_cast_fp16, y = _inversed_scores_65_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_65_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_67_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_65_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_67_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_2904_cast_fp16 = softmax(axis = var_25, x = scores_67_cast_fp16)[name = tensor<string, []>("op_2904_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_865_cast_fp16 = select(a = var_6_to_fp16, b = var_2904_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_865_cast_fp16")];
+            tensor<bool, []> x_365_transpose_x_0 = const()[name = tensor<string, []>("x_365_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_365_transpose_y_0 = const()[name = tensor<string, []>("x_365_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_176 = transpose(perm = value_33_perm_0, x = v_33_cast_fp16)[name = tensor<string, []>("transpose_176")];
+            tensor<fp16, [1, 8, 126, 128]> x_365_cast_fp16 = matmul(transpose_x = x_365_transpose_x_0, transpose_y = x_365_transpose_y_0, x = input_865_cast_fp16, y = transpose_176)[name = tensor<string, []>("x_365_cast_fp16")];
+            tensor<int32, [4]> var_2908_perm_0 = const()[name = tensor<string, []>("op_2908_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_2909 = const()[name = tensor<string, []>("op_2909"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_172 = transpose(perm = var_2908_perm_0, x = x_365_cast_fp16)[name = tensor<string, []>("transpose_172")];
+            tensor<fp16, [1, 126, 1024]> input_867_cast_fp16 = reshape(shape = var_2909, x = transpose_172)[name = tensor<string, []>("input_867_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_16_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_16_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(812999360)))];
+            tensor<fp16, [1, 126, 1024]> linear_151_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_16_self_attn_linear_out_weight_to_fp16, x = input_867_cast_fp16)[name = tensor<string, []>("linear_151_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_871_cast_fp16 = add(x = input_863_cast_fp16, y = linear_151_cast_fp16)[name = tensor<string, []>("input_871_cast_fp16")];
+            tensor<int32, [1]> x_369_axes_0 = const()[name = tensor<string, []>("x_369_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_16_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_16_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(815096576)))];
+            tensor<fp16, [1024]> model_layers_16_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_16_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(815098688)))];
+            tensor<fp16, [1, 126, 1024]> x_369_cast_fp16 = layer_norm(axes = x_369_axes_0, beta = model_layers_16_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_16_norm_conv_weight_to_fp16, x = input_871_cast_fp16)[name = tensor<string, []>("x_369_cast_fp16")];
+            tensor<int32, [3]> input_873_perm_0 = const()[name = tensor<string, []>("input_873_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_2925 = const()[name = tensor<string, []>("op_2925"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_2927 = const()[name = tensor<string, []>("op_2927"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_875_pad_type_0 = const()[name = tensor<string, []>("input_875_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_875_pad_0 = const()[name = tensor<string, []>("input_875_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_16_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_16_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(815100800)))];
+            tensor<fp16, [1, 1024, 126]> transpose_171 = transpose(perm = input_873_perm_0, x = x_369_cast_fp16)[name = tensor<string, []>("transpose_171")];
+            tensor<fp16, [1, 2048, 126]> input_875_cast_fp16 = conv(dilations = var_2927, groups = var_27, pad = input_875_pad_0, pad_type = input_875_pad_type_0, strides = var_2925, weight = model_layers_16_conv_pointwise_conv1_weight_to_fp16, x = transpose_171)[name = tensor<string, []>("input_875_cast_fp16")];
+            tensor<int32, []> x_371_split_num_splits_0 = const()[name = tensor<string, []>("x_371_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_371_split_axis_0 = const()[name = tensor<string, []>("x_371_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_371_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_371_split_cast_fp16_1 = split(axis = x_371_split_axis_0, num_splits = x_371_split_num_splits_0, x = input_875_cast_fp16)[name = tensor<string, []>("x_371_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_371_split_1_sigmoid_cast_fp16 = sigmoid(x = x_371_split_cast_fp16_1)[name = tensor<string, []>("x_371_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_371_cast_fp16 = mul(x = x_371_split_cast_fp16_0, y = x_371_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_371_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_877_cast_fp16 = select(a = var_6_to_fp16, b = x_371_cast_fp16, cond = var_323)[name = tensor<string, []>("input_877_cast_fp16")];
+            tensor<int32, [6]> input_879_pad_0 = const()[name = tensor<string, []>("input_879_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_879_mode_0 = const()[name = tensor<string, []>("input_879_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_176_to_fp16 = const()[name = tensor<string, []>("const_176_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_879_cast_fp16 = pad(constant_val = const_176_to_fp16, mode = input_879_mode_0, pad = input_879_pad_0, x = input_877_cast_fp16)[name = tensor<string, []>("input_879_cast_fp16")];
+            tensor<int32, [1]> var_2936 = const()[name = tensor<string, []>("op_2936"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_2938 = const()[name = tensor<string, []>("op_2938"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_881_pad_type_0 = const()[name = tensor<string, []>("input_881_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_881_pad_0 = const()[name = tensor<string, []>("input_881_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_883_weight_0_to_fp16 = const()[name = tensor<string, []>("input_883_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(819295168)))];
+            tensor<fp16, [1024]> input_883_bias_0_to_fp16 = const()[name = tensor<string, []>("input_883_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(819313664)))];
+            tensor<fp16, [1, 1024, 126]> input_883_cast_fp16 = conv(bias = input_883_bias_0_to_fp16, dilations = var_2938, groups = var_5, pad = input_881_pad_0, pad_type = input_881_pad_type_0, strides = var_2936, weight = input_883_weight_0_to_fp16, x = input_879_cast_fp16)[name = tensor<string, []>("input_883_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_885_cast_fp16 = silu(x = input_883_cast_fp16)[name = tensor<string, []>("input_885_cast_fp16")];
+            tensor<int32, [1]> var_2948 = const()[name = tensor<string, []>("op_2948"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_2950 = const()[name = tensor<string, []>("op_2950"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_373_pad_type_0 = const()[name = tensor<string, []>("x_373_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_373_pad_0 = const()[name = tensor<string, []>("x_373_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_16_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_16_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(819315776)))];
+            tensor<fp16, [1, 1024, 126]> x_373_cast_fp16 = conv(dilations = var_2950, groups = var_27, pad = x_373_pad_0, pad_type = x_373_pad_type_0, strides = var_2948, weight = model_layers_16_conv_pointwise_conv2_weight_to_fp16, x = input_885_cast_fp16)[name = tensor<string, []>("x_373_cast_fp16")];
+            tensor<int32, [3]> input_887_perm_0 = const()[name = tensor<string, []>("input_887_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_170 = transpose(perm = input_887_perm_0, x = x_373_cast_fp16)[name = tensor<string, []>("transpose_170")];
+            tensor<fp16, [1, 126, 1024]> input_889_cast_fp16 = add(x = input_871_cast_fp16, y = transpose_170)[name = tensor<string, []>("input_889_cast_fp16")];
+            tensor<int32, [1]> input_891_axes_0 = const()[name = tensor<string, []>("input_891_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_16_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_16_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(821412992)))];
+            tensor<fp16, [1024]> model_layers_16_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_16_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(821415104)))];
+            tensor<fp16, [1, 126, 1024]> input_891_cast_fp16 = layer_norm(axes = input_891_axes_0, beta = model_layers_16_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_16_norm_feed_forward2_weight_to_fp16, x = input_889_cast_fp16)[name = tensor<string, []>("input_891_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_16_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_16_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(821417216)))];
+            tensor<fp16, [1, 126, 4096]> linear_152_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_16_feed_forward2_linear1_weight_to_fp16, x = input_891_cast_fp16)[name = tensor<string, []>("linear_152_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_895_cast_fp16 = silu(x = linear_152_cast_fp16)[name = tensor<string, []>("input_895_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_16_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_16_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(829805888)))];
+            tensor<fp16, [1, 126, 1024]> linear_153_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_16_feed_forward2_linear2_weight_to_fp16, x = input_895_cast_fp16)[name = tensor<string, []>("linear_153_cast_fp16")];
+            tensor<fp16, []> var_2969_to_fp16 = const()[name = tensor<string, []>("op_2969_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2970_cast_fp16 = mul(x = linear_153_cast_fp16, y = var_2969_to_fp16)[name = tensor<string, []>("op_2970_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_901_cast_fp16 = add(x = input_889_cast_fp16, y = var_2970_cast_fp16)[name = tensor<string, []>("input_901_cast_fp16")];
+            tensor<int32, [1]> input_903_axes_0 = const()[name = tensor<string, []>("input_903_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_16_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_16_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(838194560)))];
+            tensor<fp16, [1024]> model_layers_16_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_16_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(838196672)))];
+            tensor<fp16, [1, 126, 1024]> input_903_cast_fp16 = layer_norm(axes = input_903_axes_0, beta = model_layers_16_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_16_norm_out_weight_to_fp16, x = input_901_cast_fp16)[name = tensor<string, []>("input_903_cast_fp16")];
+            tensor<int32, [1]> input_905_axes_0 = const()[name = tensor<string, []>("input_905_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_17_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_17_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(838198784)))];
+            tensor<fp16, [1024]> model_layers_17_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_17_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(838200896)))];
+            tensor<fp16, [1, 126, 1024]> input_905_cast_fp16 = layer_norm(axes = input_905_axes_0, beta = model_layers_17_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_17_norm_feed_forward1_weight_to_fp16, x = input_903_cast_fp16)[name = tensor<string, []>("input_905_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_17_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_17_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(838203008)))];
+            tensor<fp16, [1, 126, 4096]> linear_154_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_17_feed_forward1_linear1_weight_to_fp16, x = input_905_cast_fp16)[name = tensor<string, []>("linear_154_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_909_cast_fp16 = silu(x = linear_154_cast_fp16)[name = tensor<string, []>("input_909_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_17_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_17_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(846591680)))];
+            tensor<fp16, [1, 126, 1024]> linear_155_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_17_feed_forward1_linear2_weight_to_fp16, x = input_909_cast_fp16)[name = tensor<string, []>("linear_155_cast_fp16")];
+            tensor<fp16, []> var_2998_to_fp16 = const()[name = tensor<string, []>("op_2998_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_2999_cast_fp16 = mul(x = linear_155_cast_fp16, y = var_2998_to_fp16)[name = tensor<string, []>("op_2999_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_915_cast_fp16 = add(x = input_903_cast_fp16, y = var_2999_cast_fp16)[name = tensor<string, []>("input_915_cast_fp16")];
+            tensor<int32, [1]> query_35_axes_0 = const()[name = tensor<string, []>("query_35_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_17_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_17_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(854980352)))];
+            tensor<fp16, [1024]> model_layers_17_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_17_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(854982464)))];
+            tensor<fp16, [1, 126, 1024]> query_35_cast_fp16 = layer_norm(axes = query_35_axes_0, beta = model_layers_17_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_17_norm_self_att_weight_to_fp16, x = input_915_cast_fp16)[name = tensor<string, []>("query_35_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_17_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_17_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(854984576)))];
+            tensor<fp16, [1, 126, 1024]> linear_156_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_17_self_attn_linear_q_weight_to_fp16, x = query_35_cast_fp16)[name = tensor<string, []>("linear_156_cast_fp16")];
+            tensor<int32, [4]> var_3015 = const()[name = tensor<string, []>("op_3015"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_103_cast_fp16 = reshape(shape = var_3015, x = linear_156_cast_fp16)[name = tensor<string, []>("q_103_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_17_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_17_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(857081792)))];
+            tensor<fp16, [1, 126, 1024]> linear_157_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_17_self_attn_linear_k_weight_to_fp16, x = query_35_cast_fp16)[name = tensor<string, []>("linear_157_cast_fp16")];
+            tensor<int32, [4]> var_3019 = const()[name = tensor<string, []>("op_3019"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_69_cast_fp16 = reshape(shape = var_3019, x = linear_157_cast_fp16)[name = tensor<string, []>("k_69_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_17_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_17_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(859179008)))];
+            tensor<fp16, [1, 126, 1024]> linear_158_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_17_self_attn_linear_v_weight_to_fp16, x = query_35_cast_fp16)[name = tensor<string, []>("linear_158_cast_fp16")];
+            tensor<int32, [4]> var_3023 = const()[name = tensor<string, []>("op_3023"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_35_cast_fp16 = reshape(shape = var_3023, x = linear_158_cast_fp16)[name = tensor<string, []>("v_35_cast_fp16")];
+            tensor<int32, [4]> value_35_perm_0 = const()[name = tensor<string, []>("value_35_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_17_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_17_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(861276224)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3035_cast_fp16 = add(x = q_103_cast_fp16, y = model_layers_17_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_3035_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_17_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_17_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(861278336)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3037_cast_fp16 = add(x = q_103_cast_fp16, y = model_layers_17_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_3037_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_35_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_35_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_381_transpose_x_0 = const()[name = tensor<string, []>("x_381_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_381_transpose_y_0 = const()[name = tensor<string, []>("x_381_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_3039_to_fp16 = const()[name = tensor<string, []>("op_3039_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(861280448)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_168 = transpose(perm = q_with_bias_v_35_perm_0, x = var_3037_cast_fp16)[name = tensor<string, []>("transpose_168")];
+            tensor<fp16, [1, 8, 126, 251]> x_381_cast_fp16 = matmul(transpose_x = x_381_transpose_x_0, transpose_y = x_381_transpose_y_0, x = transpose_168, y = var_3039_to_fp16)[name = tensor<string, []>("x_381_cast_fp16")];
+            tensor<int32, [8]> x_383_pad_0 = const()[name = tensor<string, []>("x_383_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_383_mode_0 = const()[name = tensor<string, []>("x_383_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_183_to_fp16 = const()[name = tensor<string, []>("const_183_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_383_cast_fp16 = pad(constant_val = const_183_to_fp16, mode = x_383_mode_0, pad = x_383_pad_0, x = x_381_cast_fp16)[name = tensor<string, []>("x_383_cast_fp16")];
+            tensor<int32, [4]> var_3047 = const()[name = tensor<string, []>("op_3047"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_385_cast_fp16 = reshape(shape = var_3047, x = x_383_cast_fp16)[name = tensor<string, []>("x_385_cast_fp16")];
+            tensor<int32, [4]> var_3051_begin_0 = const()[name = tensor<string, []>("op_3051_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3051_end_0 = const()[name = tensor<string, []>("op_3051_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_3051_end_mask_0 = const()[name = tensor<string, []>("op_3051_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> 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 = x_385_cast_fp16)[name = tensor<string, []>("op_3051_cast_fp16")];
+            tensor<int32, [4]> var_3052 = const()[name = tensor<string, []>("op_3052"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_69_cast_fp16 = reshape(shape = var_3052, x = var_3051_cast_fp16)[name = tensor<string, []>("matrix_bd_69_cast_fp16")];
+            tensor<bool, []> matrix_ac_35_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_35_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_35_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_35_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_106_perm_0 = const()[name = tensor<string, []>("transpose_106_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_107_perm_0 = const()[name = tensor<string, []>("transpose_107_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_166 = transpose(perm = transpose_107_perm_0, x = k_69_cast_fp16)[name = tensor<string, []>("transpose_166")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_167 = transpose(perm = transpose_106_perm_0, x = var_3035_cast_fp16)[name = tensor<string, []>("transpose_167")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_35_cast_fp16 = matmul(transpose_x = matrix_ac_35_transpose_x_0, transpose_y = matrix_ac_35_transpose_y_0, x = transpose_167, y = transpose_166)[name = tensor<string, []>("matrix_ac_35_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_71_begin_0 = const()[name = tensor<string, []>("matrix_bd_71_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_71_end_0 = const()[name = tensor<string, []>("matrix_bd_71_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_71_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_71_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_71_cast_fp16 = slice_by_index(begin = matrix_bd_71_begin_0, end = matrix_bd_71_end_0, end_mask = matrix_bd_71_end_mask_0, x = matrix_bd_69_cast_fp16)[name = tensor<string, []>("matrix_bd_71_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3061_cast_fp16 = add(x = matrix_ac_35_cast_fp16, y = matrix_bd_71_cast_fp16)[name = tensor<string, []>("op_3061_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_69_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_69_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_69_cast_fp16 = mul(x = var_3061_cast_fp16, y = _inversed_scores_69_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_69_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_71_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_69_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_71_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3067_cast_fp16 = softmax(axis = var_25, x = scores_71_cast_fp16)[name = tensor<string, []>("op_3067_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_917_cast_fp16 = select(a = var_6_to_fp16, b = var_3067_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_917_cast_fp16")];
+            tensor<bool, []> x_387_transpose_x_0 = const()[name = tensor<string, []>("x_387_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_387_transpose_y_0 = const()[name = tensor<string, []>("x_387_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_169 = transpose(perm = value_35_perm_0, x = v_35_cast_fp16)[name = tensor<string, []>("transpose_169")];
+            tensor<fp16, [1, 8, 126, 128]> x_387_cast_fp16 = matmul(transpose_x = x_387_transpose_x_0, transpose_y = x_387_transpose_y_0, x = input_917_cast_fp16, y = transpose_169)[name = tensor<string, []>("x_387_cast_fp16")];
+            tensor<int32, [4]> var_3071_perm_0 = const()[name = tensor<string, []>("op_3071_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3072 = const()[name = tensor<string, []>("op_3072"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_165 = transpose(perm = var_3071_perm_0, x = x_387_cast_fp16)[name = tensor<string, []>("transpose_165")];
+            tensor<fp16, [1, 126, 1024]> input_919_cast_fp16 = reshape(shape = var_3072, x = transpose_165)[name = tensor<string, []>("input_919_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_17_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_17_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(861794560)))];
+            tensor<fp16, [1, 126, 1024]> linear_160_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_17_self_attn_linear_out_weight_to_fp16, x = input_919_cast_fp16)[name = tensor<string, []>("linear_160_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_923_cast_fp16 = add(x = input_915_cast_fp16, y = linear_160_cast_fp16)[name = tensor<string, []>("input_923_cast_fp16")];
+            tensor<int32, [1]> x_391_axes_0 = const()[name = tensor<string, []>("x_391_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_17_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_17_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(863891776)))];
+            tensor<fp16, [1024]> model_layers_17_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_17_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(863893888)))];
+            tensor<fp16, [1, 126, 1024]> x_391_cast_fp16 = layer_norm(axes = x_391_axes_0, beta = model_layers_17_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_17_norm_conv_weight_to_fp16, x = input_923_cast_fp16)[name = tensor<string, []>("x_391_cast_fp16")];
+            tensor<int32, [3]> input_925_perm_0 = const()[name = tensor<string, []>("input_925_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_3088 = const()[name = tensor<string, []>("op_3088"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_3090 = const()[name = tensor<string, []>("op_3090"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_927_pad_type_0 = const()[name = tensor<string, []>("input_927_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_927_pad_0 = const()[name = tensor<string, []>("input_927_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_17_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_17_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(863896000)))];
+            tensor<fp16, [1, 1024, 126]> transpose_164 = transpose(perm = input_925_perm_0, x = x_391_cast_fp16)[name = tensor<string, []>("transpose_164")];
+            tensor<fp16, [1, 2048, 126]> input_927_cast_fp16 = conv(dilations = var_3090, groups = var_27, pad = input_927_pad_0, pad_type = input_927_pad_type_0, strides = var_3088, weight = model_layers_17_conv_pointwise_conv1_weight_to_fp16, x = transpose_164)[name = tensor<string, []>("input_927_cast_fp16")];
+            tensor<int32, []> x_393_split_num_splits_0 = const()[name = tensor<string, []>("x_393_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_393_split_axis_0 = const()[name = tensor<string, []>("x_393_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_393_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_393_split_cast_fp16_1 = split(axis = x_393_split_axis_0, num_splits = x_393_split_num_splits_0, x = input_927_cast_fp16)[name = tensor<string, []>("x_393_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_393_split_1_sigmoid_cast_fp16 = sigmoid(x = x_393_split_cast_fp16_1)[name = tensor<string, []>("x_393_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_393_cast_fp16 = mul(x = x_393_split_cast_fp16_0, y = x_393_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_393_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_929_cast_fp16 = select(a = var_6_to_fp16, b = x_393_cast_fp16, cond = var_323)[name = tensor<string, []>("input_929_cast_fp16")];
+            tensor<int32, [6]> input_931_pad_0 = const()[name = tensor<string, []>("input_931_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_931_mode_0 = const()[name = tensor<string, []>("input_931_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_186_to_fp16 = const()[name = tensor<string, []>("const_186_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_931_cast_fp16 = pad(constant_val = const_186_to_fp16, mode = input_931_mode_0, pad = input_931_pad_0, x = input_929_cast_fp16)[name = tensor<string, []>("input_931_cast_fp16")];
+            tensor<int32, [1]> var_3099 = const()[name = tensor<string, []>("op_3099"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_3101 = const()[name = tensor<string, []>("op_3101"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_933_pad_type_0 = const()[name = tensor<string, []>("input_933_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_933_pad_0 = const()[name = tensor<string, []>("input_933_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_935_weight_0_to_fp16 = const()[name = tensor<string, []>("input_935_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(868090368)))];
+            tensor<fp16, [1024]> input_935_bias_0_to_fp16 = const()[name = tensor<string, []>("input_935_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(868108864)))];
+            tensor<fp16, [1, 1024, 126]> input_935_cast_fp16 = conv(bias = input_935_bias_0_to_fp16, dilations = var_3101, groups = var_5, pad = input_933_pad_0, pad_type = input_933_pad_type_0, strides = var_3099, weight = input_935_weight_0_to_fp16, x = input_931_cast_fp16)[name = tensor<string, []>("input_935_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_937_cast_fp16 = silu(x = input_935_cast_fp16)[name = tensor<string, []>("input_937_cast_fp16")];
+            tensor<int32, [1]> var_3111 = const()[name = tensor<string, []>("op_3111"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_3113 = const()[name = tensor<string, []>("op_3113"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_395_pad_type_0 = const()[name = tensor<string, []>("x_395_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_395_pad_0 = const()[name = tensor<string, []>("x_395_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_17_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_17_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(868110976)))];
+            tensor<fp16, [1, 1024, 126]> x_395_cast_fp16 = conv(dilations = var_3113, groups = var_27, pad = x_395_pad_0, pad_type = x_395_pad_type_0, strides = var_3111, weight = model_layers_17_conv_pointwise_conv2_weight_to_fp16, x = input_937_cast_fp16)[name = tensor<string, []>("x_395_cast_fp16")];
+            tensor<int32, [3]> input_939_perm_0 = const()[name = tensor<string, []>("input_939_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_163 = transpose(perm = input_939_perm_0, x = x_395_cast_fp16)[name = tensor<string, []>("transpose_163")];
+            tensor<fp16, [1, 126, 1024]> input_941_cast_fp16 = add(x = input_923_cast_fp16, y = transpose_163)[name = tensor<string, []>("input_941_cast_fp16")];
+            tensor<int32, [1]> input_943_axes_0 = const()[name = tensor<string, []>("input_943_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_17_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_17_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(870208192)))];
+            tensor<fp16, [1024]> model_layers_17_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_17_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(870210304)))];
+            tensor<fp16, [1, 126, 1024]> input_943_cast_fp16 = layer_norm(axes = input_943_axes_0, beta = model_layers_17_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_17_norm_feed_forward2_weight_to_fp16, x = input_941_cast_fp16)[name = tensor<string, []>("input_943_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_17_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_17_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(870212416)))];
+            tensor<fp16, [1, 126, 4096]> linear_161_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_17_feed_forward2_linear1_weight_to_fp16, x = input_943_cast_fp16)[name = tensor<string, []>("linear_161_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_947_cast_fp16 = silu(x = linear_161_cast_fp16)[name = tensor<string, []>("input_947_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_17_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_17_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(878601088)))];
+            tensor<fp16, [1, 126, 1024]> linear_162_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_17_feed_forward2_linear2_weight_to_fp16, x = input_947_cast_fp16)[name = tensor<string, []>("linear_162_cast_fp16")];
+            tensor<fp16, []> var_3132_to_fp16 = const()[name = tensor<string, []>("op_3132_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3133_cast_fp16 = mul(x = linear_162_cast_fp16, y = var_3132_to_fp16)[name = tensor<string, []>("op_3133_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_953_cast_fp16 = add(x = input_941_cast_fp16, y = var_3133_cast_fp16)[name = tensor<string, []>("input_953_cast_fp16")];
+            tensor<int32, [1]> input_955_axes_0 = const()[name = tensor<string, []>("input_955_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_17_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_17_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(886989760)))];
+            tensor<fp16, [1024]> model_layers_17_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_17_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(886991872)))];
+            tensor<fp16, [1, 126, 1024]> input_955_cast_fp16 = layer_norm(axes = input_955_axes_0, beta = model_layers_17_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_17_norm_out_weight_to_fp16, x = input_953_cast_fp16)[name = tensor<string, []>("input_955_cast_fp16")];
+            tensor<int32, [1]> input_957_axes_0 = const()[name = tensor<string, []>("input_957_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_18_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_18_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(886993984)))];
+            tensor<fp16, [1024]> model_layers_18_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_18_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(886996096)))];
+            tensor<fp16, [1, 126, 1024]> input_957_cast_fp16 = layer_norm(axes = input_957_axes_0, beta = model_layers_18_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_18_norm_feed_forward1_weight_to_fp16, x = input_955_cast_fp16)[name = tensor<string, []>("input_957_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_18_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_18_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(886998208)))];
+            tensor<fp16, [1, 126, 4096]> linear_163_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_18_feed_forward1_linear1_weight_to_fp16, x = input_957_cast_fp16)[name = tensor<string, []>("linear_163_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_961_cast_fp16 = silu(x = linear_163_cast_fp16)[name = tensor<string, []>("input_961_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_18_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_18_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(895386880)))];
+            tensor<fp16, [1, 126, 1024]> linear_164_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_18_feed_forward1_linear2_weight_to_fp16, x = input_961_cast_fp16)[name = tensor<string, []>("linear_164_cast_fp16")];
+            tensor<fp16, []> var_3161_to_fp16 = const()[name = tensor<string, []>("op_3161_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3162_cast_fp16 = mul(x = linear_164_cast_fp16, y = var_3161_to_fp16)[name = tensor<string, []>("op_3162_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_967_cast_fp16 = add(x = input_955_cast_fp16, y = var_3162_cast_fp16)[name = tensor<string, []>("input_967_cast_fp16")];
+            tensor<int32, [1]> query_37_axes_0 = const()[name = tensor<string, []>("query_37_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_18_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_18_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(903775552)))];
+            tensor<fp16, [1024]> model_layers_18_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_18_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(903777664)))];
+            tensor<fp16, [1, 126, 1024]> query_37_cast_fp16 = layer_norm(axes = query_37_axes_0, beta = model_layers_18_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_18_norm_self_att_weight_to_fp16, x = input_967_cast_fp16)[name = tensor<string, []>("query_37_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_18_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_18_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(903779776)))];
+            tensor<fp16, [1, 126, 1024]> linear_165_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_18_self_attn_linear_q_weight_to_fp16, x = query_37_cast_fp16)[name = tensor<string, []>("linear_165_cast_fp16")];
+            tensor<int32, [4]> var_3178 = const()[name = tensor<string, []>("op_3178"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_109_cast_fp16 = reshape(shape = var_3178, x = linear_165_cast_fp16)[name = tensor<string, []>("q_109_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_18_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_18_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(905876992)))];
+            tensor<fp16, [1, 126, 1024]> linear_166_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_18_self_attn_linear_k_weight_to_fp16, x = query_37_cast_fp16)[name = tensor<string, []>("linear_166_cast_fp16")];
+            tensor<int32, [4]> var_3182 = const()[name = tensor<string, []>("op_3182"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_73_cast_fp16 = reshape(shape = var_3182, x = linear_166_cast_fp16)[name = tensor<string, []>("k_73_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_18_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_18_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(907974208)))];
+            tensor<fp16, [1, 126, 1024]> linear_167_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_18_self_attn_linear_v_weight_to_fp16, x = query_37_cast_fp16)[name = tensor<string, []>("linear_167_cast_fp16")];
+            tensor<int32, [4]> var_3186 = const()[name = tensor<string, []>("op_3186"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_37_cast_fp16 = reshape(shape = var_3186, x = linear_167_cast_fp16)[name = tensor<string, []>("v_37_cast_fp16")];
+            tensor<int32, [4]> value_37_perm_0 = const()[name = tensor<string, []>("value_37_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_18_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_18_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(910071424)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3198_cast_fp16 = add(x = q_109_cast_fp16, y = model_layers_18_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_3198_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_18_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_18_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(910073536)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3200_cast_fp16 = add(x = q_109_cast_fp16, y = model_layers_18_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_3200_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_37_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_37_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_403_transpose_x_0 = const()[name = tensor<string, []>("x_403_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_403_transpose_y_0 = const()[name = tensor<string, []>("x_403_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_3202_to_fp16 = const()[name = tensor<string, []>("op_3202_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(910075648)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_161 = transpose(perm = q_with_bias_v_37_perm_0, x = var_3200_cast_fp16)[name = tensor<string, []>("transpose_161")];
+            tensor<fp16, [1, 8, 126, 251]> x_403_cast_fp16 = matmul(transpose_x = x_403_transpose_x_0, transpose_y = x_403_transpose_y_0, x = transpose_161, y = var_3202_to_fp16)[name = tensor<string, []>("x_403_cast_fp16")];
+            tensor<int32, [8]> x_405_pad_0 = const()[name = tensor<string, []>("x_405_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_405_mode_0 = const()[name = tensor<string, []>("x_405_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_193_to_fp16 = const()[name = tensor<string, []>("const_193_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_405_cast_fp16 = pad(constant_val = const_193_to_fp16, mode = x_405_mode_0, pad = x_405_pad_0, x = x_403_cast_fp16)[name = tensor<string, []>("x_405_cast_fp16")];
+            tensor<int32, [4]> var_3210 = const()[name = tensor<string, []>("op_3210"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_407_cast_fp16 = reshape(shape = var_3210, x = x_405_cast_fp16)[name = tensor<string, []>("x_407_cast_fp16")];
+            tensor<int32, [4]> var_3214_begin_0 = const()[name = tensor<string, []>("op_3214_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3214_end_0 = const()[name = tensor<string, []>("op_3214_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_3214_end_mask_0 = const()[name = tensor<string, []>("op_3214_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_3214_cast_fp16 = slice_by_index(begin = var_3214_begin_0, end = var_3214_end_0, end_mask = var_3214_end_mask_0, x = x_407_cast_fp16)[name = tensor<string, []>("op_3214_cast_fp16")];
+            tensor<int32, [4]> var_3215 = const()[name = tensor<string, []>("op_3215"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_73_cast_fp16 = reshape(shape = var_3215, x = var_3214_cast_fp16)[name = tensor<string, []>("matrix_bd_73_cast_fp16")];
+            tensor<bool, []> matrix_ac_37_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_37_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_37_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_37_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_108_perm_0 = const()[name = tensor<string, []>("transpose_108_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_109_perm_0 = const()[name = tensor<string, []>("transpose_109_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_159 = transpose(perm = transpose_109_perm_0, x = k_73_cast_fp16)[name = tensor<string, []>("transpose_159")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_160 = transpose(perm = transpose_108_perm_0, x = var_3198_cast_fp16)[name = tensor<string, []>("transpose_160")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_37_cast_fp16 = matmul(transpose_x = matrix_ac_37_transpose_x_0, transpose_y = matrix_ac_37_transpose_y_0, x = transpose_160, y = transpose_159)[name = tensor<string, []>("matrix_ac_37_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_75_begin_0 = const()[name = tensor<string, []>("matrix_bd_75_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_75_end_0 = const()[name = tensor<string, []>("matrix_bd_75_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_75_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_75_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_75_cast_fp16 = slice_by_index(begin = matrix_bd_75_begin_0, end = matrix_bd_75_end_0, end_mask = matrix_bd_75_end_mask_0, x = matrix_bd_73_cast_fp16)[name = tensor<string, []>("matrix_bd_75_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3224_cast_fp16 = add(x = matrix_ac_37_cast_fp16, y = matrix_bd_75_cast_fp16)[name = tensor<string, []>("op_3224_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_73_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_73_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_73_cast_fp16 = mul(x = var_3224_cast_fp16, y = _inversed_scores_73_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_73_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_75_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_73_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_75_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3230_cast_fp16 = softmax(axis = var_25, x = scores_75_cast_fp16)[name = tensor<string, []>("op_3230_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_969_cast_fp16 = select(a = var_6_to_fp16, b = var_3230_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_969_cast_fp16")];
+            tensor<bool, []> x_409_transpose_x_0 = const()[name = tensor<string, []>("x_409_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_409_transpose_y_0 = const()[name = tensor<string, []>("x_409_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_162 = transpose(perm = value_37_perm_0, x = v_37_cast_fp16)[name = tensor<string, []>("transpose_162")];
+            tensor<fp16, [1, 8, 126, 128]> x_409_cast_fp16 = matmul(transpose_x = x_409_transpose_x_0, transpose_y = x_409_transpose_y_0, x = input_969_cast_fp16, y = transpose_162)[name = tensor<string, []>("x_409_cast_fp16")];
+            tensor<int32, [4]> var_3234_perm_0 = const()[name = tensor<string, []>("op_3234_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3235 = const()[name = tensor<string, []>("op_3235"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_158 = transpose(perm = var_3234_perm_0, x = x_409_cast_fp16)[name = tensor<string, []>("transpose_158")];
+            tensor<fp16, [1, 126, 1024]> input_971_cast_fp16 = reshape(shape = var_3235, x = transpose_158)[name = tensor<string, []>("input_971_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_18_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_18_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(910589760)))];
+            tensor<fp16, [1, 126, 1024]> linear_169_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_18_self_attn_linear_out_weight_to_fp16, x = input_971_cast_fp16)[name = tensor<string, []>("linear_169_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_975_cast_fp16 = add(x = input_967_cast_fp16, y = linear_169_cast_fp16)[name = tensor<string, []>("input_975_cast_fp16")];
+            tensor<int32, [1]> x_413_axes_0 = const()[name = tensor<string, []>("x_413_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_18_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_18_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(912686976)))];
+            tensor<fp16, [1024]> model_layers_18_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_18_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(912689088)))];
+            tensor<fp16, [1, 126, 1024]> x_413_cast_fp16 = layer_norm(axes = x_413_axes_0, beta = model_layers_18_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_18_norm_conv_weight_to_fp16, x = input_975_cast_fp16)[name = tensor<string, []>("x_413_cast_fp16")];
+            tensor<int32, [3]> input_977_perm_0 = const()[name = tensor<string, []>("input_977_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_3251 = const()[name = tensor<string, []>("op_3251"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_3253 = const()[name = tensor<string, []>("op_3253"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_979_pad_type_0 = const()[name = tensor<string, []>("input_979_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_979_pad_0 = const()[name = tensor<string, []>("input_979_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_18_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_18_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(912691200)))];
+            tensor<fp16, [1, 1024, 126]> transpose_157 = transpose(perm = input_977_perm_0, x = x_413_cast_fp16)[name = tensor<string, []>("transpose_157")];
+            tensor<fp16, [1, 2048, 126]> input_979_cast_fp16 = conv(dilations = var_3253, groups = var_27, pad = input_979_pad_0, pad_type = input_979_pad_type_0, strides = var_3251, weight = model_layers_18_conv_pointwise_conv1_weight_to_fp16, x = transpose_157)[name = tensor<string, []>("input_979_cast_fp16")];
+            tensor<int32, []> x_415_split_num_splits_0 = const()[name = tensor<string, []>("x_415_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_415_split_axis_0 = const()[name = tensor<string, []>("x_415_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_415_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_415_split_cast_fp16_1 = split(axis = x_415_split_axis_0, num_splits = x_415_split_num_splits_0, x = input_979_cast_fp16)[name = tensor<string, []>("x_415_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_415_split_1_sigmoid_cast_fp16 = sigmoid(x = x_415_split_cast_fp16_1)[name = tensor<string, []>("x_415_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_415_cast_fp16 = mul(x = x_415_split_cast_fp16_0, y = x_415_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_415_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_981_cast_fp16 = select(a = var_6_to_fp16, b = x_415_cast_fp16, cond = var_323)[name = tensor<string, []>("input_981_cast_fp16")];
+            tensor<int32, [6]> input_983_pad_0 = const()[name = tensor<string, []>("input_983_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_983_mode_0 = const()[name = tensor<string, []>("input_983_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_196_to_fp16 = const()[name = tensor<string, []>("const_196_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_983_cast_fp16 = pad(constant_val = const_196_to_fp16, mode = input_983_mode_0, pad = input_983_pad_0, x = input_981_cast_fp16)[name = tensor<string, []>("input_983_cast_fp16")];
+            tensor<int32, [1]> var_3262 = const()[name = tensor<string, []>("op_3262"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_3264 = const()[name = tensor<string, []>("op_3264"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_985_pad_type_0 = const()[name = tensor<string, []>("input_985_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_985_pad_0 = const()[name = tensor<string, []>("input_985_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_987_weight_0_to_fp16 = const()[name = tensor<string, []>("input_987_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(916885568)))];
+            tensor<fp16, [1024]> input_987_bias_0_to_fp16 = const()[name = tensor<string, []>("input_987_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(916904064)))];
+            tensor<fp16, [1, 1024, 126]> input_987_cast_fp16 = conv(bias = input_987_bias_0_to_fp16, dilations = var_3264, groups = var_5, pad = input_985_pad_0, pad_type = input_985_pad_type_0, strides = var_3262, weight = input_987_weight_0_to_fp16, x = input_983_cast_fp16)[name = tensor<string, []>("input_987_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_989_cast_fp16 = silu(x = input_987_cast_fp16)[name = tensor<string, []>("input_989_cast_fp16")];
+            tensor<int32, [1]> var_3274 = const()[name = tensor<string, []>("op_3274"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_3276 = const()[name = tensor<string, []>("op_3276"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_417_pad_type_0 = const()[name = tensor<string, []>("x_417_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_417_pad_0 = const()[name = tensor<string, []>("x_417_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_18_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_18_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(916906176)))];
+            tensor<fp16, [1, 1024, 126]> x_417_cast_fp16 = conv(dilations = var_3276, groups = var_27, pad = x_417_pad_0, pad_type = x_417_pad_type_0, strides = var_3274, weight = model_layers_18_conv_pointwise_conv2_weight_to_fp16, x = input_989_cast_fp16)[name = tensor<string, []>("x_417_cast_fp16")];
+            tensor<int32, [3]> input_991_perm_0 = const()[name = tensor<string, []>("input_991_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_156 = transpose(perm = input_991_perm_0, x = x_417_cast_fp16)[name = tensor<string, []>("transpose_156")];
+            tensor<fp16, [1, 126, 1024]> input_993_cast_fp16 = add(x = input_975_cast_fp16, y = transpose_156)[name = tensor<string, []>("input_993_cast_fp16")];
+            tensor<int32, [1]> input_995_axes_0 = const()[name = tensor<string, []>("input_995_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_18_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_18_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(919003392)))];
+            tensor<fp16, [1024]> model_layers_18_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_18_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(919005504)))];
+            tensor<fp16, [1, 126, 1024]> input_995_cast_fp16 = layer_norm(axes = input_995_axes_0, beta = model_layers_18_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_18_norm_feed_forward2_weight_to_fp16, x = input_993_cast_fp16)[name = tensor<string, []>("input_995_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_18_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_18_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(919007616)))];
+            tensor<fp16, [1, 126, 4096]> linear_170_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_18_feed_forward2_linear1_weight_to_fp16, x = input_995_cast_fp16)[name = tensor<string, []>("linear_170_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_999_cast_fp16 = silu(x = linear_170_cast_fp16)[name = tensor<string, []>("input_999_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_18_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_18_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(927396288)))];
+            tensor<fp16, [1, 126, 1024]> linear_171_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_18_feed_forward2_linear2_weight_to_fp16, x = input_999_cast_fp16)[name = tensor<string, []>("linear_171_cast_fp16")];
+            tensor<fp16, []> var_3295_to_fp16 = const()[name = tensor<string, []>("op_3295_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3296_cast_fp16 = mul(x = linear_171_cast_fp16, y = var_3295_to_fp16)[name = tensor<string, []>("op_3296_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1005_cast_fp16 = add(x = input_993_cast_fp16, y = var_3296_cast_fp16)[name = tensor<string, []>("input_1005_cast_fp16")];
+            tensor<int32, [1]> input_1007_axes_0 = const()[name = tensor<string, []>("input_1007_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_18_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_18_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(935784960)))];
+            tensor<fp16, [1024]> model_layers_18_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_18_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(935787072)))];
+            tensor<fp16, [1, 126, 1024]> input_1007_cast_fp16 = layer_norm(axes = input_1007_axes_0, beta = model_layers_18_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_18_norm_out_weight_to_fp16, x = input_1005_cast_fp16)[name = tensor<string, []>("input_1007_cast_fp16")];
+            tensor<int32, [1]> input_1009_axes_0 = const()[name = tensor<string, []>("input_1009_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_19_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_19_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(935789184)))];
+            tensor<fp16, [1024]> model_layers_19_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_19_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(935791296)))];
+            tensor<fp16, [1, 126, 1024]> input_1009_cast_fp16 = layer_norm(axes = input_1009_axes_0, beta = model_layers_19_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_19_norm_feed_forward1_weight_to_fp16, x = input_1007_cast_fp16)[name = tensor<string, []>("input_1009_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_19_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_19_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(935793408)))];
+            tensor<fp16, [1, 126, 4096]> linear_172_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_19_feed_forward1_linear1_weight_to_fp16, x = input_1009_cast_fp16)[name = tensor<string, []>("linear_172_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_1013_cast_fp16 = silu(x = linear_172_cast_fp16)[name = tensor<string, []>("input_1013_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_19_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_19_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(944182080)))];
+            tensor<fp16, [1, 126, 1024]> linear_173_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_19_feed_forward1_linear2_weight_to_fp16, x = input_1013_cast_fp16)[name = tensor<string, []>("linear_173_cast_fp16")];
+            tensor<fp16, []> var_3324_to_fp16 = const()[name = tensor<string, []>("op_3324_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3325_cast_fp16 = mul(x = linear_173_cast_fp16, y = var_3324_to_fp16)[name = tensor<string, []>("op_3325_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1019_cast_fp16 = add(x = input_1007_cast_fp16, y = var_3325_cast_fp16)[name = tensor<string, []>("input_1019_cast_fp16")];
+            tensor<int32, [1]> query_39_axes_0 = const()[name = tensor<string, []>("query_39_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_19_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_19_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(952570752)))];
+            tensor<fp16, [1024]> model_layers_19_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_19_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(952572864)))];
+            tensor<fp16, [1, 126, 1024]> query_39_cast_fp16 = layer_norm(axes = query_39_axes_0, beta = model_layers_19_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_19_norm_self_att_weight_to_fp16, x = input_1019_cast_fp16)[name = tensor<string, []>("query_39_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_19_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_19_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(952574976)))];
+            tensor<fp16, [1, 126, 1024]> linear_174_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_19_self_attn_linear_q_weight_to_fp16, x = query_39_cast_fp16)[name = tensor<string, []>("linear_174_cast_fp16")];
+            tensor<int32, [4]> var_3341 = const()[name = tensor<string, []>("op_3341"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_115_cast_fp16 = reshape(shape = var_3341, x = linear_174_cast_fp16)[name = tensor<string, []>("q_115_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_19_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_19_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(954672192)))];
+            tensor<fp16, [1, 126, 1024]> linear_175_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_19_self_attn_linear_k_weight_to_fp16, x = query_39_cast_fp16)[name = tensor<string, []>("linear_175_cast_fp16")];
+            tensor<int32, [4]> var_3345 = const()[name = tensor<string, []>("op_3345"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_77_cast_fp16 = reshape(shape = var_3345, x = linear_175_cast_fp16)[name = tensor<string, []>("k_77_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_19_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_19_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(956769408)))];
+            tensor<fp16, [1, 126, 1024]> linear_176_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_19_self_attn_linear_v_weight_to_fp16, x = query_39_cast_fp16)[name = tensor<string, []>("linear_176_cast_fp16")];
+            tensor<int32, [4]> var_3349 = const()[name = tensor<string, []>("op_3349"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_39_cast_fp16 = reshape(shape = var_3349, x = linear_176_cast_fp16)[name = tensor<string, []>("v_39_cast_fp16")];
+            tensor<int32, [4]> value_39_perm_0 = const()[name = tensor<string, []>("value_39_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_19_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_19_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(958866624)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3361_cast_fp16 = add(x = q_115_cast_fp16, y = model_layers_19_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_3361_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_19_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_19_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(958868736)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3363_cast_fp16 = add(x = q_115_cast_fp16, y = model_layers_19_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_3363_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_39_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_39_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_425_transpose_x_0 = const()[name = tensor<string, []>("x_425_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_425_transpose_y_0 = const()[name = tensor<string, []>("x_425_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_3365_to_fp16 = const()[name = tensor<string, []>("op_3365_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(958870848)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_154 = transpose(perm = q_with_bias_v_39_perm_0, x = var_3363_cast_fp16)[name = tensor<string, []>("transpose_154")];
+            tensor<fp16, [1, 8, 126, 251]> x_425_cast_fp16 = matmul(transpose_x = x_425_transpose_x_0, transpose_y = x_425_transpose_y_0, x = transpose_154, y = var_3365_to_fp16)[name = tensor<string, []>("x_425_cast_fp16")];
+            tensor<int32, [8]> x_427_pad_0 = const()[name = tensor<string, []>("x_427_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_427_mode_0 = const()[name = tensor<string, []>("x_427_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_203_to_fp16 = const()[name = tensor<string, []>("const_203_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_427_cast_fp16 = pad(constant_val = const_203_to_fp16, mode = x_427_mode_0, pad = x_427_pad_0, x = x_425_cast_fp16)[name = tensor<string, []>("x_427_cast_fp16")];
+            tensor<int32, [4]> var_3373 = const()[name = tensor<string, []>("op_3373"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_429_cast_fp16 = reshape(shape = var_3373, x = x_427_cast_fp16)[name = tensor<string, []>("x_429_cast_fp16")];
+            tensor<int32, [4]> var_3377_begin_0 = const()[name = tensor<string, []>("op_3377_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3377_end_0 = const()[name = tensor<string, []>("op_3377_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_3377_end_mask_0 = const()[name = tensor<string, []>("op_3377_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_3377_cast_fp16 = slice_by_index(begin = var_3377_begin_0, end = var_3377_end_0, end_mask = var_3377_end_mask_0, x = x_429_cast_fp16)[name = tensor<string, []>("op_3377_cast_fp16")];
+            tensor<int32, [4]> var_3378 = const()[name = tensor<string, []>("op_3378"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_77_cast_fp16 = reshape(shape = var_3378, x = var_3377_cast_fp16)[name = tensor<string, []>("matrix_bd_77_cast_fp16")];
+            tensor<bool, []> matrix_ac_39_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_39_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_39_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_39_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_110_perm_0 = const()[name = tensor<string, []>("transpose_110_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_111_perm_0 = const()[name = tensor<string, []>("transpose_111_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_152 = transpose(perm = transpose_111_perm_0, x = k_77_cast_fp16)[name = tensor<string, []>("transpose_152")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_153 = transpose(perm = transpose_110_perm_0, x = var_3361_cast_fp16)[name = tensor<string, []>("transpose_153")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_39_cast_fp16 = matmul(transpose_x = matrix_ac_39_transpose_x_0, transpose_y = matrix_ac_39_transpose_y_0, x = transpose_153, y = transpose_152)[name = tensor<string, []>("matrix_ac_39_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_79_begin_0 = const()[name = tensor<string, []>("matrix_bd_79_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_79_end_0 = const()[name = tensor<string, []>("matrix_bd_79_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_79_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_79_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_79_cast_fp16 = slice_by_index(begin = matrix_bd_79_begin_0, end = matrix_bd_79_end_0, end_mask = matrix_bd_79_end_mask_0, x = matrix_bd_77_cast_fp16)[name = tensor<string, []>("matrix_bd_79_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3387_cast_fp16 = add(x = matrix_ac_39_cast_fp16, y = matrix_bd_79_cast_fp16)[name = tensor<string, []>("op_3387_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_77_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_77_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_77_cast_fp16 = mul(x = var_3387_cast_fp16, y = _inversed_scores_77_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_77_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_79_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_77_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_79_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3393_cast_fp16 = softmax(axis = var_25, x = scores_79_cast_fp16)[name = tensor<string, []>("op_3393_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_1021_cast_fp16 = select(a = var_6_to_fp16, b = var_3393_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_1021_cast_fp16")];
+            tensor<bool, []> x_431_transpose_x_0 = const()[name = tensor<string, []>("x_431_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_431_transpose_y_0 = const()[name = tensor<string, []>("x_431_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_155 = transpose(perm = value_39_perm_0, x = v_39_cast_fp16)[name = tensor<string, []>("transpose_155")];
+            tensor<fp16, [1, 8, 126, 128]> x_431_cast_fp16 = matmul(transpose_x = x_431_transpose_x_0, transpose_y = x_431_transpose_y_0, x = input_1021_cast_fp16, y = transpose_155)[name = tensor<string, []>("x_431_cast_fp16")];
+            tensor<int32, [4]> var_3397_perm_0 = const()[name = tensor<string, []>("op_3397_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3398 = const()[name = tensor<string, []>("op_3398"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_151 = transpose(perm = var_3397_perm_0, x = x_431_cast_fp16)[name = tensor<string, []>("transpose_151")];
+            tensor<fp16, [1, 126, 1024]> input_1023_cast_fp16 = reshape(shape = var_3398, x = transpose_151)[name = tensor<string, []>("input_1023_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_19_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_19_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(959384960)))];
+            tensor<fp16, [1, 126, 1024]> linear_178_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_19_self_attn_linear_out_weight_to_fp16, x = input_1023_cast_fp16)[name = tensor<string, []>("linear_178_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1027_cast_fp16 = add(x = input_1019_cast_fp16, y = linear_178_cast_fp16)[name = tensor<string, []>("input_1027_cast_fp16")];
+            tensor<int32, [1]> x_435_axes_0 = const()[name = tensor<string, []>("x_435_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_19_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_19_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(961482176)))];
+            tensor<fp16, [1024]> model_layers_19_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_19_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(961484288)))];
+            tensor<fp16, [1, 126, 1024]> x_435_cast_fp16 = layer_norm(axes = x_435_axes_0, beta = model_layers_19_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_19_norm_conv_weight_to_fp16, x = input_1027_cast_fp16)[name = tensor<string, []>("x_435_cast_fp16")];
+            tensor<int32, [3]> input_1029_perm_0 = const()[name = tensor<string, []>("input_1029_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_3414 = const()[name = tensor<string, []>("op_3414"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_3416 = const()[name = tensor<string, []>("op_3416"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_1031_pad_type_0 = const()[name = tensor<string, []>("input_1031_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_1031_pad_0 = const()[name = tensor<string, []>("input_1031_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_19_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_19_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(961486400)))];
+            tensor<fp16, [1, 1024, 126]> transpose_150 = transpose(perm = input_1029_perm_0, x = x_435_cast_fp16)[name = tensor<string, []>("transpose_150")];
+            tensor<fp16, [1, 2048, 126]> input_1031_cast_fp16 = conv(dilations = var_3416, groups = var_27, pad = input_1031_pad_0, pad_type = input_1031_pad_type_0, strides = var_3414, weight = model_layers_19_conv_pointwise_conv1_weight_to_fp16, x = transpose_150)[name = tensor<string, []>("input_1031_cast_fp16")];
+            tensor<int32, []> x_437_split_num_splits_0 = const()[name = tensor<string, []>("x_437_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_437_split_axis_0 = const()[name = tensor<string, []>("x_437_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_437_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_437_split_cast_fp16_1 = split(axis = x_437_split_axis_0, num_splits = x_437_split_num_splits_0, x = input_1031_cast_fp16)[name = tensor<string, []>("x_437_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_437_split_1_sigmoid_cast_fp16 = sigmoid(x = x_437_split_cast_fp16_1)[name = tensor<string, []>("x_437_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_437_cast_fp16 = mul(x = x_437_split_cast_fp16_0, y = x_437_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_437_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_1033_cast_fp16 = select(a = var_6_to_fp16, b = x_437_cast_fp16, cond = var_323)[name = tensor<string, []>("input_1033_cast_fp16")];
+            tensor<int32, [6]> input_1035_pad_0 = const()[name = tensor<string, []>("input_1035_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_1035_mode_0 = const()[name = tensor<string, []>("input_1035_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_206_to_fp16 = const()[name = tensor<string, []>("const_206_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_1035_cast_fp16 = pad(constant_val = const_206_to_fp16, mode = input_1035_mode_0, pad = input_1035_pad_0, x = input_1033_cast_fp16)[name = tensor<string, []>("input_1035_cast_fp16")];
+            tensor<int32, [1]> var_3425 = const()[name = tensor<string, []>("op_3425"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_3427 = const()[name = tensor<string, []>("op_3427"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_1037_pad_type_0 = const()[name = tensor<string, []>("input_1037_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_1037_pad_0 = const()[name = tensor<string, []>("input_1037_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_1039_weight_0_to_fp16 = const()[name = tensor<string, []>("input_1039_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(965680768)))];
+            tensor<fp16, [1024]> input_1039_bias_0_to_fp16 = const()[name = tensor<string, []>("input_1039_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(965699264)))];
+            tensor<fp16, [1, 1024, 126]> input_1039_cast_fp16 = conv(bias = input_1039_bias_0_to_fp16, dilations = var_3427, groups = var_5, pad = input_1037_pad_0, pad_type = input_1037_pad_type_0, strides = var_3425, weight = input_1039_weight_0_to_fp16, x = input_1035_cast_fp16)[name = tensor<string, []>("input_1039_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_1041_cast_fp16 = silu(x = input_1039_cast_fp16)[name = tensor<string, []>("input_1041_cast_fp16")];
+            tensor<int32, [1]> var_3437 = const()[name = tensor<string, []>("op_3437"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_3439 = const()[name = tensor<string, []>("op_3439"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_439_pad_type_0 = const()[name = tensor<string, []>("x_439_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_439_pad_0 = const()[name = tensor<string, []>("x_439_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_19_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_19_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(965701376)))];
+            tensor<fp16, [1, 1024, 126]> x_439_cast_fp16 = conv(dilations = var_3439, groups = var_27, pad = x_439_pad_0, pad_type = x_439_pad_type_0, strides = var_3437, weight = model_layers_19_conv_pointwise_conv2_weight_to_fp16, x = input_1041_cast_fp16)[name = tensor<string, []>("x_439_cast_fp16")];
+            tensor<int32, [3]> input_1043_perm_0 = const()[name = tensor<string, []>("input_1043_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_149 = transpose(perm = input_1043_perm_0, x = x_439_cast_fp16)[name = tensor<string, []>("transpose_149")];
+            tensor<fp16, [1, 126, 1024]> input_1045_cast_fp16 = add(x = input_1027_cast_fp16, y = transpose_149)[name = tensor<string, []>("input_1045_cast_fp16")];
+            tensor<int32, [1]> input_1047_axes_0 = const()[name = tensor<string, []>("input_1047_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_19_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_19_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(967798592)))];
+            tensor<fp16, [1024]> model_layers_19_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_19_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(967800704)))];
+            tensor<fp16, [1, 126, 1024]> input_1047_cast_fp16 = layer_norm(axes = input_1047_axes_0, beta = model_layers_19_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_19_norm_feed_forward2_weight_to_fp16, x = input_1045_cast_fp16)[name = tensor<string, []>("input_1047_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_19_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_19_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(967802816)))];
+            tensor<fp16, [1, 126, 4096]> linear_179_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_19_feed_forward2_linear1_weight_to_fp16, x = input_1047_cast_fp16)[name = tensor<string, []>("linear_179_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_1051_cast_fp16 = silu(x = linear_179_cast_fp16)[name = tensor<string, []>("input_1051_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_19_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_19_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(976191488)))];
+            tensor<fp16, [1, 126, 1024]> linear_180_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_19_feed_forward2_linear2_weight_to_fp16, x = input_1051_cast_fp16)[name = tensor<string, []>("linear_180_cast_fp16")];
+            tensor<fp16, []> var_3458_to_fp16 = const()[name = tensor<string, []>("op_3458_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3459_cast_fp16 = mul(x = linear_180_cast_fp16, y = var_3458_to_fp16)[name = tensor<string, []>("op_3459_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1057_cast_fp16 = add(x = input_1045_cast_fp16, y = var_3459_cast_fp16)[name = tensor<string, []>("input_1057_cast_fp16")];
+            tensor<int32, [1]> input_1059_axes_0 = const()[name = tensor<string, []>("input_1059_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_19_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_19_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(984580160)))];
+            tensor<fp16, [1024]> model_layers_19_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_19_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(984582272)))];
+            tensor<fp16, [1, 126, 1024]> input_1059_cast_fp16 = layer_norm(axes = input_1059_axes_0, beta = model_layers_19_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_19_norm_out_weight_to_fp16, x = input_1057_cast_fp16)[name = tensor<string, []>("input_1059_cast_fp16")];
+            tensor<int32, [1]> input_1061_axes_0 = const()[name = tensor<string, []>("input_1061_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_20_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_20_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(984584384)))];
+            tensor<fp16, [1024]> model_layers_20_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_20_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(984586496)))];
+            tensor<fp16, [1, 126, 1024]> input_1061_cast_fp16 = layer_norm(axes = input_1061_axes_0, beta = model_layers_20_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_20_norm_feed_forward1_weight_to_fp16, x = input_1059_cast_fp16)[name = tensor<string, []>("input_1061_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_20_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_20_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(984588608)))];
+            tensor<fp16, [1, 126, 4096]> linear_181_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_20_feed_forward1_linear1_weight_to_fp16, x = input_1061_cast_fp16)[name = tensor<string, []>("linear_181_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_1065_cast_fp16 = silu(x = linear_181_cast_fp16)[name = tensor<string, []>("input_1065_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_20_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_20_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(992977280)))];
+            tensor<fp16, [1, 126, 1024]> linear_182_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_20_feed_forward1_linear2_weight_to_fp16, x = input_1065_cast_fp16)[name = tensor<string, []>("linear_182_cast_fp16")];
+            tensor<fp16, []> var_3487_to_fp16 = const()[name = tensor<string, []>("op_3487_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3488_cast_fp16 = mul(x = linear_182_cast_fp16, y = var_3487_to_fp16)[name = tensor<string, []>("op_3488_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1071_cast_fp16 = add(x = input_1059_cast_fp16, y = var_3488_cast_fp16)[name = tensor<string, []>("input_1071_cast_fp16")];
+            tensor<int32, [1]> query_41_axes_0 = const()[name = tensor<string, []>("query_41_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_20_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_20_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1001365952)))];
+            tensor<fp16, [1024]> model_layers_20_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_20_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1001368064)))];
+            tensor<fp16, [1, 126, 1024]> query_41_cast_fp16 = layer_norm(axes = query_41_axes_0, beta = model_layers_20_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_20_norm_self_att_weight_to_fp16, x = input_1071_cast_fp16)[name = tensor<string, []>("query_41_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_20_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_20_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1001370176)))];
+            tensor<fp16, [1, 126, 1024]> linear_183_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_20_self_attn_linear_q_weight_to_fp16, x = query_41_cast_fp16)[name = tensor<string, []>("linear_183_cast_fp16")];
+            tensor<int32, [4]> var_3504 = const()[name = tensor<string, []>("op_3504"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_121_cast_fp16 = reshape(shape = var_3504, x = linear_183_cast_fp16)[name = tensor<string, []>("q_121_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_20_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_20_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1003467392)))];
+            tensor<fp16, [1, 126, 1024]> linear_184_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_20_self_attn_linear_k_weight_to_fp16, x = query_41_cast_fp16)[name = tensor<string, []>("linear_184_cast_fp16")];
+            tensor<int32, [4]> var_3508 = const()[name = tensor<string, []>("op_3508"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_81_cast_fp16 = reshape(shape = var_3508, x = linear_184_cast_fp16)[name = tensor<string, []>("k_81_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_20_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_20_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1005564608)))];
+            tensor<fp16, [1, 126, 1024]> linear_185_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_20_self_attn_linear_v_weight_to_fp16, x = query_41_cast_fp16)[name = tensor<string, []>("linear_185_cast_fp16")];
+            tensor<int32, [4]> var_3512 = const()[name = tensor<string, []>("op_3512"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_41_cast_fp16 = reshape(shape = var_3512, x = linear_185_cast_fp16)[name = tensor<string, []>("v_41_cast_fp16")];
+            tensor<int32, [4]> value_41_perm_0 = const()[name = tensor<string, []>("value_41_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_20_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_20_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1007661824)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3524_cast_fp16 = add(x = q_121_cast_fp16, y = model_layers_20_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_3524_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_20_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_20_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1007663936)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3526_cast_fp16 = add(x = q_121_cast_fp16, y = model_layers_20_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_3526_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_41_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_41_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_447_transpose_x_0 = const()[name = tensor<string, []>("x_447_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_447_transpose_y_0 = const()[name = tensor<string, []>("x_447_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_3528_to_fp16 = const()[name = tensor<string, []>("op_3528_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1007666048)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_147 = transpose(perm = q_with_bias_v_41_perm_0, x = var_3526_cast_fp16)[name = tensor<string, []>("transpose_147")];
+            tensor<fp16, [1, 8, 126, 251]> x_447_cast_fp16 = matmul(transpose_x = x_447_transpose_x_0, transpose_y = x_447_transpose_y_0, x = transpose_147, y = var_3528_to_fp16)[name = tensor<string, []>("x_447_cast_fp16")];
+            tensor<int32, [8]> x_449_pad_0 = const()[name = tensor<string, []>("x_449_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_449_mode_0 = const()[name = tensor<string, []>("x_449_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_213_to_fp16 = const()[name = tensor<string, []>("const_213_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_449_cast_fp16 = pad(constant_val = const_213_to_fp16, mode = x_449_mode_0, pad = x_449_pad_0, x = x_447_cast_fp16)[name = tensor<string, []>("x_449_cast_fp16")];
+            tensor<int32, [4]> var_3536 = const()[name = tensor<string, []>("op_3536"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_451_cast_fp16 = reshape(shape = var_3536, x = x_449_cast_fp16)[name = tensor<string, []>("x_451_cast_fp16")];
+            tensor<int32, [4]> var_3540_begin_0 = const()[name = tensor<string, []>("op_3540_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3540_end_0 = const()[name = tensor<string, []>("op_3540_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_3540_end_mask_0 = const()[name = tensor<string, []>("op_3540_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_3540_cast_fp16 = slice_by_index(begin = var_3540_begin_0, end = var_3540_end_0, end_mask = var_3540_end_mask_0, x = x_451_cast_fp16)[name = tensor<string, []>("op_3540_cast_fp16")];
+            tensor<int32, [4]> var_3541 = const()[name = tensor<string, []>("op_3541"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_81_cast_fp16 = reshape(shape = var_3541, x = var_3540_cast_fp16)[name = tensor<string, []>("matrix_bd_81_cast_fp16")];
+            tensor<bool, []> matrix_ac_41_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_41_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_41_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_41_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_112_perm_0 = const()[name = tensor<string, []>("transpose_112_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_113_perm_0 = const()[name = tensor<string, []>("transpose_113_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_145 = transpose(perm = transpose_113_perm_0, x = k_81_cast_fp16)[name = tensor<string, []>("transpose_145")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_146 = transpose(perm = transpose_112_perm_0, x = var_3524_cast_fp16)[name = tensor<string, []>("transpose_146")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_41_cast_fp16 = matmul(transpose_x = matrix_ac_41_transpose_x_0, transpose_y = matrix_ac_41_transpose_y_0, x = transpose_146, y = transpose_145)[name = tensor<string, []>("matrix_ac_41_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_83_begin_0 = const()[name = tensor<string, []>("matrix_bd_83_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_83_end_0 = const()[name = tensor<string, []>("matrix_bd_83_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_83_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_83_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_83_cast_fp16 = slice_by_index(begin = matrix_bd_83_begin_0, end = matrix_bd_83_end_0, end_mask = matrix_bd_83_end_mask_0, x = matrix_bd_81_cast_fp16)[name = tensor<string, []>("matrix_bd_83_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3550_cast_fp16 = add(x = matrix_ac_41_cast_fp16, y = matrix_bd_83_cast_fp16)[name = tensor<string, []>("op_3550_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_81_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_81_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_81_cast_fp16 = mul(x = var_3550_cast_fp16, y = _inversed_scores_81_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_81_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_83_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_81_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_83_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3556_cast_fp16 = softmax(axis = var_25, x = scores_83_cast_fp16)[name = tensor<string, []>("op_3556_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_1073_cast_fp16 = select(a = var_6_to_fp16, b = var_3556_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_1073_cast_fp16")];
+            tensor<bool, []> x_453_transpose_x_0 = const()[name = tensor<string, []>("x_453_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_453_transpose_y_0 = const()[name = tensor<string, []>("x_453_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_148 = transpose(perm = value_41_perm_0, x = v_41_cast_fp16)[name = tensor<string, []>("transpose_148")];
+            tensor<fp16, [1, 8, 126, 128]> x_453_cast_fp16 = matmul(transpose_x = x_453_transpose_x_0, transpose_y = x_453_transpose_y_0, x = input_1073_cast_fp16, y = transpose_148)[name = tensor<string, []>("x_453_cast_fp16")];
+            tensor<int32, [4]> var_3560_perm_0 = const()[name = tensor<string, []>("op_3560_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3561 = const()[name = tensor<string, []>("op_3561"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_144 = transpose(perm = var_3560_perm_0, x = x_453_cast_fp16)[name = tensor<string, []>("transpose_144")];
+            tensor<fp16, [1, 126, 1024]> input_1075_cast_fp16 = reshape(shape = var_3561, x = transpose_144)[name = tensor<string, []>("input_1075_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_20_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_20_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1008180160)))];
+            tensor<fp16, [1, 126, 1024]> linear_187_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_20_self_attn_linear_out_weight_to_fp16, x = input_1075_cast_fp16)[name = tensor<string, []>("linear_187_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1079_cast_fp16 = add(x = input_1071_cast_fp16, y = linear_187_cast_fp16)[name = tensor<string, []>("input_1079_cast_fp16")];
+            tensor<int32, [1]> x_457_axes_0 = const()[name = tensor<string, []>("x_457_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_20_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_20_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1010277376)))];
+            tensor<fp16, [1024]> model_layers_20_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_20_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1010279488)))];
+            tensor<fp16, [1, 126, 1024]> x_457_cast_fp16 = layer_norm(axes = x_457_axes_0, beta = model_layers_20_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_20_norm_conv_weight_to_fp16, x = input_1079_cast_fp16)[name = tensor<string, []>("x_457_cast_fp16")];
+            tensor<int32, [3]> input_1081_perm_0 = const()[name = tensor<string, []>("input_1081_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_3577 = const()[name = tensor<string, []>("op_3577"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_3579 = const()[name = tensor<string, []>("op_3579"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_1083_pad_type_0 = const()[name = tensor<string, []>("input_1083_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_1083_pad_0 = const()[name = tensor<string, []>("input_1083_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_20_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_20_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1010281600)))];
+            tensor<fp16, [1, 1024, 126]> transpose_143 = transpose(perm = input_1081_perm_0, x = x_457_cast_fp16)[name = tensor<string, []>("transpose_143")];
+            tensor<fp16, [1, 2048, 126]> input_1083_cast_fp16 = conv(dilations = var_3579, groups = var_27, pad = input_1083_pad_0, pad_type = input_1083_pad_type_0, strides = var_3577, weight = model_layers_20_conv_pointwise_conv1_weight_to_fp16, x = transpose_143)[name = tensor<string, []>("input_1083_cast_fp16")];
+            tensor<int32, []> x_459_split_num_splits_0 = const()[name = tensor<string, []>("x_459_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_459_split_axis_0 = const()[name = tensor<string, []>("x_459_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_459_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_459_split_cast_fp16_1 = split(axis = x_459_split_axis_0, num_splits = x_459_split_num_splits_0, x = input_1083_cast_fp16)[name = tensor<string, []>("x_459_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_459_split_1_sigmoid_cast_fp16 = sigmoid(x = x_459_split_cast_fp16_1)[name = tensor<string, []>("x_459_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_459_cast_fp16 = mul(x = x_459_split_cast_fp16_0, y = x_459_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_459_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_1085_cast_fp16 = select(a = var_6_to_fp16, b = x_459_cast_fp16, cond = var_323)[name = tensor<string, []>("input_1085_cast_fp16")];
+            tensor<int32, [6]> input_1087_pad_0 = const()[name = tensor<string, []>("input_1087_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_1087_mode_0 = const()[name = tensor<string, []>("input_1087_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_216_to_fp16 = const()[name = tensor<string, []>("const_216_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_1087_cast_fp16 = pad(constant_val = const_216_to_fp16, mode = input_1087_mode_0, pad = input_1087_pad_0, x = input_1085_cast_fp16)[name = tensor<string, []>("input_1087_cast_fp16")];
+            tensor<int32, [1]> var_3588 = const()[name = tensor<string, []>("op_3588"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_3590 = const()[name = tensor<string, []>("op_3590"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_1089_pad_type_0 = const()[name = tensor<string, []>("input_1089_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_1089_pad_0 = const()[name = tensor<string, []>("input_1089_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_1091_weight_0_to_fp16 = const()[name = tensor<string, []>("input_1091_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1014475968)))];
+            tensor<fp16, [1024]> input_1091_bias_0_to_fp16 = const()[name = tensor<string, []>("input_1091_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1014494464)))];
+            tensor<fp16, [1, 1024, 126]> input_1091_cast_fp16 = conv(bias = input_1091_bias_0_to_fp16, dilations = var_3590, groups = var_5, pad = input_1089_pad_0, pad_type = input_1089_pad_type_0, strides = var_3588, weight = input_1091_weight_0_to_fp16, x = input_1087_cast_fp16)[name = tensor<string, []>("input_1091_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_1093_cast_fp16 = silu(x = input_1091_cast_fp16)[name = tensor<string, []>("input_1093_cast_fp16")];
+            tensor<int32, [1]> var_3600 = const()[name = tensor<string, []>("op_3600"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_3602 = const()[name = tensor<string, []>("op_3602"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_461_pad_type_0 = const()[name = tensor<string, []>("x_461_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_461_pad_0 = const()[name = tensor<string, []>("x_461_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_20_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_20_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1014496576)))];
+            tensor<fp16, [1, 1024, 126]> x_461_cast_fp16 = conv(dilations = var_3602, groups = var_27, pad = x_461_pad_0, pad_type = x_461_pad_type_0, strides = var_3600, weight = model_layers_20_conv_pointwise_conv2_weight_to_fp16, x = input_1093_cast_fp16)[name = tensor<string, []>("x_461_cast_fp16")];
+            tensor<int32, [3]> input_1095_perm_0 = const()[name = tensor<string, []>("input_1095_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_142 = transpose(perm = input_1095_perm_0, x = x_461_cast_fp16)[name = tensor<string, []>("transpose_142")];
+            tensor<fp16, [1, 126, 1024]> input_1097_cast_fp16 = add(x = input_1079_cast_fp16, y = transpose_142)[name = tensor<string, []>("input_1097_cast_fp16")];
+            tensor<int32, [1]> input_1099_axes_0 = const()[name = tensor<string, []>("input_1099_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_20_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_20_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1016593792)))];
+            tensor<fp16, [1024]> model_layers_20_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_20_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1016595904)))];
+            tensor<fp16, [1, 126, 1024]> input_1099_cast_fp16 = layer_norm(axes = input_1099_axes_0, beta = model_layers_20_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_20_norm_feed_forward2_weight_to_fp16, x = input_1097_cast_fp16)[name = tensor<string, []>("input_1099_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_20_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_20_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1016598016)))];
+            tensor<fp16, [1, 126, 4096]> linear_188_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_20_feed_forward2_linear1_weight_to_fp16, x = input_1099_cast_fp16)[name = tensor<string, []>("linear_188_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_1103_cast_fp16 = silu(x = linear_188_cast_fp16)[name = tensor<string, []>("input_1103_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_20_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_20_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1024986688)))];
+            tensor<fp16, [1, 126, 1024]> linear_189_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_20_feed_forward2_linear2_weight_to_fp16, x = input_1103_cast_fp16)[name = tensor<string, []>("linear_189_cast_fp16")];
+            tensor<fp16, []> var_3621_to_fp16 = const()[name = tensor<string, []>("op_3621_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3622_cast_fp16 = mul(x = linear_189_cast_fp16, y = var_3621_to_fp16)[name = tensor<string, []>("op_3622_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1109_cast_fp16 = add(x = input_1097_cast_fp16, y = var_3622_cast_fp16)[name = tensor<string, []>("input_1109_cast_fp16")];
+            tensor<int32, [1]> input_1111_axes_0 = const()[name = tensor<string, []>("input_1111_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_20_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_20_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1033375360)))];
+            tensor<fp16, [1024]> model_layers_20_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_20_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1033377472)))];
+            tensor<fp16, [1, 126, 1024]> input_1111_cast_fp16 = layer_norm(axes = input_1111_axes_0, beta = model_layers_20_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_20_norm_out_weight_to_fp16, x = input_1109_cast_fp16)[name = tensor<string, []>("input_1111_cast_fp16")];
+            tensor<int32, [1]> input_1113_axes_0 = const()[name = tensor<string, []>("input_1113_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_21_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_21_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1033379584)))];
+            tensor<fp16, [1024]> model_layers_21_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_21_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1033381696)))];
+            tensor<fp16, [1, 126, 1024]> input_1113_cast_fp16 = layer_norm(axes = input_1113_axes_0, beta = model_layers_21_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_21_norm_feed_forward1_weight_to_fp16, x = input_1111_cast_fp16)[name = tensor<string, []>("input_1113_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_21_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_21_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1033383808)))];
+            tensor<fp16, [1, 126, 4096]> linear_190_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_21_feed_forward1_linear1_weight_to_fp16, x = input_1113_cast_fp16)[name = tensor<string, []>("linear_190_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_1117_cast_fp16 = silu(x = linear_190_cast_fp16)[name = tensor<string, []>("input_1117_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_21_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_21_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1041772480)))];
+            tensor<fp16, [1, 126, 1024]> linear_191_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_21_feed_forward1_linear2_weight_to_fp16, x = input_1117_cast_fp16)[name = tensor<string, []>("linear_191_cast_fp16")];
+            tensor<fp16, []> var_3650_to_fp16 = const()[name = tensor<string, []>("op_3650_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3651_cast_fp16 = mul(x = linear_191_cast_fp16, y = var_3650_to_fp16)[name = tensor<string, []>("op_3651_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1123_cast_fp16 = add(x = input_1111_cast_fp16, y = var_3651_cast_fp16)[name = tensor<string, []>("input_1123_cast_fp16")];
+            tensor<int32, [1]> query_43_axes_0 = const()[name = tensor<string, []>("query_43_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_21_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_21_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1050161152)))];
+            tensor<fp16, [1024]> model_layers_21_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_21_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1050163264)))];
+            tensor<fp16, [1, 126, 1024]> query_43_cast_fp16 = layer_norm(axes = query_43_axes_0, beta = model_layers_21_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_21_norm_self_att_weight_to_fp16, x = input_1123_cast_fp16)[name = tensor<string, []>("query_43_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_21_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_21_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1050165376)))];
+            tensor<fp16, [1, 126, 1024]> linear_192_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_21_self_attn_linear_q_weight_to_fp16, x = query_43_cast_fp16)[name = tensor<string, []>("linear_192_cast_fp16")];
+            tensor<int32, [4]> var_3667 = const()[name = tensor<string, []>("op_3667"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_127_cast_fp16 = reshape(shape = var_3667, x = linear_192_cast_fp16)[name = tensor<string, []>("q_127_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_21_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_21_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1052262592)))];
+            tensor<fp16, [1, 126, 1024]> linear_193_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_21_self_attn_linear_k_weight_to_fp16, x = query_43_cast_fp16)[name = tensor<string, []>("linear_193_cast_fp16")];
+            tensor<int32, [4]> var_3671 = const()[name = tensor<string, []>("op_3671"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_85_cast_fp16 = reshape(shape = var_3671, x = linear_193_cast_fp16)[name = tensor<string, []>("k_85_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_21_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_21_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1054359808)))];
+            tensor<fp16, [1, 126, 1024]> linear_194_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_21_self_attn_linear_v_weight_to_fp16, x = query_43_cast_fp16)[name = tensor<string, []>("linear_194_cast_fp16")];
+            tensor<int32, [4]> var_3675 = const()[name = tensor<string, []>("op_3675"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_43_cast_fp16 = reshape(shape = var_3675, x = linear_194_cast_fp16)[name = tensor<string, []>("v_43_cast_fp16")];
+            tensor<int32, [4]> value_43_perm_0 = const()[name = tensor<string, []>("value_43_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_21_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_21_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1056457024)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3687_cast_fp16 = add(x = q_127_cast_fp16, y = model_layers_21_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_3687_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_21_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_21_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1056459136)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3689_cast_fp16 = add(x = q_127_cast_fp16, y = model_layers_21_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_3689_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_43_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_43_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_469_transpose_x_0 = const()[name = tensor<string, []>("x_469_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_469_transpose_y_0 = const()[name = tensor<string, []>("x_469_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_3691_to_fp16 = const()[name = tensor<string, []>("op_3691_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1056461248)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_140 = transpose(perm = q_with_bias_v_43_perm_0, x = var_3689_cast_fp16)[name = tensor<string, []>("transpose_140")];
+            tensor<fp16, [1, 8, 126, 251]> x_469_cast_fp16 = matmul(transpose_x = x_469_transpose_x_0, transpose_y = x_469_transpose_y_0, x = transpose_140, y = var_3691_to_fp16)[name = tensor<string, []>("x_469_cast_fp16")];
+            tensor<int32, [8]> x_471_pad_0 = const()[name = tensor<string, []>("x_471_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_471_mode_0 = const()[name = tensor<string, []>("x_471_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_223_to_fp16 = const()[name = tensor<string, []>("const_223_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_471_cast_fp16 = pad(constant_val = const_223_to_fp16, mode = x_471_mode_0, pad = x_471_pad_0, x = x_469_cast_fp16)[name = tensor<string, []>("x_471_cast_fp16")];
+            tensor<int32, [4]> var_3699 = const()[name = tensor<string, []>("op_3699"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_473_cast_fp16 = reshape(shape = var_3699, x = x_471_cast_fp16)[name = tensor<string, []>("x_473_cast_fp16")];
+            tensor<int32, [4]> var_3703_begin_0 = const()[name = tensor<string, []>("op_3703_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3703_end_0 = const()[name = tensor<string, []>("op_3703_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_3703_end_mask_0 = const()[name = tensor<string, []>("op_3703_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_3703_cast_fp16 = slice_by_index(begin = var_3703_begin_0, end = var_3703_end_0, end_mask = var_3703_end_mask_0, x = x_473_cast_fp16)[name = tensor<string, []>("op_3703_cast_fp16")];
+            tensor<int32, [4]> var_3704 = const()[name = tensor<string, []>("op_3704"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_85_cast_fp16 = reshape(shape = var_3704, x = var_3703_cast_fp16)[name = tensor<string, []>("matrix_bd_85_cast_fp16")];
+            tensor<bool, []> matrix_ac_43_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_43_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_43_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_43_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_114_perm_0 = const()[name = tensor<string, []>("transpose_114_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_115_perm_0 = const()[name = tensor<string, []>("transpose_115_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_138 = transpose(perm = transpose_115_perm_0, x = k_85_cast_fp16)[name = tensor<string, []>("transpose_138")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_139 = transpose(perm = transpose_114_perm_0, x = var_3687_cast_fp16)[name = tensor<string, []>("transpose_139")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_43_cast_fp16 = matmul(transpose_x = matrix_ac_43_transpose_x_0, transpose_y = matrix_ac_43_transpose_y_0, x = transpose_139, y = transpose_138)[name = tensor<string, []>("matrix_ac_43_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_87_begin_0 = const()[name = tensor<string, []>("matrix_bd_87_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_87_end_0 = const()[name = tensor<string, []>("matrix_bd_87_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_87_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_87_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_87_cast_fp16 = slice_by_index(begin = matrix_bd_87_begin_0, end = matrix_bd_87_end_0, end_mask = matrix_bd_87_end_mask_0, x = matrix_bd_85_cast_fp16)[name = tensor<string, []>("matrix_bd_87_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3713_cast_fp16 = add(x = matrix_ac_43_cast_fp16, y = matrix_bd_87_cast_fp16)[name = tensor<string, []>("op_3713_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_85_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_85_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_85_cast_fp16 = mul(x = var_3713_cast_fp16, y = _inversed_scores_85_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_85_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_87_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_85_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_87_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3719_cast_fp16 = softmax(axis = var_25, x = scores_87_cast_fp16)[name = tensor<string, []>("op_3719_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_1125_cast_fp16 = select(a = var_6_to_fp16, b = var_3719_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_1125_cast_fp16")];
+            tensor<bool, []> x_475_transpose_x_0 = const()[name = tensor<string, []>("x_475_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_475_transpose_y_0 = const()[name = tensor<string, []>("x_475_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_141 = transpose(perm = value_43_perm_0, x = v_43_cast_fp16)[name = tensor<string, []>("transpose_141")];
+            tensor<fp16, [1, 8, 126, 128]> x_475_cast_fp16 = matmul(transpose_x = x_475_transpose_x_0, transpose_y = x_475_transpose_y_0, x = input_1125_cast_fp16, y = transpose_141)[name = tensor<string, []>("x_475_cast_fp16")];
+            tensor<int32, [4]> var_3723_perm_0 = const()[name = tensor<string, []>("op_3723_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3724 = const()[name = tensor<string, []>("op_3724"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_137 = transpose(perm = var_3723_perm_0, x = x_475_cast_fp16)[name = tensor<string, []>("transpose_137")];
+            tensor<fp16, [1, 126, 1024]> input_1127_cast_fp16 = reshape(shape = var_3724, x = transpose_137)[name = tensor<string, []>("input_1127_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_21_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_21_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1056975360)))];
+            tensor<fp16, [1, 126, 1024]> linear_196_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_21_self_attn_linear_out_weight_to_fp16, x = input_1127_cast_fp16)[name = tensor<string, []>("linear_196_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1131_cast_fp16 = add(x = input_1123_cast_fp16, y = linear_196_cast_fp16)[name = tensor<string, []>("input_1131_cast_fp16")];
+            tensor<int32, [1]> x_479_axes_0 = const()[name = tensor<string, []>("x_479_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_21_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_21_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1059072576)))];
+            tensor<fp16, [1024]> model_layers_21_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_21_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1059074688)))];
+            tensor<fp16, [1, 126, 1024]> x_479_cast_fp16 = layer_norm(axes = x_479_axes_0, beta = model_layers_21_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_21_norm_conv_weight_to_fp16, x = input_1131_cast_fp16)[name = tensor<string, []>("x_479_cast_fp16")];
+            tensor<int32, [3]> input_1133_perm_0 = const()[name = tensor<string, []>("input_1133_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_3740 = const()[name = tensor<string, []>("op_3740"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_3742 = const()[name = tensor<string, []>("op_3742"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_1135_pad_type_0 = const()[name = tensor<string, []>("input_1135_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_1135_pad_0 = const()[name = tensor<string, []>("input_1135_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_21_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_21_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1059076800)))];
+            tensor<fp16, [1, 1024, 126]> transpose_136 = transpose(perm = input_1133_perm_0, x = x_479_cast_fp16)[name = tensor<string, []>("transpose_136")];
+            tensor<fp16, [1, 2048, 126]> input_1135_cast_fp16 = conv(dilations = var_3742, groups = var_27, pad = input_1135_pad_0, pad_type = input_1135_pad_type_0, strides = var_3740, weight = model_layers_21_conv_pointwise_conv1_weight_to_fp16, x = transpose_136)[name = tensor<string, []>("input_1135_cast_fp16")];
+            tensor<int32, []> x_481_split_num_splits_0 = const()[name = tensor<string, []>("x_481_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_481_split_axis_0 = const()[name = tensor<string, []>("x_481_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_481_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_481_split_cast_fp16_1 = split(axis = x_481_split_axis_0, num_splits = x_481_split_num_splits_0, x = input_1135_cast_fp16)[name = tensor<string, []>("x_481_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_481_split_1_sigmoid_cast_fp16 = sigmoid(x = x_481_split_cast_fp16_1)[name = tensor<string, []>("x_481_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_481_cast_fp16 = mul(x = x_481_split_cast_fp16_0, y = x_481_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_481_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_1137_cast_fp16 = select(a = var_6_to_fp16, b = x_481_cast_fp16, cond = var_323)[name = tensor<string, []>("input_1137_cast_fp16")];
+            tensor<int32, [6]> input_1139_pad_0 = const()[name = tensor<string, []>("input_1139_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_1139_mode_0 = const()[name = tensor<string, []>("input_1139_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_226_to_fp16 = const()[name = tensor<string, []>("const_226_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_1139_cast_fp16 = pad(constant_val = const_226_to_fp16, mode = input_1139_mode_0, pad = input_1139_pad_0, x = input_1137_cast_fp16)[name = tensor<string, []>("input_1139_cast_fp16")];
+            tensor<int32, [1]> var_3751 = const()[name = tensor<string, []>("op_3751"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_3753 = const()[name = tensor<string, []>("op_3753"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_1141_pad_type_0 = const()[name = tensor<string, []>("input_1141_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_1141_pad_0 = const()[name = tensor<string, []>("input_1141_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_1143_weight_0_to_fp16 = const()[name = tensor<string, []>("input_1143_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1063271168)))];
+            tensor<fp16, [1024]> input_1143_bias_0_to_fp16 = const()[name = tensor<string, []>("input_1143_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1063289664)))];
+            tensor<fp16, [1, 1024, 126]> input_1143_cast_fp16 = conv(bias = input_1143_bias_0_to_fp16, dilations = var_3753, groups = var_5, pad = input_1141_pad_0, pad_type = input_1141_pad_type_0, strides = var_3751, weight = input_1143_weight_0_to_fp16, x = input_1139_cast_fp16)[name = tensor<string, []>("input_1143_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_1145_cast_fp16 = silu(x = input_1143_cast_fp16)[name = tensor<string, []>("input_1145_cast_fp16")];
+            tensor<int32, [1]> var_3763 = const()[name = tensor<string, []>("op_3763"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_3765 = const()[name = tensor<string, []>("op_3765"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_483_pad_type_0 = const()[name = tensor<string, []>("x_483_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_483_pad_0 = const()[name = tensor<string, []>("x_483_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_21_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_21_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1063291776)))];
+            tensor<fp16, [1, 1024, 126]> x_483_cast_fp16 = conv(dilations = var_3765, groups = var_27, pad = x_483_pad_0, pad_type = x_483_pad_type_0, strides = var_3763, weight = model_layers_21_conv_pointwise_conv2_weight_to_fp16, x = input_1145_cast_fp16)[name = tensor<string, []>("x_483_cast_fp16")];
+            tensor<int32, [3]> input_1147_perm_0 = const()[name = tensor<string, []>("input_1147_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_135 = transpose(perm = input_1147_perm_0, x = x_483_cast_fp16)[name = tensor<string, []>("transpose_135")];
+            tensor<fp16, [1, 126, 1024]> input_1149_cast_fp16 = add(x = input_1131_cast_fp16, y = transpose_135)[name = tensor<string, []>("input_1149_cast_fp16")];
+            tensor<int32, [1]> input_1151_axes_0 = const()[name = tensor<string, []>("input_1151_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_21_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_21_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1065388992)))];
+            tensor<fp16, [1024]> model_layers_21_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_21_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1065391104)))];
+            tensor<fp16, [1, 126, 1024]> input_1151_cast_fp16 = layer_norm(axes = input_1151_axes_0, beta = model_layers_21_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_21_norm_feed_forward2_weight_to_fp16, x = input_1149_cast_fp16)[name = tensor<string, []>("input_1151_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_21_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_21_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1065393216)))];
+            tensor<fp16, [1, 126, 4096]> linear_197_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_21_feed_forward2_linear1_weight_to_fp16, x = input_1151_cast_fp16)[name = tensor<string, []>("linear_197_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_1155_cast_fp16 = silu(x = linear_197_cast_fp16)[name = tensor<string, []>("input_1155_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_21_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_21_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1073781888)))];
+            tensor<fp16, [1, 126, 1024]> linear_198_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_21_feed_forward2_linear2_weight_to_fp16, x = input_1155_cast_fp16)[name = tensor<string, []>("linear_198_cast_fp16")];
+            tensor<fp16, []> var_3784_to_fp16 = const()[name = tensor<string, []>("op_3784_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3785_cast_fp16 = mul(x = linear_198_cast_fp16, y = var_3784_to_fp16)[name = tensor<string, []>("op_3785_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1161_cast_fp16 = add(x = input_1149_cast_fp16, y = var_3785_cast_fp16)[name = tensor<string, []>("input_1161_cast_fp16")];
+            tensor<int32, [1]> input_1163_axes_0 = const()[name = tensor<string, []>("input_1163_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_21_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_21_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1082170560)))];
+            tensor<fp16, [1024]> model_layers_21_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_21_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1082172672)))];
+            tensor<fp16, [1, 126, 1024]> input_1163_cast_fp16 = layer_norm(axes = input_1163_axes_0, beta = model_layers_21_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_21_norm_out_weight_to_fp16, x = input_1161_cast_fp16)[name = tensor<string, []>("input_1163_cast_fp16")];
+            tensor<int32, [1]> input_1165_axes_0 = const()[name = tensor<string, []>("input_1165_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_22_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_22_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1082174784)))];
+            tensor<fp16, [1024]> model_layers_22_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_22_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1082176896)))];
+            tensor<fp16, [1, 126, 1024]> input_1165_cast_fp16 = layer_norm(axes = input_1165_axes_0, beta = model_layers_22_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_22_norm_feed_forward1_weight_to_fp16, x = input_1163_cast_fp16)[name = tensor<string, []>("input_1165_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_22_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_22_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1082179008)))];
+            tensor<fp16, [1, 126, 4096]> linear_199_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_22_feed_forward1_linear1_weight_to_fp16, x = input_1165_cast_fp16)[name = tensor<string, []>("linear_199_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_1169_cast_fp16 = silu(x = linear_199_cast_fp16)[name = tensor<string, []>("input_1169_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_22_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_22_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1090567680)))];
+            tensor<fp16, [1, 126, 1024]> linear_200_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_22_feed_forward1_linear2_weight_to_fp16, x = input_1169_cast_fp16)[name = tensor<string, []>("linear_200_cast_fp16")];
+            tensor<fp16, []> var_3813_to_fp16 = const()[name = tensor<string, []>("op_3813_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3814_cast_fp16 = mul(x = linear_200_cast_fp16, y = var_3813_to_fp16)[name = tensor<string, []>("op_3814_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1175_cast_fp16 = add(x = input_1163_cast_fp16, y = var_3814_cast_fp16)[name = tensor<string, []>("input_1175_cast_fp16")];
+            tensor<int32, [1]> query_45_axes_0 = const()[name = tensor<string, []>("query_45_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_22_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_22_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1098956352)))];
+            tensor<fp16, [1024]> model_layers_22_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_22_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1098958464)))];
+            tensor<fp16, [1, 126, 1024]> query_45_cast_fp16 = layer_norm(axes = query_45_axes_0, beta = model_layers_22_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_22_norm_self_att_weight_to_fp16, x = input_1175_cast_fp16)[name = tensor<string, []>("query_45_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_22_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_22_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1098960576)))];
+            tensor<fp16, [1, 126, 1024]> linear_201_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_22_self_attn_linear_q_weight_to_fp16, x = query_45_cast_fp16)[name = tensor<string, []>("linear_201_cast_fp16")];
+            tensor<int32, [4]> var_3830 = const()[name = tensor<string, []>("op_3830"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_133_cast_fp16 = reshape(shape = var_3830, x = linear_201_cast_fp16)[name = tensor<string, []>("q_133_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_22_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_22_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1101057792)))];
+            tensor<fp16, [1, 126, 1024]> linear_202_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_22_self_attn_linear_k_weight_to_fp16, x = query_45_cast_fp16)[name = tensor<string, []>("linear_202_cast_fp16")];
+            tensor<int32, [4]> var_3834 = const()[name = tensor<string, []>("op_3834"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_89_cast_fp16 = reshape(shape = var_3834, x = linear_202_cast_fp16)[name = tensor<string, []>("k_89_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_22_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_22_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1103155008)))];
+            tensor<fp16, [1, 126, 1024]> linear_203_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_22_self_attn_linear_v_weight_to_fp16, x = query_45_cast_fp16)[name = tensor<string, []>("linear_203_cast_fp16")];
+            tensor<int32, [4]> var_3838 = const()[name = tensor<string, []>("op_3838"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_45_cast_fp16 = reshape(shape = var_3838, x = linear_203_cast_fp16)[name = tensor<string, []>("v_45_cast_fp16")];
+            tensor<int32, [4]> value_45_perm_0 = const()[name = tensor<string, []>("value_45_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_22_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_22_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1105252224)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3850_cast_fp16 = add(x = q_133_cast_fp16, y = model_layers_22_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_3850_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_22_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_22_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1105254336)))];
+            tensor<fp16, [1, 126, 8, 128]> var_3852_cast_fp16 = add(x = q_133_cast_fp16, y = model_layers_22_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_3852_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_45_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_45_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_491_transpose_x_0 = const()[name = tensor<string, []>("x_491_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_491_transpose_y_0 = const()[name = tensor<string, []>("x_491_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_3854_to_fp16 = const()[name = tensor<string, []>("op_3854_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1105256448)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_133 = transpose(perm = q_with_bias_v_45_perm_0, x = var_3852_cast_fp16)[name = tensor<string, []>("transpose_133")];
+            tensor<fp16, [1, 8, 126, 251]> x_491_cast_fp16 = matmul(transpose_x = x_491_transpose_x_0, transpose_y = x_491_transpose_y_0, x = transpose_133, y = var_3854_to_fp16)[name = tensor<string, []>("x_491_cast_fp16")];
+            tensor<int32, [8]> x_493_pad_0 = const()[name = tensor<string, []>("x_493_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_493_mode_0 = const()[name = tensor<string, []>("x_493_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_233_to_fp16 = const()[name = tensor<string, []>("const_233_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_493_cast_fp16 = pad(constant_val = const_233_to_fp16, mode = x_493_mode_0, pad = x_493_pad_0, x = x_491_cast_fp16)[name = tensor<string, []>("x_493_cast_fp16")];
+            tensor<int32, [4]> var_3862 = const()[name = tensor<string, []>("op_3862"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_495_cast_fp16 = reshape(shape = var_3862, x = x_493_cast_fp16)[name = tensor<string, []>("x_495_cast_fp16")];
+            tensor<int32, [4]> var_3866_begin_0 = const()[name = tensor<string, []>("op_3866_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_3866_end_0 = const()[name = tensor<string, []>("op_3866_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_3866_end_mask_0 = const()[name = tensor<string, []>("op_3866_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> 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 = x_495_cast_fp16)[name = tensor<string, []>("op_3866_cast_fp16")];
+            tensor<int32, [4]> var_3867 = const()[name = tensor<string, []>("op_3867"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_89_cast_fp16 = reshape(shape = var_3867, x = var_3866_cast_fp16)[name = tensor<string, []>("matrix_bd_89_cast_fp16")];
+            tensor<bool, []> matrix_ac_45_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_45_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_45_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_45_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_116_perm_0 = const()[name = tensor<string, []>("transpose_116_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_117_perm_0 = const()[name = tensor<string, []>("transpose_117_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_131 = transpose(perm = transpose_117_perm_0, x = k_89_cast_fp16)[name = tensor<string, []>("transpose_131")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_132 = transpose(perm = transpose_116_perm_0, x = var_3850_cast_fp16)[name = tensor<string, []>("transpose_132")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_45_cast_fp16 = matmul(transpose_x = matrix_ac_45_transpose_x_0, transpose_y = matrix_ac_45_transpose_y_0, x = transpose_132, y = transpose_131)[name = tensor<string, []>("matrix_ac_45_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_91_begin_0 = const()[name = tensor<string, []>("matrix_bd_91_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_91_end_0 = const()[name = tensor<string, []>("matrix_bd_91_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_91_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_91_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_91_cast_fp16 = slice_by_index(begin = matrix_bd_91_begin_0, end = matrix_bd_91_end_0, end_mask = matrix_bd_91_end_mask_0, x = matrix_bd_89_cast_fp16)[name = tensor<string, []>("matrix_bd_91_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3876_cast_fp16 = add(x = matrix_ac_45_cast_fp16, y = matrix_bd_91_cast_fp16)[name = tensor<string, []>("op_3876_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_89_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_89_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_89_cast_fp16 = mul(x = var_3876_cast_fp16, y = _inversed_scores_89_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_89_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_91_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_89_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_91_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_3882_cast_fp16 = softmax(axis = var_25, x = scores_91_cast_fp16)[name = tensor<string, []>("op_3882_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_1177_cast_fp16 = select(a = var_6_to_fp16, b = var_3882_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_1177_cast_fp16")];
+            tensor<bool, []> x_497_transpose_x_0 = const()[name = tensor<string, []>("x_497_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_497_transpose_y_0 = const()[name = tensor<string, []>("x_497_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_134 = transpose(perm = value_45_perm_0, x = v_45_cast_fp16)[name = tensor<string, []>("transpose_134")];
+            tensor<fp16, [1, 8, 126, 128]> x_497_cast_fp16 = matmul(transpose_x = x_497_transpose_x_0, transpose_y = x_497_transpose_y_0, x = input_1177_cast_fp16, y = transpose_134)[name = tensor<string, []>("x_497_cast_fp16")];
+            tensor<int32, [4]> var_3886_perm_0 = const()[name = tensor<string, []>("op_3886_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_3887 = const()[name = tensor<string, []>("op_3887"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_130 = transpose(perm = var_3886_perm_0, x = x_497_cast_fp16)[name = tensor<string, []>("transpose_130")];
+            tensor<fp16, [1, 126, 1024]> input_1179_cast_fp16 = reshape(shape = var_3887, x = transpose_130)[name = tensor<string, []>("input_1179_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_22_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_22_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1105770560)))];
+            tensor<fp16, [1, 126, 1024]> linear_205_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_22_self_attn_linear_out_weight_to_fp16, x = input_1179_cast_fp16)[name = tensor<string, []>("linear_205_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1183_cast_fp16 = add(x = input_1175_cast_fp16, y = linear_205_cast_fp16)[name = tensor<string, []>("input_1183_cast_fp16")];
+            tensor<int32, [1]> x_501_axes_0 = const()[name = tensor<string, []>("x_501_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_22_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_22_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1107867776)))];
+            tensor<fp16, [1024]> model_layers_22_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_22_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1107869888)))];
+            tensor<fp16, [1, 126, 1024]> x_501_cast_fp16 = layer_norm(axes = x_501_axes_0, beta = model_layers_22_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_22_norm_conv_weight_to_fp16, x = input_1183_cast_fp16)[name = tensor<string, []>("x_501_cast_fp16")];
+            tensor<int32, [3]> input_1185_perm_0 = const()[name = tensor<string, []>("input_1185_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_3903 = const()[name = tensor<string, []>("op_3903"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_3905 = const()[name = tensor<string, []>("op_3905"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_1187_pad_type_0 = const()[name = tensor<string, []>("input_1187_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_1187_pad_0 = const()[name = tensor<string, []>("input_1187_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_22_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_22_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1107872000)))];
+            tensor<fp16, [1, 1024, 126]> transpose_129 = transpose(perm = input_1185_perm_0, x = x_501_cast_fp16)[name = tensor<string, []>("transpose_129")];
+            tensor<fp16, [1, 2048, 126]> input_1187_cast_fp16 = conv(dilations = var_3905, groups = var_27, pad = input_1187_pad_0, pad_type = input_1187_pad_type_0, strides = var_3903, weight = model_layers_22_conv_pointwise_conv1_weight_to_fp16, x = transpose_129)[name = tensor<string, []>("input_1187_cast_fp16")];
+            tensor<int32, []> x_503_split_num_splits_0 = const()[name = tensor<string, []>("x_503_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_503_split_axis_0 = const()[name = tensor<string, []>("x_503_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_503_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_503_split_cast_fp16_1 = split(axis = x_503_split_axis_0, num_splits = x_503_split_num_splits_0, x = input_1187_cast_fp16)[name = tensor<string, []>("x_503_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_503_split_1_sigmoid_cast_fp16 = sigmoid(x = x_503_split_cast_fp16_1)[name = tensor<string, []>("x_503_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_503_cast_fp16 = mul(x = x_503_split_cast_fp16_0, y = x_503_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_503_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_1189_cast_fp16 = select(a = var_6_to_fp16, b = x_503_cast_fp16, cond = var_323)[name = tensor<string, []>("input_1189_cast_fp16")];
+            tensor<int32, [6]> input_1191_pad_0 = const()[name = tensor<string, []>("input_1191_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_1191_mode_0 = const()[name = tensor<string, []>("input_1191_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_236_to_fp16 = const()[name = tensor<string, []>("const_236_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_1191_cast_fp16 = pad(constant_val = const_236_to_fp16, mode = input_1191_mode_0, pad = input_1191_pad_0, x = input_1189_cast_fp16)[name = tensor<string, []>("input_1191_cast_fp16")];
+            tensor<int32, [1]> var_3914 = const()[name = tensor<string, []>("op_3914"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_3916 = const()[name = tensor<string, []>("op_3916"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_1193_pad_type_0 = const()[name = tensor<string, []>("input_1193_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_1193_pad_0 = const()[name = tensor<string, []>("input_1193_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_1195_weight_0_to_fp16 = const()[name = tensor<string, []>("input_1195_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1112066368)))];
+            tensor<fp16, [1024]> input_1195_bias_0_to_fp16 = const()[name = tensor<string, []>("input_1195_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1112084864)))];
+            tensor<fp16, [1, 1024, 126]> input_1195_cast_fp16 = conv(bias = input_1195_bias_0_to_fp16, dilations = var_3916, groups = var_5, pad = input_1193_pad_0, pad_type = input_1193_pad_type_0, strides = var_3914, weight = input_1195_weight_0_to_fp16, x = input_1191_cast_fp16)[name = tensor<string, []>("input_1195_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_1197_cast_fp16 = silu(x = input_1195_cast_fp16)[name = tensor<string, []>("input_1197_cast_fp16")];
+            tensor<int32, [1]> var_3926 = const()[name = tensor<string, []>("op_3926"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_3928 = const()[name = tensor<string, []>("op_3928"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_505_pad_type_0 = const()[name = tensor<string, []>("x_505_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_505_pad_0 = const()[name = tensor<string, []>("x_505_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_22_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_22_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1112086976)))];
+            tensor<fp16, [1, 1024, 126]> x_505_cast_fp16 = conv(dilations = var_3928, groups = var_27, pad = x_505_pad_0, pad_type = x_505_pad_type_0, strides = var_3926, weight = model_layers_22_conv_pointwise_conv2_weight_to_fp16, x = input_1197_cast_fp16)[name = tensor<string, []>("x_505_cast_fp16")];
+            tensor<int32, [3]> input_1199_perm_0 = const()[name = tensor<string, []>("input_1199_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_128 = transpose(perm = input_1199_perm_0, x = x_505_cast_fp16)[name = tensor<string, []>("transpose_128")];
+            tensor<fp16, [1, 126, 1024]> input_1201_cast_fp16 = add(x = input_1183_cast_fp16, y = transpose_128)[name = tensor<string, []>("input_1201_cast_fp16")];
+            tensor<int32, [1]> input_1203_axes_0 = const()[name = tensor<string, []>("input_1203_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_22_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_22_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1114184192)))];
+            tensor<fp16, [1024]> model_layers_22_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_22_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1114186304)))];
+            tensor<fp16, [1, 126, 1024]> input_1203_cast_fp16 = layer_norm(axes = input_1203_axes_0, beta = model_layers_22_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_22_norm_feed_forward2_weight_to_fp16, x = input_1201_cast_fp16)[name = tensor<string, []>("input_1203_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_22_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_22_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1114188416)))];
+            tensor<fp16, [1, 126, 4096]> linear_206_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_22_feed_forward2_linear1_weight_to_fp16, x = input_1203_cast_fp16)[name = tensor<string, []>("linear_206_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_1207_cast_fp16 = silu(x = linear_206_cast_fp16)[name = tensor<string, []>("input_1207_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_22_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_22_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1122577088)))];
+            tensor<fp16, [1, 126, 1024]> linear_207_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_22_feed_forward2_linear2_weight_to_fp16, x = input_1207_cast_fp16)[name = tensor<string, []>("linear_207_cast_fp16")];
+            tensor<fp16, []> var_3947_to_fp16 = const()[name = tensor<string, []>("op_3947_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3948_cast_fp16 = mul(x = linear_207_cast_fp16, y = var_3947_to_fp16)[name = tensor<string, []>("op_3948_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1213_cast_fp16 = add(x = input_1201_cast_fp16, y = var_3948_cast_fp16)[name = tensor<string, []>("input_1213_cast_fp16")];
+            tensor<int32, [1]> input_1215_axes_0 = const()[name = tensor<string, []>("input_1215_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_22_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_22_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1130965760)))];
+            tensor<fp16, [1024]> model_layers_22_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_22_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1130967872)))];
+            tensor<fp16, [1, 126, 1024]> input_1215_cast_fp16 = layer_norm(axes = input_1215_axes_0, beta = model_layers_22_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_22_norm_out_weight_to_fp16, x = input_1213_cast_fp16)[name = tensor<string, []>("input_1215_cast_fp16")];
+            tensor<int32, [1]> input_1217_axes_0 = const()[name = tensor<string, []>("input_1217_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_23_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_23_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1130969984)))];
+            tensor<fp16, [1024]> model_layers_23_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_23_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1130972096)))];
+            tensor<fp16, [1, 126, 1024]> input_1217_cast_fp16 = layer_norm(axes = input_1217_axes_0, beta = model_layers_23_norm_feed_forward1_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_23_norm_feed_forward1_weight_to_fp16, x = input_1215_cast_fp16)[name = tensor<string, []>("input_1217_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_23_feed_forward1_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_23_feed_forward1_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1130974208)))];
+            tensor<fp16, [1, 126, 4096]> linear_208_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_23_feed_forward1_linear1_weight_to_fp16, x = input_1217_cast_fp16)[name = tensor<string, []>("linear_208_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_1221_cast_fp16 = silu(x = linear_208_cast_fp16)[name = tensor<string, []>("input_1221_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_23_feed_forward1_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_23_feed_forward1_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1139362880)))];
+            tensor<fp16, [1, 126, 1024]> linear_209_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_23_feed_forward1_linear2_weight_to_fp16, x = input_1221_cast_fp16)[name = tensor<string, []>("linear_209_cast_fp16")];
+            tensor<fp16, []> var_3976_to_fp16 = const()[name = tensor<string, []>("op_3976_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_3977_cast_fp16 = mul(x = linear_209_cast_fp16, y = var_3976_to_fp16)[name = tensor<string, []>("op_3977_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1227_cast_fp16 = add(x = input_1215_cast_fp16, y = var_3977_cast_fp16)[name = tensor<string, []>("input_1227_cast_fp16")];
+            tensor<int32, [1]> query_axes_0 = const()[name = tensor<string, []>("query_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_23_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_23_norm_self_att_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1147751552)))];
+            tensor<fp16, [1024]> model_layers_23_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_23_norm_self_att_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1147753664)))];
+            tensor<fp16, [1, 126, 1024]> query_cast_fp16 = layer_norm(axes = query_axes_0, beta = model_layers_23_norm_self_att_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_23_norm_self_att_weight_to_fp16, x = input_1227_cast_fp16)[name = tensor<string, []>("query_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_23_self_attn_linear_q_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_23_self_attn_linear_q_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1147755776)))];
+            tensor<fp16, [1, 126, 1024]> linear_210_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_23_self_attn_linear_q_weight_to_fp16, x = query_cast_fp16)[name = tensor<string, []>("linear_210_cast_fp16")];
+            tensor<int32, [4]> var_3993 = const()[name = tensor<string, []>("op_3993"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> q_139_cast_fp16 = reshape(shape = var_3993, x = linear_210_cast_fp16)[name = tensor<string, []>("q_139_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_23_self_attn_linear_k_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_23_self_attn_linear_k_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1149852992)))];
+            tensor<fp16, [1, 126, 1024]> linear_211_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_23_self_attn_linear_k_weight_to_fp16, x = query_cast_fp16)[name = tensor<string, []>("linear_211_cast_fp16")];
+            tensor<int32, [4]> var_3997 = const()[name = tensor<string, []>("op_3997"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> k_93_cast_fp16 = reshape(shape = var_3997, x = linear_211_cast_fp16)[name = tensor<string, []>("k_93_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_23_self_attn_linear_v_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_23_self_attn_linear_v_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1151950208)))];
+            tensor<fp16, [1, 126, 1024]> linear_212_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_23_self_attn_linear_v_weight_to_fp16, x = query_cast_fp16)[name = tensor<string, []>("linear_212_cast_fp16")];
+            tensor<int32, [4]> var_4001 = const()[name = tensor<string, []>("op_4001"), val = tensor<int32, [4]>([1, -1, 8, 128])];
+            tensor<fp16, [1, 126, 8, 128]> v_cast_fp16 = reshape(shape = var_4001, x = linear_212_cast_fp16)[name = tensor<string, []>("v_cast_fp16")];
+            tensor<int32, [4]> value_perm_0 = const()[name = tensor<string, []>("value_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<fp16, [8, 128]> model_layers_23_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("model_layers_23_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1154047424)))];
+            tensor<fp16, [1, 126, 8, 128]> var_4013_cast_fp16 = add(x = q_139_cast_fp16, y = model_layers_23_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_4013_cast_fp16")];
+            tensor<fp16, [8, 128]> model_layers_23_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("model_layers_23_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 128]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1154049536)))];
+            tensor<fp16, [1, 126, 8, 128]> var_4015_cast_fp16 = add(x = q_139_cast_fp16, y = model_layers_23_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_4015_cast_fp16")];
+            tensor<int32, [4]> q_with_bias_v_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<bool, []> x_513_transpose_x_0 = const()[name = tensor<string, []>("x_513_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_513_transpose_y_0 = const()[name = tensor<string, []>("x_513_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 128, 251]> var_4017_to_fp16 = const()[name = tensor<string, []>("op_4017_to_fp16"), val = tensor<fp16, [1, 8, 128, 251]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1154051648)))];
+            tensor<fp16, [1, 8, 126, 128]> transpose_126 = transpose(perm = q_with_bias_v_perm_0, x = var_4015_cast_fp16)[name = tensor<string, []>("transpose_126")];
+            tensor<fp16, [1, 8, 126, 251]> x_513_cast_fp16 = matmul(transpose_x = x_513_transpose_x_0, transpose_y = x_513_transpose_y_0, x = transpose_126, y = var_4017_to_fp16)[name = tensor<string, []>("x_513_cast_fp16")];
+            tensor<int32, [8]> x_515_pad_0 = const()[name = tensor<string, []>("x_515_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
+            tensor<string, []> x_515_mode_0 = const()[name = tensor<string, []>("x_515_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_243_to_fp16 = const()[name = tensor<string, []>("const_243_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 8, 126, 252]> x_515_cast_fp16 = pad(constant_val = const_243_to_fp16, mode = x_515_mode_0, pad = x_515_pad_0, x = x_513_cast_fp16)[name = tensor<string, []>("x_515_cast_fp16")];
+            tensor<int32, [4]> var_4025 = const()[name = tensor<string, []>("op_4025"), val = tensor<int32, [4]>([1, 8, -1, 126])];
+            tensor<fp16, [1, 8, 252, 126]> x_517_cast_fp16 = reshape(shape = var_4025, x = x_515_cast_fp16)[name = tensor<string, []>("x_517_cast_fp16")];
+            tensor<int32, [4]> var_4029_begin_0 = const()[name = tensor<string, []>("op_4029_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
+            tensor<int32, [4]> var_4029_end_0 = const()[name = tensor<string, []>("op_4029_end_0"), val = tensor<int32, [4]>([1, 8, 252, 126])];
+            tensor<bool, [4]> var_4029_end_mask_0 = const()[name = tensor<string, []>("op_4029_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
+            tensor<fp16, [1, 8, 251, 126]> var_4029_cast_fp16 = slice_by_index(begin = var_4029_begin_0, end = var_4029_end_0, end_mask = var_4029_end_mask_0, x = x_517_cast_fp16)[name = tensor<string, []>("op_4029_cast_fp16")];
+            tensor<int32, [4]> var_4030 = const()[name = tensor<string, []>("op_4030"), val = tensor<int32, [4]>([1, 8, 126, 251])];
+            tensor<fp16, [1, 8, 126, 251]> matrix_bd_93_cast_fp16 = reshape(shape = var_4030, x = var_4029_cast_fp16)[name = tensor<string, []>("matrix_bd_93_cast_fp16")];
+            tensor<bool, []> matrix_ac_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> matrix_ac_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<int32, [4]> transpose_118_perm_0 = const()[name = tensor<string, []>("transpose_118_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
+            tensor<int32, [4]> transpose_119_perm_0 = const()[name = tensor<string, []>("transpose_119_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
+            tensor<fp16, [1, 8, 128, 126]> transpose_124 = transpose(perm = transpose_119_perm_0, x = k_93_cast_fp16)[name = tensor<string, []>("transpose_124")];
+            tensor<fp16, [1, 8, 126, 128]> transpose_125 = transpose(perm = transpose_118_perm_0, x = var_4013_cast_fp16)[name = tensor<string, []>("transpose_125")];
+            tensor<fp16, [1, 8, 126, 126]> matrix_ac_cast_fp16 = matmul(transpose_x = matrix_ac_transpose_x_0, transpose_y = matrix_ac_transpose_y_0, x = transpose_125, y = transpose_124)[name = tensor<string, []>("matrix_ac_cast_fp16")];
+            tensor<int32, [4]> matrix_bd_begin_0 = const()[name = tensor<string, []>("matrix_bd_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
+            tensor<int32, [4]> matrix_bd_end_0 = const()[name = tensor<string, []>("matrix_bd_end_0"), val = tensor<int32, [4]>([1, 8, 126, 126])];
+            tensor<bool, [4]> matrix_bd_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
+            tensor<fp16, [1, 8, 126, 126]> matrix_bd_cast_fp16 = slice_by_index(begin = matrix_bd_begin_0, end = matrix_bd_end_0, end_mask = matrix_bd_end_mask_0, x = matrix_bd_93_cast_fp16)[name = tensor<string, []>("matrix_bd_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_4039_cast_fp16 = add(x = matrix_ac_cast_fp16, y = matrix_bd_cast_fp16)[name = tensor<string, []>("op_4039_cast_fp16")];
+            tensor<fp16, []> _inversed_scores_93_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_93_y_0_to_fp16"), val = tensor<fp16, []>(0x1.6ap-4)];
+            tensor<fp16, [1, 8, 126, 126]> _inversed_scores_93_cast_fp16 = mul(x = var_4039_cast_fp16, y = _inversed_scores_93_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_93_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> scores_cast_fp16 = select(a = var_7_to_fp16, b = _inversed_scores_93_cast_fp16, cond = mask_3)[name = tensor<string, []>("scores_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> var_4045_cast_fp16 = softmax(axis = var_25, x = scores_cast_fp16)[name = tensor<string, []>("op_4045_cast_fp16")];
+            tensor<fp16, [1, 8, 126, 126]> input_1229_cast_fp16 = select(a = var_6_to_fp16, b = var_4045_cast_fp16, cond = mask_3)[name = tensor<string, []>("input_1229_cast_fp16")];
+            tensor<bool, []> x_519_transpose_x_0 = const()[name = tensor<string, []>("x_519_transpose_x_0"), val = tensor<bool, []>(false)];
+            tensor<bool, []> x_519_transpose_y_0 = const()[name = tensor<string, []>("x_519_transpose_y_0"), val = tensor<bool, []>(false)];
+            tensor<fp16, [1, 8, 126, 128]> transpose_127 = transpose(perm = value_perm_0, x = v_cast_fp16)[name = tensor<string, []>("transpose_127")];
+            tensor<fp16, [1, 8, 126, 128]> x_519_cast_fp16 = matmul(transpose_x = x_519_transpose_x_0, transpose_y = x_519_transpose_y_0, x = input_1229_cast_fp16, y = transpose_127)[name = tensor<string, []>("x_519_cast_fp16")];
+            tensor<int32, [4]> var_4049_perm_0 = const()[name = tensor<string, []>("op_4049_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
+            tensor<int32, [3]> var_4050 = const()[name = tensor<string, []>("op_4050"), val = tensor<int32, [3]>([1, -1, 1024])];
+            tensor<fp16, [1, 126, 8, 128]> transpose_123 = transpose(perm = var_4049_perm_0, x = x_519_cast_fp16)[name = tensor<string, []>("transpose_123")];
+            tensor<fp16, [1, 126, 1024]> input_1231_cast_fp16 = reshape(shape = var_4050, x = transpose_123)[name = tensor<string, []>("input_1231_cast_fp16")];
+            tensor<fp16, [1024, 1024]> model_layers_23_self_attn_linear_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_23_self_attn_linear_out_weight_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1154565760)))];
+            tensor<fp16, [1, 126, 1024]> linear_214_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_23_self_attn_linear_out_weight_to_fp16, x = input_1231_cast_fp16)[name = tensor<string, []>("linear_214_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_1235_cast_fp16 = add(x = input_1227_cast_fp16, y = linear_214_cast_fp16)[name = tensor<string, []>("input_1235_cast_fp16")];
+            tensor<int32, [1]> x_523_axes_0 = const()[name = tensor<string, []>("x_523_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_23_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_23_norm_conv_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1156662976)))];
+            tensor<fp16, [1024]> model_layers_23_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_23_norm_conv_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1156665088)))];
+            tensor<fp16, [1, 126, 1024]> x_523_cast_fp16 = layer_norm(axes = x_523_axes_0, beta = model_layers_23_norm_conv_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_23_norm_conv_weight_to_fp16, x = input_1235_cast_fp16)[name = tensor<string, []>("x_523_cast_fp16")];
+            tensor<int32, [3]> input_1237_perm_0 = const()[name = tensor<string, []>("input_1237_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<int32, [1]> var_4066 = const()[name = tensor<string, []>("op_4066"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_4068 = const()[name = tensor<string, []>("op_4068"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_1239_pad_type_0 = const()[name = tensor<string, []>("input_1239_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_1239_pad_0 = const()[name = tensor<string, []>("input_1239_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [2048, 1024, 1]> model_layers_23_conv_pointwise_conv1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_23_conv_pointwise_conv1_weight_to_fp16"), val = tensor<fp16, [2048, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1156667200)))];
+            tensor<fp16, [1, 1024, 126]> transpose_122 = transpose(perm = input_1237_perm_0, x = x_523_cast_fp16)[name = tensor<string, []>("transpose_122")];
+            tensor<fp16, [1, 2048, 126]> input_1239_cast_fp16 = conv(dilations = var_4068, groups = var_27, pad = input_1239_pad_0, pad_type = input_1239_pad_type_0, strides = var_4066, weight = model_layers_23_conv_pointwise_conv1_weight_to_fp16, x = transpose_122)[name = tensor<string, []>("input_1239_cast_fp16")];
+            tensor<int32, []> x_525_split_num_splits_0 = const()[name = tensor<string, []>("x_525_split_num_splits_0"), val = tensor<int32, []>(2)];
+            tensor<int32, []> x_525_split_axis_0 = const()[name = tensor<string, []>("x_525_split_axis_0"), val = tensor<int32, []>(1)];
+            tensor<fp16, [1, 1024, 126]> x_525_split_cast_fp16_0, tensor<fp16, [1, 1024, 126]> x_525_split_cast_fp16_1 = split(axis = x_525_split_axis_0, num_splits = x_525_split_num_splits_0, x = input_1239_cast_fp16)[name = tensor<string, []>("x_525_split_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_525_split_1_sigmoid_cast_fp16 = sigmoid(x = x_525_split_cast_fp16_1)[name = tensor<string, []>("x_525_split_1_sigmoid_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> x_525_cast_fp16 = mul(x = x_525_split_cast_fp16_0, y = x_525_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_525_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_1241_cast_fp16 = select(a = var_6_to_fp16, b = x_525_cast_fp16, cond = var_323)[name = tensor<string, []>("input_1241_cast_fp16")];
+            tensor<int32, [6]> input_1243_pad_0 = const()[name = tensor<string, []>("input_1243_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
+            tensor<string, []> input_1243_mode_0 = const()[name = tensor<string, []>("input_1243_mode_0"), val = tensor<string, []>("constant")];
+            tensor<fp16, []> const_246_to_fp16 = const()[name = tensor<string, []>("const_246_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
+            tensor<fp16, [1, 1024, 134]> input_1243_cast_fp16 = pad(constant_val = const_246_to_fp16, mode = input_1243_mode_0, pad = input_1243_pad_0, x = input_1241_cast_fp16)[name = tensor<string, []>("input_1243_cast_fp16")];
+            tensor<int32, [1]> var_4077 = const()[name = tensor<string, []>("op_4077"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_4079 = const()[name = tensor<string, []>("op_4079"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> input_1245_pad_type_0 = const()[name = tensor<string, []>("input_1245_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> input_1245_pad_0 = const()[name = tensor<string, []>("input_1245_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1, 9]> input_1247_weight_0_to_fp16 = const()[name = tensor<string, []>("input_1247_weight_0_to_fp16"), val = tensor<fp16, [1024, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1160861568)))];
+            tensor<fp16, [1024]> input_1247_bias_0_to_fp16 = const()[name = tensor<string, []>("input_1247_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1160880064)))];
+            tensor<fp16, [1, 1024, 126]> input_1247_cast_fp16 = conv(bias = input_1247_bias_0_to_fp16, dilations = var_4079, groups = var_5, pad = input_1245_pad_0, pad_type = input_1245_pad_type_0, strides = var_4077, weight = input_1247_weight_0_to_fp16, x = input_1243_cast_fp16)[name = tensor<string, []>("input_1247_cast_fp16")];
+            tensor<fp16, [1, 1024, 126]> input_1249_cast_fp16 = silu(x = input_1247_cast_fp16)[name = tensor<string, []>("input_1249_cast_fp16")];
+            tensor<int32, [1]> var_4089 = const()[name = tensor<string, []>("op_4089"), val = tensor<int32, [1]>([1])];
+            tensor<int32, [1]> var_4091 = const()[name = tensor<string, []>("op_4091"), val = tensor<int32, [1]>([1])];
+            tensor<string, []> x_527_pad_type_0 = const()[name = tensor<string, []>("x_527_pad_type_0"), val = tensor<string, []>("custom")];
+            tensor<int32, [2]> x_527_pad_0 = const()[name = tensor<string, []>("x_527_pad_0"), val = tensor<int32, [2]>([0, 0])];
+            tensor<fp16, [1024, 1024, 1]> model_layers_23_conv_pointwise_conv2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_23_conv_pointwise_conv2_weight_to_fp16"), val = tensor<fp16, [1024, 1024, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1160882176)))];
+            tensor<fp16, [1, 1024, 126]> x_527_cast_fp16 = conv(dilations = var_4091, groups = var_27, pad = x_527_pad_0, pad_type = x_527_pad_type_0, strides = var_4089, weight = model_layers_23_conv_pointwise_conv2_weight_to_fp16, x = input_1249_cast_fp16)[name = tensor<string, []>("x_527_cast_fp16")];
+            tensor<int32, [3]> input_1251_perm_0 = const()[name = tensor<string, []>("input_1251_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<fp16, [1, 126, 1024]> transpose_121 = transpose(perm = input_1251_perm_0, x = x_527_cast_fp16)[name = tensor<string, []>("transpose_121")];
+            tensor<fp16, [1, 126, 1024]> input_1253_cast_fp16 = add(x = input_1235_cast_fp16, y = transpose_121)[name = tensor<string, []>("input_1253_cast_fp16")];
+            tensor<int32, [1]> input_1255_axes_0 = const()[name = tensor<string, []>("input_1255_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_23_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_23_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1162979392)))];
+            tensor<fp16, [1024]> model_layers_23_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_23_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1162981504)))];
+            tensor<fp16, [1, 126, 1024]> input_1255_cast_fp16 = layer_norm(axes = input_1255_axes_0, beta = model_layers_23_norm_feed_forward2_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_23_norm_feed_forward2_weight_to_fp16, x = input_1253_cast_fp16)[name = tensor<string, []>("input_1255_cast_fp16")];
+            tensor<fp16, [4096, 1024]> model_layers_23_feed_forward2_linear1_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_23_feed_forward2_linear1_weight_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1162983616)))];
+            tensor<fp16, [1, 126, 4096]> linear_215_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = model_layers_23_feed_forward2_linear1_weight_to_fp16, x = input_1255_cast_fp16)[name = tensor<string, []>("linear_215_cast_fp16")];
+            tensor<fp16, [1, 126, 4096]> input_1259_cast_fp16 = silu(x = linear_215_cast_fp16)[name = tensor<string, []>("input_1259_cast_fp16")];
+            tensor<fp16, [1024, 4096]> model_layers_23_feed_forward2_linear2_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_23_feed_forward2_linear2_weight_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1171372288)))];
+            tensor<fp16, [1, 126, 1024]> linear_216_cast_fp16 = linear(bias = linear_2_bias_0_to_fp16, weight = model_layers_23_feed_forward2_linear2_weight_to_fp16, x = input_1259_cast_fp16)[name = tensor<string, []>("linear_216_cast_fp16")];
+            tensor<fp16, []> var_4110_to_fp16 = const()[name = tensor<string, []>("op_4110_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
+            tensor<fp16, [1, 126, 1024]> var_4111_cast_fp16 = mul(x = linear_216_cast_fp16, y = var_4110_to_fp16)[name = tensor<string, []>("op_4111_cast_fp16")];
+            tensor<fp16, [1, 126, 1024]> input_cast_fp16 = add(x = input_1253_cast_fp16, y = var_4111_cast_fp16)[name = tensor<string, []>("input_cast_fp16")];
+            tensor<int32, [1]> audio_signal_axes_0 = const()[name = tensor<string, []>("audio_signal_axes_0"), val = tensor<int32, [1]>([-1])];
+            tensor<fp16, [1024]> model_layers_23_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("model_layers_23_norm_out_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1179760960)))];
+            tensor<fp16, [1024]> model_layers_23_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("model_layers_23_norm_out_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1179763072)))];
+            tensor<fp16, [1, 126, 1024]> audio_signal_cast_fp16 = layer_norm(axes = audio_signal_axes_0, beta = model_layers_23_norm_out_bias_to_fp16, epsilon = var_4_to_fp16, gamma = model_layers_23_norm_out_weight_to_fp16, x = input_cast_fp16)[name = tensor<string, []>("audio_signal_cast_fp16")];
+            tensor<int32, [3]> obj_1_perm_0 = const()[name = tensor<string, []>("obj_1_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
+            tensor<string, []> obj_1_cast_fp16_to_fp32_dtype_0 = const()[name = tensor<string, []>("obj_1_cast_fp16_to_fp32_dtype_0"), val = tensor<string, []>("fp32")];
+            tensor<fp16, [1, 1024, 126]> transpose_120 = transpose(perm = obj_1_perm_0, x = audio_signal_cast_fp16)[name = tensor<string, []>("transpose_120")];
+            tensor<fp32, [1, 1024, 126]> encoder_output = cast(dtype = obj_1_cast_fp16_to_fp32_dtype_0, x = transpose_120)[name = tensor<string, []>("cast_227")];
+        } -> (encoder_output, encoder_output_length);
+}
\ No newline at end of file