diff --git "a/whisper.cpp/encoder.mlmodelc/ggml-large-v3-encoder.mlmodelc/model.mil" "b/whisper.cpp/encoder.mlmodelc/ggml-large-v3-encoder.mlmodelc/model.mil" new file mode 100644--- /dev/null +++ "b/whisper.cpp/encoder.mlmodelc/ggml-large-v3-encoder.mlmodelc/model.mil" @@ -0,0 +1,1896 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "5.33.5"}, {"coremlc-version", "1877.40.3"}, {"coremltools-component-torch", "1.11.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "7.1"}})] +{ + func main(tensor logmel_data) { + tensor var_72 = const()[name = tensor("op_72"), val = tensor(1)]; + tensor var_80 = const()[name = tensor("op_80"), val = tensor([1])]; + tensor var_82 = const()[name = tensor("op_82"), val = tensor([1])]; + tensor var_84_pad_type_0 = const()[name = tensor("op_84_pad_type_0"), val = tensor("custom")]; + tensor var_84_pad_0 = const()[name = tensor("op_84_pad_0"), val = tensor([1, 1])]; + tensor logmel_data_to_fp16_dtype_0 = const()[name = tensor("logmel_data_to_fp16_dtype_0"), val = tensor("fp16")]; + tensor weight_3_to_fp16 = const()[name = tensor("weight_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor bias_3_to_fp16 = const()[name = tensor("bias_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(983168)))]; + tensor cast_193 = cast(dtype = logmel_data_to_fp16_dtype_0, x = logmel_data)[name = tensor("cast_193")]; + tensor var_84_cast_fp16 = conv(bias = bias_3_to_fp16, dilations = var_82, groups = var_72, pad = var_84_pad_0, pad_type = var_84_pad_type_0, strides = var_80, weight = weight_3_to_fp16, x = cast_193)[name = tensor("op_84_cast_fp16")]; + tensor input_1_mode_0 = const()[name = tensor("input_1_mode_0"), val = tensor("EXACT")]; + tensor input_1_cast_fp16 = gelu(mode = input_1_mode_0, x = var_84_cast_fp16)[name = tensor("input_1_cast_fp16")]; + tensor var_88 = const()[name = tensor("op_88"), val = tensor(1)]; + tensor var_97 = const()[name = tensor("op_97"), val = tensor([2])]; + tensor var_99 = const()[name = tensor("op_99"), val = tensor([1])]; + tensor var_101_pad_type_0 = const()[name = tensor("op_101_pad_type_0"), val = tensor("custom")]; + tensor var_101_pad_0 = const()[name = tensor("op_101_pad_0"), val = tensor([1, 1])]; + tensor weight_7_to_fp16 = const()[name = tensor("weight_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(985792)))]; + tensor bias_7_to_fp16 = const()[name = tensor("bias_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10816256)))]; + tensor var_101_cast_fp16 = conv(bias = bias_7_to_fp16, dilations = var_99, groups = var_88, pad = var_101_pad_0, pad_type = var_101_pad_type_0, strides = var_97, weight = weight_7_to_fp16, x = input_1_cast_fp16)[name = tensor("op_101_cast_fp16")]; + tensor x_3_mode_0 = const()[name = tensor("x_3_mode_0"), val = tensor("EXACT")]; + tensor x_3_cast_fp16 = gelu(mode = x_3_mode_0, x = var_101_cast_fp16)[name = tensor("x_3_cast_fp16")]; + tensor var_106 = const()[name = tensor("op_106"), val = tensor([0, 2, 1])]; + tensor positional_embedding_to_fp16 = const()[name = tensor("positional_embedding_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10818880)))]; + tensor transpose_320 = transpose(perm = var_106, x = x_3_cast_fp16)[name = tensor("transpose_320")]; + tensor var_109_cast_fp16 = add(x = transpose_320, y = positional_embedding_to_fp16)[name = tensor("op_109_cast_fp16")]; + tensor var_122 = const()[name = tensor("op_122"), val = tensor(-1)]; + tensor var_139_axes_0 = const()[name = tensor("op_139_axes_0"), val = tensor([-1])]; + tensor blocks_0_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_0_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14658944)))]; + tensor blocks_0_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_0_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14661568)))]; + tensor var_128_to_fp16 = const()[name = tensor("op_128_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_139_cast_fp16 = layer_norm(axes = var_139_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_128_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_109_cast_fp16)[name = tensor("op_139_cast_fp16")]; + tensor var_150_to_fp16 = const()[name = tensor("op_150_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14664192)))]; + tensor var_151_to_fp16 = const()[name = tensor("op_151_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17941056)))]; + tensor linear_0_cast_fp16 = linear(bias = var_151_to_fp16, weight = var_150_to_fp16, x = var_139_cast_fp16)[name = tensor("linear_0_cast_fp16")]; + tensor var_154_to_fp16 = const()[name = tensor("op_154_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17943680)))]; + tensor linear_1_bias_0_to_fp16 = const()[name = tensor("linear_1_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21220544)))]; + tensor linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_154_to_fp16, x = var_139_cast_fp16)[name = tensor("linear_1_cast_fp16")]; + tensor var_158_to_fp16 = const()[name = tensor("op_158_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21223168)))]; + tensor var_159_to_fp16 = const()[name = tensor("op_159_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24500032)))]; + tensor linear_2_cast_fp16 = linear(bias = var_159_to_fp16, weight = var_158_to_fp16, x = var_139_cast_fp16)[name = tensor("linear_2_cast_fp16")]; + tensor var_167 = const()[name = tensor("op_167"), val = tensor([1, 1500, 20, -1])]; + tensor var_168_cast_fp16 = reshape(shape = var_167, x = linear_0_cast_fp16)[name = tensor("op_168_cast_fp16")]; + tensor const_224_to_fp16 = const()[name = tensor("const_224_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_3_cast_fp16 = mul(x = var_168_cast_fp16, y = const_224_to_fp16)[name = tensor("q_3_cast_fp16")]; + tensor var_174 = const()[name = tensor("op_174"), val = tensor([1, 1500, 20, -1])]; + tensor var_175_cast_fp16 = reshape(shape = var_174, x = linear_1_cast_fp16)[name = tensor("op_175_cast_fp16")]; + tensor const_225_to_fp16 = const()[name = tensor("const_225_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_3_cast_fp16 = mul(x = var_175_cast_fp16, y = const_225_to_fp16)[name = tensor("k_3_cast_fp16")]; + tensor var_181 = const()[name = tensor("op_181"), val = tensor([1, 1500, 20, -1])]; + tensor var_182_cast_fp16 = reshape(shape = var_181, x = linear_2_cast_fp16)[name = tensor("op_182_cast_fp16")]; + tensor var_183 = const()[name = tensor("op_183"), val = tensor([0, 2, 1, 3])]; + tensor qk_1_transpose_x_0 = const()[name = tensor("qk_1_transpose_x_0"), val = tensor(false)]; + tensor qk_1_transpose_y_0 = const()[name = tensor("qk_1_transpose_y_0"), val = tensor(false)]; + tensor transpose_128_perm_0 = const()[name = tensor("transpose_128_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_129_perm_0 = const()[name = tensor("transpose_129_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_317 = transpose(perm = transpose_129_perm_0, x = k_3_cast_fp16)[name = tensor("transpose_317")]; + tensor transpose_318 = transpose(perm = transpose_128_perm_0, x = q_3_cast_fp16)[name = tensor("transpose_318")]; + tensor qk_1_cast_fp16 = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_318, y = transpose_317)[name = tensor("qk_1_cast_fp16")]; + tensor var_187_cast_fp16 = softmax(axis = var_122, x = qk_1_cast_fp16)[name = tensor("op_187_cast_fp16")]; + tensor var_189_transpose_x_0 = const()[name = tensor("op_189_transpose_x_0"), val = tensor(false)]; + tensor var_189_transpose_y_0 = const()[name = tensor("op_189_transpose_y_0"), val = tensor(false)]; + tensor transpose_319 = transpose(perm = var_183, x = var_182_cast_fp16)[name = tensor("transpose_319")]; + tensor var_189_cast_fp16 = matmul(transpose_x = var_189_transpose_x_0, transpose_y = var_189_transpose_y_0, x = var_187_cast_fp16, y = transpose_319)[name = tensor("op_189_cast_fp16")]; + tensor var_190 = const()[name = tensor("op_190"), val = tensor([0, 2, 1, 3])]; + tensor concat_0 = const()[name = tensor("concat_0"), val = tensor([1, 1500, 1280])]; + tensor transpose_316 = transpose(perm = var_190, x = var_189_cast_fp16)[name = tensor("transpose_316")]; + tensor x_11_cast_fp16 = reshape(shape = concat_0, x = transpose_316)[name = tensor("x_11_cast_fp16")]; + tensor var_195_to_fp16 = const()[name = tensor("op_195_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24502656)))]; + tensor var_196_to_fp16 = const()[name = tensor("op_196_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27779520)))]; + tensor linear_3_cast_fp16 = linear(bias = var_196_to_fp16, weight = var_195_to_fp16, x = x_11_cast_fp16)[name = tensor("linear_3_cast_fp16")]; + tensor x_13_cast_fp16 = add(x = var_109_cast_fp16, y = linear_3_cast_fp16)[name = tensor("x_13_cast_fp16")]; + tensor var_203_axes_0 = const()[name = tensor("op_203_axes_0"), val = tensor([-1])]; + tensor blocks_0_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27782144)))]; + tensor blocks_0_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27784768)))]; + tensor var_203_cast_fp16 = layer_norm(axes = var_203_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_128_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast_fp16)[name = tensor("op_203_cast_fp16")]; + tensor var_212_to_fp16 = const()[name = tensor("op_212_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27787392)))]; + tensor var_213_to_fp16 = const()[name = tensor("op_213_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40894656)))]; + tensor linear_4_cast_fp16 = linear(bias = var_213_to_fp16, weight = var_212_to_fp16, x = var_203_cast_fp16)[name = tensor("linear_4_cast_fp16")]; + tensor x_17_mode_0 = const()[name = tensor("x_17_mode_0"), val = tensor("EXACT")]; + tensor x_17_cast_fp16 = gelu(mode = x_17_mode_0, x = linear_4_cast_fp16)[name = tensor("x_17_cast_fp16")]; + tensor var_218_to_fp16 = const()[name = tensor("op_218_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40904960)))]; + tensor var_219_to_fp16 = const()[name = tensor("op_219_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(54012224)))]; + tensor linear_5_cast_fp16 = linear(bias = var_219_to_fp16, weight = var_218_to_fp16, x = x_17_cast_fp16)[name = tensor("linear_5_cast_fp16")]; + tensor x_19_cast_fp16 = add(x = x_13_cast_fp16, y = linear_5_cast_fp16)[name = tensor("x_19_cast_fp16")]; + tensor var_229 = const()[name = tensor("op_229"), val = tensor(-1)]; + tensor var_246_axes_0 = const()[name = tensor("op_246_axes_0"), val = tensor([-1])]; + tensor blocks_1_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_1_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(54014848)))]; + tensor blocks_1_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_1_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(54017472)))]; + tensor var_235_to_fp16 = const()[name = tensor("op_235_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_246_cast_fp16 = layer_norm(axes = var_246_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_235_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast_fp16)[name = tensor("op_246_cast_fp16")]; + tensor var_257_to_fp16 = const()[name = tensor("op_257_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(54020096)))]; + tensor var_258_to_fp16 = const()[name = tensor("op_258_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57296960)))]; + tensor linear_6_cast_fp16 = linear(bias = var_258_to_fp16, weight = var_257_to_fp16, x = var_246_cast_fp16)[name = tensor("linear_6_cast_fp16")]; + tensor var_261_to_fp16 = const()[name = tensor("op_261_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57299584)))]; + tensor linear_7_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_261_to_fp16, x = var_246_cast_fp16)[name = tensor("linear_7_cast_fp16")]; + tensor var_265_to_fp16 = const()[name = tensor("op_265_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60576448)))]; + tensor var_266_to_fp16 = const()[name = tensor("op_266_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63853312)))]; + tensor linear_8_cast_fp16 = linear(bias = var_266_to_fp16, weight = var_265_to_fp16, x = var_246_cast_fp16)[name = tensor("linear_8_cast_fp16")]; + tensor var_274 = const()[name = tensor("op_274"), val = tensor([1, 1500, 20, -1])]; + tensor var_275_cast_fp16 = reshape(shape = var_274, x = linear_6_cast_fp16)[name = tensor("op_275_cast_fp16")]; + tensor const_226_to_fp16 = const()[name = tensor("const_226_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_7_cast_fp16 = mul(x = var_275_cast_fp16, y = const_226_to_fp16)[name = tensor("q_7_cast_fp16")]; + tensor var_281 = const()[name = tensor("op_281"), val = tensor([1, 1500, 20, -1])]; + tensor var_282_cast_fp16 = reshape(shape = var_281, x = linear_7_cast_fp16)[name = tensor("op_282_cast_fp16")]; + tensor const_227_to_fp16 = const()[name = tensor("const_227_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_7_cast_fp16 = mul(x = var_282_cast_fp16, y = const_227_to_fp16)[name = tensor("k_7_cast_fp16")]; + tensor var_288 = const()[name = tensor("op_288"), val = tensor([1, 1500, 20, -1])]; + tensor var_289_cast_fp16 = reshape(shape = var_288, x = linear_8_cast_fp16)[name = tensor("op_289_cast_fp16")]; + tensor var_290 = const()[name = tensor("op_290"), val = tensor([0, 2, 1, 3])]; + tensor qk_3_transpose_x_0 = const()[name = tensor("qk_3_transpose_x_0"), val = tensor(false)]; + tensor qk_3_transpose_y_0 = const()[name = tensor("qk_3_transpose_y_0"), val = tensor(false)]; + tensor transpose_130_perm_0 = const()[name = tensor("transpose_130_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_131_perm_0 = const()[name = tensor("transpose_131_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_313 = transpose(perm = transpose_131_perm_0, x = k_7_cast_fp16)[name = tensor("transpose_313")]; + tensor transpose_314 = transpose(perm = transpose_130_perm_0, x = q_7_cast_fp16)[name = tensor("transpose_314")]; + tensor qk_3_cast_fp16 = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_314, y = transpose_313)[name = tensor("qk_3_cast_fp16")]; + tensor var_294_cast_fp16 = softmax(axis = var_229, x = qk_3_cast_fp16)[name = tensor("op_294_cast_fp16")]; + tensor var_296_transpose_x_0 = const()[name = tensor("op_296_transpose_x_0"), val = tensor(false)]; + tensor var_296_transpose_y_0 = const()[name = tensor("op_296_transpose_y_0"), val = tensor(false)]; + tensor transpose_315 = transpose(perm = var_290, x = var_289_cast_fp16)[name = tensor("transpose_315")]; + tensor var_296_cast_fp16 = matmul(transpose_x = var_296_transpose_x_0, transpose_y = var_296_transpose_y_0, x = var_294_cast_fp16, y = transpose_315)[name = tensor("op_296_cast_fp16")]; + tensor var_297 = const()[name = tensor("op_297"), val = tensor([0, 2, 1, 3])]; + tensor concat_1 = const()[name = tensor("concat_1"), val = tensor([1, 1500, 1280])]; + tensor transpose_312 = transpose(perm = var_297, x = var_296_cast_fp16)[name = tensor("transpose_312")]; + tensor x_23_cast_fp16 = reshape(shape = concat_1, x = transpose_312)[name = tensor("x_23_cast_fp16")]; + tensor var_302_to_fp16 = const()[name = tensor("op_302_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63855936)))]; + tensor var_303_to_fp16 = const()[name = tensor("op_303_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67132800)))]; + tensor linear_9_cast_fp16 = linear(bias = var_303_to_fp16, weight = var_302_to_fp16, x = x_23_cast_fp16)[name = tensor("linear_9_cast_fp16")]; + tensor x_25_cast_fp16 = add(x = x_19_cast_fp16, y = linear_9_cast_fp16)[name = tensor("x_25_cast_fp16")]; + tensor var_310_axes_0 = const()[name = tensor("op_310_axes_0"), val = tensor([-1])]; + tensor blocks_1_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67135424)))]; + tensor blocks_1_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67138048)))]; + tensor var_310_cast_fp16 = layer_norm(axes = var_310_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_235_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast_fp16)[name = tensor("op_310_cast_fp16")]; + tensor var_319_to_fp16 = const()[name = tensor("op_319_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67140672)))]; + tensor var_320_to_fp16 = const()[name = tensor("op_320_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80247936)))]; + tensor linear_10_cast_fp16 = linear(bias = var_320_to_fp16, weight = var_319_to_fp16, x = var_310_cast_fp16)[name = tensor("linear_10_cast_fp16")]; + tensor x_29_mode_0 = const()[name = tensor("x_29_mode_0"), val = tensor("EXACT")]; + tensor x_29_cast_fp16 = gelu(mode = x_29_mode_0, x = linear_10_cast_fp16)[name = tensor("x_29_cast_fp16")]; + tensor var_325_to_fp16 = const()[name = tensor("op_325_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80258240)))]; + tensor var_326_to_fp16 = const()[name = tensor("op_326_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93365504)))]; + tensor linear_11_cast_fp16 = linear(bias = var_326_to_fp16, weight = var_325_to_fp16, x = x_29_cast_fp16)[name = tensor("linear_11_cast_fp16")]; + tensor x_31_cast_fp16 = add(x = x_25_cast_fp16, y = linear_11_cast_fp16)[name = tensor("x_31_cast_fp16")]; + tensor var_336 = const()[name = tensor("op_336"), val = tensor(-1)]; + tensor var_353_axes_0 = const()[name = tensor("op_353_axes_0"), val = tensor([-1])]; + tensor blocks_2_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_2_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93368128)))]; + tensor blocks_2_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_2_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93370752)))]; + tensor var_342_to_fp16 = const()[name = tensor("op_342_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_353_cast_fp16 = layer_norm(axes = var_353_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_342_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast_fp16)[name = tensor("op_353_cast_fp16")]; + tensor var_364_to_fp16 = const()[name = tensor("op_364_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93373376)))]; + tensor var_365_to_fp16 = const()[name = tensor("op_365_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96650240)))]; + tensor linear_12_cast_fp16 = linear(bias = var_365_to_fp16, weight = var_364_to_fp16, x = var_353_cast_fp16)[name = tensor("linear_12_cast_fp16")]; + tensor var_368_to_fp16 = const()[name = tensor("op_368_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96652864)))]; + tensor linear_13_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_368_to_fp16, x = var_353_cast_fp16)[name = tensor("linear_13_cast_fp16")]; + tensor var_372_to_fp16 = const()[name = tensor("op_372_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99929728)))]; + tensor var_373_to_fp16 = const()[name = tensor("op_373_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103206592)))]; + tensor linear_14_cast_fp16 = linear(bias = var_373_to_fp16, weight = var_372_to_fp16, x = var_353_cast_fp16)[name = tensor("linear_14_cast_fp16")]; + tensor var_381 = const()[name = tensor("op_381"), val = tensor([1, 1500, 20, -1])]; + tensor var_382_cast_fp16 = reshape(shape = var_381, x = linear_12_cast_fp16)[name = tensor("op_382_cast_fp16")]; + tensor const_228_to_fp16 = const()[name = tensor("const_228_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_11_cast_fp16 = mul(x = var_382_cast_fp16, y = const_228_to_fp16)[name = tensor("q_11_cast_fp16")]; + tensor var_388 = const()[name = tensor("op_388"), val = tensor([1, 1500, 20, -1])]; + tensor var_389_cast_fp16 = reshape(shape = var_388, x = linear_13_cast_fp16)[name = tensor("op_389_cast_fp16")]; + tensor const_229_to_fp16 = const()[name = tensor("const_229_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_11_cast_fp16 = mul(x = var_389_cast_fp16, y = const_229_to_fp16)[name = tensor("k_11_cast_fp16")]; + tensor var_395 = const()[name = tensor("op_395"), val = tensor([1, 1500, 20, -1])]; + tensor var_396_cast_fp16 = reshape(shape = var_395, x = linear_14_cast_fp16)[name = tensor("op_396_cast_fp16")]; + tensor var_397 = const()[name = tensor("op_397"), val = tensor([0, 2, 1, 3])]; + tensor qk_5_transpose_x_0 = const()[name = tensor("qk_5_transpose_x_0"), val = tensor(false)]; + tensor qk_5_transpose_y_0 = const()[name = tensor("qk_5_transpose_y_0"), val = tensor(false)]; + tensor transpose_132_perm_0 = const()[name = tensor("transpose_132_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_133_perm_0 = const()[name = tensor("transpose_133_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_309 = transpose(perm = transpose_133_perm_0, x = k_11_cast_fp16)[name = tensor("transpose_309")]; + tensor transpose_310 = transpose(perm = transpose_132_perm_0, x = q_11_cast_fp16)[name = tensor("transpose_310")]; + tensor qk_5_cast_fp16 = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_310, y = transpose_309)[name = tensor("qk_5_cast_fp16")]; + tensor var_401_cast_fp16 = softmax(axis = var_336, x = qk_5_cast_fp16)[name = tensor("op_401_cast_fp16")]; + tensor var_403_transpose_x_0 = const()[name = tensor("op_403_transpose_x_0"), val = tensor(false)]; + tensor var_403_transpose_y_0 = const()[name = tensor("op_403_transpose_y_0"), val = tensor(false)]; + tensor transpose_311 = transpose(perm = var_397, x = var_396_cast_fp16)[name = tensor("transpose_311")]; + tensor var_403_cast_fp16 = matmul(transpose_x = var_403_transpose_x_0, transpose_y = var_403_transpose_y_0, x = var_401_cast_fp16, y = transpose_311)[name = tensor("op_403_cast_fp16")]; + tensor var_404 = const()[name = tensor("op_404"), val = tensor([0, 2, 1, 3])]; + tensor concat_2 = const()[name = tensor("concat_2"), val = tensor([1, 1500, 1280])]; + tensor transpose_308 = transpose(perm = var_404, x = var_403_cast_fp16)[name = tensor("transpose_308")]; + tensor x_35_cast_fp16 = reshape(shape = concat_2, x = transpose_308)[name = tensor("x_35_cast_fp16")]; + tensor var_409_to_fp16 = const()[name = tensor("op_409_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103209216)))]; + tensor var_410_to_fp16 = const()[name = tensor("op_410_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106486080)))]; + tensor linear_15_cast_fp16 = linear(bias = var_410_to_fp16, weight = var_409_to_fp16, x = x_35_cast_fp16)[name = tensor("linear_15_cast_fp16")]; + tensor x_37_cast_fp16 = add(x = x_31_cast_fp16, y = linear_15_cast_fp16)[name = tensor("x_37_cast_fp16")]; + tensor var_417_axes_0 = const()[name = tensor("op_417_axes_0"), val = tensor([-1])]; + tensor blocks_2_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106488704)))]; + tensor blocks_2_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106491328)))]; + tensor var_417_cast_fp16 = layer_norm(axes = var_417_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_342_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast_fp16)[name = tensor("op_417_cast_fp16")]; + tensor var_426_to_fp16 = const()[name = tensor("op_426_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106493952)))]; + tensor var_427_to_fp16 = const()[name = tensor("op_427_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119601216)))]; + tensor linear_16_cast_fp16 = linear(bias = var_427_to_fp16, weight = var_426_to_fp16, x = var_417_cast_fp16)[name = tensor("linear_16_cast_fp16")]; + tensor x_41_mode_0 = const()[name = tensor("x_41_mode_0"), val = tensor("EXACT")]; + tensor x_41_cast_fp16 = gelu(mode = x_41_mode_0, x = linear_16_cast_fp16)[name = tensor("x_41_cast_fp16")]; + tensor var_432_to_fp16 = const()[name = tensor("op_432_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119611520)))]; + tensor var_433_to_fp16 = const()[name = tensor("op_433_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132718784)))]; + tensor linear_17_cast_fp16 = linear(bias = var_433_to_fp16, weight = var_432_to_fp16, x = x_41_cast_fp16)[name = tensor("linear_17_cast_fp16")]; + tensor x_43_cast_fp16 = add(x = x_37_cast_fp16, y = linear_17_cast_fp16)[name = tensor("x_43_cast_fp16")]; + tensor var_443 = const()[name = tensor("op_443"), val = tensor(-1)]; + tensor var_460_axes_0 = const()[name = tensor("op_460_axes_0"), val = tensor([-1])]; + tensor blocks_3_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_3_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132721408)))]; + tensor blocks_3_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_3_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132724032)))]; + tensor var_449_to_fp16 = const()[name = tensor("op_449_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_460_cast_fp16 = layer_norm(axes = var_460_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_449_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast_fp16)[name = tensor("op_460_cast_fp16")]; + tensor var_471_to_fp16 = const()[name = tensor("op_471_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132726656)))]; + tensor var_472_to_fp16 = const()[name = tensor("op_472_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136003520)))]; + tensor linear_18_cast_fp16 = linear(bias = var_472_to_fp16, weight = var_471_to_fp16, x = var_460_cast_fp16)[name = tensor("linear_18_cast_fp16")]; + tensor var_475_to_fp16 = const()[name = tensor("op_475_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136006144)))]; + tensor linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_475_to_fp16, x = var_460_cast_fp16)[name = tensor("linear_19_cast_fp16")]; + tensor var_479_to_fp16 = const()[name = tensor("op_479_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(139283008)))]; + tensor var_480_to_fp16 = const()[name = tensor("op_480_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142559872)))]; + tensor linear_20_cast_fp16 = linear(bias = var_480_to_fp16, weight = var_479_to_fp16, x = var_460_cast_fp16)[name = tensor("linear_20_cast_fp16")]; + tensor var_488 = const()[name = tensor("op_488"), val = tensor([1, 1500, 20, -1])]; + tensor var_489_cast_fp16 = reshape(shape = var_488, x = linear_18_cast_fp16)[name = tensor("op_489_cast_fp16")]; + tensor const_230_to_fp16 = const()[name = tensor("const_230_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_15_cast_fp16 = mul(x = var_489_cast_fp16, y = const_230_to_fp16)[name = tensor("q_15_cast_fp16")]; + tensor var_495 = const()[name = tensor("op_495"), val = tensor([1, 1500, 20, -1])]; + tensor var_496_cast_fp16 = reshape(shape = var_495, x = linear_19_cast_fp16)[name = tensor("op_496_cast_fp16")]; + tensor const_231_to_fp16 = const()[name = tensor("const_231_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_15_cast_fp16 = mul(x = var_496_cast_fp16, y = const_231_to_fp16)[name = tensor("k_15_cast_fp16")]; + tensor var_502 = const()[name = tensor("op_502"), val = tensor([1, 1500, 20, -1])]; + tensor var_503_cast_fp16 = reshape(shape = var_502, x = linear_20_cast_fp16)[name = tensor("op_503_cast_fp16")]; + tensor var_504 = const()[name = tensor("op_504"), val = tensor([0, 2, 1, 3])]; + tensor qk_7_transpose_x_0 = const()[name = tensor("qk_7_transpose_x_0"), val = tensor(false)]; + tensor qk_7_transpose_y_0 = const()[name = tensor("qk_7_transpose_y_0"), val = tensor(false)]; + tensor transpose_134_perm_0 = const()[name = tensor("transpose_134_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_135_perm_0 = const()[name = tensor("transpose_135_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_305 = transpose(perm = transpose_135_perm_0, x = k_15_cast_fp16)[name = tensor("transpose_305")]; + tensor transpose_306 = transpose(perm = transpose_134_perm_0, x = q_15_cast_fp16)[name = tensor("transpose_306")]; + tensor qk_7_cast_fp16 = matmul(transpose_x = qk_7_transpose_x_0, transpose_y = qk_7_transpose_y_0, x = transpose_306, y = transpose_305)[name = tensor("qk_7_cast_fp16")]; + tensor var_508_cast_fp16 = softmax(axis = var_443, x = qk_7_cast_fp16)[name = tensor("op_508_cast_fp16")]; + tensor var_510_transpose_x_0 = const()[name = tensor("op_510_transpose_x_0"), val = tensor(false)]; + tensor var_510_transpose_y_0 = const()[name = tensor("op_510_transpose_y_0"), val = tensor(false)]; + tensor transpose_307 = transpose(perm = var_504, x = var_503_cast_fp16)[name = tensor("transpose_307")]; + tensor var_510_cast_fp16 = matmul(transpose_x = var_510_transpose_x_0, transpose_y = var_510_transpose_y_0, x = var_508_cast_fp16, y = transpose_307)[name = tensor("op_510_cast_fp16")]; + tensor var_511 = const()[name = tensor("op_511"), val = tensor([0, 2, 1, 3])]; + tensor concat_3 = const()[name = tensor("concat_3"), val = tensor([1, 1500, 1280])]; + tensor transpose_304 = transpose(perm = var_511, x = var_510_cast_fp16)[name = tensor("transpose_304")]; + tensor x_47_cast_fp16 = reshape(shape = concat_3, x = transpose_304)[name = tensor("x_47_cast_fp16")]; + tensor var_516_to_fp16 = const()[name = tensor("op_516_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142562496)))]; + tensor var_517_to_fp16 = const()[name = tensor("op_517_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145839360)))]; + tensor linear_21_cast_fp16 = linear(bias = var_517_to_fp16, weight = var_516_to_fp16, x = x_47_cast_fp16)[name = tensor("linear_21_cast_fp16")]; + tensor x_49_cast_fp16 = add(x = x_43_cast_fp16, y = linear_21_cast_fp16)[name = tensor("x_49_cast_fp16")]; + tensor var_524_axes_0 = const()[name = tensor("op_524_axes_0"), val = tensor([-1])]; + tensor blocks_3_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145841984)))]; + tensor blocks_3_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145844608)))]; + tensor var_524_cast_fp16 = layer_norm(axes = var_524_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_449_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast_fp16)[name = tensor("op_524_cast_fp16")]; + tensor var_533_to_fp16 = const()[name = tensor("op_533_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145847232)))]; + tensor var_534_to_fp16 = const()[name = tensor("op_534_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158954496)))]; + tensor linear_22_cast_fp16 = linear(bias = var_534_to_fp16, weight = var_533_to_fp16, x = var_524_cast_fp16)[name = tensor("linear_22_cast_fp16")]; + tensor x_53_mode_0 = const()[name = tensor("x_53_mode_0"), val = tensor("EXACT")]; + tensor x_53_cast_fp16 = gelu(mode = x_53_mode_0, x = linear_22_cast_fp16)[name = tensor("x_53_cast_fp16")]; + tensor var_539_to_fp16 = const()[name = tensor("op_539_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158964800)))]; + tensor var_540_to_fp16 = const()[name = tensor("op_540_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(172072064)))]; + tensor linear_23_cast_fp16 = linear(bias = var_540_to_fp16, weight = var_539_to_fp16, x = x_53_cast_fp16)[name = tensor("linear_23_cast_fp16")]; + tensor x_55_cast_fp16 = add(x = x_49_cast_fp16, y = linear_23_cast_fp16)[name = tensor("x_55_cast_fp16")]; + tensor var_550 = const()[name = tensor("op_550"), val = tensor(-1)]; + tensor var_567_axes_0 = const()[name = tensor("op_567_axes_0"), val = tensor([-1])]; + tensor blocks_4_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_4_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(172074688)))]; + tensor blocks_4_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_4_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(172077312)))]; + tensor var_556_to_fp16 = const()[name = tensor("op_556_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_567_cast_fp16 = layer_norm(axes = var_567_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_556_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast_fp16)[name = tensor("op_567_cast_fp16")]; + tensor var_578_to_fp16 = const()[name = tensor("op_578_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(172079936)))]; + tensor var_579_to_fp16 = const()[name = tensor("op_579_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175356800)))]; + tensor linear_24_cast_fp16 = linear(bias = var_579_to_fp16, weight = var_578_to_fp16, x = var_567_cast_fp16)[name = tensor("linear_24_cast_fp16")]; + tensor var_582_to_fp16 = const()[name = tensor("op_582_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175359424)))]; + tensor linear_25_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_582_to_fp16, x = var_567_cast_fp16)[name = tensor("linear_25_cast_fp16")]; + tensor var_586_to_fp16 = const()[name = tensor("op_586_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178636288)))]; + tensor var_587_to_fp16 = const()[name = tensor("op_587_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(181913152)))]; + tensor linear_26_cast_fp16 = linear(bias = var_587_to_fp16, weight = var_586_to_fp16, x = var_567_cast_fp16)[name = tensor("linear_26_cast_fp16")]; + tensor var_595 = const()[name = tensor("op_595"), val = tensor([1, 1500, 20, -1])]; + tensor var_596_cast_fp16 = reshape(shape = var_595, x = linear_24_cast_fp16)[name = tensor("op_596_cast_fp16")]; + tensor const_232_to_fp16 = const()[name = tensor("const_232_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_19_cast_fp16 = mul(x = var_596_cast_fp16, y = const_232_to_fp16)[name = tensor("q_19_cast_fp16")]; + tensor var_602 = const()[name = tensor("op_602"), val = tensor([1, 1500, 20, -1])]; + tensor var_603_cast_fp16 = reshape(shape = var_602, x = linear_25_cast_fp16)[name = tensor("op_603_cast_fp16")]; + tensor const_233_to_fp16 = const()[name = tensor("const_233_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_19_cast_fp16 = mul(x = var_603_cast_fp16, y = const_233_to_fp16)[name = tensor("k_19_cast_fp16")]; + tensor var_609 = const()[name = tensor("op_609"), val = tensor([1, 1500, 20, -1])]; + tensor var_610_cast_fp16 = reshape(shape = var_609, x = linear_26_cast_fp16)[name = tensor("op_610_cast_fp16")]; + tensor var_611 = const()[name = tensor("op_611"), val = tensor([0, 2, 1, 3])]; + tensor qk_9_transpose_x_0 = const()[name = tensor("qk_9_transpose_x_0"), val = tensor(false)]; + tensor qk_9_transpose_y_0 = const()[name = tensor("qk_9_transpose_y_0"), val = tensor(false)]; + tensor transpose_136_perm_0 = const()[name = tensor("transpose_136_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_137_perm_0 = const()[name = tensor("transpose_137_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_301 = transpose(perm = transpose_137_perm_0, x = k_19_cast_fp16)[name = tensor("transpose_301")]; + tensor transpose_302 = transpose(perm = transpose_136_perm_0, x = q_19_cast_fp16)[name = tensor("transpose_302")]; + tensor qk_9_cast_fp16 = matmul(transpose_x = qk_9_transpose_x_0, transpose_y = qk_9_transpose_y_0, x = transpose_302, y = transpose_301)[name = tensor("qk_9_cast_fp16")]; + tensor var_615_cast_fp16 = softmax(axis = var_550, x = qk_9_cast_fp16)[name = tensor("op_615_cast_fp16")]; + tensor var_617_transpose_x_0 = const()[name = tensor("op_617_transpose_x_0"), val = tensor(false)]; + tensor var_617_transpose_y_0 = const()[name = tensor("op_617_transpose_y_0"), val = tensor(false)]; + tensor transpose_303 = transpose(perm = var_611, x = var_610_cast_fp16)[name = tensor("transpose_303")]; + tensor var_617_cast_fp16 = matmul(transpose_x = var_617_transpose_x_0, transpose_y = var_617_transpose_y_0, x = var_615_cast_fp16, y = transpose_303)[name = tensor("op_617_cast_fp16")]; + tensor var_618 = const()[name = tensor("op_618"), val = tensor([0, 2, 1, 3])]; + tensor concat_4 = const()[name = tensor("concat_4"), val = tensor([1, 1500, 1280])]; + tensor transpose_300 = transpose(perm = var_618, x = var_617_cast_fp16)[name = tensor("transpose_300")]; + tensor x_59_cast_fp16 = reshape(shape = concat_4, x = transpose_300)[name = tensor("x_59_cast_fp16")]; + tensor var_623_to_fp16 = const()[name = tensor("op_623_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(181915776)))]; + tensor var_624_to_fp16 = const()[name = tensor("op_624_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(185192640)))]; + tensor linear_27_cast_fp16 = linear(bias = var_624_to_fp16, weight = var_623_to_fp16, x = x_59_cast_fp16)[name = tensor("linear_27_cast_fp16")]; + tensor x_61_cast_fp16 = add(x = x_55_cast_fp16, y = linear_27_cast_fp16)[name = tensor("x_61_cast_fp16")]; + tensor var_631_axes_0 = const()[name = tensor("op_631_axes_0"), val = tensor([-1])]; + tensor blocks_4_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(185195264)))]; + tensor blocks_4_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(185197888)))]; + tensor var_631_cast_fp16 = layer_norm(axes = var_631_axes_0, beta = blocks_4_mlp_ln_bias_to_fp16, epsilon = var_556_to_fp16, gamma = blocks_4_mlp_ln_weight_to_fp16, x = x_61_cast_fp16)[name = tensor("op_631_cast_fp16")]; + tensor var_640_to_fp16 = const()[name = tensor("op_640_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(185200512)))]; + tensor var_641_to_fp16 = const()[name = tensor("op_641_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198307776)))]; + tensor linear_28_cast_fp16 = linear(bias = var_641_to_fp16, weight = var_640_to_fp16, x = var_631_cast_fp16)[name = tensor("linear_28_cast_fp16")]; + tensor x_65_mode_0 = const()[name = tensor("x_65_mode_0"), val = tensor("EXACT")]; + tensor x_65_cast_fp16 = gelu(mode = x_65_mode_0, x = linear_28_cast_fp16)[name = tensor("x_65_cast_fp16")]; + tensor var_646_to_fp16 = const()[name = tensor("op_646_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198318080)))]; + tensor var_647_to_fp16 = const()[name = tensor("op_647_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211425344)))]; + tensor linear_29_cast_fp16 = linear(bias = var_647_to_fp16, weight = var_646_to_fp16, x = x_65_cast_fp16)[name = tensor("linear_29_cast_fp16")]; + tensor x_67_cast_fp16 = add(x = x_61_cast_fp16, y = linear_29_cast_fp16)[name = tensor("x_67_cast_fp16")]; + tensor var_657 = const()[name = tensor("op_657"), val = tensor(-1)]; + tensor var_674_axes_0 = const()[name = tensor("op_674_axes_0"), val = tensor([-1])]; + tensor blocks_5_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_5_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211427968)))]; + tensor blocks_5_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_5_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211430592)))]; + tensor var_663_to_fp16 = const()[name = tensor("op_663_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_674_cast_fp16 = layer_norm(axes = var_674_axes_0, beta = blocks_5_attn_ln_bias_to_fp16, epsilon = var_663_to_fp16, gamma = blocks_5_attn_ln_weight_to_fp16, x = x_67_cast_fp16)[name = tensor("op_674_cast_fp16")]; + tensor var_685_to_fp16 = const()[name = tensor("op_685_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211433216)))]; + tensor var_686_to_fp16 = const()[name = tensor("op_686_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214710080)))]; + tensor linear_30_cast_fp16 = linear(bias = var_686_to_fp16, weight = var_685_to_fp16, x = var_674_cast_fp16)[name = tensor("linear_30_cast_fp16")]; + tensor var_689_to_fp16 = const()[name = tensor("op_689_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214712704)))]; + tensor linear_31_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_689_to_fp16, x = var_674_cast_fp16)[name = tensor("linear_31_cast_fp16")]; + tensor var_693_to_fp16 = const()[name = tensor("op_693_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217989568)))]; + tensor var_694_to_fp16 = const()[name = tensor("op_694_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(221266432)))]; + tensor linear_32_cast_fp16 = linear(bias = var_694_to_fp16, weight = var_693_to_fp16, x = var_674_cast_fp16)[name = tensor("linear_32_cast_fp16")]; + tensor var_702 = const()[name = tensor("op_702"), val = tensor([1, 1500, 20, -1])]; + tensor var_703_cast_fp16 = reshape(shape = var_702, x = linear_30_cast_fp16)[name = tensor("op_703_cast_fp16")]; + tensor const_234_to_fp16 = const()[name = tensor("const_234_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_23_cast_fp16 = mul(x = var_703_cast_fp16, y = const_234_to_fp16)[name = tensor("q_23_cast_fp16")]; + tensor var_709 = const()[name = tensor("op_709"), val = tensor([1, 1500, 20, -1])]; + tensor var_710_cast_fp16 = reshape(shape = var_709, x = linear_31_cast_fp16)[name = tensor("op_710_cast_fp16")]; + tensor const_235_to_fp16 = const()[name = tensor("const_235_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_23_cast_fp16 = mul(x = var_710_cast_fp16, y = const_235_to_fp16)[name = tensor("k_23_cast_fp16")]; + tensor var_716 = const()[name = tensor("op_716"), val = tensor([1, 1500, 20, -1])]; + tensor var_717_cast_fp16 = reshape(shape = var_716, x = linear_32_cast_fp16)[name = tensor("op_717_cast_fp16")]; + tensor var_718 = const()[name = tensor("op_718"), val = tensor([0, 2, 1, 3])]; + tensor qk_11_transpose_x_0 = const()[name = tensor("qk_11_transpose_x_0"), val = tensor(false)]; + tensor qk_11_transpose_y_0 = const()[name = tensor("qk_11_transpose_y_0"), val = tensor(false)]; + tensor transpose_138_perm_0 = const()[name = tensor("transpose_138_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_139_perm_0 = const()[name = tensor("transpose_139_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_297 = transpose(perm = transpose_139_perm_0, x = k_23_cast_fp16)[name = tensor("transpose_297")]; + tensor transpose_298 = transpose(perm = transpose_138_perm_0, x = q_23_cast_fp16)[name = tensor("transpose_298")]; + tensor qk_11_cast_fp16 = matmul(transpose_x = qk_11_transpose_x_0, transpose_y = qk_11_transpose_y_0, x = transpose_298, y = transpose_297)[name = tensor("qk_11_cast_fp16")]; + tensor var_722_cast_fp16 = softmax(axis = var_657, x = qk_11_cast_fp16)[name = tensor("op_722_cast_fp16")]; + tensor var_724_transpose_x_0 = const()[name = tensor("op_724_transpose_x_0"), val = tensor(false)]; + tensor var_724_transpose_y_0 = const()[name = tensor("op_724_transpose_y_0"), val = tensor(false)]; + tensor transpose_299 = transpose(perm = var_718, x = var_717_cast_fp16)[name = tensor("transpose_299")]; + tensor var_724_cast_fp16 = matmul(transpose_x = var_724_transpose_x_0, transpose_y = var_724_transpose_y_0, x = var_722_cast_fp16, y = transpose_299)[name = tensor("op_724_cast_fp16")]; + tensor var_725 = const()[name = tensor("op_725"), val = tensor([0, 2, 1, 3])]; + tensor concat_5 = const()[name = tensor("concat_5"), val = tensor([1, 1500, 1280])]; + tensor transpose_296 = transpose(perm = var_725, x = var_724_cast_fp16)[name = tensor("transpose_296")]; + tensor x_71_cast_fp16 = reshape(shape = concat_5, x = transpose_296)[name = tensor("x_71_cast_fp16")]; + tensor var_730_to_fp16 = const()[name = tensor("op_730_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(221269056)))]; + tensor var_731_to_fp16 = const()[name = tensor("op_731_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224545920)))]; + tensor linear_33_cast_fp16 = linear(bias = var_731_to_fp16, weight = var_730_to_fp16, x = x_71_cast_fp16)[name = tensor("linear_33_cast_fp16")]; + tensor x_73_cast_fp16 = add(x = x_67_cast_fp16, y = linear_33_cast_fp16)[name = tensor("x_73_cast_fp16")]; + tensor var_738_axes_0 = const()[name = tensor("op_738_axes_0"), val = tensor([-1])]; + tensor blocks_5_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224548544)))]; + tensor blocks_5_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224551168)))]; + tensor var_738_cast_fp16 = layer_norm(axes = var_738_axes_0, beta = blocks_5_mlp_ln_bias_to_fp16, epsilon = var_663_to_fp16, gamma = blocks_5_mlp_ln_weight_to_fp16, x = x_73_cast_fp16)[name = tensor("op_738_cast_fp16")]; + tensor var_747_to_fp16 = const()[name = tensor("op_747_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224553792)))]; + tensor var_748_to_fp16 = const()[name = tensor("op_748_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237661056)))]; + tensor linear_34_cast_fp16 = linear(bias = var_748_to_fp16, weight = var_747_to_fp16, x = var_738_cast_fp16)[name = tensor("linear_34_cast_fp16")]; + tensor x_77_mode_0 = const()[name = tensor("x_77_mode_0"), val = tensor("EXACT")]; + tensor x_77_cast_fp16 = gelu(mode = x_77_mode_0, x = linear_34_cast_fp16)[name = tensor("x_77_cast_fp16")]; + tensor var_753_to_fp16 = const()[name = tensor("op_753_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237671360)))]; + tensor var_754_to_fp16 = const()[name = tensor("op_754_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250778624)))]; + tensor linear_35_cast_fp16 = linear(bias = var_754_to_fp16, weight = var_753_to_fp16, x = x_77_cast_fp16)[name = tensor("linear_35_cast_fp16")]; + tensor x_79_cast_fp16 = add(x = x_73_cast_fp16, y = linear_35_cast_fp16)[name = tensor("x_79_cast_fp16")]; + tensor var_764 = const()[name = tensor("op_764"), val = tensor(-1)]; + tensor var_781_axes_0 = const()[name = tensor("op_781_axes_0"), val = tensor([-1])]; + tensor blocks_6_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_6_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250781248)))]; + tensor blocks_6_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_6_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250783872)))]; + tensor var_770_to_fp16 = const()[name = tensor("op_770_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_781_cast_fp16 = layer_norm(axes = var_781_axes_0, beta = blocks_6_attn_ln_bias_to_fp16, epsilon = var_770_to_fp16, gamma = blocks_6_attn_ln_weight_to_fp16, x = x_79_cast_fp16)[name = tensor("op_781_cast_fp16")]; + tensor var_792_to_fp16 = const()[name = tensor("op_792_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250786496)))]; + tensor var_793_to_fp16 = const()[name = tensor("op_793_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(254063360)))]; + tensor linear_36_cast_fp16 = linear(bias = var_793_to_fp16, weight = var_792_to_fp16, x = var_781_cast_fp16)[name = tensor("linear_36_cast_fp16")]; + tensor var_796_to_fp16 = const()[name = tensor("op_796_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(254065984)))]; + tensor linear_37_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_796_to_fp16, x = var_781_cast_fp16)[name = tensor("linear_37_cast_fp16")]; + tensor var_800_to_fp16 = const()[name = tensor("op_800_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(257342848)))]; + tensor var_801_to_fp16 = const()[name = tensor("op_801_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260619712)))]; + tensor linear_38_cast_fp16 = linear(bias = var_801_to_fp16, weight = var_800_to_fp16, x = var_781_cast_fp16)[name = tensor("linear_38_cast_fp16")]; + tensor var_809 = const()[name = tensor("op_809"), val = tensor([1, 1500, 20, -1])]; + tensor var_810_cast_fp16 = reshape(shape = var_809, x = linear_36_cast_fp16)[name = tensor("op_810_cast_fp16")]; + tensor const_236_to_fp16 = const()[name = tensor("const_236_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_27_cast_fp16 = mul(x = var_810_cast_fp16, y = const_236_to_fp16)[name = tensor("q_27_cast_fp16")]; + tensor var_816 = const()[name = tensor("op_816"), val = tensor([1, 1500, 20, -1])]; + tensor var_817_cast_fp16 = reshape(shape = var_816, x = linear_37_cast_fp16)[name = tensor("op_817_cast_fp16")]; + tensor const_237_to_fp16 = const()[name = tensor("const_237_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_27_cast_fp16 = mul(x = var_817_cast_fp16, y = const_237_to_fp16)[name = tensor("k_27_cast_fp16")]; + tensor var_823 = const()[name = tensor("op_823"), val = tensor([1, 1500, 20, -1])]; + tensor var_824_cast_fp16 = reshape(shape = var_823, x = linear_38_cast_fp16)[name = tensor("op_824_cast_fp16")]; + tensor var_825 = const()[name = tensor("op_825"), val = tensor([0, 2, 1, 3])]; + tensor qk_13_transpose_x_0 = const()[name = tensor("qk_13_transpose_x_0"), val = tensor(false)]; + tensor qk_13_transpose_y_0 = const()[name = tensor("qk_13_transpose_y_0"), val = tensor(false)]; + tensor transpose_140_perm_0 = const()[name = tensor("transpose_140_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_141_perm_0 = const()[name = tensor("transpose_141_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_293 = transpose(perm = transpose_141_perm_0, x = k_27_cast_fp16)[name = tensor("transpose_293")]; + tensor transpose_294 = transpose(perm = transpose_140_perm_0, x = q_27_cast_fp16)[name = tensor("transpose_294")]; + tensor qk_13_cast_fp16 = matmul(transpose_x = qk_13_transpose_x_0, transpose_y = qk_13_transpose_y_0, x = transpose_294, y = transpose_293)[name = tensor("qk_13_cast_fp16")]; + tensor var_829_cast_fp16 = softmax(axis = var_764, x = qk_13_cast_fp16)[name = tensor("op_829_cast_fp16")]; + tensor var_831_transpose_x_0 = const()[name = tensor("op_831_transpose_x_0"), val = tensor(false)]; + tensor var_831_transpose_y_0 = const()[name = tensor("op_831_transpose_y_0"), val = tensor(false)]; + tensor transpose_295 = transpose(perm = var_825, x = var_824_cast_fp16)[name = tensor("transpose_295")]; + tensor var_831_cast_fp16 = matmul(transpose_x = var_831_transpose_x_0, transpose_y = var_831_transpose_y_0, x = var_829_cast_fp16, y = transpose_295)[name = tensor("op_831_cast_fp16")]; + tensor var_832 = const()[name = tensor("op_832"), val = tensor([0, 2, 1, 3])]; + tensor concat_6 = const()[name = tensor("concat_6"), val = tensor([1, 1500, 1280])]; + tensor transpose_292 = transpose(perm = var_832, x = var_831_cast_fp16)[name = tensor("transpose_292")]; + tensor x_83_cast_fp16 = reshape(shape = concat_6, x = transpose_292)[name = tensor("x_83_cast_fp16")]; + tensor var_837_to_fp16 = const()[name = tensor("op_837_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260622336)))]; + tensor var_838_to_fp16 = const()[name = tensor("op_838_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263899200)))]; + tensor linear_39_cast_fp16 = linear(bias = var_838_to_fp16, weight = var_837_to_fp16, x = x_83_cast_fp16)[name = tensor("linear_39_cast_fp16")]; + tensor x_85_cast_fp16 = add(x = x_79_cast_fp16, y = linear_39_cast_fp16)[name = tensor("x_85_cast_fp16")]; + tensor var_845_axes_0 = const()[name = tensor("op_845_axes_0"), val = tensor([-1])]; + tensor blocks_6_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_6_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263901824)))]; + tensor blocks_6_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_6_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263904448)))]; + tensor var_845_cast_fp16 = layer_norm(axes = var_845_axes_0, beta = blocks_6_mlp_ln_bias_to_fp16, epsilon = var_770_to_fp16, gamma = blocks_6_mlp_ln_weight_to_fp16, x = x_85_cast_fp16)[name = tensor("op_845_cast_fp16")]; + tensor var_854_to_fp16 = const()[name = tensor("op_854_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263907072)))]; + tensor var_855_to_fp16 = const()[name = tensor("op_855_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(277014336)))]; + tensor linear_40_cast_fp16 = linear(bias = var_855_to_fp16, weight = var_854_to_fp16, x = var_845_cast_fp16)[name = tensor("linear_40_cast_fp16")]; + tensor x_89_mode_0 = const()[name = tensor("x_89_mode_0"), val = tensor("EXACT")]; + tensor x_89_cast_fp16 = gelu(mode = x_89_mode_0, x = linear_40_cast_fp16)[name = tensor("x_89_cast_fp16")]; + tensor var_860_to_fp16 = const()[name = tensor("op_860_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(277024640)))]; + tensor var_861_to_fp16 = const()[name = tensor("op_861_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(290131904)))]; + tensor linear_41_cast_fp16 = linear(bias = var_861_to_fp16, weight = var_860_to_fp16, x = x_89_cast_fp16)[name = tensor("linear_41_cast_fp16")]; + tensor x_91_cast_fp16 = add(x = x_85_cast_fp16, y = linear_41_cast_fp16)[name = tensor("x_91_cast_fp16")]; + tensor var_871 = const()[name = tensor("op_871"), val = tensor(-1)]; + tensor var_888_axes_0 = const()[name = tensor("op_888_axes_0"), val = tensor([-1])]; + tensor blocks_7_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_7_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(290134528)))]; + tensor blocks_7_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_7_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(290137152)))]; + tensor var_877_to_fp16 = const()[name = tensor("op_877_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_888_cast_fp16 = layer_norm(axes = var_888_axes_0, beta = blocks_7_attn_ln_bias_to_fp16, epsilon = var_877_to_fp16, gamma = blocks_7_attn_ln_weight_to_fp16, x = x_91_cast_fp16)[name = tensor("op_888_cast_fp16")]; + tensor var_899_to_fp16 = const()[name = tensor("op_899_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(290139776)))]; + tensor var_900_to_fp16 = const()[name = tensor("op_900_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293416640)))]; + tensor linear_42_cast_fp16 = linear(bias = var_900_to_fp16, weight = var_899_to_fp16, x = var_888_cast_fp16)[name = tensor("linear_42_cast_fp16")]; + tensor var_903_to_fp16 = const()[name = tensor("op_903_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293419264)))]; + tensor linear_43_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_903_to_fp16, x = var_888_cast_fp16)[name = tensor("linear_43_cast_fp16")]; + tensor var_907_to_fp16 = const()[name = tensor("op_907_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(296696128)))]; + tensor var_908_to_fp16 = const()[name = tensor("op_908_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(299972992)))]; + tensor linear_44_cast_fp16 = linear(bias = var_908_to_fp16, weight = var_907_to_fp16, x = var_888_cast_fp16)[name = tensor("linear_44_cast_fp16")]; + tensor var_916 = const()[name = tensor("op_916"), val = tensor([1, 1500, 20, -1])]; + tensor var_917_cast_fp16 = reshape(shape = var_916, x = linear_42_cast_fp16)[name = tensor("op_917_cast_fp16")]; + tensor const_238_to_fp16 = const()[name = tensor("const_238_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_31_cast_fp16 = mul(x = var_917_cast_fp16, y = const_238_to_fp16)[name = tensor("q_31_cast_fp16")]; + tensor var_923 = const()[name = tensor("op_923"), val = tensor([1, 1500, 20, -1])]; + tensor var_924_cast_fp16 = reshape(shape = var_923, x = linear_43_cast_fp16)[name = tensor("op_924_cast_fp16")]; + tensor const_239_to_fp16 = const()[name = tensor("const_239_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_31_cast_fp16 = mul(x = var_924_cast_fp16, y = const_239_to_fp16)[name = tensor("k_31_cast_fp16")]; + tensor var_930 = const()[name = tensor("op_930"), val = tensor([1, 1500, 20, -1])]; + tensor var_931_cast_fp16 = reshape(shape = var_930, x = linear_44_cast_fp16)[name = tensor("op_931_cast_fp16")]; + tensor var_932 = const()[name = tensor("op_932"), val = tensor([0, 2, 1, 3])]; + tensor qk_15_transpose_x_0 = const()[name = tensor("qk_15_transpose_x_0"), val = tensor(false)]; + tensor qk_15_transpose_y_0 = const()[name = tensor("qk_15_transpose_y_0"), val = tensor(false)]; + tensor transpose_142_perm_0 = const()[name = tensor("transpose_142_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_143_perm_0 = const()[name = tensor("transpose_143_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_289 = transpose(perm = transpose_143_perm_0, x = k_31_cast_fp16)[name = tensor("transpose_289")]; + tensor transpose_290 = transpose(perm = transpose_142_perm_0, x = q_31_cast_fp16)[name = tensor("transpose_290")]; + tensor qk_15_cast_fp16 = matmul(transpose_x = qk_15_transpose_x_0, transpose_y = qk_15_transpose_y_0, x = transpose_290, y = transpose_289)[name = tensor("qk_15_cast_fp16")]; + tensor var_936_cast_fp16 = softmax(axis = var_871, x = qk_15_cast_fp16)[name = tensor("op_936_cast_fp16")]; + tensor var_938_transpose_x_0 = const()[name = tensor("op_938_transpose_x_0"), val = tensor(false)]; + tensor var_938_transpose_y_0 = const()[name = tensor("op_938_transpose_y_0"), val = tensor(false)]; + tensor transpose_291 = transpose(perm = var_932, x = var_931_cast_fp16)[name = tensor("transpose_291")]; + tensor var_938_cast_fp16 = matmul(transpose_x = var_938_transpose_x_0, transpose_y = var_938_transpose_y_0, x = var_936_cast_fp16, y = transpose_291)[name = tensor("op_938_cast_fp16")]; + tensor var_939 = const()[name = tensor("op_939"), val = tensor([0, 2, 1, 3])]; + tensor concat_7 = const()[name = tensor("concat_7"), val = tensor([1, 1500, 1280])]; + tensor transpose_288 = transpose(perm = var_939, x = var_938_cast_fp16)[name = tensor("transpose_288")]; + tensor x_95_cast_fp16 = reshape(shape = concat_7, x = transpose_288)[name = tensor("x_95_cast_fp16")]; + tensor var_944_to_fp16 = const()[name = tensor("op_944_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(299975616)))]; + tensor var_945_to_fp16 = const()[name = tensor("op_945_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303252480)))]; + tensor linear_45_cast_fp16 = linear(bias = var_945_to_fp16, weight = var_944_to_fp16, x = x_95_cast_fp16)[name = tensor("linear_45_cast_fp16")]; + tensor x_97_cast_fp16 = add(x = x_91_cast_fp16, y = linear_45_cast_fp16)[name = tensor("x_97_cast_fp16")]; + tensor var_952_axes_0 = const()[name = tensor("op_952_axes_0"), val = tensor([-1])]; + tensor blocks_7_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_7_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303255104)))]; + tensor blocks_7_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_7_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303257728)))]; + tensor var_952_cast_fp16 = layer_norm(axes = var_952_axes_0, beta = blocks_7_mlp_ln_bias_to_fp16, epsilon = var_877_to_fp16, gamma = blocks_7_mlp_ln_weight_to_fp16, x = x_97_cast_fp16)[name = tensor("op_952_cast_fp16")]; + tensor var_961_to_fp16 = const()[name = tensor("op_961_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303260352)))]; + tensor var_962_to_fp16 = const()[name = tensor("op_962_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316367616)))]; + tensor linear_46_cast_fp16 = linear(bias = var_962_to_fp16, weight = var_961_to_fp16, x = var_952_cast_fp16)[name = tensor("linear_46_cast_fp16")]; + tensor x_101_mode_0 = const()[name = tensor("x_101_mode_0"), val = tensor("EXACT")]; + tensor x_101_cast_fp16 = gelu(mode = x_101_mode_0, x = linear_46_cast_fp16)[name = tensor("x_101_cast_fp16")]; + tensor var_967_to_fp16 = const()[name = tensor("op_967_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316377920)))]; + tensor var_968_to_fp16 = const()[name = tensor("op_968_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329485184)))]; + tensor linear_47_cast_fp16 = linear(bias = var_968_to_fp16, weight = var_967_to_fp16, x = x_101_cast_fp16)[name = tensor("linear_47_cast_fp16")]; + tensor x_103_cast_fp16 = add(x = x_97_cast_fp16, y = linear_47_cast_fp16)[name = tensor("x_103_cast_fp16")]; + tensor var_978 = const()[name = tensor("op_978"), val = tensor(-1)]; + tensor var_995_axes_0 = const()[name = tensor("op_995_axes_0"), val = tensor([-1])]; + tensor blocks_8_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_8_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329487808)))]; + tensor blocks_8_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_8_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329490432)))]; + tensor var_984_to_fp16 = const()[name = tensor("op_984_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_995_cast_fp16 = layer_norm(axes = var_995_axes_0, beta = blocks_8_attn_ln_bias_to_fp16, epsilon = var_984_to_fp16, gamma = blocks_8_attn_ln_weight_to_fp16, x = x_103_cast_fp16)[name = tensor("op_995_cast_fp16")]; + tensor var_1006_to_fp16 = const()[name = tensor("op_1006_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329493056)))]; + tensor var_1007_to_fp16 = const()[name = tensor("op_1007_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(332769920)))]; + tensor linear_48_cast_fp16 = linear(bias = var_1007_to_fp16, weight = var_1006_to_fp16, x = var_995_cast_fp16)[name = tensor("linear_48_cast_fp16")]; + tensor var_1010_to_fp16 = const()[name = tensor("op_1010_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(332772544)))]; + tensor linear_49_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1010_to_fp16, x = var_995_cast_fp16)[name = tensor("linear_49_cast_fp16")]; + tensor var_1014_to_fp16 = const()[name = tensor("op_1014_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(336049408)))]; + tensor var_1015_to_fp16 = const()[name = tensor("op_1015_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339326272)))]; + tensor linear_50_cast_fp16 = linear(bias = var_1015_to_fp16, weight = var_1014_to_fp16, x = var_995_cast_fp16)[name = tensor("linear_50_cast_fp16")]; + tensor var_1023 = const()[name = tensor("op_1023"), val = tensor([1, 1500, 20, -1])]; + tensor var_1024_cast_fp16 = reshape(shape = var_1023, x = linear_48_cast_fp16)[name = tensor("op_1024_cast_fp16")]; + tensor const_240_to_fp16 = const()[name = tensor("const_240_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_35_cast_fp16 = mul(x = var_1024_cast_fp16, y = const_240_to_fp16)[name = tensor("q_35_cast_fp16")]; + tensor var_1030 = const()[name = tensor("op_1030"), val = tensor([1, 1500, 20, -1])]; + tensor var_1031_cast_fp16 = reshape(shape = var_1030, x = linear_49_cast_fp16)[name = tensor("op_1031_cast_fp16")]; + tensor const_241_to_fp16 = const()[name = tensor("const_241_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_35_cast_fp16 = mul(x = var_1031_cast_fp16, y = const_241_to_fp16)[name = tensor("k_35_cast_fp16")]; + tensor var_1037 = const()[name = tensor("op_1037"), val = tensor([1, 1500, 20, -1])]; + tensor var_1038_cast_fp16 = reshape(shape = var_1037, x = linear_50_cast_fp16)[name = tensor("op_1038_cast_fp16")]; + tensor var_1039 = const()[name = tensor("op_1039"), val = tensor([0, 2, 1, 3])]; + tensor qk_17_transpose_x_0 = const()[name = tensor("qk_17_transpose_x_0"), val = tensor(false)]; + tensor qk_17_transpose_y_0 = const()[name = tensor("qk_17_transpose_y_0"), val = tensor(false)]; + tensor transpose_144_perm_0 = const()[name = tensor("transpose_144_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_145_perm_0 = const()[name = tensor("transpose_145_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_285 = transpose(perm = transpose_145_perm_0, x = k_35_cast_fp16)[name = tensor("transpose_285")]; + tensor transpose_286 = transpose(perm = transpose_144_perm_0, x = q_35_cast_fp16)[name = tensor("transpose_286")]; + tensor qk_17_cast_fp16 = matmul(transpose_x = qk_17_transpose_x_0, transpose_y = qk_17_transpose_y_0, x = transpose_286, y = transpose_285)[name = tensor("qk_17_cast_fp16")]; + tensor var_1043_cast_fp16 = softmax(axis = var_978, x = qk_17_cast_fp16)[name = tensor("op_1043_cast_fp16")]; + tensor var_1045_transpose_x_0 = const()[name = tensor("op_1045_transpose_x_0"), val = tensor(false)]; + tensor var_1045_transpose_y_0 = const()[name = tensor("op_1045_transpose_y_0"), val = tensor(false)]; + tensor transpose_287 = transpose(perm = var_1039, x = var_1038_cast_fp16)[name = tensor("transpose_287")]; + tensor var_1045_cast_fp16 = matmul(transpose_x = var_1045_transpose_x_0, transpose_y = var_1045_transpose_y_0, x = var_1043_cast_fp16, y = transpose_287)[name = tensor("op_1045_cast_fp16")]; + tensor var_1046 = const()[name = tensor("op_1046"), val = tensor([0, 2, 1, 3])]; + tensor concat_8 = const()[name = tensor("concat_8"), val = tensor([1, 1500, 1280])]; + tensor transpose_284 = transpose(perm = var_1046, x = var_1045_cast_fp16)[name = tensor("transpose_284")]; + tensor x_107_cast_fp16 = reshape(shape = concat_8, x = transpose_284)[name = tensor("x_107_cast_fp16")]; + tensor var_1051_to_fp16 = const()[name = tensor("op_1051_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339328896)))]; + tensor var_1052_to_fp16 = const()[name = tensor("op_1052_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342605760)))]; + tensor linear_51_cast_fp16 = linear(bias = var_1052_to_fp16, weight = var_1051_to_fp16, x = x_107_cast_fp16)[name = tensor("linear_51_cast_fp16")]; + tensor x_109_cast_fp16 = add(x = x_103_cast_fp16, y = linear_51_cast_fp16)[name = tensor("x_109_cast_fp16")]; + tensor var_1059_axes_0 = const()[name = tensor("op_1059_axes_0"), val = tensor([-1])]; + tensor blocks_8_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_8_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342608384)))]; + tensor blocks_8_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_8_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342611008)))]; + tensor var_1059_cast_fp16 = layer_norm(axes = var_1059_axes_0, beta = blocks_8_mlp_ln_bias_to_fp16, epsilon = var_984_to_fp16, gamma = blocks_8_mlp_ln_weight_to_fp16, x = x_109_cast_fp16)[name = tensor("op_1059_cast_fp16")]; + tensor var_1068_to_fp16 = const()[name = tensor("op_1068_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342613632)))]; + tensor var_1069_to_fp16 = const()[name = tensor("op_1069_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(355720896)))]; + tensor linear_52_cast_fp16 = linear(bias = var_1069_to_fp16, weight = var_1068_to_fp16, x = var_1059_cast_fp16)[name = tensor("linear_52_cast_fp16")]; + tensor x_113_mode_0 = const()[name = tensor("x_113_mode_0"), val = tensor("EXACT")]; + tensor x_113_cast_fp16 = gelu(mode = x_113_mode_0, x = linear_52_cast_fp16)[name = tensor("x_113_cast_fp16")]; + tensor var_1074_to_fp16 = const()[name = tensor("op_1074_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(355731200)))]; + tensor var_1075_to_fp16 = const()[name = tensor("op_1075_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368838464)))]; + tensor linear_53_cast_fp16 = linear(bias = var_1075_to_fp16, weight = var_1074_to_fp16, x = x_113_cast_fp16)[name = tensor("linear_53_cast_fp16")]; + tensor x_115_cast_fp16 = add(x = x_109_cast_fp16, y = linear_53_cast_fp16)[name = tensor("x_115_cast_fp16")]; + tensor var_1085 = const()[name = tensor("op_1085"), val = tensor(-1)]; + tensor var_1102_axes_0 = const()[name = tensor("op_1102_axes_0"), val = tensor([-1])]; + tensor blocks_9_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_9_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368841088)))]; + tensor blocks_9_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_9_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368843712)))]; + tensor var_1091_to_fp16 = const()[name = tensor("op_1091_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1102_cast_fp16 = layer_norm(axes = var_1102_axes_0, beta = blocks_9_attn_ln_bias_to_fp16, epsilon = var_1091_to_fp16, gamma = blocks_9_attn_ln_weight_to_fp16, x = x_115_cast_fp16)[name = tensor("op_1102_cast_fp16")]; + tensor var_1113_to_fp16 = const()[name = tensor("op_1113_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368846336)))]; + tensor var_1114_to_fp16 = const()[name = tensor("op_1114_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(372123200)))]; + tensor linear_54_cast_fp16 = linear(bias = var_1114_to_fp16, weight = var_1113_to_fp16, x = var_1102_cast_fp16)[name = tensor("linear_54_cast_fp16")]; + tensor var_1117_to_fp16 = const()[name = tensor("op_1117_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(372125824)))]; + tensor linear_55_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1117_to_fp16, x = var_1102_cast_fp16)[name = tensor("linear_55_cast_fp16")]; + tensor var_1121_to_fp16 = const()[name = tensor("op_1121_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(375402688)))]; + tensor var_1122_to_fp16 = const()[name = tensor("op_1122_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(378679552)))]; + tensor linear_56_cast_fp16 = linear(bias = var_1122_to_fp16, weight = var_1121_to_fp16, x = var_1102_cast_fp16)[name = tensor("linear_56_cast_fp16")]; + tensor var_1130 = const()[name = tensor("op_1130"), val = tensor([1, 1500, 20, -1])]; + tensor var_1131_cast_fp16 = reshape(shape = var_1130, x = linear_54_cast_fp16)[name = tensor("op_1131_cast_fp16")]; + tensor const_242_to_fp16 = const()[name = tensor("const_242_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_39_cast_fp16 = mul(x = var_1131_cast_fp16, y = const_242_to_fp16)[name = tensor("q_39_cast_fp16")]; + tensor var_1137 = const()[name = tensor("op_1137"), val = tensor([1, 1500, 20, -1])]; + tensor var_1138_cast_fp16 = reshape(shape = var_1137, x = linear_55_cast_fp16)[name = tensor("op_1138_cast_fp16")]; + tensor const_243_to_fp16 = const()[name = tensor("const_243_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_39_cast_fp16 = mul(x = var_1138_cast_fp16, y = const_243_to_fp16)[name = tensor("k_39_cast_fp16")]; + tensor var_1144 = const()[name = tensor("op_1144"), val = tensor([1, 1500, 20, -1])]; + tensor var_1145_cast_fp16 = reshape(shape = var_1144, x = linear_56_cast_fp16)[name = tensor("op_1145_cast_fp16")]; + tensor var_1146 = const()[name = tensor("op_1146"), val = tensor([0, 2, 1, 3])]; + tensor qk_19_transpose_x_0 = const()[name = tensor("qk_19_transpose_x_0"), val = tensor(false)]; + tensor qk_19_transpose_y_0 = const()[name = tensor("qk_19_transpose_y_0"), val = tensor(false)]; + tensor transpose_146_perm_0 = const()[name = tensor("transpose_146_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_147_perm_0 = const()[name = tensor("transpose_147_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_281 = transpose(perm = transpose_147_perm_0, x = k_39_cast_fp16)[name = tensor("transpose_281")]; + tensor transpose_282 = transpose(perm = transpose_146_perm_0, x = q_39_cast_fp16)[name = tensor("transpose_282")]; + tensor qk_19_cast_fp16 = matmul(transpose_x = qk_19_transpose_x_0, transpose_y = qk_19_transpose_y_0, x = transpose_282, y = transpose_281)[name = tensor("qk_19_cast_fp16")]; + tensor var_1150_cast_fp16 = softmax(axis = var_1085, x = qk_19_cast_fp16)[name = tensor("op_1150_cast_fp16")]; + tensor var_1152_transpose_x_0 = const()[name = tensor("op_1152_transpose_x_0"), val = tensor(false)]; + tensor var_1152_transpose_y_0 = const()[name = tensor("op_1152_transpose_y_0"), val = tensor(false)]; + tensor transpose_283 = transpose(perm = var_1146, x = var_1145_cast_fp16)[name = tensor("transpose_283")]; + tensor var_1152_cast_fp16 = matmul(transpose_x = var_1152_transpose_x_0, transpose_y = var_1152_transpose_y_0, x = var_1150_cast_fp16, y = transpose_283)[name = tensor("op_1152_cast_fp16")]; + tensor var_1153 = const()[name = tensor("op_1153"), val = tensor([0, 2, 1, 3])]; + tensor concat_9 = const()[name = tensor("concat_9"), val = tensor([1, 1500, 1280])]; + tensor transpose_280 = transpose(perm = var_1153, x = var_1152_cast_fp16)[name = tensor("transpose_280")]; + tensor x_119_cast_fp16 = reshape(shape = concat_9, x = transpose_280)[name = tensor("x_119_cast_fp16")]; + tensor var_1158_to_fp16 = const()[name = tensor("op_1158_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(378682176)))]; + tensor var_1159_to_fp16 = const()[name = tensor("op_1159_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381959040)))]; + tensor linear_57_cast_fp16 = linear(bias = var_1159_to_fp16, weight = var_1158_to_fp16, x = x_119_cast_fp16)[name = tensor("linear_57_cast_fp16")]; + tensor x_121_cast_fp16 = add(x = x_115_cast_fp16, y = linear_57_cast_fp16)[name = tensor("x_121_cast_fp16")]; + tensor var_1166_axes_0 = const()[name = tensor("op_1166_axes_0"), val = tensor([-1])]; + tensor blocks_9_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_9_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381961664)))]; + tensor blocks_9_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_9_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381964288)))]; + tensor var_1166_cast_fp16 = layer_norm(axes = var_1166_axes_0, beta = blocks_9_mlp_ln_bias_to_fp16, epsilon = var_1091_to_fp16, gamma = blocks_9_mlp_ln_weight_to_fp16, x = x_121_cast_fp16)[name = tensor("op_1166_cast_fp16")]; + tensor var_1175_to_fp16 = const()[name = tensor("op_1175_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381966912)))]; + tensor var_1176_to_fp16 = const()[name = tensor("op_1176_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(395074176)))]; + tensor linear_58_cast_fp16 = linear(bias = var_1176_to_fp16, weight = var_1175_to_fp16, x = var_1166_cast_fp16)[name = tensor("linear_58_cast_fp16")]; + tensor x_125_mode_0 = const()[name = tensor("x_125_mode_0"), val = tensor("EXACT")]; + tensor x_125_cast_fp16 = gelu(mode = x_125_mode_0, x = linear_58_cast_fp16)[name = tensor("x_125_cast_fp16")]; + tensor var_1181_to_fp16 = const()[name = tensor("op_1181_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(395084480)))]; + tensor var_1182_to_fp16 = const()[name = tensor("op_1182_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(408191744)))]; + tensor linear_59_cast_fp16 = linear(bias = var_1182_to_fp16, weight = var_1181_to_fp16, x = x_125_cast_fp16)[name = tensor("linear_59_cast_fp16")]; + tensor x_127_cast_fp16 = add(x = x_121_cast_fp16, y = linear_59_cast_fp16)[name = tensor("x_127_cast_fp16")]; + tensor var_1192 = const()[name = tensor("op_1192"), val = tensor(-1)]; + tensor var_1209_axes_0 = const()[name = tensor("op_1209_axes_0"), val = tensor([-1])]; + tensor blocks_10_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_10_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(408194368)))]; + tensor blocks_10_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_10_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(408196992)))]; + tensor var_1198_to_fp16 = const()[name = tensor("op_1198_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1209_cast_fp16 = layer_norm(axes = var_1209_axes_0, beta = blocks_10_attn_ln_bias_to_fp16, epsilon = var_1198_to_fp16, gamma = blocks_10_attn_ln_weight_to_fp16, x = x_127_cast_fp16)[name = tensor("op_1209_cast_fp16")]; + tensor var_1220_to_fp16 = const()[name = tensor("op_1220_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(408199616)))]; + tensor var_1221_to_fp16 = const()[name = tensor("op_1221_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411476480)))]; + tensor linear_60_cast_fp16 = linear(bias = var_1221_to_fp16, weight = var_1220_to_fp16, x = var_1209_cast_fp16)[name = tensor("linear_60_cast_fp16")]; + tensor var_1224_to_fp16 = const()[name = tensor("op_1224_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411479104)))]; + tensor linear_61_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1224_to_fp16, x = var_1209_cast_fp16)[name = tensor("linear_61_cast_fp16")]; + tensor var_1228_to_fp16 = const()[name = tensor("op_1228_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(414755968)))]; + tensor var_1229_to_fp16 = const()[name = tensor("op_1229_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(418032832)))]; + tensor linear_62_cast_fp16 = linear(bias = var_1229_to_fp16, weight = var_1228_to_fp16, x = var_1209_cast_fp16)[name = tensor("linear_62_cast_fp16")]; + tensor var_1237 = const()[name = tensor("op_1237"), val = tensor([1, 1500, 20, -1])]; + tensor var_1238_cast_fp16 = reshape(shape = var_1237, x = linear_60_cast_fp16)[name = tensor("op_1238_cast_fp16")]; + tensor const_244_to_fp16 = const()[name = tensor("const_244_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_43_cast_fp16 = mul(x = var_1238_cast_fp16, y = const_244_to_fp16)[name = tensor("q_43_cast_fp16")]; + tensor var_1244 = const()[name = tensor("op_1244"), val = tensor([1, 1500, 20, -1])]; + tensor var_1245_cast_fp16 = reshape(shape = var_1244, x = linear_61_cast_fp16)[name = tensor("op_1245_cast_fp16")]; + tensor const_245_to_fp16 = const()[name = tensor("const_245_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_43_cast_fp16 = mul(x = var_1245_cast_fp16, y = const_245_to_fp16)[name = tensor("k_43_cast_fp16")]; + tensor var_1251 = const()[name = tensor("op_1251"), val = tensor([1, 1500, 20, -1])]; + tensor var_1252_cast_fp16 = reshape(shape = var_1251, x = linear_62_cast_fp16)[name = tensor("op_1252_cast_fp16")]; + tensor var_1253 = const()[name = tensor("op_1253"), val = tensor([0, 2, 1, 3])]; + tensor qk_21_transpose_x_0 = const()[name = tensor("qk_21_transpose_x_0"), val = tensor(false)]; + tensor qk_21_transpose_y_0 = const()[name = tensor("qk_21_transpose_y_0"), val = tensor(false)]; + tensor transpose_148_perm_0 = const()[name = tensor("transpose_148_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_149_perm_0 = const()[name = tensor("transpose_149_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_277 = transpose(perm = transpose_149_perm_0, x = k_43_cast_fp16)[name = tensor("transpose_277")]; + tensor transpose_278 = transpose(perm = transpose_148_perm_0, x = q_43_cast_fp16)[name = tensor("transpose_278")]; + tensor qk_21_cast_fp16 = matmul(transpose_x = qk_21_transpose_x_0, transpose_y = qk_21_transpose_y_0, x = transpose_278, y = transpose_277)[name = tensor("qk_21_cast_fp16")]; + tensor var_1257_cast_fp16 = softmax(axis = var_1192, x = qk_21_cast_fp16)[name = tensor("op_1257_cast_fp16")]; + tensor var_1259_transpose_x_0 = const()[name = tensor("op_1259_transpose_x_0"), val = tensor(false)]; + tensor var_1259_transpose_y_0 = const()[name = tensor("op_1259_transpose_y_0"), val = tensor(false)]; + tensor transpose_279 = transpose(perm = var_1253, x = var_1252_cast_fp16)[name = tensor("transpose_279")]; + tensor var_1259_cast_fp16 = matmul(transpose_x = var_1259_transpose_x_0, transpose_y = var_1259_transpose_y_0, x = var_1257_cast_fp16, y = transpose_279)[name = tensor("op_1259_cast_fp16")]; + tensor var_1260 = const()[name = tensor("op_1260"), val = tensor([0, 2, 1, 3])]; + tensor concat_10 = const()[name = tensor("concat_10"), val = tensor([1, 1500, 1280])]; + tensor transpose_276 = transpose(perm = var_1260, x = var_1259_cast_fp16)[name = tensor("transpose_276")]; + tensor x_131_cast_fp16 = reshape(shape = concat_10, x = transpose_276)[name = tensor("x_131_cast_fp16")]; + tensor var_1265_to_fp16 = const()[name = tensor("op_1265_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(418035456)))]; + tensor var_1266_to_fp16 = const()[name = tensor("op_1266_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(421312320)))]; + tensor linear_63_cast_fp16 = linear(bias = var_1266_to_fp16, weight = var_1265_to_fp16, x = x_131_cast_fp16)[name = tensor("linear_63_cast_fp16")]; + tensor x_133_cast_fp16 = add(x = x_127_cast_fp16, y = linear_63_cast_fp16)[name = tensor("x_133_cast_fp16")]; + tensor var_1273_axes_0 = const()[name = tensor("op_1273_axes_0"), val = tensor([-1])]; + tensor blocks_10_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_10_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(421314944)))]; + tensor blocks_10_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_10_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(421317568)))]; + tensor var_1273_cast_fp16 = layer_norm(axes = var_1273_axes_0, beta = blocks_10_mlp_ln_bias_to_fp16, epsilon = var_1198_to_fp16, gamma = blocks_10_mlp_ln_weight_to_fp16, x = x_133_cast_fp16)[name = tensor("op_1273_cast_fp16")]; + tensor var_1282_to_fp16 = const()[name = tensor("op_1282_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(421320192)))]; + tensor var_1283_to_fp16 = const()[name = tensor("op_1283_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(434427456)))]; + tensor linear_64_cast_fp16 = linear(bias = var_1283_to_fp16, weight = var_1282_to_fp16, x = var_1273_cast_fp16)[name = tensor("linear_64_cast_fp16")]; + tensor x_137_mode_0 = const()[name = tensor("x_137_mode_0"), val = tensor("EXACT")]; + tensor x_137_cast_fp16 = gelu(mode = x_137_mode_0, x = linear_64_cast_fp16)[name = tensor("x_137_cast_fp16")]; + tensor var_1288_to_fp16 = const()[name = tensor("op_1288_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(434437760)))]; + tensor var_1289_to_fp16 = const()[name = tensor("op_1289_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447545024)))]; + tensor linear_65_cast_fp16 = linear(bias = var_1289_to_fp16, weight = var_1288_to_fp16, x = x_137_cast_fp16)[name = tensor("linear_65_cast_fp16")]; + tensor x_139_cast_fp16 = add(x = x_133_cast_fp16, y = linear_65_cast_fp16)[name = tensor("x_139_cast_fp16")]; + tensor var_1299 = const()[name = tensor("op_1299"), val = tensor(-1)]; + tensor var_1316_axes_0 = const()[name = tensor("op_1316_axes_0"), val = tensor([-1])]; + tensor blocks_11_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_11_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447547648)))]; + tensor blocks_11_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_11_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447550272)))]; + tensor var_1305_to_fp16 = const()[name = tensor("op_1305_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1316_cast_fp16 = layer_norm(axes = var_1316_axes_0, beta = blocks_11_attn_ln_bias_to_fp16, epsilon = var_1305_to_fp16, gamma = blocks_11_attn_ln_weight_to_fp16, x = x_139_cast_fp16)[name = tensor("op_1316_cast_fp16")]; + tensor var_1327_to_fp16 = const()[name = tensor("op_1327_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447552896)))]; + tensor var_1328_to_fp16 = const()[name = tensor("op_1328_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(450829760)))]; + tensor linear_66_cast_fp16 = linear(bias = var_1328_to_fp16, weight = var_1327_to_fp16, x = var_1316_cast_fp16)[name = tensor("linear_66_cast_fp16")]; + tensor var_1331_to_fp16 = const()[name = tensor("op_1331_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(450832384)))]; + tensor linear_67_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1331_to_fp16, x = var_1316_cast_fp16)[name = tensor("linear_67_cast_fp16")]; + tensor var_1335_to_fp16 = const()[name = tensor("op_1335_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(454109248)))]; + tensor var_1336_to_fp16 = const()[name = tensor("op_1336_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(457386112)))]; + tensor linear_68_cast_fp16 = linear(bias = var_1336_to_fp16, weight = var_1335_to_fp16, x = var_1316_cast_fp16)[name = tensor("linear_68_cast_fp16")]; + tensor var_1344 = const()[name = tensor("op_1344"), val = tensor([1, 1500, 20, -1])]; + tensor var_1345_cast_fp16 = reshape(shape = var_1344, x = linear_66_cast_fp16)[name = tensor("op_1345_cast_fp16")]; + tensor const_246_to_fp16 = const()[name = tensor("const_246_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_47_cast_fp16 = mul(x = var_1345_cast_fp16, y = const_246_to_fp16)[name = tensor("q_47_cast_fp16")]; + tensor var_1351 = const()[name = tensor("op_1351"), val = tensor([1, 1500, 20, -1])]; + tensor var_1352_cast_fp16 = reshape(shape = var_1351, x = linear_67_cast_fp16)[name = tensor("op_1352_cast_fp16")]; + tensor const_247_to_fp16 = const()[name = tensor("const_247_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_47_cast_fp16 = mul(x = var_1352_cast_fp16, y = const_247_to_fp16)[name = tensor("k_47_cast_fp16")]; + tensor var_1358 = const()[name = tensor("op_1358"), val = tensor([1, 1500, 20, -1])]; + tensor var_1359_cast_fp16 = reshape(shape = var_1358, x = linear_68_cast_fp16)[name = tensor("op_1359_cast_fp16")]; + tensor var_1360 = const()[name = tensor("op_1360"), val = tensor([0, 2, 1, 3])]; + tensor qk_23_transpose_x_0 = const()[name = tensor("qk_23_transpose_x_0"), val = tensor(false)]; + tensor qk_23_transpose_y_0 = const()[name = tensor("qk_23_transpose_y_0"), val = tensor(false)]; + tensor transpose_150_perm_0 = const()[name = tensor("transpose_150_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_151_perm_0 = const()[name = tensor("transpose_151_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_273 = transpose(perm = transpose_151_perm_0, x = k_47_cast_fp16)[name = tensor("transpose_273")]; + tensor transpose_274 = transpose(perm = transpose_150_perm_0, x = q_47_cast_fp16)[name = tensor("transpose_274")]; + tensor qk_23_cast_fp16 = matmul(transpose_x = qk_23_transpose_x_0, transpose_y = qk_23_transpose_y_0, x = transpose_274, y = transpose_273)[name = tensor("qk_23_cast_fp16")]; + tensor var_1364_cast_fp16 = softmax(axis = var_1299, x = qk_23_cast_fp16)[name = tensor("op_1364_cast_fp16")]; + tensor var_1366_transpose_x_0 = const()[name = tensor("op_1366_transpose_x_0"), val = tensor(false)]; + tensor var_1366_transpose_y_0 = const()[name = tensor("op_1366_transpose_y_0"), val = tensor(false)]; + tensor transpose_275 = transpose(perm = var_1360, x = var_1359_cast_fp16)[name = tensor("transpose_275")]; + tensor var_1366_cast_fp16 = matmul(transpose_x = var_1366_transpose_x_0, transpose_y = var_1366_transpose_y_0, x = var_1364_cast_fp16, y = transpose_275)[name = tensor("op_1366_cast_fp16")]; + tensor var_1367 = const()[name = tensor("op_1367"), val = tensor([0, 2, 1, 3])]; + tensor concat_11 = const()[name = tensor("concat_11"), val = tensor([1, 1500, 1280])]; + tensor transpose_272 = transpose(perm = var_1367, x = var_1366_cast_fp16)[name = tensor("transpose_272")]; + tensor x_143_cast_fp16 = reshape(shape = concat_11, x = transpose_272)[name = tensor("x_143_cast_fp16")]; + tensor var_1372_to_fp16 = const()[name = tensor("op_1372_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(457388736)))]; + tensor var_1373_to_fp16 = const()[name = tensor("op_1373_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460665600)))]; + tensor linear_69_cast_fp16 = linear(bias = var_1373_to_fp16, weight = var_1372_to_fp16, x = x_143_cast_fp16)[name = tensor("linear_69_cast_fp16")]; + tensor x_145_cast_fp16 = add(x = x_139_cast_fp16, y = linear_69_cast_fp16)[name = tensor("x_145_cast_fp16")]; + tensor var_1380_axes_0 = const()[name = tensor("op_1380_axes_0"), val = tensor([-1])]; + tensor blocks_11_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_11_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460668224)))]; + tensor blocks_11_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_11_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460670848)))]; + tensor var_1380_cast_fp16 = layer_norm(axes = var_1380_axes_0, beta = blocks_11_mlp_ln_bias_to_fp16, epsilon = var_1305_to_fp16, gamma = blocks_11_mlp_ln_weight_to_fp16, x = x_145_cast_fp16)[name = tensor("op_1380_cast_fp16")]; + tensor var_1389_to_fp16 = const()[name = tensor("op_1389_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460673472)))]; + tensor var_1390_to_fp16 = const()[name = tensor("op_1390_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473780736)))]; + tensor linear_70_cast_fp16 = linear(bias = var_1390_to_fp16, weight = var_1389_to_fp16, x = var_1380_cast_fp16)[name = tensor("linear_70_cast_fp16")]; + tensor x_149_mode_0 = const()[name = tensor("x_149_mode_0"), val = tensor("EXACT")]; + tensor x_149_cast_fp16 = gelu(mode = x_149_mode_0, x = linear_70_cast_fp16)[name = tensor("x_149_cast_fp16")]; + tensor var_1395_to_fp16 = const()[name = tensor("op_1395_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473791040)))]; + tensor var_1396_to_fp16 = const()[name = tensor("op_1396_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486898304)))]; + tensor linear_71_cast_fp16 = linear(bias = var_1396_to_fp16, weight = var_1395_to_fp16, x = x_149_cast_fp16)[name = tensor("linear_71_cast_fp16")]; + tensor x_151_cast_fp16 = add(x = x_145_cast_fp16, y = linear_71_cast_fp16)[name = tensor("x_151_cast_fp16")]; + tensor var_1406 = const()[name = tensor("op_1406"), val = tensor(-1)]; + tensor var_1423_axes_0 = const()[name = tensor("op_1423_axes_0"), val = tensor([-1])]; + tensor blocks_12_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_12_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486900928)))]; + tensor blocks_12_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_12_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486903552)))]; + tensor var_1412_to_fp16 = const()[name = tensor("op_1412_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1423_cast_fp16 = layer_norm(axes = var_1423_axes_0, beta = blocks_12_attn_ln_bias_to_fp16, epsilon = var_1412_to_fp16, gamma = blocks_12_attn_ln_weight_to_fp16, x = x_151_cast_fp16)[name = tensor("op_1423_cast_fp16")]; + tensor var_1434_to_fp16 = const()[name = tensor("op_1434_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486906176)))]; + tensor var_1435_to_fp16 = const()[name = tensor("op_1435_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(490183040)))]; + tensor linear_72_cast_fp16 = linear(bias = var_1435_to_fp16, weight = var_1434_to_fp16, x = var_1423_cast_fp16)[name = tensor("linear_72_cast_fp16")]; + tensor var_1438_to_fp16 = const()[name = tensor("op_1438_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(490185664)))]; + tensor linear_73_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1438_to_fp16, x = var_1423_cast_fp16)[name = tensor("linear_73_cast_fp16")]; + tensor var_1442_to_fp16 = const()[name = tensor("op_1442_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(493462528)))]; + tensor var_1443_to_fp16 = const()[name = tensor("op_1443_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496739392)))]; + tensor linear_74_cast_fp16 = linear(bias = var_1443_to_fp16, weight = var_1442_to_fp16, x = var_1423_cast_fp16)[name = tensor("linear_74_cast_fp16")]; + tensor var_1451 = const()[name = tensor("op_1451"), val = tensor([1, 1500, 20, -1])]; + tensor var_1452_cast_fp16 = reshape(shape = var_1451, x = linear_72_cast_fp16)[name = tensor("op_1452_cast_fp16")]; + tensor const_248_to_fp16 = const()[name = tensor("const_248_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_51_cast_fp16 = mul(x = var_1452_cast_fp16, y = const_248_to_fp16)[name = tensor("q_51_cast_fp16")]; + tensor var_1458 = const()[name = tensor("op_1458"), val = tensor([1, 1500, 20, -1])]; + tensor var_1459_cast_fp16 = reshape(shape = var_1458, x = linear_73_cast_fp16)[name = tensor("op_1459_cast_fp16")]; + tensor const_249_to_fp16 = const()[name = tensor("const_249_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_51_cast_fp16 = mul(x = var_1459_cast_fp16, y = const_249_to_fp16)[name = tensor("k_51_cast_fp16")]; + tensor var_1465 = const()[name = tensor("op_1465"), val = tensor([1, 1500, 20, -1])]; + tensor var_1466_cast_fp16 = reshape(shape = var_1465, x = linear_74_cast_fp16)[name = tensor("op_1466_cast_fp16")]; + tensor var_1467 = const()[name = tensor("op_1467"), val = tensor([0, 2, 1, 3])]; + tensor qk_25_transpose_x_0 = const()[name = tensor("qk_25_transpose_x_0"), val = tensor(false)]; + tensor qk_25_transpose_y_0 = const()[name = tensor("qk_25_transpose_y_0"), val = tensor(false)]; + tensor transpose_152_perm_0 = const()[name = tensor("transpose_152_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_153_perm_0 = const()[name = tensor("transpose_153_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_269 = transpose(perm = transpose_153_perm_0, x = k_51_cast_fp16)[name = tensor("transpose_269")]; + tensor transpose_270 = transpose(perm = transpose_152_perm_0, x = q_51_cast_fp16)[name = tensor("transpose_270")]; + tensor qk_25_cast_fp16 = matmul(transpose_x = qk_25_transpose_x_0, transpose_y = qk_25_transpose_y_0, x = transpose_270, y = transpose_269)[name = tensor("qk_25_cast_fp16")]; + tensor var_1471_cast_fp16 = softmax(axis = var_1406, x = qk_25_cast_fp16)[name = tensor("op_1471_cast_fp16")]; + tensor var_1473_transpose_x_0 = const()[name = tensor("op_1473_transpose_x_0"), val = tensor(false)]; + tensor var_1473_transpose_y_0 = const()[name = tensor("op_1473_transpose_y_0"), val = tensor(false)]; + tensor transpose_271 = transpose(perm = var_1467, x = var_1466_cast_fp16)[name = tensor("transpose_271")]; + tensor var_1473_cast_fp16 = matmul(transpose_x = var_1473_transpose_x_0, transpose_y = var_1473_transpose_y_0, x = var_1471_cast_fp16, y = transpose_271)[name = tensor("op_1473_cast_fp16")]; + tensor var_1474 = const()[name = tensor("op_1474"), val = tensor([0, 2, 1, 3])]; + tensor concat_12 = const()[name = tensor("concat_12"), val = tensor([1, 1500, 1280])]; + tensor transpose_268 = transpose(perm = var_1474, x = var_1473_cast_fp16)[name = tensor("transpose_268")]; + tensor x_155_cast_fp16 = reshape(shape = concat_12, x = transpose_268)[name = tensor("x_155_cast_fp16")]; + tensor var_1479_to_fp16 = const()[name = tensor("op_1479_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496742016)))]; + tensor var_1480_to_fp16 = const()[name = tensor("op_1480_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(500018880)))]; + tensor linear_75_cast_fp16 = linear(bias = var_1480_to_fp16, weight = var_1479_to_fp16, x = x_155_cast_fp16)[name = tensor("linear_75_cast_fp16")]; + tensor x_157_cast_fp16 = add(x = x_151_cast_fp16, y = linear_75_cast_fp16)[name = tensor("x_157_cast_fp16")]; + tensor var_1487_axes_0 = const()[name = tensor("op_1487_axes_0"), val = tensor([-1])]; + tensor blocks_12_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_12_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(500021504)))]; + tensor blocks_12_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_12_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(500024128)))]; + tensor var_1487_cast_fp16 = layer_norm(axes = var_1487_axes_0, beta = blocks_12_mlp_ln_bias_to_fp16, epsilon = var_1412_to_fp16, gamma = blocks_12_mlp_ln_weight_to_fp16, x = x_157_cast_fp16)[name = tensor("op_1487_cast_fp16")]; + tensor var_1496_to_fp16 = const()[name = tensor("op_1496_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(500026752)))]; + tensor var_1497_to_fp16 = const()[name = tensor("op_1497_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(513134016)))]; + tensor linear_76_cast_fp16 = linear(bias = var_1497_to_fp16, weight = var_1496_to_fp16, x = var_1487_cast_fp16)[name = tensor("linear_76_cast_fp16")]; + tensor x_161_mode_0 = const()[name = tensor("x_161_mode_0"), val = tensor("EXACT")]; + tensor x_161_cast_fp16 = gelu(mode = x_161_mode_0, x = linear_76_cast_fp16)[name = tensor("x_161_cast_fp16")]; + tensor var_1502_to_fp16 = const()[name = tensor("op_1502_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(513144320)))]; + tensor var_1503_to_fp16 = const()[name = tensor("op_1503_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(526251584)))]; + tensor linear_77_cast_fp16 = linear(bias = var_1503_to_fp16, weight = var_1502_to_fp16, x = x_161_cast_fp16)[name = tensor("linear_77_cast_fp16")]; + tensor x_163_cast_fp16 = add(x = x_157_cast_fp16, y = linear_77_cast_fp16)[name = tensor("x_163_cast_fp16")]; + tensor var_1513 = const()[name = tensor("op_1513"), val = tensor(-1)]; + tensor var_1530_axes_0 = const()[name = tensor("op_1530_axes_0"), val = tensor([-1])]; + tensor blocks_13_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_13_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(526254208)))]; + tensor blocks_13_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_13_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(526256832)))]; + tensor var_1519_to_fp16 = const()[name = tensor("op_1519_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1530_cast_fp16 = layer_norm(axes = var_1530_axes_0, beta = blocks_13_attn_ln_bias_to_fp16, epsilon = var_1519_to_fp16, gamma = blocks_13_attn_ln_weight_to_fp16, x = x_163_cast_fp16)[name = tensor("op_1530_cast_fp16")]; + tensor var_1541_to_fp16 = const()[name = tensor("op_1541_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(526259456)))]; + tensor var_1542_to_fp16 = const()[name = tensor("op_1542_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529536320)))]; + tensor linear_78_cast_fp16 = linear(bias = var_1542_to_fp16, weight = var_1541_to_fp16, x = var_1530_cast_fp16)[name = tensor("linear_78_cast_fp16")]; + tensor var_1545_to_fp16 = const()[name = tensor("op_1545_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529538944)))]; + tensor linear_79_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1545_to_fp16, x = var_1530_cast_fp16)[name = tensor("linear_79_cast_fp16")]; + tensor var_1549_to_fp16 = const()[name = tensor("op_1549_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(532815808)))]; + tensor var_1550_to_fp16 = const()[name = tensor("op_1550_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(536092672)))]; + tensor linear_80_cast_fp16 = linear(bias = var_1550_to_fp16, weight = var_1549_to_fp16, x = var_1530_cast_fp16)[name = tensor("linear_80_cast_fp16")]; + tensor var_1558 = const()[name = tensor("op_1558"), val = tensor([1, 1500, 20, -1])]; + tensor var_1559_cast_fp16 = reshape(shape = var_1558, x = linear_78_cast_fp16)[name = tensor("op_1559_cast_fp16")]; + tensor const_250_to_fp16 = const()[name = tensor("const_250_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_55_cast_fp16 = mul(x = var_1559_cast_fp16, y = const_250_to_fp16)[name = tensor("q_55_cast_fp16")]; + tensor var_1565 = const()[name = tensor("op_1565"), val = tensor([1, 1500, 20, -1])]; + tensor var_1566_cast_fp16 = reshape(shape = var_1565, x = linear_79_cast_fp16)[name = tensor("op_1566_cast_fp16")]; + tensor const_251_to_fp16 = const()[name = tensor("const_251_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_55_cast_fp16 = mul(x = var_1566_cast_fp16, y = const_251_to_fp16)[name = tensor("k_55_cast_fp16")]; + tensor var_1572 = const()[name = tensor("op_1572"), val = tensor([1, 1500, 20, -1])]; + tensor var_1573_cast_fp16 = reshape(shape = var_1572, x = linear_80_cast_fp16)[name = tensor("op_1573_cast_fp16")]; + tensor var_1574 = const()[name = tensor("op_1574"), val = tensor([0, 2, 1, 3])]; + tensor qk_27_transpose_x_0 = const()[name = tensor("qk_27_transpose_x_0"), val = tensor(false)]; + tensor qk_27_transpose_y_0 = const()[name = tensor("qk_27_transpose_y_0"), val = tensor(false)]; + tensor transpose_154_perm_0 = const()[name = tensor("transpose_154_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_155_perm_0 = const()[name = tensor("transpose_155_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_265 = transpose(perm = transpose_155_perm_0, x = k_55_cast_fp16)[name = tensor("transpose_265")]; + tensor transpose_266 = transpose(perm = transpose_154_perm_0, x = q_55_cast_fp16)[name = tensor("transpose_266")]; + tensor qk_27_cast_fp16 = matmul(transpose_x = qk_27_transpose_x_0, transpose_y = qk_27_transpose_y_0, x = transpose_266, y = transpose_265)[name = tensor("qk_27_cast_fp16")]; + tensor var_1578_cast_fp16 = softmax(axis = var_1513, x = qk_27_cast_fp16)[name = tensor("op_1578_cast_fp16")]; + tensor var_1580_transpose_x_0 = const()[name = tensor("op_1580_transpose_x_0"), val = tensor(false)]; + tensor var_1580_transpose_y_0 = const()[name = tensor("op_1580_transpose_y_0"), val = tensor(false)]; + tensor transpose_267 = transpose(perm = var_1574, x = var_1573_cast_fp16)[name = tensor("transpose_267")]; + tensor var_1580_cast_fp16 = matmul(transpose_x = var_1580_transpose_x_0, transpose_y = var_1580_transpose_y_0, x = var_1578_cast_fp16, y = transpose_267)[name = tensor("op_1580_cast_fp16")]; + tensor var_1581 = const()[name = tensor("op_1581"), val = tensor([0, 2, 1, 3])]; + tensor concat_13 = const()[name = tensor("concat_13"), val = tensor([1, 1500, 1280])]; + tensor transpose_264 = transpose(perm = var_1581, x = var_1580_cast_fp16)[name = tensor("transpose_264")]; + tensor x_167_cast_fp16 = reshape(shape = concat_13, x = transpose_264)[name = tensor("x_167_cast_fp16")]; + tensor var_1586_to_fp16 = const()[name = tensor("op_1586_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(536095296)))]; + tensor var_1587_to_fp16 = const()[name = tensor("op_1587_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(539372160)))]; + tensor linear_81_cast_fp16 = linear(bias = var_1587_to_fp16, weight = var_1586_to_fp16, x = x_167_cast_fp16)[name = tensor("linear_81_cast_fp16")]; + tensor x_169_cast_fp16 = add(x = x_163_cast_fp16, y = linear_81_cast_fp16)[name = tensor("x_169_cast_fp16")]; + tensor var_1594_axes_0 = const()[name = tensor("op_1594_axes_0"), val = tensor([-1])]; + tensor blocks_13_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_13_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(539374784)))]; + tensor blocks_13_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_13_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(539377408)))]; + tensor var_1594_cast_fp16 = layer_norm(axes = var_1594_axes_0, beta = blocks_13_mlp_ln_bias_to_fp16, epsilon = var_1519_to_fp16, gamma = blocks_13_mlp_ln_weight_to_fp16, x = x_169_cast_fp16)[name = tensor("op_1594_cast_fp16")]; + tensor var_1603_to_fp16 = const()[name = tensor("op_1603_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(539380032)))]; + tensor var_1604_to_fp16 = const()[name = tensor("op_1604_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552487296)))]; + tensor linear_82_cast_fp16 = linear(bias = var_1604_to_fp16, weight = var_1603_to_fp16, x = var_1594_cast_fp16)[name = tensor("linear_82_cast_fp16")]; + tensor x_173_mode_0 = const()[name = tensor("x_173_mode_0"), val = tensor("EXACT")]; + tensor x_173_cast_fp16 = gelu(mode = x_173_mode_0, x = linear_82_cast_fp16)[name = tensor("x_173_cast_fp16")]; + tensor var_1609_to_fp16 = const()[name = tensor("op_1609_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552497600)))]; + tensor var_1610_to_fp16 = const()[name = tensor("op_1610_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565604864)))]; + tensor linear_83_cast_fp16 = linear(bias = var_1610_to_fp16, weight = var_1609_to_fp16, x = x_173_cast_fp16)[name = tensor("linear_83_cast_fp16")]; + tensor x_175_cast_fp16 = add(x = x_169_cast_fp16, y = linear_83_cast_fp16)[name = tensor("x_175_cast_fp16")]; + tensor var_1620 = const()[name = tensor("op_1620"), val = tensor(-1)]; + tensor var_1637_axes_0 = const()[name = tensor("op_1637_axes_0"), val = tensor([-1])]; + tensor blocks_14_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_14_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565607488)))]; + tensor blocks_14_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_14_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565610112)))]; + tensor var_1626_to_fp16 = const()[name = tensor("op_1626_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1637_cast_fp16 = layer_norm(axes = var_1637_axes_0, beta = blocks_14_attn_ln_bias_to_fp16, epsilon = var_1626_to_fp16, gamma = blocks_14_attn_ln_weight_to_fp16, x = x_175_cast_fp16)[name = tensor("op_1637_cast_fp16")]; + tensor var_1648_to_fp16 = const()[name = tensor("op_1648_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565612736)))]; + tensor var_1649_to_fp16 = const()[name = tensor("op_1649_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568889600)))]; + tensor linear_84_cast_fp16 = linear(bias = var_1649_to_fp16, weight = var_1648_to_fp16, x = var_1637_cast_fp16)[name = tensor("linear_84_cast_fp16")]; + tensor var_1652_to_fp16 = const()[name = tensor("op_1652_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568892224)))]; + tensor linear_85_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1652_to_fp16, x = var_1637_cast_fp16)[name = tensor("linear_85_cast_fp16")]; + tensor var_1656_to_fp16 = const()[name = tensor("op_1656_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(572169088)))]; + tensor var_1657_to_fp16 = const()[name = tensor("op_1657_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(575445952)))]; + tensor linear_86_cast_fp16 = linear(bias = var_1657_to_fp16, weight = var_1656_to_fp16, x = var_1637_cast_fp16)[name = tensor("linear_86_cast_fp16")]; + tensor var_1665 = const()[name = tensor("op_1665"), val = tensor([1, 1500, 20, -1])]; + tensor var_1666_cast_fp16 = reshape(shape = var_1665, x = linear_84_cast_fp16)[name = tensor("op_1666_cast_fp16")]; + tensor const_252_to_fp16 = const()[name = tensor("const_252_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_59_cast_fp16 = mul(x = var_1666_cast_fp16, y = const_252_to_fp16)[name = tensor("q_59_cast_fp16")]; + tensor var_1672 = const()[name = tensor("op_1672"), val = tensor([1, 1500, 20, -1])]; + tensor var_1673_cast_fp16 = reshape(shape = var_1672, x = linear_85_cast_fp16)[name = tensor("op_1673_cast_fp16")]; + tensor const_253_to_fp16 = const()[name = tensor("const_253_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_59_cast_fp16 = mul(x = var_1673_cast_fp16, y = const_253_to_fp16)[name = tensor("k_59_cast_fp16")]; + tensor var_1679 = const()[name = tensor("op_1679"), val = tensor([1, 1500, 20, -1])]; + tensor var_1680_cast_fp16 = reshape(shape = var_1679, x = linear_86_cast_fp16)[name = tensor("op_1680_cast_fp16")]; + tensor var_1681 = const()[name = tensor("op_1681"), val = tensor([0, 2, 1, 3])]; + tensor qk_29_transpose_x_0 = const()[name = tensor("qk_29_transpose_x_0"), val = tensor(false)]; + tensor qk_29_transpose_y_0 = const()[name = tensor("qk_29_transpose_y_0"), val = tensor(false)]; + tensor transpose_156_perm_0 = const()[name = tensor("transpose_156_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_157_perm_0 = const()[name = tensor("transpose_157_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_261 = transpose(perm = transpose_157_perm_0, x = k_59_cast_fp16)[name = tensor("transpose_261")]; + tensor transpose_262 = transpose(perm = transpose_156_perm_0, x = q_59_cast_fp16)[name = tensor("transpose_262")]; + tensor qk_29_cast_fp16 = matmul(transpose_x = qk_29_transpose_x_0, transpose_y = qk_29_transpose_y_0, x = transpose_262, y = transpose_261)[name = tensor("qk_29_cast_fp16")]; + tensor var_1685_cast_fp16 = softmax(axis = var_1620, x = qk_29_cast_fp16)[name = tensor("op_1685_cast_fp16")]; + tensor var_1687_transpose_x_0 = const()[name = tensor("op_1687_transpose_x_0"), val = tensor(false)]; + tensor var_1687_transpose_y_0 = const()[name = tensor("op_1687_transpose_y_0"), val = tensor(false)]; + tensor transpose_263 = transpose(perm = var_1681, x = var_1680_cast_fp16)[name = tensor("transpose_263")]; + tensor var_1687_cast_fp16 = matmul(transpose_x = var_1687_transpose_x_0, transpose_y = var_1687_transpose_y_0, x = var_1685_cast_fp16, y = transpose_263)[name = tensor("op_1687_cast_fp16")]; + tensor var_1688 = const()[name = tensor("op_1688"), val = tensor([0, 2, 1, 3])]; + tensor concat_14 = const()[name = tensor("concat_14"), val = tensor([1, 1500, 1280])]; + tensor transpose_260 = transpose(perm = var_1688, x = var_1687_cast_fp16)[name = tensor("transpose_260")]; + tensor x_179_cast_fp16 = reshape(shape = concat_14, x = transpose_260)[name = tensor("x_179_cast_fp16")]; + tensor var_1693_to_fp16 = const()[name = tensor("op_1693_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(575448576)))]; + tensor var_1694_to_fp16 = const()[name = tensor("op_1694_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578725440)))]; + tensor linear_87_cast_fp16 = linear(bias = var_1694_to_fp16, weight = var_1693_to_fp16, x = x_179_cast_fp16)[name = tensor("linear_87_cast_fp16")]; + tensor x_181_cast_fp16 = add(x = x_175_cast_fp16, y = linear_87_cast_fp16)[name = tensor("x_181_cast_fp16")]; + tensor var_1701_axes_0 = const()[name = tensor("op_1701_axes_0"), val = tensor([-1])]; + tensor blocks_14_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_14_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578728064)))]; + tensor blocks_14_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_14_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578730688)))]; + tensor var_1701_cast_fp16 = layer_norm(axes = var_1701_axes_0, beta = blocks_14_mlp_ln_bias_to_fp16, epsilon = var_1626_to_fp16, gamma = blocks_14_mlp_ln_weight_to_fp16, x = x_181_cast_fp16)[name = tensor("op_1701_cast_fp16")]; + tensor var_1710_to_fp16 = const()[name = tensor("op_1710_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578733312)))]; + tensor var_1711_to_fp16 = const()[name = tensor("op_1711_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(591840576)))]; + tensor linear_88_cast_fp16 = linear(bias = var_1711_to_fp16, weight = var_1710_to_fp16, x = var_1701_cast_fp16)[name = tensor("linear_88_cast_fp16")]; + tensor x_185_mode_0 = const()[name = tensor("x_185_mode_0"), val = tensor("EXACT")]; + tensor x_185_cast_fp16 = gelu(mode = x_185_mode_0, x = linear_88_cast_fp16)[name = tensor("x_185_cast_fp16")]; + tensor var_1716_to_fp16 = const()[name = tensor("op_1716_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(591850880)))]; + tensor var_1717_to_fp16 = const()[name = tensor("op_1717_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604958144)))]; + tensor linear_89_cast_fp16 = linear(bias = var_1717_to_fp16, weight = var_1716_to_fp16, x = x_185_cast_fp16)[name = tensor("linear_89_cast_fp16")]; + tensor x_187_cast_fp16 = add(x = x_181_cast_fp16, y = linear_89_cast_fp16)[name = tensor("x_187_cast_fp16")]; + tensor var_1727 = const()[name = tensor("op_1727"), val = tensor(-1)]; + tensor var_1744_axes_0 = const()[name = tensor("op_1744_axes_0"), val = tensor([-1])]; + tensor blocks_15_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_15_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604960768)))]; + tensor blocks_15_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_15_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604963392)))]; + tensor var_1733_to_fp16 = const()[name = tensor("op_1733_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1744_cast_fp16 = layer_norm(axes = var_1744_axes_0, beta = blocks_15_attn_ln_bias_to_fp16, epsilon = var_1733_to_fp16, gamma = blocks_15_attn_ln_weight_to_fp16, x = x_187_cast_fp16)[name = tensor("op_1744_cast_fp16")]; + tensor var_1755_to_fp16 = const()[name = tensor("op_1755_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604966016)))]; + tensor var_1756_to_fp16 = const()[name = tensor("op_1756_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(608242880)))]; + tensor linear_90_cast_fp16 = linear(bias = var_1756_to_fp16, weight = var_1755_to_fp16, x = var_1744_cast_fp16)[name = tensor("linear_90_cast_fp16")]; + tensor var_1759_to_fp16 = const()[name = tensor("op_1759_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(608245504)))]; + tensor linear_91_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1759_to_fp16, x = var_1744_cast_fp16)[name = tensor("linear_91_cast_fp16")]; + tensor var_1763_to_fp16 = const()[name = tensor("op_1763_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(611522368)))]; + tensor var_1764_to_fp16 = const()[name = tensor("op_1764_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614799232)))]; + tensor linear_92_cast_fp16 = linear(bias = var_1764_to_fp16, weight = var_1763_to_fp16, x = var_1744_cast_fp16)[name = tensor("linear_92_cast_fp16")]; + tensor var_1772 = const()[name = tensor("op_1772"), val = tensor([1, 1500, 20, -1])]; + tensor var_1773_cast_fp16 = reshape(shape = var_1772, x = linear_90_cast_fp16)[name = tensor("op_1773_cast_fp16")]; + tensor const_254_to_fp16 = const()[name = tensor("const_254_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_63_cast_fp16 = mul(x = var_1773_cast_fp16, y = const_254_to_fp16)[name = tensor("q_63_cast_fp16")]; + tensor var_1779 = const()[name = tensor("op_1779"), val = tensor([1, 1500, 20, -1])]; + tensor var_1780_cast_fp16 = reshape(shape = var_1779, x = linear_91_cast_fp16)[name = tensor("op_1780_cast_fp16")]; + tensor const_255_to_fp16 = const()[name = tensor("const_255_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_63_cast_fp16 = mul(x = var_1780_cast_fp16, y = const_255_to_fp16)[name = tensor("k_63_cast_fp16")]; + tensor var_1786 = const()[name = tensor("op_1786"), val = tensor([1, 1500, 20, -1])]; + tensor var_1787_cast_fp16 = reshape(shape = var_1786, x = linear_92_cast_fp16)[name = tensor("op_1787_cast_fp16")]; + tensor var_1788 = const()[name = tensor("op_1788"), val = tensor([0, 2, 1, 3])]; + tensor qk_31_transpose_x_0 = const()[name = tensor("qk_31_transpose_x_0"), val = tensor(false)]; + tensor qk_31_transpose_y_0 = const()[name = tensor("qk_31_transpose_y_0"), val = tensor(false)]; + tensor transpose_158_perm_0 = const()[name = tensor("transpose_158_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_159_perm_0 = const()[name = tensor("transpose_159_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_257 = transpose(perm = transpose_159_perm_0, x = k_63_cast_fp16)[name = tensor("transpose_257")]; + tensor transpose_258 = transpose(perm = transpose_158_perm_0, x = q_63_cast_fp16)[name = tensor("transpose_258")]; + tensor qk_31_cast_fp16 = matmul(transpose_x = qk_31_transpose_x_0, transpose_y = qk_31_transpose_y_0, x = transpose_258, y = transpose_257)[name = tensor("qk_31_cast_fp16")]; + tensor var_1792_cast_fp16 = softmax(axis = var_1727, x = qk_31_cast_fp16)[name = tensor("op_1792_cast_fp16")]; + tensor var_1794_transpose_x_0 = const()[name = tensor("op_1794_transpose_x_0"), val = tensor(false)]; + tensor var_1794_transpose_y_0 = const()[name = tensor("op_1794_transpose_y_0"), val = tensor(false)]; + tensor transpose_259 = transpose(perm = var_1788, x = var_1787_cast_fp16)[name = tensor("transpose_259")]; + tensor var_1794_cast_fp16 = matmul(transpose_x = var_1794_transpose_x_0, transpose_y = var_1794_transpose_y_0, x = var_1792_cast_fp16, y = transpose_259)[name = tensor("op_1794_cast_fp16")]; + tensor var_1795 = const()[name = tensor("op_1795"), val = tensor([0, 2, 1, 3])]; + tensor concat_15 = const()[name = tensor("concat_15"), val = tensor([1, 1500, 1280])]; + tensor transpose_256 = transpose(perm = var_1795, x = var_1794_cast_fp16)[name = tensor("transpose_256")]; + tensor x_191_cast_fp16 = reshape(shape = concat_15, x = transpose_256)[name = tensor("x_191_cast_fp16")]; + tensor var_1800_to_fp16 = const()[name = tensor("op_1800_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614801856)))]; + tensor var_1801_to_fp16 = const()[name = tensor("op_1801_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(618078720)))]; + tensor linear_93_cast_fp16 = linear(bias = var_1801_to_fp16, weight = var_1800_to_fp16, x = x_191_cast_fp16)[name = tensor("linear_93_cast_fp16")]; + tensor x_193_cast_fp16 = add(x = x_187_cast_fp16, y = linear_93_cast_fp16)[name = tensor("x_193_cast_fp16")]; + tensor var_1808_axes_0 = const()[name = tensor("op_1808_axes_0"), val = tensor([-1])]; + tensor blocks_15_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_15_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(618081344)))]; + tensor blocks_15_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_15_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(618083968)))]; + tensor var_1808_cast_fp16 = layer_norm(axes = var_1808_axes_0, beta = blocks_15_mlp_ln_bias_to_fp16, epsilon = var_1733_to_fp16, gamma = blocks_15_mlp_ln_weight_to_fp16, x = x_193_cast_fp16)[name = tensor("op_1808_cast_fp16")]; + tensor var_1817_to_fp16 = const()[name = tensor("op_1817_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(618086592)))]; + tensor var_1818_to_fp16 = const()[name = tensor("op_1818_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(631193856)))]; + tensor linear_94_cast_fp16 = linear(bias = var_1818_to_fp16, weight = var_1817_to_fp16, x = var_1808_cast_fp16)[name = tensor("linear_94_cast_fp16")]; + tensor x_197_mode_0 = const()[name = tensor("x_197_mode_0"), val = tensor("EXACT")]; + tensor x_197_cast_fp16 = gelu(mode = x_197_mode_0, x = linear_94_cast_fp16)[name = tensor("x_197_cast_fp16")]; + tensor var_1823_to_fp16 = const()[name = tensor("op_1823_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(631204160)))]; + tensor var_1824_to_fp16 = const()[name = tensor("op_1824_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(644311424)))]; + tensor linear_95_cast_fp16 = linear(bias = var_1824_to_fp16, weight = var_1823_to_fp16, x = x_197_cast_fp16)[name = tensor("linear_95_cast_fp16")]; + tensor x_199_cast_fp16 = add(x = x_193_cast_fp16, y = linear_95_cast_fp16)[name = tensor("x_199_cast_fp16")]; + tensor var_1834 = const()[name = tensor("op_1834"), val = tensor(-1)]; + tensor var_1851_axes_0 = const()[name = tensor("op_1851_axes_0"), val = tensor([-1])]; + tensor blocks_16_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_16_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(644314048)))]; + tensor blocks_16_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_16_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(644316672)))]; + tensor var_1840_to_fp16 = const()[name = tensor("op_1840_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1851_cast_fp16 = layer_norm(axes = var_1851_axes_0, beta = blocks_16_attn_ln_bias_to_fp16, epsilon = var_1840_to_fp16, gamma = blocks_16_attn_ln_weight_to_fp16, x = x_199_cast_fp16)[name = tensor("op_1851_cast_fp16")]; + tensor var_1862_to_fp16 = const()[name = tensor("op_1862_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(644319296)))]; + tensor var_1863_to_fp16 = const()[name = tensor("op_1863_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647596160)))]; + tensor linear_96_cast_fp16 = linear(bias = var_1863_to_fp16, weight = var_1862_to_fp16, x = var_1851_cast_fp16)[name = tensor("linear_96_cast_fp16")]; + tensor var_1866_to_fp16 = const()[name = tensor("op_1866_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647598784)))]; + tensor linear_97_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1866_to_fp16, x = var_1851_cast_fp16)[name = tensor("linear_97_cast_fp16")]; + tensor var_1870_to_fp16 = const()[name = tensor("op_1870_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(650875648)))]; + tensor var_1871_to_fp16 = const()[name = tensor("op_1871_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(654152512)))]; + tensor linear_98_cast_fp16 = linear(bias = var_1871_to_fp16, weight = var_1870_to_fp16, x = var_1851_cast_fp16)[name = tensor("linear_98_cast_fp16")]; + tensor var_1879 = const()[name = tensor("op_1879"), val = tensor([1, 1500, 20, -1])]; + tensor var_1880_cast_fp16 = reshape(shape = var_1879, x = linear_96_cast_fp16)[name = tensor("op_1880_cast_fp16")]; + tensor const_256_to_fp16 = const()[name = tensor("const_256_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_67_cast_fp16 = mul(x = var_1880_cast_fp16, y = const_256_to_fp16)[name = tensor("q_67_cast_fp16")]; + tensor var_1886 = const()[name = tensor("op_1886"), val = tensor([1, 1500, 20, -1])]; + tensor var_1887_cast_fp16 = reshape(shape = var_1886, x = linear_97_cast_fp16)[name = tensor("op_1887_cast_fp16")]; + tensor const_257_to_fp16 = const()[name = tensor("const_257_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_67_cast_fp16 = mul(x = var_1887_cast_fp16, y = const_257_to_fp16)[name = tensor("k_67_cast_fp16")]; + tensor var_1893 = const()[name = tensor("op_1893"), val = tensor([1, 1500, 20, -1])]; + tensor var_1894_cast_fp16 = reshape(shape = var_1893, x = linear_98_cast_fp16)[name = tensor("op_1894_cast_fp16")]; + tensor var_1895 = const()[name = tensor("op_1895"), val = tensor([0, 2, 1, 3])]; + tensor qk_33_transpose_x_0 = const()[name = tensor("qk_33_transpose_x_0"), val = tensor(false)]; + tensor qk_33_transpose_y_0 = const()[name = tensor("qk_33_transpose_y_0"), val = tensor(false)]; + tensor transpose_160_perm_0 = const()[name = tensor("transpose_160_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_161_perm_0 = const()[name = tensor("transpose_161_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_253 = transpose(perm = transpose_161_perm_0, x = k_67_cast_fp16)[name = tensor("transpose_253")]; + tensor transpose_254 = transpose(perm = transpose_160_perm_0, x = q_67_cast_fp16)[name = tensor("transpose_254")]; + tensor qk_33_cast_fp16 = matmul(transpose_x = qk_33_transpose_x_0, transpose_y = qk_33_transpose_y_0, x = transpose_254, y = transpose_253)[name = tensor("qk_33_cast_fp16")]; + tensor var_1899_cast_fp16 = softmax(axis = var_1834, x = qk_33_cast_fp16)[name = tensor("op_1899_cast_fp16")]; + tensor var_1901_transpose_x_0 = const()[name = tensor("op_1901_transpose_x_0"), val = tensor(false)]; + tensor var_1901_transpose_y_0 = const()[name = tensor("op_1901_transpose_y_0"), val = tensor(false)]; + tensor transpose_255 = transpose(perm = var_1895, x = var_1894_cast_fp16)[name = tensor("transpose_255")]; + tensor var_1901_cast_fp16 = matmul(transpose_x = var_1901_transpose_x_0, transpose_y = var_1901_transpose_y_0, x = var_1899_cast_fp16, y = transpose_255)[name = tensor("op_1901_cast_fp16")]; + tensor var_1902 = const()[name = tensor("op_1902"), val = tensor([0, 2, 1, 3])]; + tensor concat_16 = const()[name = tensor("concat_16"), val = tensor([1, 1500, 1280])]; + tensor transpose_252 = transpose(perm = var_1902, x = var_1901_cast_fp16)[name = tensor("transpose_252")]; + tensor x_203_cast_fp16 = reshape(shape = concat_16, x = transpose_252)[name = tensor("x_203_cast_fp16")]; + tensor var_1907_to_fp16 = const()[name = tensor("op_1907_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(654155136)))]; + tensor var_1908_to_fp16 = const()[name = tensor("op_1908_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657432000)))]; + tensor linear_99_cast_fp16 = linear(bias = var_1908_to_fp16, weight = var_1907_to_fp16, x = x_203_cast_fp16)[name = tensor("linear_99_cast_fp16")]; + tensor x_205_cast_fp16 = add(x = x_199_cast_fp16, y = linear_99_cast_fp16)[name = tensor("x_205_cast_fp16")]; + tensor var_1915_axes_0 = const()[name = tensor("op_1915_axes_0"), val = tensor([-1])]; + tensor blocks_16_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_16_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657434624)))]; + tensor blocks_16_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_16_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657437248)))]; + tensor var_1915_cast_fp16 = layer_norm(axes = var_1915_axes_0, beta = blocks_16_mlp_ln_bias_to_fp16, epsilon = var_1840_to_fp16, gamma = blocks_16_mlp_ln_weight_to_fp16, x = x_205_cast_fp16)[name = tensor("op_1915_cast_fp16")]; + tensor var_1924_to_fp16 = const()[name = tensor("op_1924_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657439872)))]; + tensor var_1925_to_fp16 = const()[name = tensor("op_1925_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(670547136)))]; + tensor linear_100_cast_fp16 = linear(bias = var_1925_to_fp16, weight = var_1924_to_fp16, x = var_1915_cast_fp16)[name = tensor("linear_100_cast_fp16")]; + tensor x_209_mode_0 = const()[name = tensor("x_209_mode_0"), val = tensor("EXACT")]; + tensor x_209_cast_fp16 = gelu(mode = x_209_mode_0, x = linear_100_cast_fp16)[name = tensor("x_209_cast_fp16")]; + tensor var_1930_to_fp16 = const()[name = tensor("op_1930_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(670557440)))]; + tensor var_1931_to_fp16 = const()[name = tensor("op_1931_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(683664704)))]; + tensor linear_101_cast_fp16 = linear(bias = var_1931_to_fp16, weight = var_1930_to_fp16, x = x_209_cast_fp16)[name = tensor("linear_101_cast_fp16")]; + tensor x_211_cast_fp16 = add(x = x_205_cast_fp16, y = linear_101_cast_fp16)[name = tensor("x_211_cast_fp16")]; + tensor var_1941 = const()[name = tensor("op_1941"), val = tensor(-1)]; + tensor var_1958_axes_0 = const()[name = tensor("op_1958_axes_0"), val = tensor([-1])]; + tensor blocks_17_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_17_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(683667328)))]; + tensor blocks_17_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_17_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(683669952)))]; + tensor var_1947_to_fp16 = const()[name = tensor("op_1947_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1958_cast_fp16 = layer_norm(axes = var_1958_axes_0, beta = blocks_17_attn_ln_bias_to_fp16, epsilon = var_1947_to_fp16, gamma = blocks_17_attn_ln_weight_to_fp16, x = x_211_cast_fp16)[name = tensor("op_1958_cast_fp16")]; + tensor var_1969_to_fp16 = const()[name = tensor("op_1969_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(683672576)))]; + tensor var_1970_to_fp16 = const()[name = tensor("op_1970_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(686949440)))]; + tensor linear_102_cast_fp16 = linear(bias = var_1970_to_fp16, weight = var_1969_to_fp16, x = var_1958_cast_fp16)[name = tensor("linear_102_cast_fp16")]; + tensor var_1973_to_fp16 = const()[name = tensor("op_1973_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(686952064)))]; + tensor linear_103_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1973_to_fp16, x = var_1958_cast_fp16)[name = tensor("linear_103_cast_fp16")]; + tensor var_1977_to_fp16 = const()[name = tensor("op_1977_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(690228928)))]; + tensor var_1978_to_fp16 = const()[name = tensor("op_1978_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(693505792)))]; + tensor linear_104_cast_fp16 = linear(bias = var_1978_to_fp16, weight = var_1977_to_fp16, x = var_1958_cast_fp16)[name = tensor("linear_104_cast_fp16")]; + tensor var_1986 = const()[name = tensor("op_1986"), val = tensor([1, 1500, 20, -1])]; + tensor var_1987_cast_fp16 = reshape(shape = var_1986, x = linear_102_cast_fp16)[name = tensor("op_1987_cast_fp16")]; + tensor const_258_to_fp16 = const()[name = tensor("const_258_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_71_cast_fp16 = mul(x = var_1987_cast_fp16, y = const_258_to_fp16)[name = tensor("q_71_cast_fp16")]; + tensor var_1993 = const()[name = tensor("op_1993"), val = tensor([1, 1500, 20, -1])]; + tensor var_1994_cast_fp16 = reshape(shape = var_1993, x = linear_103_cast_fp16)[name = tensor("op_1994_cast_fp16")]; + tensor const_259_to_fp16 = const()[name = tensor("const_259_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_71_cast_fp16 = mul(x = var_1994_cast_fp16, y = const_259_to_fp16)[name = tensor("k_71_cast_fp16")]; + tensor var_2000 = const()[name = tensor("op_2000"), val = tensor([1, 1500, 20, -1])]; + tensor var_2001_cast_fp16 = reshape(shape = var_2000, x = linear_104_cast_fp16)[name = tensor("op_2001_cast_fp16")]; + tensor var_2002 = const()[name = tensor("op_2002"), val = tensor([0, 2, 1, 3])]; + tensor qk_35_transpose_x_0 = const()[name = tensor("qk_35_transpose_x_0"), val = tensor(false)]; + tensor qk_35_transpose_y_0 = const()[name = tensor("qk_35_transpose_y_0"), val = tensor(false)]; + tensor transpose_162_perm_0 = const()[name = tensor("transpose_162_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_163_perm_0 = const()[name = tensor("transpose_163_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_249 = transpose(perm = transpose_163_perm_0, x = k_71_cast_fp16)[name = tensor("transpose_249")]; + tensor transpose_250 = transpose(perm = transpose_162_perm_0, x = q_71_cast_fp16)[name = tensor("transpose_250")]; + tensor qk_35_cast_fp16 = matmul(transpose_x = qk_35_transpose_x_0, transpose_y = qk_35_transpose_y_0, x = transpose_250, y = transpose_249)[name = tensor("qk_35_cast_fp16")]; + tensor var_2006_cast_fp16 = softmax(axis = var_1941, x = qk_35_cast_fp16)[name = tensor("op_2006_cast_fp16")]; + tensor var_2008_transpose_x_0 = const()[name = tensor("op_2008_transpose_x_0"), val = tensor(false)]; + tensor var_2008_transpose_y_0 = const()[name = tensor("op_2008_transpose_y_0"), val = tensor(false)]; + tensor transpose_251 = transpose(perm = var_2002, x = var_2001_cast_fp16)[name = tensor("transpose_251")]; + tensor var_2008_cast_fp16 = matmul(transpose_x = var_2008_transpose_x_0, transpose_y = var_2008_transpose_y_0, x = var_2006_cast_fp16, y = transpose_251)[name = tensor("op_2008_cast_fp16")]; + tensor var_2009 = const()[name = tensor("op_2009"), val = tensor([0, 2, 1, 3])]; + tensor concat_17 = const()[name = tensor("concat_17"), val = tensor([1, 1500, 1280])]; + tensor transpose_248 = transpose(perm = var_2009, x = var_2008_cast_fp16)[name = tensor("transpose_248")]; + tensor x_215_cast_fp16 = reshape(shape = concat_17, x = transpose_248)[name = tensor("x_215_cast_fp16")]; + tensor var_2014_to_fp16 = const()[name = tensor("op_2014_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(693508416)))]; + tensor var_2015_to_fp16 = const()[name = tensor("op_2015_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696785280)))]; + tensor linear_105_cast_fp16 = linear(bias = var_2015_to_fp16, weight = var_2014_to_fp16, x = x_215_cast_fp16)[name = tensor("linear_105_cast_fp16")]; + tensor x_217_cast_fp16 = add(x = x_211_cast_fp16, y = linear_105_cast_fp16)[name = tensor("x_217_cast_fp16")]; + tensor var_2022_axes_0 = const()[name = tensor("op_2022_axes_0"), val = tensor([-1])]; + tensor blocks_17_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_17_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696787904)))]; + tensor blocks_17_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_17_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696790528)))]; + tensor var_2022_cast_fp16 = layer_norm(axes = var_2022_axes_0, beta = blocks_17_mlp_ln_bias_to_fp16, epsilon = var_1947_to_fp16, gamma = blocks_17_mlp_ln_weight_to_fp16, x = x_217_cast_fp16)[name = tensor("op_2022_cast_fp16")]; + tensor var_2031_to_fp16 = const()[name = tensor("op_2031_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696793152)))]; + tensor var_2032_to_fp16 = const()[name = tensor("op_2032_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(709900416)))]; + tensor linear_106_cast_fp16 = linear(bias = var_2032_to_fp16, weight = var_2031_to_fp16, x = var_2022_cast_fp16)[name = tensor("linear_106_cast_fp16")]; + tensor x_221_mode_0 = const()[name = tensor("x_221_mode_0"), val = tensor("EXACT")]; + tensor x_221_cast_fp16 = gelu(mode = x_221_mode_0, x = linear_106_cast_fp16)[name = tensor("x_221_cast_fp16")]; + tensor var_2037_to_fp16 = const()[name = tensor("op_2037_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(709910720)))]; + tensor var_2038_to_fp16 = const()[name = tensor("op_2038_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(723017984)))]; + tensor linear_107_cast_fp16 = linear(bias = var_2038_to_fp16, weight = var_2037_to_fp16, x = x_221_cast_fp16)[name = tensor("linear_107_cast_fp16")]; + tensor x_223_cast_fp16 = add(x = x_217_cast_fp16, y = linear_107_cast_fp16)[name = tensor("x_223_cast_fp16")]; + tensor var_2048 = const()[name = tensor("op_2048"), val = tensor(-1)]; + tensor var_2065_axes_0 = const()[name = tensor("op_2065_axes_0"), val = tensor([-1])]; + tensor blocks_18_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_18_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(723020608)))]; + tensor blocks_18_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_18_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(723023232)))]; + tensor var_2054_to_fp16 = const()[name = tensor("op_2054_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2065_cast_fp16 = layer_norm(axes = var_2065_axes_0, beta = blocks_18_attn_ln_bias_to_fp16, epsilon = var_2054_to_fp16, gamma = blocks_18_attn_ln_weight_to_fp16, x = x_223_cast_fp16)[name = tensor("op_2065_cast_fp16")]; + tensor var_2076_to_fp16 = const()[name = tensor("op_2076_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(723025856)))]; + tensor var_2077_to_fp16 = const()[name = tensor("op_2077_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(726302720)))]; + tensor linear_108_cast_fp16 = linear(bias = var_2077_to_fp16, weight = var_2076_to_fp16, x = var_2065_cast_fp16)[name = tensor("linear_108_cast_fp16")]; + tensor var_2080_to_fp16 = const()[name = tensor("op_2080_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(726305344)))]; + tensor linear_109_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2080_to_fp16, x = var_2065_cast_fp16)[name = tensor("linear_109_cast_fp16")]; + tensor var_2084_to_fp16 = const()[name = tensor("op_2084_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(729582208)))]; + tensor var_2085_to_fp16 = const()[name = tensor("op_2085_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(732859072)))]; + tensor linear_110_cast_fp16 = linear(bias = var_2085_to_fp16, weight = var_2084_to_fp16, x = var_2065_cast_fp16)[name = tensor("linear_110_cast_fp16")]; + tensor var_2093 = const()[name = tensor("op_2093"), val = tensor([1, 1500, 20, -1])]; + tensor var_2094_cast_fp16 = reshape(shape = var_2093, x = linear_108_cast_fp16)[name = tensor("op_2094_cast_fp16")]; + tensor const_260_to_fp16 = const()[name = tensor("const_260_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_75_cast_fp16 = mul(x = var_2094_cast_fp16, y = const_260_to_fp16)[name = tensor("q_75_cast_fp16")]; + tensor var_2100 = const()[name = tensor("op_2100"), val = tensor([1, 1500, 20, -1])]; + tensor var_2101_cast_fp16 = reshape(shape = var_2100, x = linear_109_cast_fp16)[name = tensor("op_2101_cast_fp16")]; + tensor const_261_to_fp16 = const()[name = tensor("const_261_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_75_cast_fp16 = mul(x = var_2101_cast_fp16, y = const_261_to_fp16)[name = tensor("k_75_cast_fp16")]; + tensor var_2107 = const()[name = tensor("op_2107"), val = tensor([1, 1500, 20, -1])]; + tensor var_2108_cast_fp16 = reshape(shape = var_2107, x = linear_110_cast_fp16)[name = tensor("op_2108_cast_fp16")]; + tensor var_2109 = const()[name = tensor("op_2109"), val = tensor([0, 2, 1, 3])]; + tensor qk_37_transpose_x_0 = const()[name = tensor("qk_37_transpose_x_0"), val = tensor(false)]; + tensor qk_37_transpose_y_0 = const()[name = tensor("qk_37_transpose_y_0"), val = tensor(false)]; + tensor transpose_164_perm_0 = const()[name = tensor("transpose_164_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_165_perm_0 = const()[name = tensor("transpose_165_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_245 = transpose(perm = transpose_165_perm_0, x = k_75_cast_fp16)[name = tensor("transpose_245")]; + tensor transpose_246 = transpose(perm = transpose_164_perm_0, x = q_75_cast_fp16)[name = tensor("transpose_246")]; + tensor qk_37_cast_fp16 = matmul(transpose_x = qk_37_transpose_x_0, transpose_y = qk_37_transpose_y_0, x = transpose_246, y = transpose_245)[name = tensor("qk_37_cast_fp16")]; + tensor var_2113_cast_fp16 = softmax(axis = var_2048, x = qk_37_cast_fp16)[name = tensor("op_2113_cast_fp16")]; + tensor var_2115_transpose_x_0 = const()[name = tensor("op_2115_transpose_x_0"), val = tensor(false)]; + tensor var_2115_transpose_y_0 = const()[name = tensor("op_2115_transpose_y_0"), val = tensor(false)]; + tensor transpose_247 = transpose(perm = var_2109, x = var_2108_cast_fp16)[name = tensor("transpose_247")]; + tensor var_2115_cast_fp16 = matmul(transpose_x = var_2115_transpose_x_0, transpose_y = var_2115_transpose_y_0, x = var_2113_cast_fp16, y = transpose_247)[name = tensor("op_2115_cast_fp16")]; + tensor var_2116 = const()[name = tensor("op_2116"), val = tensor([0, 2, 1, 3])]; + tensor concat_18 = const()[name = tensor("concat_18"), val = tensor([1, 1500, 1280])]; + tensor transpose_244 = transpose(perm = var_2116, x = var_2115_cast_fp16)[name = tensor("transpose_244")]; + tensor x_227_cast_fp16 = reshape(shape = concat_18, x = transpose_244)[name = tensor("x_227_cast_fp16")]; + tensor var_2121_to_fp16 = const()[name = tensor("op_2121_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(732861696)))]; + tensor var_2122_to_fp16 = const()[name = tensor("op_2122_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(736138560)))]; + tensor linear_111_cast_fp16 = linear(bias = var_2122_to_fp16, weight = var_2121_to_fp16, x = x_227_cast_fp16)[name = tensor("linear_111_cast_fp16")]; + tensor x_229_cast_fp16 = add(x = x_223_cast_fp16, y = linear_111_cast_fp16)[name = tensor("x_229_cast_fp16")]; + tensor var_2129_axes_0 = const()[name = tensor("op_2129_axes_0"), val = tensor([-1])]; + tensor blocks_18_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_18_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(736141184)))]; + tensor blocks_18_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_18_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(736143808)))]; + tensor var_2129_cast_fp16 = layer_norm(axes = var_2129_axes_0, beta = blocks_18_mlp_ln_bias_to_fp16, epsilon = var_2054_to_fp16, gamma = blocks_18_mlp_ln_weight_to_fp16, x = x_229_cast_fp16)[name = tensor("op_2129_cast_fp16")]; + tensor var_2138_to_fp16 = const()[name = tensor("op_2138_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(736146432)))]; + tensor var_2139_to_fp16 = const()[name = tensor("op_2139_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(749253696)))]; + tensor linear_112_cast_fp16 = linear(bias = var_2139_to_fp16, weight = var_2138_to_fp16, x = var_2129_cast_fp16)[name = tensor("linear_112_cast_fp16")]; + tensor x_233_mode_0 = const()[name = tensor("x_233_mode_0"), val = tensor("EXACT")]; + tensor x_233_cast_fp16 = gelu(mode = x_233_mode_0, x = linear_112_cast_fp16)[name = tensor("x_233_cast_fp16")]; + tensor var_2144_to_fp16 = const()[name = tensor("op_2144_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(749264000)))]; + tensor var_2145_to_fp16 = const()[name = tensor("op_2145_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762371264)))]; + tensor linear_113_cast_fp16 = linear(bias = var_2145_to_fp16, weight = var_2144_to_fp16, x = x_233_cast_fp16)[name = tensor("linear_113_cast_fp16")]; + tensor x_235_cast_fp16 = add(x = x_229_cast_fp16, y = linear_113_cast_fp16)[name = tensor("x_235_cast_fp16")]; + tensor var_2155 = const()[name = tensor("op_2155"), val = tensor(-1)]; + tensor var_2172_axes_0 = const()[name = tensor("op_2172_axes_0"), val = tensor([-1])]; + tensor blocks_19_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_19_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762373888)))]; + tensor blocks_19_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_19_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762376512)))]; + tensor var_2161_to_fp16 = const()[name = tensor("op_2161_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2172_cast_fp16 = layer_norm(axes = var_2172_axes_0, beta = blocks_19_attn_ln_bias_to_fp16, epsilon = var_2161_to_fp16, gamma = blocks_19_attn_ln_weight_to_fp16, x = x_235_cast_fp16)[name = tensor("op_2172_cast_fp16")]; + tensor var_2183_to_fp16 = const()[name = tensor("op_2183_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762379136)))]; + tensor var_2184_to_fp16 = const()[name = tensor("op_2184_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(765656000)))]; + tensor linear_114_cast_fp16 = linear(bias = var_2184_to_fp16, weight = var_2183_to_fp16, x = var_2172_cast_fp16)[name = tensor("linear_114_cast_fp16")]; + tensor var_2187_to_fp16 = const()[name = tensor("op_2187_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(765658624)))]; + tensor linear_115_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2187_to_fp16, x = var_2172_cast_fp16)[name = tensor("linear_115_cast_fp16")]; + tensor var_2191_to_fp16 = const()[name = tensor("op_2191_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(768935488)))]; + tensor var_2192_to_fp16 = const()[name = tensor("op_2192_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(772212352)))]; + tensor linear_116_cast_fp16 = linear(bias = var_2192_to_fp16, weight = var_2191_to_fp16, x = var_2172_cast_fp16)[name = tensor("linear_116_cast_fp16")]; + tensor var_2200 = const()[name = tensor("op_2200"), val = tensor([1, 1500, 20, -1])]; + tensor var_2201_cast_fp16 = reshape(shape = var_2200, x = linear_114_cast_fp16)[name = tensor("op_2201_cast_fp16")]; + tensor const_262_to_fp16 = const()[name = tensor("const_262_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_79_cast_fp16 = mul(x = var_2201_cast_fp16, y = const_262_to_fp16)[name = tensor("q_79_cast_fp16")]; + tensor var_2207 = const()[name = tensor("op_2207"), val = tensor([1, 1500, 20, -1])]; + tensor var_2208_cast_fp16 = reshape(shape = var_2207, x = linear_115_cast_fp16)[name = tensor("op_2208_cast_fp16")]; + tensor const_263_to_fp16 = const()[name = tensor("const_263_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_79_cast_fp16 = mul(x = var_2208_cast_fp16, y = const_263_to_fp16)[name = tensor("k_79_cast_fp16")]; + tensor var_2214 = const()[name = tensor("op_2214"), val = tensor([1, 1500, 20, -1])]; + tensor var_2215_cast_fp16 = reshape(shape = var_2214, x = linear_116_cast_fp16)[name = tensor("op_2215_cast_fp16")]; + tensor var_2216 = const()[name = tensor("op_2216"), val = tensor([0, 2, 1, 3])]; + tensor qk_39_transpose_x_0 = const()[name = tensor("qk_39_transpose_x_0"), val = tensor(false)]; + tensor qk_39_transpose_y_0 = const()[name = tensor("qk_39_transpose_y_0"), val = tensor(false)]; + tensor transpose_166_perm_0 = const()[name = tensor("transpose_166_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_167_perm_0 = const()[name = tensor("transpose_167_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_241 = transpose(perm = transpose_167_perm_0, x = k_79_cast_fp16)[name = tensor("transpose_241")]; + tensor transpose_242 = transpose(perm = transpose_166_perm_0, x = q_79_cast_fp16)[name = tensor("transpose_242")]; + tensor qk_39_cast_fp16 = matmul(transpose_x = qk_39_transpose_x_0, transpose_y = qk_39_transpose_y_0, x = transpose_242, y = transpose_241)[name = tensor("qk_39_cast_fp16")]; + tensor var_2220_cast_fp16 = softmax(axis = var_2155, x = qk_39_cast_fp16)[name = tensor("op_2220_cast_fp16")]; + tensor var_2222_transpose_x_0 = const()[name = tensor("op_2222_transpose_x_0"), val = tensor(false)]; + tensor var_2222_transpose_y_0 = const()[name = tensor("op_2222_transpose_y_0"), val = tensor(false)]; + tensor transpose_243 = transpose(perm = var_2216, x = var_2215_cast_fp16)[name = tensor("transpose_243")]; + tensor var_2222_cast_fp16 = matmul(transpose_x = var_2222_transpose_x_0, transpose_y = var_2222_transpose_y_0, x = var_2220_cast_fp16, y = transpose_243)[name = tensor("op_2222_cast_fp16")]; + tensor var_2223 = const()[name = tensor("op_2223"), val = tensor([0, 2, 1, 3])]; + tensor concat_19 = const()[name = tensor("concat_19"), val = tensor([1, 1500, 1280])]; + tensor transpose_240 = transpose(perm = var_2223, x = var_2222_cast_fp16)[name = tensor("transpose_240")]; + tensor x_239_cast_fp16 = reshape(shape = concat_19, x = transpose_240)[name = tensor("x_239_cast_fp16")]; + tensor var_2228_to_fp16 = const()[name = tensor("op_2228_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(772214976)))]; + tensor var_2229_to_fp16 = const()[name = tensor("op_2229_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775491840)))]; + tensor linear_117_cast_fp16 = linear(bias = var_2229_to_fp16, weight = var_2228_to_fp16, x = x_239_cast_fp16)[name = tensor("linear_117_cast_fp16")]; + tensor x_241_cast_fp16 = add(x = x_235_cast_fp16, y = linear_117_cast_fp16)[name = tensor("x_241_cast_fp16")]; + tensor var_2236_axes_0 = const()[name = tensor("op_2236_axes_0"), val = tensor([-1])]; + tensor blocks_19_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_19_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775494464)))]; + tensor blocks_19_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_19_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775497088)))]; + tensor var_2236_cast_fp16 = layer_norm(axes = var_2236_axes_0, beta = blocks_19_mlp_ln_bias_to_fp16, epsilon = var_2161_to_fp16, gamma = blocks_19_mlp_ln_weight_to_fp16, x = x_241_cast_fp16)[name = tensor("op_2236_cast_fp16")]; + tensor var_2245_to_fp16 = const()[name = tensor("op_2245_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775499712)))]; + tensor var_2246_to_fp16 = const()[name = tensor("op_2246_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(788606976)))]; + tensor linear_118_cast_fp16 = linear(bias = var_2246_to_fp16, weight = var_2245_to_fp16, x = var_2236_cast_fp16)[name = tensor("linear_118_cast_fp16")]; + tensor x_245_mode_0 = const()[name = tensor("x_245_mode_0"), val = tensor("EXACT")]; + tensor x_245_cast_fp16 = gelu(mode = x_245_mode_0, x = linear_118_cast_fp16)[name = tensor("x_245_cast_fp16")]; + tensor var_2251_to_fp16 = const()[name = tensor("op_2251_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(788617280)))]; + tensor var_2252_to_fp16 = const()[name = tensor("op_2252_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(801724544)))]; + tensor linear_119_cast_fp16 = linear(bias = var_2252_to_fp16, weight = var_2251_to_fp16, x = x_245_cast_fp16)[name = tensor("linear_119_cast_fp16")]; + tensor x_247_cast_fp16 = add(x = x_241_cast_fp16, y = linear_119_cast_fp16)[name = tensor("x_247_cast_fp16")]; + tensor var_2262 = const()[name = tensor("op_2262"), val = tensor(-1)]; + tensor var_2279_axes_0 = const()[name = tensor("op_2279_axes_0"), val = tensor([-1])]; + tensor blocks_20_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_20_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(801727168)))]; + tensor blocks_20_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_20_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(801729792)))]; + tensor var_2268_to_fp16 = const()[name = tensor("op_2268_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2279_cast_fp16 = layer_norm(axes = var_2279_axes_0, beta = blocks_20_attn_ln_bias_to_fp16, epsilon = var_2268_to_fp16, gamma = blocks_20_attn_ln_weight_to_fp16, x = x_247_cast_fp16)[name = tensor("op_2279_cast_fp16")]; + tensor var_2290_to_fp16 = const()[name = tensor("op_2290_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(801732416)))]; + tensor var_2291_to_fp16 = const()[name = tensor("op_2291_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(805009280)))]; + tensor linear_120_cast_fp16 = linear(bias = var_2291_to_fp16, weight = var_2290_to_fp16, x = var_2279_cast_fp16)[name = tensor("linear_120_cast_fp16")]; + tensor var_2294_to_fp16 = const()[name = tensor("op_2294_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(805011904)))]; + tensor linear_121_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2294_to_fp16, x = var_2279_cast_fp16)[name = tensor("linear_121_cast_fp16")]; + tensor var_2298_to_fp16 = const()[name = tensor("op_2298_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(808288768)))]; + tensor var_2299_to_fp16 = const()[name = tensor("op_2299_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(811565632)))]; + tensor linear_122_cast_fp16 = linear(bias = var_2299_to_fp16, weight = var_2298_to_fp16, x = var_2279_cast_fp16)[name = tensor("linear_122_cast_fp16")]; + tensor var_2307 = const()[name = tensor("op_2307"), val = tensor([1, 1500, 20, -1])]; + tensor var_2308_cast_fp16 = reshape(shape = var_2307, x = linear_120_cast_fp16)[name = tensor("op_2308_cast_fp16")]; + tensor const_264_to_fp16 = const()[name = tensor("const_264_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_83_cast_fp16 = mul(x = var_2308_cast_fp16, y = const_264_to_fp16)[name = tensor("q_83_cast_fp16")]; + tensor var_2314 = const()[name = tensor("op_2314"), val = tensor([1, 1500, 20, -1])]; + tensor var_2315_cast_fp16 = reshape(shape = var_2314, x = linear_121_cast_fp16)[name = tensor("op_2315_cast_fp16")]; + tensor const_265_to_fp16 = const()[name = tensor("const_265_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_83_cast_fp16 = mul(x = var_2315_cast_fp16, y = const_265_to_fp16)[name = tensor("k_83_cast_fp16")]; + tensor var_2321 = const()[name = tensor("op_2321"), val = tensor([1, 1500, 20, -1])]; + tensor var_2322_cast_fp16 = reshape(shape = var_2321, x = linear_122_cast_fp16)[name = tensor("op_2322_cast_fp16")]; + tensor var_2323 = const()[name = tensor("op_2323"), val = tensor([0, 2, 1, 3])]; + tensor qk_41_transpose_x_0 = const()[name = tensor("qk_41_transpose_x_0"), val = tensor(false)]; + tensor qk_41_transpose_y_0 = const()[name = tensor("qk_41_transpose_y_0"), val = tensor(false)]; + tensor transpose_168_perm_0 = const()[name = tensor("transpose_168_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_169_perm_0 = const()[name = tensor("transpose_169_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_237 = transpose(perm = transpose_169_perm_0, x = k_83_cast_fp16)[name = tensor("transpose_237")]; + tensor transpose_238 = transpose(perm = transpose_168_perm_0, x = q_83_cast_fp16)[name = tensor("transpose_238")]; + tensor qk_41_cast_fp16 = matmul(transpose_x = qk_41_transpose_x_0, transpose_y = qk_41_transpose_y_0, x = transpose_238, y = transpose_237)[name = tensor("qk_41_cast_fp16")]; + tensor var_2327_cast_fp16 = softmax(axis = var_2262, x = qk_41_cast_fp16)[name = tensor("op_2327_cast_fp16")]; + tensor var_2329_transpose_x_0 = const()[name = tensor("op_2329_transpose_x_0"), val = tensor(false)]; + tensor var_2329_transpose_y_0 = const()[name = tensor("op_2329_transpose_y_0"), val = tensor(false)]; + tensor transpose_239 = transpose(perm = var_2323, x = var_2322_cast_fp16)[name = tensor("transpose_239")]; + tensor var_2329_cast_fp16 = matmul(transpose_x = var_2329_transpose_x_0, transpose_y = var_2329_transpose_y_0, x = var_2327_cast_fp16, y = transpose_239)[name = tensor("op_2329_cast_fp16")]; + tensor var_2330 = const()[name = tensor("op_2330"), val = tensor([0, 2, 1, 3])]; + tensor concat_20 = const()[name = tensor("concat_20"), val = tensor([1, 1500, 1280])]; + tensor transpose_236 = transpose(perm = var_2330, x = var_2329_cast_fp16)[name = tensor("transpose_236")]; + tensor x_251_cast_fp16 = reshape(shape = concat_20, x = transpose_236)[name = tensor("x_251_cast_fp16")]; + tensor var_2335_to_fp16 = const()[name = tensor("op_2335_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(811568256)))]; + tensor var_2336_to_fp16 = const()[name = tensor("op_2336_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814845120)))]; + tensor linear_123_cast_fp16 = linear(bias = var_2336_to_fp16, weight = var_2335_to_fp16, x = x_251_cast_fp16)[name = tensor("linear_123_cast_fp16")]; + tensor x_253_cast_fp16 = add(x = x_247_cast_fp16, y = linear_123_cast_fp16)[name = tensor("x_253_cast_fp16")]; + tensor var_2343_axes_0 = const()[name = tensor("op_2343_axes_0"), val = tensor([-1])]; + tensor blocks_20_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_20_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814847744)))]; + tensor blocks_20_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_20_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814850368)))]; + tensor var_2343_cast_fp16 = layer_norm(axes = var_2343_axes_0, beta = blocks_20_mlp_ln_bias_to_fp16, epsilon = var_2268_to_fp16, gamma = blocks_20_mlp_ln_weight_to_fp16, x = x_253_cast_fp16)[name = tensor("op_2343_cast_fp16")]; + tensor var_2352_to_fp16 = const()[name = tensor("op_2352_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814852992)))]; + tensor var_2353_to_fp16 = const()[name = tensor("op_2353_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(827960256)))]; + tensor linear_124_cast_fp16 = linear(bias = var_2353_to_fp16, weight = var_2352_to_fp16, x = var_2343_cast_fp16)[name = tensor("linear_124_cast_fp16")]; + tensor x_257_mode_0 = const()[name = tensor("x_257_mode_0"), val = tensor("EXACT")]; + tensor x_257_cast_fp16 = gelu(mode = x_257_mode_0, x = linear_124_cast_fp16)[name = tensor("x_257_cast_fp16")]; + tensor var_2358_to_fp16 = const()[name = tensor("op_2358_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(827970560)))]; + tensor var_2359_to_fp16 = const()[name = tensor("op_2359_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(841077824)))]; + tensor linear_125_cast_fp16 = linear(bias = var_2359_to_fp16, weight = var_2358_to_fp16, x = x_257_cast_fp16)[name = tensor("linear_125_cast_fp16")]; + tensor x_259_cast_fp16 = add(x = x_253_cast_fp16, y = linear_125_cast_fp16)[name = tensor("x_259_cast_fp16")]; + tensor var_2369 = const()[name = tensor("op_2369"), val = tensor(-1)]; + tensor var_2386_axes_0 = const()[name = tensor("op_2386_axes_0"), val = tensor([-1])]; + tensor blocks_21_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_21_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(841080448)))]; + tensor blocks_21_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_21_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(841083072)))]; + tensor var_2375_to_fp16 = const()[name = tensor("op_2375_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2386_cast_fp16 = layer_norm(axes = var_2386_axes_0, beta = blocks_21_attn_ln_bias_to_fp16, epsilon = var_2375_to_fp16, gamma = blocks_21_attn_ln_weight_to_fp16, x = x_259_cast_fp16)[name = tensor("op_2386_cast_fp16")]; + tensor var_2397_to_fp16 = const()[name = tensor("op_2397_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(841085696)))]; + tensor var_2398_to_fp16 = const()[name = tensor("op_2398_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(844362560)))]; + tensor linear_126_cast_fp16 = linear(bias = var_2398_to_fp16, weight = var_2397_to_fp16, x = var_2386_cast_fp16)[name = tensor("linear_126_cast_fp16")]; + tensor var_2401_to_fp16 = const()[name = tensor("op_2401_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(844365184)))]; + tensor linear_127_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2401_to_fp16, x = var_2386_cast_fp16)[name = tensor("linear_127_cast_fp16")]; + tensor var_2405_to_fp16 = const()[name = tensor("op_2405_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(847642048)))]; + tensor var_2406_to_fp16 = const()[name = tensor("op_2406_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(850918912)))]; + tensor linear_128_cast_fp16 = linear(bias = var_2406_to_fp16, weight = var_2405_to_fp16, x = var_2386_cast_fp16)[name = tensor("linear_128_cast_fp16")]; + tensor var_2414 = const()[name = tensor("op_2414"), val = tensor([1, 1500, 20, -1])]; + tensor var_2415_cast_fp16 = reshape(shape = var_2414, x = linear_126_cast_fp16)[name = tensor("op_2415_cast_fp16")]; + tensor const_266_to_fp16 = const()[name = tensor("const_266_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_87_cast_fp16 = mul(x = var_2415_cast_fp16, y = const_266_to_fp16)[name = tensor("q_87_cast_fp16")]; + tensor var_2421 = const()[name = tensor("op_2421"), val = tensor([1, 1500, 20, -1])]; + tensor var_2422_cast_fp16 = reshape(shape = var_2421, x = linear_127_cast_fp16)[name = tensor("op_2422_cast_fp16")]; + tensor const_267_to_fp16 = const()[name = tensor("const_267_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_87_cast_fp16 = mul(x = var_2422_cast_fp16, y = const_267_to_fp16)[name = tensor("k_87_cast_fp16")]; + tensor var_2428 = const()[name = tensor("op_2428"), val = tensor([1, 1500, 20, -1])]; + tensor var_2429_cast_fp16 = reshape(shape = var_2428, x = linear_128_cast_fp16)[name = tensor("op_2429_cast_fp16")]; + tensor var_2430 = const()[name = tensor("op_2430"), val = tensor([0, 2, 1, 3])]; + tensor qk_43_transpose_x_0 = const()[name = tensor("qk_43_transpose_x_0"), val = tensor(false)]; + tensor qk_43_transpose_y_0 = const()[name = tensor("qk_43_transpose_y_0"), val = tensor(false)]; + tensor transpose_170_perm_0 = const()[name = tensor("transpose_170_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_171_perm_0 = const()[name = tensor("transpose_171_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_233 = transpose(perm = transpose_171_perm_0, x = k_87_cast_fp16)[name = tensor("transpose_233")]; + tensor transpose_234 = transpose(perm = transpose_170_perm_0, x = q_87_cast_fp16)[name = tensor("transpose_234")]; + tensor qk_43_cast_fp16 = matmul(transpose_x = qk_43_transpose_x_0, transpose_y = qk_43_transpose_y_0, x = transpose_234, y = transpose_233)[name = tensor("qk_43_cast_fp16")]; + tensor var_2434_cast_fp16 = softmax(axis = var_2369, x = qk_43_cast_fp16)[name = tensor("op_2434_cast_fp16")]; + tensor var_2436_transpose_x_0 = const()[name = tensor("op_2436_transpose_x_0"), val = tensor(false)]; + tensor var_2436_transpose_y_0 = const()[name = tensor("op_2436_transpose_y_0"), val = tensor(false)]; + tensor transpose_235 = transpose(perm = var_2430, x = var_2429_cast_fp16)[name = tensor("transpose_235")]; + tensor var_2436_cast_fp16 = matmul(transpose_x = var_2436_transpose_x_0, transpose_y = var_2436_transpose_y_0, x = var_2434_cast_fp16, y = transpose_235)[name = tensor("op_2436_cast_fp16")]; + tensor var_2437 = const()[name = tensor("op_2437"), val = tensor([0, 2, 1, 3])]; + tensor concat_21 = const()[name = tensor("concat_21"), val = tensor([1, 1500, 1280])]; + tensor transpose_232 = transpose(perm = var_2437, x = var_2436_cast_fp16)[name = tensor("transpose_232")]; + tensor x_263_cast_fp16 = reshape(shape = concat_21, x = transpose_232)[name = tensor("x_263_cast_fp16")]; + tensor var_2442_to_fp16 = const()[name = tensor("op_2442_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(850921536)))]; + tensor var_2443_to_fp16 = const()[name = tensor("op_2443_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(854198400)))]; + tensor linear_129_cast_fp16 = linear(bias = var_2443_to_fp16, weight = var_2442_to_fp16, x = x_263_cast_fp16)[name = tensor("linear_129_cast_fp16")]; + tensor x_265_cast_fp16 = add(x = x_259_cast_fp16, y = linear_129_cast_fp16)[name = tensor("x_265_cast_fp16")]; + tensor var_2450_axes_0 = const()[name = tensor("op_2450_axes_0"), val = tensor([-1])]; + tensor blocks_21_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_21_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(854201024)))]; + tensor blocks_21_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_21_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(854203648)))]; + tensor var_2450_cast_fp16 = layer_norm(axes = var_2450_axes_0, beta = blocks_21_mlp_ln_bias_to_fp16, epsilon = var_2375_to_fp16, gamma = blocks_21_mlp_ln_weight_to_fp16, x = x_265_cast_fp16)[name = tensor("op_2450_cast_fp16")]; + tensor var_2459_to_fp16 = const()[name = tensor("op_2459_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(854206272)))]; + tensor var_2460_to_fp16 = const()[name = tensor("op_2460_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(867313536)))]; + tensor linear_130_cast_fp16 = linear(bias = var_2460_to_fp16, weight = var_2459_to_fp16, x = var_2450_cast_fp16)[name = tensor("linear_130_cast_fp16")]; + tensor x_269_mode_0 = const()[name = tensor("x_269_mode_0"), val = tensor("EXACT")]; + tensor x_269_cast_fp16 = gelu(mode = x_269_mode_0, x = linear_130_cast_fp16)[name = tensor("x_269_cast_fp16")]; + tensor var_2465_to_fp16 = const()[name = tensor("op_2465_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(867323840)))]; + tensor var_2466_to_fp16 = const()[name = tensor("op_2466_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880431104)))]; + tensor linear_131_cast_fp16 = linear(bias = var_2466_to_fp16, weight = var_2465_to_fp16, x = x_269_cast_fp16)[name = tensor("linear_131_cast_fp16")]; + tensor x_271_cast_fp16 = add(x = x_265_cast_fp16, y = linear_131_cast_fp16)[name = tensor("x_271_cast_fp16")]; + tensor var_2476 = const()[name = tensor("op_2476"), val = tensor(-1)]; + tensor var_2493_axes_0 = const()[name = tensor("op_2493_axes_0"), val = tensor([-1])]; + tensor blocks_22_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_22_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880433728)))]; + tensor blocks_22_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_22_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880436352)))]; + tensor var_2482_to_fp16 = const()[name = tensor("op_2482_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2493_cast_fp16 = layer_norm(axes = var_2493_axes_0, beta = blocks_22_attn_ln_bias_to_fp16, epsilon = var_2482_to_fp16, gamma = blocks_22_attn_ln_weight_to_fp16, x = x_271_cast_fp16)[name = tensor("op_2493_cast_fp16")]; + tensor var_2504_to_fp16 = const()[name = tensor("op_2504_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880438976)))]; + tensor var_2505_to_fp16 = const()[name = tensor("op_2505_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(883715840)))]; + tensor linear_132_cast_fp16 = linear(bias = var_2505_to_fp16, weight = var_2504_to_fp16, x = var_2493_cast_fp16)[name = tensor("linear_132_cast_fp16")]; + tensor var_2508_to_fp16 = const()[name = tensor("op_2508_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(883718464)))]; + tensor linear_133_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2508_to_fp16, x = var_2493_cast_fp16)[name = tensor("linear_133_cast_fp16")]; + tensor var_2512_to_fp16 = const()[name = tensor("op_2512_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(886995328)))]; + tensor var_2513_to_fp16 = const()[name = tensor("op_2513_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(890272192)))]; + tensor linear_134_cast_fp16 = linear(bias = var_2513_to_fp16, weight = var_2512_to_fp16, x = var_2493_cast_fp16)[name = tensor("linear_134_cast_fp16")]; + tensor var_2521 = const()[name = tensor("op_2521"), val = tensor([1, 1500, 20, -1])]; + tensor var_2522_cast_fp16 = reshape(shape = var_2521, x = linear_132_cast_fp16)[name = tensor("op_2522_cast_fp16")]; + tensor const_268_to_fp16 = const()[name = tensor("const_268_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_91_cast_fp16 = mul(x = var_2522_cast_fp16, y = const_268_to_fp16)[name = tensor("q_91_cast_fp16")]; + tensor var_2528 = const()[name = tensor("op_2528"), val = tensor([1, 1500, 20, -1])]; + tensor var_2529_cast_fp16 = reshape(shape = var_2528, x = linear_133_cast_fp16)[name = tensor("op_2529_cast_fp16")]; + tensor const_269_to_fp16 = const()[name = tensor("const_269_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_91_cast_fp16 = mul(x = var_2529_cast_fp16, y = const_269_to_fp16)[name = tensor("k_91_cast_fp16")]; + tensor var_2535 = const()[name = tensor("op_2535"), val = tensor([1, 1500, 20, -1])]; + tensor var_2536_cast_fp16 = reshape(shape = var_2535, x = linear_134_cast_fp16)[name = tensor("op_2536_cast_fp16")]; + tensor var_2537 = const()[name = tensor("op_2537"), val = tensor([0, 2, 1, 3])]; + tensor qk_45_transpose_x_0 = const()[name = tensor("qk_45_transpose_x_0"), val = tensor(false)]; + tensor qk_45_transpose_y_0 = const()[name = tensor("qk_45_transpose_y_0"), val = tensor(false)]; + tensor transpose_172_perm_0 = const()[name = tensor("transpose_172_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_173_perm_0 = const()[name = tensor("transpose_173_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_229 = transpose(perm = transpose_173_perm_0, x = k_91_cast_fp16)[name = tensor("transpose_229")]; + tensor transpose_230 = transpose(perm = transpose_172_perm_0, x = q_91_cast_fp16)[name = tensor("transpose_230")]; + tensor qk_45_cast_fp16 = matmul(transpose_x = qk_45_transpose_x_0, transpose_y = qk_45_transpose_y_0, x = transpose_230, y = transpose_229)[name = tensor("qk_45_cast_fp16")]; + tensor var_2541_cast_fp16 = softmax(axis = var_2476, x = qk_45_cast_fp16)[name = tensor("op_2541_cast_fp16")]; + tensor var_2543_transpose_x_0 = const()[name = tensor("op_2543_transpose_x_0"), val = tensor(false)]; + tensor var_2543_transpose_y_0 = const()[name = tensor("op_2543_transpose_y_0"), val = tensor(false)]; + tensor transpose_231 = transpose(perm = var_2537, x = var_2536_cast_fp16)[name = tensor("transpose_231")]; + tensor var_2543_cast_fp16 = matmul(transpose_x = var_2543_transpose_x_0, transpose_y = var_2543_transpose_y_0, x = var_2541_cast_fp16, y = transpose_231)[name = tensor("op_2543_cast_fp16")]; + tensor var_2544 = const()[name = tensor("op_2544"), val = tensor([0, 2, 1, 3])]; + tensor concat_22 = const()[name = tensor("concat_22"), val = tensor([1, 1500, 1280])]; + tensor transpose_228 = transpose(perm = var_2544, x = var_2543_cast_fp16)[name = tensor("transpose_228")]; + tensor x_275_cast_fp16 = reshape(shape = concat_22, x = transpose_228)[name = tensor("x_275_cast_fp16")]; + tensor var_2549_to_fp16 = const()[name = tensor("op_2549_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(890274816)))]; + tensor var_2550_to_fp16 = const()[name = tensor("op_2550_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893551680)))]; + tensor linear_135_cast_fp16 = linear(bias = var_2550_to_fp16, weight = var_2549_to_fp16, x = x_275_cast_fp16)[name = tensor("linear_135_cast_fp16")]; + tensor x_277_cast_fp16 = add(x = x_271_cast_fp16, y = linear_135_cast_fp16)[name = tensor("x_277_cast_fp16")]; + tensor var_2557_axes_0 = const()[name = tensor("op_2557_axes_0"), val = tensor([-1])]; + tensor blocks_22_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_22_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893554304)))]; + tensor blocks_22_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_22_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893556928)))]; + tensor var_2557_cast_fp16 = layer_norm(axes = var_2557_axes_0, beta = blocks_22_mlp_ln_bias_to_fp16, epsilon = var_2482_to_fp16, gamma = blocks_22_mlp_ln_weight_to_fp16, x = x_277_cast_fp16)[name = tensor("op_2557_cast_fp16")]; + tensor var_2566_to_fp16 = const()[name = tensor("op_2566_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893559552)))]; + tensor var_2567_to_fp16 = const()[name = tensor("op_2567_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(906666816)))]; + tensor linear_136_cast_fp16 = linear(bias = var_2567_to_fp16, weight = var_2566_to_fp16, x = var_2557_cast_fp16)[name = tensor("linear_136_cast_fp16")]; + tensor x_281_mode_0 = const()[name = tensor("x_281_mode_0"), val = tensor("EXACT")]; + tensor x_281_cast_fp16 = gelu(mode = x_281_mode_0, x = linear_136_cast_fp16)[name = tensor("x_281_cast_fp16")]; + tensor var_2572_to_fp16 = const()[name = tensor("op_2572_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(906677120)))]; + tensor var_2573_to_fp16 = const()[name = tensor("op_2573_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(919784384)))]; + tensor linear_137_cast_fp16 = linear(bias = var_2573_to_fp16, weight = var_2572_to_fp16, x = x_281_cast_fp16)[name = tensor("linear_137_cast_fp16")]; + tensor x_283_cast_fp16 = add(x = x_277_cast_fp16, y = linear_137_cast_fp16)[name = tensor("x_283_cast_fp16")]; + tensor var_2583 = const()[name = tensor("op_2583"), val = tensor(-1)]; + tensor var_2600_axes_0 = const()[name = tensor("op_2600_axes_0"), val = tensor([-1])]; + tensor blocks_23_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_23_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(919787008)))]; + tensor blocks_23_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_23_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(919789632)))]; + tensor var_2589_to_fp16 = const()[name = tensor("op_2589_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2600_cast_fp16 = layer_norm(axes = var_2600_axes_0, beta = blocks_23_attn_ln_bias_to_fp16, epsilon = var_2589_to_fp16, gamma = blocks_23_attn_ln_weight_to_fp16, x = x_283_cast_fp16)[name = tensor("op_2600_cast_fp16")]; + tensor var_2611_to_fp16 = const()[name = tensor("op_2611_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(919792256)))]; + tensor var_2612_to_fp16 = const()[name = tensor("op_2612_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(923069120)))]; + tensor linear_138_cast_fp16 = linear(bias = var_2612_to_fp16, weight = var_2611_to_fp16, x = var_2600_cast_fp16)[name = tensor("linear_138_cast_fp16")]; + tensor var_2615_to_fp16 = const()[name = tensor("op_2615_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(923071744)))]; + tensor linear_139_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2615_to_fp16, x = var_2600_cast_fp16)[name = tensor("linear_139_cast_fp16")]; + tensor var_2619_to_fp16 = const()[name = tensor("op_2619_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(926348608)))]; + tensor var_2620_to_fp16 = const()[name = tensor("op_2620_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(929625472)))]; + tensor linear_140_cast_fp16 = linear(bias = var_2620_to_fp16, weight = var_2619_to_fp16, x = var_2600_cast_fp16)[name = tensor("linear_140_cast_fp16")]; + tensor var_2628 = const()[name = tensor("op_2628"), val = tensor([1, 1500, 20, -1])]; + tensor var_2629_cast_fp16 = reshape(shape = var_2628, x = linear_138_cast_fp16)[name = tensor("op_2629_cast_fp16")]; + tensor const_270_to_fp16 = const()[name = tensor("const_270_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_95_cast_fp16 = mul(x = var_2629_cast_fp16, y = const_270_to_fp16)[name = tensor("q_95_cast_fp16")]; + tensor var_2635 = const()[name = tensor("op_2635"), val = tensor([1, 1500, 20, -1])]; + tensor var_2636_cast_fp16 = reshape(shape = var_2635, x = linear_139_cast_fp16)[name = tensor("op_2636_cast_fp16")]; + tensor const_271_to_fp16 = const()[name = tensor("const_271_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_95_cast_fp16 = mul(x = var_2636_cast_fp16, y = const_271_to_fp16)[name = tensor("k_95_cast_fp16")]; + tensor var_2642 = const()[name = tensor("op_2642"), val = tensor([1, 1500, 20, -1])]; + tensor var_2643_cast_fp16 = reshape(shape = var_2642, x = linear_140_cast_fp16)[name = tensor("op_2643_cast_fp16")]; + tensor var_2644 = const()[name = tensor("op_2644"), val = tensor([0, 2, 1, 3])]; + tensor qk_47_transpose_x_0 = const()[name = tensor("qk_47_transpose_x_0"), val = tensor(false)]; + tensor qk_47_transpose_y_0 = const()[name = tensor("qk_47_transpose_y_0"), val = tensor(false)]; + tensor transpose_174_perm_0 = const()[name = tensor("transpose_174_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_175_perm_0 = const()[name = tensor("transpose_175_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_225 = transpose(perm = transpose_175_perm_0, x = k_95_cast_fp16)[name = tensor("transpose_225")]; + tensor transpose_226 = transpose(perm = transpose_174_perm_0, x = q_95_cast_fp16)[name = tensor("transpose_226")]; + tensor qk_47_cast_fp16 = matmul(transpose_x = qk_47_transpose_x_0, transpose_y = qk_47_transpose_y_0, x = transpose_226, y = transpose_225)[name = tensor("qk_47_cast_fp16")]; + tensor var_2648_cast_fp16 = softmax(axis = var_2583, x = qk_47_cast_fp16)[name = tensor("op_2648_cast_fp16")]; + tensor var_2650_transpose_x_0 = const()[name = tensor("op_2650_transpose_x_0"), val = tensor(false)]; + tensor var_2650_transpose_y_0 = const()[name = tensor("op_2650_transpose_y_0"), val = tensor(false)]; + tensor transpose_227 = transpose(perm = var_2644, x = var_2643_cast_fp16)[name = tensor("transpose_227")]; + tensor var_2650_cast_fp16 = matmul(transpose_x = var_2650_transpose_x_0, transpose_y = var_2650_transpose_y_0, x = var_2648_cast_fp16, y = transpose_227)[name = tensor("op_2650_cast_fp16")]; + tensor var_2651 = const()[name = tensor("op_2651"), val = tensor([0, 2, 1, 3])]; + tensor concat_23 = const()[name = tensor("concat_23"), val = tensor([1, 1500, 1280])]; + tensor transpose_224 = transpose(perm = var_2651, x = var_2650_cast_fp16)[name = tensor("transpose_224")]; + tensor x_287_cast_fp16 = reshape(shape = concat_23, x = transpose_224)[name = tensor("x_287_cast_fp16")]; + tensor var_2656_to_fp16 = const()[name = tensor("op_2656_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(929628096)))]; + tensor var_2657_to_fp16 = const()[name = tensor("op_2657_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(932904960)))]; + tensor linear_141_cast_fp16 = linear(bias = var_2657_to_fp16, weight = var_2656_to_fp16, x = x_287_cast_fp16)[name = tensor("linear_141_cast_fp16")]; + tensor x_289_cast_fp16 = add(x = x_283_cast_fp16, y = linear_141_cast_fp16)[name = tensor("x_289_cast_fp16")]; + tensor var_2664_axes_0 = const()[name = tensor("op_2664_axes_0"), val = tensor([-1])]; + tensor blocks_23_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_23_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(932907584)))]; + tensor blocks_23_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_23_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(932910208)))]; + tensor var_2664_cast_fp16 = layer_norm(axes = var_2664_axes_0, beta = blocks_23_mlp_ln_bias_to_fp16, epsilon = var_2589_to_fp16, gamma = blocks_23_mlp_ln_weight_to_fp16, x = x_289_cast_fp16)[name = tensor("op_2664_cast_fp16")]; + tensor var_2673_to_fp16 = const()[name = tensor("op_2673_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(932912832)))]; + tensor var_2674_to_fp16 = const()[name = tensor("op_2674_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(946020096)))]; + tensor linear_142_cast_fp16 = linear(bias = var_2674_to_fp16, weight = var_2673_to_fp16, x = var_2664_cast_fp16)[name = tensor("linear_142_cast_fp16")]; + tensor x_293_mode_0 = const()[name = tensor("x_293_mode_0"), val = tensor("EXACT")]; + tensor x_293_cast_fp16 = gelu(mode = x_293_mode_0, x = linear_142_cast_fp16)[name = tensor("x_293_cast_fp16")]; + tensor var_2679_to_fp16 = const()[name = tensor("op_2679_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(946030400)))]; + tensor var_2680_to_fp16 = const()[name = tensor("op_2680_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(959137664)))]; + tensor linear_143_cast_fp16 = linear(bias = var_2680_to_fp16, weight = var_2679_to_fp16, x = x_293_cast_fp16)[name = tensor("linear_143_cast_fp16")]; + tensor x_295_cast_fp16 = add(x = x_289_cast_fp16, y = linear_143_cast_fp16)[name = tensor("x_295_cast_fp16")]; + tensor var_2690 = const()[name = tensor("op_2690"), val = tensor(-1)]; + tensor var_2707_axes_0 = const()[name = tensor("op_2707_axes_0"), val = tensor([-1])]; + tensor blocks_24_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_24_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(959140288)))]; + tensor blocks_24_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_24_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(959142912)))]; + tensor var_2696_to_fp16 = const()[name = tensor("op_2696_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2707_cast_fp16 = layer_norm(axes = var_2707_axes_0, beta = blocks_24_attn_ln_bias_to_fp16, epsilon = var_2696_to_fp16, gamma = blocks_24_attn_ln_weight_to_fp16, x = x_295_cast_fp16)[name = tensor("op_2707_cast_fp16")]; + tensor var_2718_to_fp16 = const()[name = tensor("op_2718_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(959145536)))]; + tensor var_2719_to_fp16 = const()[name = tensor("op_2719_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(962422400)))]; + tensor linear_144_cast_fp16 = linear(bias = var_2719_to_fp16, weight = var_2718_to_fp16, x = var_2707_cast_fp16)[name = tensor("linear_144_cast_fp16")]; + tensor var_2722_to_fp16 = const()[name = tensor("op_2722_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(962425024)))]; + tensor linear_145_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2722_to_fp16, x = var_2707_cast_fp16)[name = tensor("linear_145_cast_fp16")]; + tensor var_2726_to_fp16 = const()[name = tensor("op_2726_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(965701888)))]; + tensor var_2727_to_fp16 = const()[name = tensor("op_2727_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(968978752)))]; + tensor linear_146_cast_fp16 = linear(bias = var_2727_to_fp16, weight = var_2726_to_fp16, x = var_2707_cast_fp16)[name = tensor("linear_146_cast_fp16")]; + tensor var_2735 = const()[name = tensor("op_2735"), val = tensor([1, 1500, 20, -1])]; + tensor var_2736_cast_fp16 = reshape(shape = var_2735, x = linear_144_cast_fp16)[name = tensor("op_2736_cast_fp16")]; + tensor const_272_to_fp16 = const()[name = tensor("const_272_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_99_cast_fp16 = mul(x = var_2736_cast_fp16, y = const_272_to_fp16)[name = tensor("q_99_cast_fp16")]; + tensor var_2742 = const()[name = tensor("op_2742"), val = tensor([1, 1500, 20, -1])]; + tensor var_2743_cast_fp16 = reshape(shape = var_2742, x = linear_145_cast_fp16)[name = tensor("op_2743_cast_fp16")]; + tensor const_273_to_fp16 = const()[name = tensor("const_273_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_99_cast_fp16 = mul(x = var_2743_cast_fp16, y = const_273_to_fp16)[name = tensor("k_99_cast_fp16")]; + tensor var_2749 = const()[name = tensor("op_2749"), val = tensor([1, 1500, 20, -1])]; + tensor var_2750_cast_fp16 = reshape(shape = var_2749, x = linear_146_cast_fp16)[name = tensor("op_2750_cast_fp16")]; + tensor var_2751 = const()[name = tensor("op_2751"), val = tensor([0, 2, 1, 3])]; + tensor qk_49_transpose_x_0 = const()[name = tensor("qk_49_transpose_x_0"), val = tensor(false)]; + tensor qk_49_transpose_y_0 = const()[name = tensor("qk_49_transpose_y_0"), val = tensor(false)]; + tensor transpose_176_perm_0 = const()[name = tensor("transpose_176_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_177_perm_0 = const()[name = tensor("transpose_177_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_221 = transpose(perm = transpose_177_perm_0, x = k_99_cast_fp16)[name = tensor("transpose_221")]; + tensor transpose_222 = transpose(perm = transpose_176_perm_0, x = q_99_cast_fp16)[name = tensor("transpose_222")]; + tensor qk_49_cast_fp16 = matmul(transpose_x = qk_49_transpose_x_0, transpose_y = qk_49_transpose_y_0, x = transpose_222, y = transpose_221)[name = tensor("qk_49_cast_fp16")]; + tensor var_2755_cast_fp16 = softmax(axis = var_2690, x = qk_49_cast_fp16)[name = tensor("op_2755_cast_fp16")]; + tensor var_2757_transpose_x_0 = const()[name = tensor("op_2757_transpose_x_0"), val = tensor(false)]; + tensor var_2757_transpose_y_0 = const()[name = tensor("op_2757_transpose_y_0"), val = tensor(false)]; + tensor transpose_223 = transpose(perm = var_2751, x = var_2750_cast_fp16)[name = tensor("transpose_223")]; + tensor var_2757_cast_fp16 = matmul(transpose_x = var_2757_transpose_x_0, transpose_y = var_2757_transpose_y_0, x = var_2755_cast_fp16, y = transpose_223)[name = tensor("op_2757_cast_fp16")]; + tensor var_2758 = const()[name = tensor("op_2758"), val = tensor([0, 2, 1, 3])]; + tensor concat_24 = const()[name = tensor("concat_24"), val = tensor([1, 1500, 1280])]; + tensor transpose_220 = transpose(perm = var_2758, x = var_2757_cast_fp16)[name = tensor("transpose_220")]; + tensor x_299_cast_fp16 = reshape(shape = concat_24, x = transpose_220)[name = tensor("x_299_cast_fp16")]; + tensor var_2763_to_fp16 = const()[name = tensor("op_2763_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(968981376)))]; + tensor var_2764_to_fp16 = const()[name = tensor("op_2764_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(972258240)))]; + tensor linear_147_cast_fp16 = linear(bias = var_2764_to_fp16, weight = var_2763_to_fp16, x = x_299_cast_fp16)[name = tensor("linear_147_cast_fp16")]; + tensor x_301_cast_fp16 = add(x = x_295_cast_fp16, y = linear_147_cast_fp16)[name = tensor("x_301_cast_fp16")]; + tensor var_2771_axes_0 = const()[name = tensor("op_2771_axes_0"), val = tensor([-1])]; + tensor blocks_24_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_24_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(972260864)))]; + tensor blocks_24_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_24_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(972263488)))]; + tensor var_2771_cast_fp16 = layer_norm(axes = var_2771_axes_0, beta = blocks_24_mlp_ln_bias_to_fp16, epsilon = var_2696_to_fp16, gamma = blocks_24_mlp_ln_weight_to_fp16, x = x_301_cast_fp16)[name = tensor("op_2771_cast_fp16")]; + tensor var_2780_to_fp16 = const()[name = tensor("op_2780_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(972266112)))]; + tensor var_2781_to_fp16 = const()[name = tensor("op_2781_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(985373376)))]; + tensor linear_148_cast_fp16 = linear(bias = var_2781_to_fp16, weight = var_2780_to_fp16, x = var_2771_cast_fp16)[name = tensor("linear_148_cast_fp16")]; + tensor x_305_mode_0 = const()[name = tensor("x_305_mode_0"), val = tensor("EXACT")]; + tensor x_305_cast_fp16 = gelu(mode = x_305_mode_0, x = linear_148_cast_fp16)[name = tensor("x_305_cast_fp16")]; + tensor var_2786_to_fp16 = const()[name = tensor("op_2786_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(985383680)))]; + tensor var_2787_to_fp16 = const()[name = tensor("op_2787_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(998490944)))]; + tensor linear_149_cast_fp16 = linear(bias = var_2787_to_fp16, weight = var_2786_to_fp16, x = x_305_cast_fp16)[name = tensor("linear_149_cast_fp16")]; + tensor x_307_cast_fp16 = add(x = x_301_cast_fp16, y = linear_149_cast_fp16)[name = tensor("x_307_cast_fp16")]; + tensor var_2797 = const()[name = tensor("op_2797"), val = tensor(-1)]; + tensor var_2814_axes_0 = const()[name = tensor("op_2814_axes_0"), val = tensor([-1])]; + tensor blocks_25_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_25_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(998493568)))]; + tensor blocks_25_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_25_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(998496192)))]; + tensor var_2803_to_fp16 = const()[name = tensor("op_2803_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2814_cast_fp16 = layer_norm(axes = var_2814_axes_0, beta = blocks_25_attn_ln_bias_to_fp16, epsilon = var_2803_to_fp16, gamma = blocks_25_attn_ln_weight_to_fp16, x = x_307_cast_fp16)[name = tensor("op_2814_cast_fp16")]; + tensor var_2825_to_fp16 = const()[name = tensor("op_2825_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(998498816)))]; + tensor var_2826_to_fp16 = const()[name = tensor("op_2826_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1001775680)))]; + tensor linear_150_cast_fp16 = linear(bias = var_2826_to_fp16, weight = var_2825_to_fp16, x = var_2814_cast_fp16)[name = tensor("linear_150_cast_fp16")]; + tensor var_2829_to_fp16 = const()[name = tensor("op_2829_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1001778304)))]; + tensor linear_151_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2829_to_fp16, x = var_2814_cast_fp16)[name = tensor("linear_151_cast_fp16")]; + tensor var_2833_to_fp16 = const()[name = tensor("op_2833_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1005055168)))]; + tensor var_2834_to_fp16 = const()[name = tensor("op_2834_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1008332032)))]; + tensor linear_152_cast_fp16 = linear(bias = var_2834_to_fp16, weight = var_2833_to_fp16, x = var_2814_cast_fp16)[name = tensor("linear_152_cast_fp16")]; + tensor var_2842 = const()[name = tensor("op_2842"), val = tensor([1, 1500, 20, -1])]; + tensor var_2843_cast_fp16 = reshape(shape = var_2842, x = linear_150_cast_fp16)[name = tensor("op_2843_cast_fp16")]; + tensor const_274_to_fp16 = const()[name = tensor("const_274_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_103_cast_fp16 = mul(x = var_2843_cast_fp16, y = const_274_to_fp16)[name = tensor("q_103_cast_fp16")]; + tensor var_2849 = const()[name = tensor("op_2849"), val = tensor([1, 1500, 20, -1])]; + tensor var_2850_cast_fp16 = reshape(shape = var_2849, x = linear_151_cast_fp16)[name = tensor("op_2850_cast_fp16")]; + tensor const_275_to_fp16 = const()[name = tensor("const_275_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_103_cast_fp16 = mul(x = var_2850_cast_fp16, y = const_275_to_fp16)[name = tensor("k_103_cast_fp16")]; + tensor var_2856 = const()[name = tensor("op_2856"), val = tensor([1, 1500, 20, -1])]; + tensor var_2857_cast_fp16 = reshape(shape = var_2856, x = linear_152_cast_fp16)[name = tensor("op_2857_cast_fp16")]; + tensor var_2858 = const()[name = tensor("op_2858"), val = tensor([0, 2, 1, 3])]; + tensor qk_51_transpose_x_0 = const()[name = tensor("qk_51_transpose_x_0"), val = tensor(false)]; + tensor qk_51_transpose_y_0 = const()[name = tensor("qk_51_transpose_y_0"), val = tensor(false)]; + tensor transpose_178_perm_0 = const()[name = tensor("transpose_178_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_179_perm_0 = const()[name = tensor("transpose_179_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_217 = transpose(perm = transpose_179_perm_0, x = k_103_cast_fp16)[name = tensor("transpose_217")]; + tensor transpose_218 = transpose(perm = transpose_178_perm_0, x = q_103_cast_fp16)[name = tensor("transpose_218")]; + tensor qk_51_cast_fp16 = matmul(transpose_x = qk_51_transpose_x_0, transpose_y = qk_51_transpose_y_0, x = transpose_218, y = transpose_217)[name = tensor("qk_51_cast_fp16")]; + tensor var_2862_cast_fp16 = softmax(axis = var_2797, x = qk_51_cast_fp16)[name = tensor("op_2862_cast_fp16")]; + tensor var_2864_transpose_x_0 = const()[name = tensor("op_2864_transpose_x_0"), val = tensor(false)]; + tensor var_2864_transpose_y_0 = const()[name = tensor("op_2864_transpose_y_0"), val = tensor(false)]; + tensor transpose_219 = transpose(perm = var_2858, x = var_2857_cast_fp16)[name = tensor("transpose_219")]; + tensor var_2864_cast_fp16 = matmul(transpose_x = var_2864_transpose_x_0, transpose_y = var_2864_transpose_y_0, x = var_2862_cast_fp16, y = transpose_219)[name = tensor("op_2864_cast_fp16")]; + tensor var_2865 = const()[name = tensor("op_2865"), val = tensor([0, 2, 1, 3])]; + tensor concat_25 = const()[name = tensor("concat_25"), val = tensor([1, 1500, 1280])]; + tensor transpose_216 = transpose(perm = var_2865, x = var_2864_cast_fp16)[name = tensor("transpose_216")]; + tensor x_311_cast_fp16 = reshape(shape = concat_25, x = transpose_216)[name = tensor("x_311_cast_fp16")]; + tensor var_2870_to_fp16 = const()[name = tensor("op_2870_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1008334656)))]; + tensor var_2871_to_fp16 = const()[name = tensor("op_2871_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1011611520)))]; + tensor linear_153_cast_fp16 = linear(bias = var_2871_to_fp16, weight = var_2870_to_fp16, x = x_311_cast_fp16)[name = tensor("linear_153_cast_fp16")]; + tensor x_313_cast_fp16 = add(x = x_307_cast_fp16, y = linear_153_cast_fp16)[name = tensor("x_313_cast_fp16")]; + tensor var_2878_axes_0 = const()[name = tensor("op_2878_axes_0"), val = tensor([-1])]; + tensor blocks_25_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_25_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1011614144)))]; + tensor blocks_25_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_25_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1011616768)))]; + tensor var_2878_cast_fp16 = layer_norm(axes = var_2878_axes_0, beta = blocks_25_mlp_ln_bias_to_fp16, epsilon = var_2803_to_fp16, gamma = blocks_25_mlp_ln_weight_to_fp16, x = x_313_cast_fp16)[name = tensor("op_2878_cast_fp16")]; + tensor var_2887_to_fp16 = const()[name = tensor("op_2887_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1011619392)))]; + tensor var_2888_to_fp16 = const()[name = tensor("op_2888_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1024726656)))]; + tensor linear_154_cast_fp16 = linear(bias = var_2888_to_fp16, weight = var_2887_to_fp16, x = var_2878_cast_fp16)[name = tensor("linear_154_cast_fp16")]; + tensor x_317_mode_0 = const()[name = tensor("x_317_mode_0"), val = tensor("EXACT")]; + tensor x_317_cast_fp16 = gelu(mode = x_317_mode_0, x = linear_154_cast_fp16)[name = tensor("x_317_cast_fp16")]; + tensor var_2893_to_fp16 = const()[name = tensor("op_2893_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1024736960)))]; + tensor var_2894_to_fp16 = const()[name = tensor("op_2894_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1037844224)))]; + tensor linear_155_cast_fp16 = linear(bias = var_2894_to_fp16, weight = var_2893_to_fp16, x = x_317_cast_fp16)[name = tensor("linear_155_cast_fp16")]; + tensor x_319_cast_fp16 = add(x = x_313_cast_fp16, y = linear_155_cast_fp16)[name = tensor("x_319_cast_fp16")]; + tensor var_2904 = const()[name = tensor("op_2904"), val = tensor(-1)]; + tensor var_2921_axes_0 = const()[name = tensor("op_2921_axes_0"), val = tensor([-1])]; + tensor blocks_26_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_26_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1037846848)))]; + tensor blocks_26_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_26_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1037849472)))]; + tensor var_2910_to_fp16 = const()[name = tensor("op_2910_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2921_cast_fp16 = layer_norm(axes = var_2921_axes_0, beta = blocks_26_attn_ln_bias_to_fp16, epsilon = var_2910_to_fp16, gamma = blocks_26_attn_ln_weight_to_fp16, x = x_319_cast_fp16)[name = tensor("op_2921_cast_fp16")]; + tensor var_2932_to_fp16 = const()[name = tensor("op_2932_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1037852096)))]; + tensor var_2933_to_fp16 = const()[name = tensor("op_2933_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1041128960)))]; + tensor linear_156_cast_fp16 = linear(bias = var_2933_to_fp16, weight = var_2932_to_fp16, x = var_2921_cast_fp16)[name = tensor("linear_156_cast_fp16")]; + tensor var_2936_to_fp16 = const()[name = tensor("op_2936_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1041131584)))]; + tensor linear_157_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2936_to_fp16, x = var_2921_cast_fp16)[name = tensor("linear_157_cast_fp16")]; + tensor var_2940_to_fp16 = const()[name = tensor("op_2940_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1044408448)))]; + tensor var_2941_to_fp16 = const()[name = tensor("op_2941_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1047685312)))]; + tensor linear_158_cast_fp16 = linear(bias = var_2941_to_fp16, weight = var_2940_to_fp16, x = var_2921_cast_fp16)[name = tensor("linear_158_cast_fp16")]; + tensor var_2949 = const()[name = tensor("op_2949"), val = tensor([1, 1500, 20, -1])]; + tensor var_2950_cast_fp16 = reshape(shape = var_2949, x = linear_156_cast_fp16)[name = tensor("op_2950_cast_fp16")]; + tensor const_276_to_fp16 = const()[name = tensor("const_276_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_107_cast_fp16 = mul(x = var_2950_cast_fp16, y = const_276_to_fp16)[name = tensor("q_107_cast_fp16")]; + tensor var_2956 = const()[name = tensor("op_2956"), val = tensor([1, 1500, 20, -1])]; + tensor var_2957_cast_fp16 = reshape(shape = var_2956, x = linear_157_cast_fp16)[name = tensor("op_2957_cast_fp16")]; + tensor const_277_to_fp16 = const()[name = tensor("const_277_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_107_cast_fp16 = mul(x = var_2957_cast_fp16, y = const_277_to_fp16)[name = tensor("k_107_cast_fp16")]; + tensor var_2963 = const()[name = tensor("op_2963"), val = tensor([1, 1500, 20, -1])]; + tensor var_2964_cast_fp16 = reshape(shape = var_2963, x = linear_158_cast_fp16)[name = tensor("op_2964_cast_fp16")]; + tensor var_2965 = const()[name = tensor("op_2965"), val = tensor([0, 2, 1, 3])]; + tensor qk_53_transpose_x_0 = const()[name = tensor("qk_53_transpose_x_0"), val = tensor(false)]; + tensor qk_53_transpose_y_0 = const()[name = tensor("qk_53_transpose_y_0"), val = tensor(false)]; + tensor transpose_180_perm_0 = const()[name = tensor("transpose_180_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_181_perm_0 = const()[name = tensor("transpose_181_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_213 = transpose(perm = transpose_181_perm_0, x = k_107_cast_fp16)[name = tensor("transpose_213")]; + tensor transpose_214 = transpose(perm = transpose_180_perm_0, x = q_107_cast_fp16)[name = tensor("transpose_214")]; + tensor qk_53_cast_fp16 = matmul(transpose_x = qk_53_transpose_x_0, transpose_y = qk_53_transpose_y_0, x = transpose_214, y = transpose_213)[name = tensor("qk_53_cast_fp16")]; + tensor var_2969_cast_fp16 = softmax(axis = var_2904, x = qk_53_cast_fp16)[name = tensor("op_2969_cast_fp16")]; + tensor var_2971_transpose_x_0 = const()[name = tensor("op_2971_transpose_x_0"), val = tensor(false)]; + tensor var_2971_transpose_y_0 = const()[name = tensor("op_2971_transpose_y_0"), val = tensor(false)]; + tensor transpose_215 = transpose(perm = var_2965, x = var_2964_cast_fp16)[name = tensor("transpose_215")]; + tensor var_2971_cast_fp16 = matmul(transpose_x = var_2971_transpose_x_0, transpose_y = var_2971_transpose_y_0, x = var_2969_cast_fp16, y = transpose_215)[name = tensor("op_2971_cast_fp16")]; + tensor var_2972 = const()[name = tensor("op_2972"), val = tensor([0, 2, 1, 3])]; + tensor concat_26 = const()[name = tensor("concat_26"), val = tensor([1, 1500, 1280])]; + tensor transpose_212 = transpose(perm = var_2972, x = var_2971_cast_fp16)[name = tensor("transpose_212")]; + tensor x_323_cast_fp16 = reshape(shape = concat_26, x = transpose_212)[name = tensor("x_323_cast_fp16")]; + tensor var_2977_to_fp16 = const()[name = tensor("op_2977_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1047687936)))]; + tensor var_2978_to_fp16 = const()[name = tensor("op_2978_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1050964800)))]; + tensor linear_159_cast_fp16 = linear(bias = var_2978_to_fp16, weight = var_2977_to_fp16, x = x_323_cast_fp16)[name = tensor("linear_159_cast_fp16")]; + tensor x_325_cast_fp16 = add(x = x_319_cast_fp16, y = linear_159_cast_fp16)[name = tensor("x_325_cast_fp16")]; + tensor var_2985_axes_0 = const()[name = tensor("op_2985_axes_0"), val = tensor([-1])]; + tensor blocks_26_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_26_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1050967424)))]; + tensor blocks_26_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_26_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1050970048)))]; + tensor var_2985_cast_fp16 = layer_norm(axes = var_2985_axes_0, beta = blocks_26_mlp_ln_bias_to_fp16, epsilon = var_2910_to_fp16, gamma = blocks_26_mlp_ln_weight_to_fp16, x = x_325_cast_fp16)[name = tensor("op_2985_cast_fp16")]; + tensor var_2994_to_fp16 = const()[name = tensor("op_2994_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1050972672)))]; + tensor var_2995_to_fp16 = const()[name = tensor("op_2995_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1064079936)))]; + tensor linear_160_cast_fp16 = linear(bias = var_2995_to_fp16, weight = var_2994_to_fp16, x = var_2985_cast_fp16)[name = tensor("linear_160_cast_fp16")]; + tensor x_329_mode_0 = const()[name = tensor("x_329_mode_0"), val = tensor("EXACT")]; + tensor x_329_cast_fp16 = gelu(mode = x_329_mode_0, x = linear_160_cast_fp16)[name = tensor("x_329_cast_fp16")]; + tensor var_3000_to_fp16 = const()[name = tensor("op_3000_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1064090240)))]; + tensor var_3001_to_fp16 = const()[name = tensor("op_3001_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1077197504)))]; + tensor linear_161_cast_fp16 = linear(bias = var_3001_to_fp16, weight = var_3000_to_fp16, x = x_329_cast_fp16)[name = tensor("linear_161_cast_fp16")]; + tensor x_331_cast_fp16 = add(x = x_325_cast_fp16, y = linear_161_cast_fp16)[name = tensor("x_331_cast_fp16")]; + tensor var_3011 = const()[name = tensor("op_3011"), val = tensor(-1)]; + tensor var_3028_axes_0 = const()[name = tensor("op_3028_axes_0"), val = tensor([-1])]; + tensor blocks_27_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_27_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1077200128)))]; + tensor blocks_27_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_27_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1077202752)))]; + tensor var_3017_to_fp16 = const()[name = tensor("op_3017_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3028_cast_fp16 = layer_norm(axes = var_3028_axes_0, beta = blocks_27_attn_ln_bias_to_fp16, epsilon = var_3017_to_fp16, gamma = blocks_27_attn_ln_weight_to_fp16, x = x_331_cast_fp16)[name = tensor("op_3028_cast_fp16")]; + tensor var_3039_to_fp16 = const()[name = tensor("op_3039_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1077205376)))]; + tensor var_3040_to_fp16 = const()[name = tensor("op_3040_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1080482240)))]; + tensor linear_162_cast_fp16 = linear(bias = var_3040_to_fp16, weight = var_3039_to_fp16, x = var_3028_cast_fp16)[name = tensor("linear_162_cast_fp16")]; + tensor var_3043_to_fp16 = const()[name = tensor("op_3043_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1080484864)))]; + tensor linear_163_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3043_to_fp16, x = var_3028_cast_fp16)[name = tensor("linear_163_cast_fp16")]; + tensor var_3047_to_fp16 = const()[name = tensor("op_3047_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1083761728)))]; + tensor var_3048_to_fp16 = const()[name = tensor("op_3048_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1087038592)))]; + tensor linear_164_cast_fp16 = linear(bias = var_3048_to_fp16, weight = var_3047_to_fp16, x = var_3028_cast_fp16)[name = tensor("linear_164_cast_fp16")]; + tensor var_3056 = const()[name = tensor("op_3056"), val = tensor([1, 1500, 20, -1])]; + tensor var_3057_cast_fp16 = reshape(shape = var_3056, x = linear_162_cast_fp16)[name = tensor("op_3057_cast_fp16")]; + tensor const_278_to_fp16 = const()[name = tensor("const_278_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_111_cast_fp16 = mul(x = var_3057_cast_fp16, y = const_278_to_fp16)[name = tensor("q_111_cast_fp16")]; + tensor var_3063 = const()[name = tensor("op_3063"), val = tensor([1, 1500, 20, -1])]; + tensor var_3064_cast_fp16 = reshape(shape = var_3063, x = linear_163_cast_fp16)[name = tensor("op_3064_cast_fp16")]; + tensor const_279_to_fp16 = const()[name = tensor("const_279_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_111_cast_fp16 = mul(x = var_3064_cast_fp16, y = const_279_to_fp16)[name = tensor("k_111_cast_fp16")]; + tensor var_3070 = const()[name = tensor("op_3070"), val = tensor([1, 1500, 20, -1])]; + tensor var_3071_cast_fp16 = reshape(shape = var_3070, x = linear_164_cast_fp16)[name = tensor("op_3071_cast_fp16")]; + tensor var_3072 = const()[name = tensor("op_3072"), val = tensor([0, 2, 1, 3])]; + tensor qk_55_transpose_x_0 = const()[name = tensor("qk_55_transpose_x_0"), val = tensor(false)]; + tensor qk_55_transpose_y_0 = const()[name = tensor("qk_55_transpose_y_0"), val = tensor(false)]; + tensor transpose_182_perm_0 = const()[name = tensor("transpose_182_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_183_perm_0 = const()[name = tensor("transpose_183_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_209 = transpose(perm = transpose_183_perm_0, x = k_111_cast_fp16)[name = tensor("transpose_209")]; + tensor transpose_210 = transpose(perm = transpose_182_perm_0, x = q_111_cast_fp16)[name = tensor("transpose_210")]; + tensor qk_55_cast_fp16 = matmul(transpose_x = qk_55_transpose_x_0, transpose_y = qk_55_transpose_y_0, x = transpose_210, y = transpose_209)[name = tensor("qk_55_cast_fp16")]; + tensor var_3076_cast_fp16 = softmax(axis = var_3011, x = qk_55_cast_fp16)[name = tensor("op_3076_cast_fp16")]; + tensor var_3078_transpose_x_0 = const()[name = tensor("op_3078_transpose_x_0"), val = tensor(false)]; + tensor var_3078_transpose_y_0 = const()[name = tensor("op_3078_transpose_y_0"), val = tensor(false)]; + tensor transpose_211 = transpose(perm = var_3072, x = var_3071_cast_fp16)[name = tensor("transpose_211")]; + tensor var_3078_cast_fp16 = matmul(transpose_x = var_3078_transpose_x_0, transpose_y = var_3078_transpose_y_0, x = var_3076_cast_fp16, y = transpose_211)[name = tensor("op_3078_cast_fp16")]; + tensor var_3079 = const()[name = tensor("op_3079"), val = tensor([0, 2, 1, 3])]; + tensor concat_27 = const()[name = tensor("concat_27"), val = tensor([1, 1500, 1280])]; + tensor transpose_208 = transpose(perm = var_3079, x = var_3078_cast_fp16)[name = tensor("transpose_208")]; + tensor x_335_cast_fp16 = reshape(shape = concat_27, x = transpose_208)[name = tensor("x_335_cast_fp16")]; + tensor var_3084_to_fp16 = const()[name = tensor("op_3084_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1087041216)))]; + tensor var_3085_to_fp16 = const()[name = tensor("op_3085_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1090318080)))]; + tensor linear_165_cast_fp16 = linear(bias = var_3085_to_fp16, weight = var_3084_to_fp16, x = x_335_cast_fp16)[name = tensor("linear_165_cast_fp16")]; + tensor x_337_cast_fp16 = add(x = x_331_cast_fp16, y = linear_165_cast_fp16)[name = tensor("x_337_cast_fp16")]; + tensor var_3092_axes_0 = const()[name = tensor("op_3092_axes_0"), val = tensor([-1])]; + tensor blocks_27_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_27_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1090320704)))]; + tensor blocks_27_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_27_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1090323328)))]; + tensor var_3092_cast_fp16 = layer_norm(axes = var_3092_axes_0, beta = blocks_27_mlp_ln_bias_to_fp16, epsilon = var_3017_to_fp16, gamma = blocks_27_mlp_ln_weight_to_fp16, x = x_337_cast_fp16)[name = tensor("op_3092_cast_fp16")]; + tensor var_3101_to_fp16 = const()[name = tensor("op_3101_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1090325952)))]; + tensor var_3102_to_fp16 = const()[name = tensor("op_3102_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1103433216)))]; + tensor linear_166_cast_fp16 = linear(bias = var_3102_to_fp16, weight = var_3101_to_fp16, x = var_3092_cast_fp16)[name = tensor("linear_166_cast_fp16")]; + tensor x_341_mode_0 = const()[name = tensor("x_341_mode_0"), val = tensor("EXACT")]; + tensor x_341_cast_fp16 = gelu(mode = x_341_mode_0, x = linear_166_cast_fp16)[name = tensor("x_341_cast_fp16")]; + tensor var_3107_to_fp16 = const()[name = tensor("op_3107_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1103443520)))]; + tensor var_3108_to_fp16 = const()[name = tensor("op_3108_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116550784)))]; + tensor linear_167_cast_fp16 = linear(bias = var_3108_to_fp16, weight = var_3107_to_fp16, x = x_341_cast_fp16)[name = tensor("linear_167_cast_fp16")]; + tensor x_343_cast_fp16 = add(x = x_337_cast_fp16, y = linear_167_cast_fp16)[name = tensor("x_343_cast_fp16")]; + tensor var_3118 = const()[name = tensor("op_3118"), val = tensor(-1)]; + tensor var_3135_axes_0 = const()[name = tensor("op_3135_axes_0"), val = tensor([-1])]; + tensor blocks_28_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_28_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116553408)))]; + tensor blocks_28_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_28_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116556032)))]; + tensor var_3124_to_fp16 = const()[name = tensor("op_3124_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3135_cast_fp16 = layer_norm(axes = var_3135_axes_0, beta = blocks_28_attn_ln_bias_to_fp16, epsilon = var_3124_to_fp16, gamma = blocks_28_attn_ln_weight_to_fp16, x = x_343_cast_fp16)[name = tensor("op_3135_cast_fp16")]; + tensor var_3146_to_fp16 = const()[name = tensor("op_3146_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116558656)))]; + tensor var_3147_to_fp16 = const()[name = tensor("op_3147_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1119835520)))]; + tensor linear_168_cast_fp16 = linear(bias = var_3147_to_fp16, weight = var_3146_to_fp16, x = var_3135_cast_fp16)[name = tensor("linear_168_cast_fp16")]; + tensor var_3150_to_fp16 = const()[name = tensor("op_3150_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1119838144)))]; + tensor linear_169_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3150_to_fp16, x = var_3135_cast_fp16)[name = tensor("linear_169_cast_fp16")]; + tensor var_3154_to_fp16 = const()[name = tensor("op_3154_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1123115008)))]; + tensor var_3155_to_fp16 = const()[name = tensor("op_3155_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1126391872)))]; + tensor linear_170_cast_fp16 = linear(bias = var_3155_to_fp16, weight = var_3154_to_fp16, x = var_3135_cast_fp16)[name = tensor("linear_170_cast_fp16")]; + tensor var_3163 = const()[name = tensor("op_3163"), val = tensor([1, 1500, 20, -1])]; + tensor var_3164_cast_fp16 = reshape(shape = var_3163, x = linear_168_cast_fp16)[name = tensor("op_3164_cast_fp16")]; + tensor const_280_to_fp16 = const()[name = tensor("const_280_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_115_cast_fp16 = mul(x = var_3164_cast_fp16, y = const_280_to_fp16)[name = tensor("q_115_cast_fp16")]; + tensor var_3170 = const()[name = tensor("op_3170"), val = tensor([1, 1500, 20, -1])]; + tensor var_3171_cast_fp16 = reshape(shape = var_3170, x = linear_169_cast_fp16)[name = tensor("op_3171_cast_fp16")]; + tensor const_281_to_fp16 = const()[name = tensor("const_281_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_115_cast_fp16 = mul(x = var_3171_cast_fp16, y = const_281_to_fp16)[name = tensor("k_115_cast_fp16")]; + tensor var_3177 = const()[name = tensor("op_3177"), val = tensor([1, 1500, 20, -1])]; + tensor var_3178_cast_fp16 = reshape(shape = var_3177, x = linear_170_cast_fp16)[name = tensor("op_3178_cast_fp16")]; + tensor var_3179 = const()[name = tensor("op_3179"), val = tensor([0, 2, 1, 3])]; + tensor qk_57_transpose_x_0 = const()[name = tensor("qk_57_transpose_x_0"), val = tensor(false)]; + tensor qk_57_transpose_y_0 = const()[name = tensor("qk_57_transpose_y_0"), val = tensor(false)]; + tensor transpose_184_perm_0 = const()[name = tensor("transpose_184_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_185_perm_0 = const()[name = tensor("transpose_185_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_205 = transpose(perm = transpose_185_perm_0, x = k_115_cast_fp16)[name = tensor("transpose_205")]; + tensor transpose_206 = transpose(perm = transpose_184_perm_0, x = q_115_cast_fp16)[name = tensor("transpose_206")]; + tensor qk_57_cast_fp16 = matmul(transpose_x = qk_57_transpose_x_0, transpose_y = qk_57_transpose_y_0, x = transpose_206, y = transpose_205)[name = tensor("qk_57_cast_fp16")]; + tensor var_3183_cast_fp16 = softmax(axis = var_3118, x = qk_57_cast_fp16)[name = tensor("op_3183_cast_fp16")]; + tensor var_3185_transpose_x_0 = const()[name = tensor("op_3185_transpose_x_0"), val = tensor(false)]; + tensor var_3185_transpose_y_0 = const()[name = tensor("op_3185_transpose_y_0"), val = tensor(false)]; + tensor transpose_207 = transpose(perm = var_3179, x = var_3178_cast_fp16)[name = tensor("transpose_207")]; + tensor var_3185_cast_fp16 = matmul(transpose_x = var_3185_transpose_x_0, transpose_y = var_3185_transpose_y_0, x = var_3183_cast_fp16, y = transpose_207)[name = tensor("op_3185_cast_fp16")]; + tensor var_3186 = const()[name = tensor("op_3186"), val = tensor([0, 2, 1, 3])]; + tensor concat_28 = const()[name = tensor("concat_28"), val = tensor([1, 1500, 1280])]; + tensor transpose_204 = transpose(perm = var_3186, x = var_3185_cast_fp16)[name = tensor("transpose_204")]; + tensor x_347_cast_fp16 = reshape(shape = concat_28, x = transpose_204)[name = tensor("x_347_cast_fp16")]; + tensor var_3191_to_fp16 = const()[name = tensor("op_3191_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1126394496)))]; + tensor var_3192_to_fp16 = const()[name = tensor("op_3192_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1129671360)))]; + tensor linear_171_cast_fp16 = linear(bias = var_3192_to_fp16, weight = var_3191_to_fp16, x = x_347_cast_fp16)[name = tensor("linear_171_cast_fp16")]; + tensor x_349_cast_fp16 = add(x = x_343_cast_fp16, y = linear_171_cast_fp16)[name = tensor("x_349_cast_fp16")]; + tensor var_3199_axes_0 = const()[name = tensor("op_3199_axes_0"), val = tensor([-1])]; + tensor blocks_28_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_28_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1129673984)))]; + tensor blocks_28_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_28_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1129676608)))]; + tensor var_3199_cast_fp16 = layer_norm(axes = var_3199_axes_0, beta = blocks_28_mlp_ln_bias_to_fp16, epsilon = var_3124_to_fp16, gamma = blocks_28_mlp_ln_weight_to_fp16, x = x_349_cast_fp16)[name = tensor("op_3199_cast_fp16")]; + tensor var_3208_to_fp16 = const()[name = tensor("op_3208_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1129679232)))]; + tensor var_3209_to_fp16 = const()[name = tensor("op_3209_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1142786496)))]; + tensor linear_172_cast_fp16 = linear(bias = var_3209_to_fp16, weight = var_3208_to_fp16, x = var_3199_cast_fp16)[name = tensor("linear_172_cast_fp16")]; + tensor x_353_mode_0 = const()[name = tensor("x_353_mode_0"), val = tensor("EXACT")]; + tensor x_353_cast_fp16 = gelu(mode = x_353_mode_0, x = linear_172_cast_fp16)[name = tensor("x_353_cast_fp16")]; + tensor var_3214_to_fp16 = const()[name = tensor("op_3214_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1142796800)))]; + tensor var_3215_to_fp16 = const()[name = tensor("op_3215_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1155904064)))]; + tensor linear_173_cast_fp16 = linear(bias = var_3215_to_fp16, weight = var_3214_to_fp16, x = x_353_cast_fp16)[name = tensor("linear_173_cast_fp16")]; + tensor x_355_cast_fp16 = add(x = x_349_cast_fp16, y = linear_173_cast_fp16)[name = tensor("x_355_cast_fp16")]; + tensor var_3225 = const()[name = tensor("op_3225"), val = tensor(-1)]; + tensor var_3242_axes_0 = const()[name = tensor("op_3242_axes_0"), val = tensor([-1])]; + tensor blocks_29_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_29_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1155906688)))]; + tensor blocks_29_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_29_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1155909312)))]; + tensor var_3231_to_fp16 = const()[name = tensor("op_3231_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3242_cast_fp16 = layer_norm(axes = var_3242_axes_0, beta = blocks_29_attn_ln_bias_to_fp16, epsilon = var_3231_to_fp16, gamma = blocks_29_attn_ln_weight_to_fp16, x = x_355_cast_fp16)[name = tensor("op_3242_cast_fp16")]; + tensor var_3253_to_fp16 = const()[name = tensor("op_3253_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1155911936)))]; + tensor var_3254_to_fp16 = const()[name = tensor("op_3254_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1159188800)))]; + tensor linear_174_cast_fp16 = linear(bias = var_3254_to_fp16, weight = var_3253_to_fp16, x = var_3242_cast_fp16)[name = tensor("linear_174_cast_fp16")]; + tensor var_3257_to_fp16 = const()[name = tensor("op_3257_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1159191424)))]; + tensor linear_175_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3257_to_fp16, x = var_3242_cast_fp16)[name = tensor("linear_175_cast_fp16")]; + tensor var_3261_to_fp16 = const()[name = tensor("op_3261_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1162468288)))]; + tensor var_3262_to_fp16 = const()[name = tensor("op_3262_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1165745152)))]; + tensor linear_176_cast_fp16 = linear(bias = var_3262_to_fp16, weight = var_3261_to_fp16, x = var_3242_cast_fp16)[name = tensor("linear_176_cast_fp16")]; + tensor var_3270 = const()[name = tensor("op_3270"), val = tensor([1, 1500, 20, -1])]; + tensor var_3271_cast_fp16 = reshape(shape = var_3270, x = linear_174_cast_fp16)[name = tensor("op_3271_cast_fp16")]; + tensor const_282_to_fp16 = const()[name = tensor("const_282_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_119_cast_fp16 = mul(x = var_3271_cast_fp16, y = const_282_to_fp16)[name = tensor("q_119_cast_fp16")]; + tensor var_3277 = const()[name = tensor("op_3277"), val = tensor([1, 1500, 20, -1])]; + tensor var_3278_cast_fp16 = reshape(shape = var_3277, x = linear_175_cast_fp16)[name = tensor("op_3278_cast_fp16")]; + tensor const_283_to_fp16 = const()[name = tensor("const_283_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_119_cast_fp16 = mul(x = var_3278_cast_fp16, y = const_283_to_fp16)[name = tensor("k_119_cast_fp16")]; + tensor var_3284 = const()[name = tensor("op_3284"), val = tensor([1, 1500, 20, -1])]; + tensor var_3285_cast_fp16 = reshape(shape = var_3284, x = linear_176_cast_fp16)[name = tensor("op_3285_cast_fp16")]; + tensor var_3286 = const()[name = tensor("op_3286"), val = tensor([0, 2, 1, 3])]; + tensor qk_59_transpose_x_0 = const()[name = tensor("qk_59_transpose_x_0"), val = tensor(false)]; + tensor qk_59_transpose_y_0 = const()[name = tensor("qk_59_transpose_y_0"), val = tensor(false)]; + tensor transpose_186_perm_0 = const()[name = tensor("transpose_186_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_187_perm_0 = const()[name = tensor("transpose_187_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_201 = transpose(perm = transpose_187_perm_0, x = k_119_cast_fp16)[name = tensor("transpose_201")]; + tensor transpose_202 = transpose(perm = transpose_186_perm_0, x = q_119_cast_fp16)[name = tensor("transpose_202")]; + tensor qk_59_cast_fp16 = matmul(transpose_x = qk_59_transpose_x_0, transpose_y = qk_59_transpose_y_0, x = transpose_202, y = transpose_201)[name = tensor("qk_59_cast_fp16")]; + tensor var_3290_cast_fp16 = softmax(axis = var_3225, x = qk_59_cast_fp16)[name = tensor("op_3290_cast_fp16")]; + tensor var_3292_transpose_x_0 = const()[name = tensor("op_3292_transpose_x_0"), val = tensor(false)]; + tensor var_3292_transpose_y_0 = const()[name = tensor("op_3292_transpose_y_0"), val = tensor(false)]; + tensor transpose_203 = transpose(perm = var_3286, x = var_3285_cast_fp16)[name = tensor("transpose_203")]; + tensor var_3292_cast_fp16 = matmul(transpose_x = var_3292_transpose_x_0, transpose_y = var_3292_transpose_y_0, x = var_3290_cast_fp16, y = transpose_203)[name = tensor("op_3292_cast_fp16")]; + tensor var_3293 = const()[name = tensor("op_3293"), val = tensor([0, 2, 1, 3])]; + tensor concat_29 = const()[name = tensor("concat_29"), val = tensor([1, 1500, 1280])]; + tensor transpose_200 = transpose(perm = var_3293, x = var_3292_cast_fp16)[name = tensor("transpose_200")]; + tensor x_359_cast_fp16 = reshape(shape = concat_29, x = transpose_200)[name = tensor("x_359_cast_fp16")]; + tensor var_3298_to_fp16 = const()[name = tensor("op_3298_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1165747776)))]; + tensor var_3299_to_fp16 = const()[name = tensor("op_3299_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1169024640)))]; + tensor linear_177_cast_fp16 = linear(bias = var_3299_to_fp16, weight = var_3298_to_fp16, x = x_359_cast_fp16)[name = tensor("linear_177_cast_fp16")]; + tensor x_361_cast_fp16 = add(x = x_355_cast_fp16, y = linear_177_cast_fp16)[name = tensor("x_361_cast_fp16")]; + tensor var_3306_axes_0 = const()[name = tensor("op_3306_axes_0"), val = tensor([-1])]; + tensor blocks_29_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_29_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1169027264)))]; + tensor blocks_29_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_29_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1169029888)))]; + tensor var_3306_cast_fp16 = layer_norm(axes = var_3306_axes_0, beta = blocks_29_mlp_ln_bias_to_fp16, epsilon = var_3231_to_fp16, gamma = blocks_29_mlp_ln_weight_to_fp16, x = x_361_cast_fp16)[name = tensor("op_3306_cast_fp16")]; + tensor var_3315_to_fp16 = const()[name = tensor("op_3315_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1169032512)))]; + tensor var_3316_to_fp16 = const()[name = tensor("op_3316_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1182139776)))]; + tensor linear_178_cast_fp16 = linear(bias = var_3316_to_fp16, weight = var_3315_to_fp16, x = var_3306_cast_fp16)[name = tensor("linear_178_cast_fp16")]; + tensor x_365_mode_0 = const()[name = tensor("x_365_mode_0"), val = tensor("EXACT")]; + tensor x_365_cast_fp16 = gelu(mode = x_365_mode_0, x = linear_178_cast_fp16)[name = tensor("x_365_cast_fp16")]; + tensor var_3321_to_fp16 = const()[name = tensor("op_3321_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1182150080)))]; + tensor var_3322_to_fp16 = const()[name = tensor("op_3322_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1195257344)))]; + tensor linear_179_cast_fp16 = linear(bias = var_3322_to_fp16, weight = var_3321_to_fp16, x = x_365_cast_fp16)[name = tensor("linear_179_cast_fp16")]; + tensor x_367_cast_fp16 = add(x = x_361_cast_fp16, y = linear_179_cast_fp16)[name = tensor("x_367_cast_fp16")]; + tensor var_3332 = const()[name = tensor("op_3332"), val = tensor(-1)]; + tensor var_3349_axes_0 = const()[name = tensor("op_3349_axes_0"), val = tensor([-1])]; + tensor blocks_30_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_30_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1195259968)))]; + tensor blocks_30_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_30_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1195262592)))]; + tensor var_3338_to_fp16 = const()[name = tensor("op_3338_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3349_cast_fp16 = layer_norm(axes = var_3349_axes_0, beta = blocks_30_attn_ln_bias_to_fp16, epsilon = var_3338_to_fp16, gamma = blocks_30_attn_ln_weight_to_fp16, x = x_367_cast_fp16)[name = tensor("op_3349_cast_fp16")]; + tensor var_3360_to_fp16 = const()[name = tensor("op_3360_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1195265216)))]; + tensor var_3361_to_fp16 = const()[name = tensor("op_3361_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1198542080)))]; + tensor linear_180_cast_fp16 = linear(bias = var_3361_to_fp16, weight = var_3360_to_fp16, x = var_3349_cast_fp16)[name = tensor("linear_180_cast_fp16")]; + tensor var_3364_to_fp16 = const()[name = tensor("op_3364_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1198544704)))]; + tensor linear_181_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3364_to_fp16, x = var_3349_cast_fp16)[name = tensor("linear_181_cast_fp16")]; + tensor var_3368_to_fp16 = const()[name = tensor("op_3368_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1201821568)))]; + tensor var_3369_to_fp16 = const()[name = tensor("op_3369_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1205098432)))]; + tensor linear_182_cast_fp16 = linear(bias = var_3369_to_fp16, weight = var_3368_to_fp16, x = var_3349_cast_fp16)[name = tensor("linear_182_cast_fp16")]; + tensor var_3377 = const()[name = tensor("op_3377"), val = tensor([1, 1500, 20, -1])]; + tensor var_3378_cast_fp16 = reshape(shape = var_3377, x = linear_180_cast_fp16)[name = tensor("op_3378_cast_fp16")]; + tensor const_284_to_fp16 = const()[name = tensor("const_284_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_123_cast_fp16 = mul(x = var_3378_cast_fp16, y = const_284_to_fp16)[name = tensor("q_123_cast_fp16")]; + tensor var_3384 = const()[name = tensor("op_3384"), val = tensor([1, 1500, 20, -1])]; + tensor var_3385_cast_fp16 = reshape(shape = var_3384, x = linear_181_cast_fp16)[name = tensor("op_3385_cast_fp16")]; + tensor const_285_to_fp16 = const()[name = tensor("const_285_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_123_cast_fp16 = mul(x = var_3385_cast_fp16, y = const_285_to_fp16)[name = tensor("k_123_cast_fp16")]; + tensor var_3391 = const()[name = tensor("op_3391"), val = tensor([1, 1500, 20, -1])]; + tensor var_3392_cast_fp16 = reshape(shape = var_3391, x = linear_182_cast_fp16)[name = tensor("op_3392_cast_fp16")]; + tensor var_3393 = const()[name = tensor("op_3393"), val = tensor([0, 2, 1, 3])]; + tensor qk_61_transpose_x_0 = const()[name = tensor("qk_61_transpose_x_0"), val = tensor(false)]; + tensor qk_61_transpose_y_0 = const()[name = tensor("qk_61_transpose_y_0"), val = tensor(false)]; + tensor transpose_188_perm_0 = const()[name = tensor("transpose_188_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_189_perm_0 = const()[name = tensor("transpose_189_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_197 = transpose(perm = transpose_189_perm_0, x = k_123_cast_fp16)[name = tensor("transpose_197")]; + tensor transpose_198 = transpose(perm = transpose_188_perm_0, x = q_123_cast_fp16)[name = tensor("transpose_198")]; + tensor qk_61_cast_fp16 = matmul(transpose_x = qk_61_transpose_x_0, transpose_y = qk_61_transpose_y_0, x = transpose_198, y = transpose_197)[name = tensor("qk_61_cast_fp16")]; + tensor var_3397_cast_fp16 = softmax(axis = var_3332, x = qk_61_cast_fp16)[name = tensor("op_3397_cast_fp16")]; + tensor var_3399_transpose_x_0 = const()[name = tensor("op_3399_transpose_x_0"), val = tensor(false)]; + tensor var_3399_transpose_y_0 = const()[name = tensor("op_3399_transpose_y_0"), val = tensor(false)]; + tensor transpose_199 = transpose(perm = var_3393, x = var_3392_cast_fp16)[name = tensor("transpose_199")]; + tensor var_3399_cast_fp16 = matmul(transpose_x = var_3399_transpose_x_0, transpose_y = var_3399_transpose_y_0, x = var_3397_cast_fp16, y = transpose_199)[name = tensor("op_3399_cast_fp16")]; + tensor var_3400 = const()[name = tensor("op_3400"), val = tensor([0, 2, 1, 3])]; + tensor concat_30 = const()[name = tensor("concat_30"), val = tensor([1, 1500, 1280])]; + tensor transpose_196 = transpose(perm = var_3400, x = var_3399_cast_fp16)[name = tensor("transpose_196")]; + tensor x_371_cast_fp16 = reshape(shape = concat_30, x = transpose_196)[name = tensor("x_371_cast_fp16")]; + tensor var_3405_to_fp16 = const()[name = tensor("op_3405_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1205101056)))]; + tensor var_3406_to_fp16 = const()[name = tensor("op_3406_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208377920)))]; + tensor linear_183_cast_fp16 = linear(bias = var_3406_to_fp16, weight = var_3405_to_fp16, x = x_371_cast_fp16)[name = tensor("linear_183_cast_fp16")]; + tensor x_373_cast_fp16 = add(x = x_367_cast_fp16, y = linear_183_cast_fp16)[name = tensor("x_373_cast_fp16")]; + tensor var_3413_axes_0 = const()[name = tensor("op_3413_axes_0"), val = tensor([-1])]; + tensor blocks_30_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_30_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208380544)))]; + tensor blocks_30_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_30_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208383168)))]; + tensor var_3413_cast_fp16 = layer_norm(axes = var_3413_axes_0, beta = blocks_30_mlp_ln_bias_to_fp16, epsilon = var_3338_to_fp16, gamma = blocks_30_mlp_ln_weight_to_fp16, x = x_373_cast_fp16)[name = tensor("op_3413_cast_fp16")]; + tensor var_3422_to_fp16 = const()[name = tensor("op_3422_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208385792)))]; + tensor var_3423_to_fp16 = const()[name = tensor("op_3423_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1221493056)))]; + tensor linear_184_cast_fp16 = linear(bias = var_3423_to_fp16, weight = var_3422_to_fp16, x = var_3413_cast_fp16)[name = tensor("linear_184_cast_fp16")]; + tensor x_377_mode_0 = const()[name = tensor("x_377_mode_0"), val = tensor("EXACT")]; + tensor x_377_cast_fp16 = gelu(mode = x_377_mode_0, x = linear_184_cast_fp16)[name = tensor("x_377_cast_fp16")]; + tensor var_3428_to_fp16 = const()[name = tensor("op_3428_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1221503360)))]; + tensor var_3429_to_fp16 = const()[name = tensor("op_3429_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1234610624)))]; + tensor linear_185_cast_fp16 = linear(bias = var_3429_to_fp16, weight = var_3428_to_fp16, x = x_377_cast_fp16)[name = tensor("linear_185_cast_fp16")]; + tensor x_379_cast_fp16 = add(x = x_373_cast_fp16, y = linear_185_cast_fp16)[name = tensor("x_379_cast_fp16")]; + tensor var_3439 = const()[name = tensor("op_3439"), val = tensor(-1)]; + tensor var_3456_axes_0 = const()[name = tensor("op_3456_axes_0"), val = tensor([-1])]; + tensor blocks_31_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_31_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1234613248)))]; + tensor blocks_31_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_31_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1234615872)))]; + tensor var_3445_to_fp16 = const()[name = tensor("op_3445_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3456_cast_fp16 = layer_norm(axes = var_3456_axes_0, beta = blocks_31_attn_ln_bias_to_fp16, epsilon = var_3445_to_fp16, gamma = blocks_31_attn_ln_weight_to_fp16, x = x_379_cast_fp16)[name = tensor("op_3456_cast_fp16")]; + tensor var_3467_to_fp16 = const()[name = tensor("op_3467_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1234618496)))]; + tensor var_3468_to_fp16 = const()[name = tensor("op_3468_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1237895360)))]; + tensor linear_186_cast_fp16 = linear(bias = var_3468_to_fp16, weight = var_3467_to_fp16, x = var_3456_cast_fp16)[name = tensor("linear_186_cast_fp16")]; + tensor var_3471_to_fp16 = const()[name = tensor("op_3471_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1237897984)))]; + tensor linear_187_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3471_to_fp16, x = var_3456_cast_fp16)[name = tensor("linear_187_cast_fp16")]; + tensor var_3475_to_fp16 = const()[name = tensor("op_3475_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1241174848)))]; + tensor var_3476_to_fp16 = const()[name = tensor("op_3476_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1244451712)))]; + tensor linear_188_cast_fp16 = linear(bias = var_3476_to_fp16, weight = var_3475_to_fp16, x = var_3456_cast_fp16)[name = tensor("linear_188_cast_fp16")]; + tensor var_3484 = const()[name = tensor("op_3484"), val = tensor([1, 1500, 20, -1])]; + tensor var_3485_cast_fp16 = reshape(shape = var_3484, x = linear_186_cast_fp16)[name = tensor("op_3485_cast_fp16")]; + tensor const_286_to_fp16 = const()[name = tensor("const_286_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_cast_fp16 = mul(x = var_3485_cast_fp16, y = const_286_to_fp16)[name = tensor("q_cast_fp16")]; + tensor var_3491 = const()[name = tensor("op_3491"), val = tensor([1, 1500, 20, -1])]; + tensor var_3492_cast_fp16 = reshape(shape = var_3491, x = linear_187_cast_fp16)[name = tensor("op_3492_cast_fp16")]; + tensor const_287_to_fp16 = const()[name = tensor("const_287_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_cast_fp16 = mul(x = var_3492_cast_fp16, y = const_287_to_fp16)[name = tensor("k_cast_fp16")]; + tensor var_3498 = const()[name = tensor("op_3498"), val = tensor([1, 1500, 20, -1])]; + tensor var_3499_cast_fp16 = reshape(shape = var_3498, x = linear_188_cast_fp16)[name = tensor("op_3499_cast_fp16")]; + tensor var_3500 = const()[name = tensor("op_3500"), val = tensor([0, 2, 1, 3])]; + tensor qk_transpose_x_0 = const()[name = tensor("qk_transpose_x_0"), val = tensor(false)]; + tensor qk_transpose_y_0 = const()[name = tensor("qk_transpose_y_0"), val = tensor(false)]; + tensor transpose_190_perm_0 = const()[name = tensor("transpose_190_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_191_perm_0 = const()[name = tensor("transpose_191_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_193 = transpose(perm = transpose_191_perm_0, x = k_cast_fp16)[name = tensor("transpose_193")]; + tensor transpose_194 = transpose(perm = transpose_190_perm_0, x = q_cast_fp16)[name = tensor("transpose_194")]; + tensor qk_cast_fp16 = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_194, y = transpose_193)[name = tensor("qk_cast_fp16")]; + tensor var_3504_cast_fp16 = softmax(axis = var_3439, x = qk_cast_fp16)[name = tensor("op_3504_cast_fp16")]; + tensor var_3506_transpose_x_0 = const()[name = tensor("op_3506_transpose_x_0"), val = tensor(false)]; + tensor var_3506_transpose_y_0 = const()[name = tensor("op_3506_transpose_y_0"), val = tensor(false)]; + tensor transpose_195 = transpose(perm = var_3500, x = var_3499_cast_fp16)[name = tensor("transpose_195")]; + tensor var_3506_cast_fp16 = matmul(transpose_x = var_3506_transpose_x_0, transpose_y = var_3506_transpose_y_0, x = var_3504_cast_fp16, y = transpose_195)[name = tensor("op_3506_cast_fp16")]; + tensor var_3507 = const()[name = tensor("op_3507"), val = tensor([0, 2, 1, 3])]; + tensor concat_31 = const()[name = tensor("concat_31"), val = tensor([1, 1500, 1280])]; + tensor transpose_192 = transpose(perm = var_3507, x = var_3506_cast_fp16)[name = tensor("transpose_192")]; + tensor x_383_cast_fp16 = reshape(shape = concat_31, x = transpose_192)[name = tensor("x_383_cast_fp16")]; + tensor var_3512_to_fp16 = const()[name = tensor("op_3512_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1244454336)))]; + tensor var_3513_to_fp16 = const()[name = tensor("op_3513_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1247731200)))]; + tensor linear_189_cast_fp16 = linear(bias = var_3513_to_fp16, weight = var_3512_to_fp16, x = x_383_cast_fp16)[name = tensor("linear_189_cast_fp16")]; + tensor x_385_cast_fp16 = add(x = x_379_cast_fp16, y = linear_189_cast_fp16)[name = tensor("x_385_cast_fp16")]; + tensor var_3520_axes_0 = const()[name = tensor("op_3520_axes_0"), val = tensor([-1])]; + tensor blocks_31_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_31_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1247733824)))]; + tensor blocks_31_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_31_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1247736448)))]; + tensor var_3520_cast_fp16 = layer_norm(axes = var_3520_axes_0, beta = blocks_31_mlp_ln_bias_to_fp16, epsilon = var_3445_to_fp16, gamma = blocks_31_mlp_ln_weight_to_fp16, x = x_385_cast_fp16)[name = tensor("op_3520_cast_fp16")]; + tensor var_3529_to_fp16 = const()[name = tensor("op_3529_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1247739072)))]; + tensor var_3530_to_fp16 = const()[name = tensor("op_3530_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1260846336)))]; + tensor linear_190_cast_fp16 = linear(bias = var_3530_to_fp16, weight = var_3529_to_fp16, x = var_3520_cast_fp16)[name = tensor("linear_190_cast_fp16")]; + tensor x_389_mode_0 = const()[name = tensor("x_389_mode_0"), val = tensor("EXACT")]; + tensor x_389_cast_fp16 = gelu(mode = x_389_mode_0, x = linear_190_cast_fp16)[name = tensor("x_389_cast_fp16")]; + tensor var_3535_to_fp16 = const()[name = tensor("op_3535_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1260856640)))]; + tensor var_3536_to_fp16 = const()[name = tensor("op_3536_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1273963904)))]; + tensor linear_191_cast_fp16 = linear(bias = var_3536_to_fp16, weight = var_3535_to_fp16, x = x_389_cast_fp16)[name = tensor("linear_191_cast_fp16")]; + tensor x_cast_fp16 = add(x = x_385_cast_fp16, y = linear_191_cast_fp16)[name = tensor("x_cast_fp16")]; + tensor var_3550_axes_0 = const()[name = tensor("op_3550_axes_0"), val = tensor([-1])]; + tensor ln_post_weight_to_fp16 = const()[name = tensor("ln_post_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1273966528)))]; + tensor ln_post_bias_to_fp16 = const()[name = tensor("ln_post_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1273969152)))]; + tensor var_3541_to_fp16 = const()[name = tensor("op_3541_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3550_cast_fp16 = layer_norm(axes = var_3550_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_3541_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast_fp16)[name = tensor("op_3550_cast_fp16")]; + tensor var_3550_cast_fp16_to_fp32_dtype_0 = const()[name = tensor("op_3550_cast_fp16_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor output = cast(dtype = var_3550_cast_fp16_to_fp32_dtype_0, x = var_3550_cast_fp16)[name = tensor("cast_192")]; + } -> (output); +} \ No newline at end of file