diff --git "a/ggml-large-v2-encoder.mlmodelc/model.mil" "b/ggml-large-v2-encoder.mlmodelc/model.mil" new file mode 100644--- /dev/null +++ "b/ggml-large-v2-encoder.mlmodelc/model.mil" @@ -0,0 +1,1927 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "5.33.5"}, {"coremlc-version", "1877.40.3"}})] +{ + 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(614528)))]; + tensor cast_967 = cast(dtype = logmel_data_to_fp16_dtype_0, x = logmel_data); + tensor var_84_cast = 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_967); + tensor input_1_mode_0 = const()[name = tensor("input_1_mode_0"), val = tensor("EXACT")]; + tensor input_1_cast = gelu(mode = input_1_mode_0, x = var_84_cast); + 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(617152)))]; + tensor bias_7_to_fp16 = const()[name = tensor("bias_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10447616)))]; + tensor var_101_cast = 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); + tensor x_3_mode_0 = const()[name = tensor("x_3_mode_0"), val = tensor("EXACT")]; + tensor x_3_cast = gelu(mode = x_3_mode_0, x = var_101_cast); + 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(10450240)))]; + tensor transpose_256 = transpose(perm = var_106, x = x_3_cast); + tensor var_109_cast = add(x = transpose_256, y = positional_embedding_to_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(14290304)))]; + 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(14292928)))]; + tensor var_128_to_fp16 = const()[name = tensor("op_128_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_139_cast = 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); + tensor var_150_to_fp16 = const()[name = tensor("op_150_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14295552)))]; + tensor var_151_to_fp16 = const()[name = tensor("op_151_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17572416)))]; + tensor q_1_cast = linear(bias = var_151_to_fp16, weight = var_150_to_fp16, x = var_139_cast); + tensor var_154_to_fp16 = const()[name = tensor("op_154_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17575040)))]; + tensor k_1_bias_0_to_fp16 = const()[name = tensor("k_1_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20851904)))]; + tensor k_1_cast = linear(bias = k_1_bias_0_to_fp16, weight = var_154_to_fp16, x = var_139_cast); + tensor var_158_to_fp16 = const()[name = tensor("op_158_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20854528)))]; + tensor var_159_to_fp16 = const()[name = tensor("op_159_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24131392)))]; + tensor v_1_cast = linear(bias = var_159_to_fp16, weight = var_158_to_fp16, x = var_139_cast); + tensor var_167 = const()[name = tensor("op_167"), val = tensor([1, 1500, 20, -1])]; + tensor var_168_cast = reshape(shape = var_167, x = q_1_cast); + tensor const_224_to_fp16 = const()[name = tensor("const_224_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_3_cast = mul(x = var_168_cast, y = const_224_to_fp16); + tensor var_174 = const()[name = tensor("op_174"), val = tensor([1, 1500, 20, -1])]; + tensor var_175_cast = reshape(shape = var_174, x = k_1_cast); + tensor const_225_to_fp16 = const()[name = tensor("const_225_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_3_cast = mul(x = var_175_cast, y = const_225_to_fp16); + tensor var_181 = const()[name = tensor("op_181"), val = tensor([1, 1500, 20, -1])]; + tensor var_182_cast = reshape(shape = var_181, x = v_1_cast); + 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_64_perm_0 = const()[name = tensor("transpose_64_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_65_perm_0 = const()[name = tensor("transpose_65_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_253 = transpose(perm = transpose_65_perm_0, x = k_3_cast); + tensor transpose_254 = transpose(perm = transpose_64_perm_0, x = q_3_cast); + tensor qk_1_cast = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_254, y = transpose_253); + tensor var_187_cast = softmax(axis = var_122, x = qk_1_cast); + 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_255 = transpose(perm = var_183, x = var_182_cast); + tensor var_189_cast = matmul(transpose_x = var_189_transpose_x_0, transpose_y = var_189_transpose_y_0, x = var_187_cast, y = transpose_255); + 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_252 = transpose(perm = var_190, x = var_189_cast); + tensor x_11_cast = reshape(shape = concat_0, x = transpose_252); + tensor var_195_to_fp16 = const()[name = tensor("op_195_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24134016)))]; + tensor var_196_to_fp16 = const()[name = tensor("op_196_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27410880)))]; + tensor var_197_cast = linear(bias = var_196_to_fp16, weight = var_195_to_fp16, x = x_11_cast); + tensor x_13_cast = add(x = var_109_cast, y = var_197_cast); + 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(27413504)))]; + 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(27416128)))]; + tensor var_203_cast = 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); + tensor var_212_to_fp16 = const()[name = tensor("op_212_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27418752)))]; + tensor var_213_to_fp16 = const()[name = tensor("op_213_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40526016)))]; + tensor input_9_cast = linear(bias = var_213_to_fp16, weight = var_212_to_fp16, x = var_203_cast); + tensor x_17_mode_0 = const()[name = tensor("x_17_mode_0"), val = tensor("EXACT")]; + tensor x_17_cast = gelu(mode = x_17_mode_0, x = input_9_cast); + tensor var_218_to_fp16 = const()[name = tensor("op_218_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40536320)))]; + tensor var_219_to_fp16 = const()[name = tensor("op_219_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53643584)))]; + tensor var_220_cast = linear(bias = var_219_to_fp16, weight = var_218_to_fp16, x = x_17_cast); + tensor x_19_cast = add(x = x_13_cast, y = var_220_cast); + 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(53646208)))]; + 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(53648832)))]; + tensor var_235_to_fp16 = const()[name = tensor("op_235_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_246_cast = 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); + tensor var_257_to_fp16 = const()[name = tensor("op_257_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53651456)))]; + tensor var_258_to_fp16 = const()[name = tensor("op_258_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56928320)))]; + tensor q_5_cast = linear(bias = var_258_to_fp16, weight = var_257_to_fp16, x = var_246_cast); + tensor var_261_to_fp16 = const()[name = tensor("op_261_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56930944)))]; + tensor k_5_bias_0_to_fp16 = const()[name = tensor("k_5_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60207808)))]; + tensor k_5_cast = linear(bias = k_5_bias_0_to_fp16, weight = var_261_to_fp16, x = var_246_cast); + tensor var_265_to_fp16 = const()[name = tensor("op_265_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60210432)))]; + tensor var_266_to_fp16 = const()[name = tensor("op_266_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63487296)))]; + tensor v_5_cast = linear(bias = var_266_to_fp16, weight = var_265_to_fp16, x = var_246_cast); + tensor var_274 = const()[name = tensor("op_274"), val = tensor([1, 1500, 20, -1])]; + tensor var_275_cast = reshape(shape = var_274, x = q_5_cast); + tensor const_226_to_fp16 = const()[name = tensor("const_226_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_7_cast = mul(x = var_275_cast, y = const_226_to_fp16); + tensor var_281 = const()[name = tensor("op_281"), val = tensor([1, 1500, 20, -1])]; + tensor var_282_cast = reshape(shape = var_281, x = k_5_cast); + tensor const_227_to_fp16 = const()[name = tensor("const_227_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_7_cast = mul(x = var_282_cast, y = const_227_to_fp16); + tensor var_288 = const()[name = tensor("op_288"), val = tensor([1, 1500, 20, -1])]; + tensor var_289_cast = reshape(shape = var_288, x = v_5_cast); + 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_66_perm_0 = const()[name = tensor("transpose_66_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_67_perm_0 = const()[name = tensor("transpose_67_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_249 = transpose(perm = transpose_67_perm_0, x = k_7_cast); + tensor transpose_250 = transpose(perm = transpose_66_perm_0, x = q_7_cast); + tensor qk_3_cast = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_250, y = transpose_249); + tensor var_294_cast = softmax(axis = var_229, x = qk_3_cast); + 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_251 = transpose(perm = var_290, x = var_289_cast); + tensor var_296_cast = matmul(transpose_x = var_296_transpose_x_0, transpose_y = var_296_transpose_y_0, x = var_294_cast, y = transpose_251); + 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_248 = transpose(perm = var_297, x = var_296_cast); + tensor x_23_cast = reshape(shape = concat_1, x = transpose_248); + tensor var_302_to_fp16 = const()[name = tensor("op_302_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63489920)))]; + tensor var_303_to_fp16 = const()[name = tensor("op_303_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66766784)))]; + tensor var_304_cast = linear(bias = var_303_to_fp16, weight = var_302_to_fp16, x = x_23_cast); + tensor x_25_cast = add(x = x_19_cast, y = var_304_cast); + 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(66769408)))]; + 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(66772032)))]; + tensor var_310_cast = 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); + tensor var_319_to_fp16 = const()[name = tensor("op_319_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66774656)))]; + tensor var_320_to_fp16 = const()[name = tensor("op_320_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79881920)))]; + tensor input_17_cast = linear(bias = var_320_to_fp16, weight = var_319_to_fp16, x = var_310_cast); + tensor x_29_mode_0 = const()[name = tensor("x_29_mode_0"), val = tensor("EXACT")]; + tensor x_29_cast = gelu(mode = x_29_mode_0, x = input_17_cast); + tensor var_325_to_fp16 = const()[name = tensor("op_325_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79892224)))]; + tensor var_326_to_fp16 = const()[name = tensor("op_326_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92999488)))]; + tensor var_327_cast = linear(bias = var_326_to_fp16, weight = var_325_to_fp16, x = x_29_cast); + tensor x_31_cast = add(x = x_25_cast, y = var_327_cast); + 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(93002112)))]; + 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(93004736)))]; + tensor var_342_to_fp16 = const()[name = tensor("op_342_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_353_cast = 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); + tensor var_364_to_fp16 = const()[name = tensor("op_364_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93007360)))]; + tensor var_365_to_fp16 = const()[name = tensor("op_365_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96284224)))]; + tensor q_9_cast = linear(bias = var_365_to_fp16, weight = var_364_to_fp16, x = var_353_cast); + tensor var_368_to_fp16 = const()[name = tensor("op_368_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96286848)))]; + tensor k_9_bias_0_to_fp16 = const()[name = tensor("k_9_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99563712)))]; + tensor k_9_cast = linear(bias = k_9_bias_0_to_fp16, weight = var_368_to_fp16, x = var_353_cast); + tensor var_372_to_fp16 = const()[name = tensor("op_372_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99566336)))]; + tensor var_373_to_fp16 = const()[name = tensor("op_373_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102843200)))]; + tensor v_9_cast = linear(bias = var_373_to_fp16, weight = var_372_to_fp16, x = var_353_cast); + tensor var_381 = const()[name = tensor("op_381"), val = tensor([1, 1500, 20, -1])]; + tensor var_382_cast = reshape(shape = var_381, x = q_9_cast); + tensor const_228_to_fp16 = const()[name = tensor("const_228_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_11_cast = mul(x = var_382_cast, y = const_228_to_fp16); + tensor var_388 = const()[name = tensor("op_388"), val = tensor([1, 1500, 20, -1])]; + tensor var_389_cast = reshape(shape = var_388, x = k_9_cast); + tensor const_229_to_fp16 = const()[name = tensor("const_229_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_11_cast = mul(x = var_389_cast, y = const_229_to_fp16); + tensor var_395 = const()[name = tensor("op_395"), val = tensor([1, 1500, 20, -1])]; + tensor var_396_cast = reshape(shape = var_395, x = v_9_cast); + 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_68_perm_0 = const()[name = tensor("transpose_68_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_69_perm_0 = const()[name = tensor("transpose_69_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_245 = transpose(perm = transpose_69_perm_0, x = k_11_cast); + tensor transpose_246 = transpose(perm = transpose_68_perm_0, x = q_11_cast); + tensor qk_5_cast = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_246, y = transpose_245); + tensor var_401_cast = softmax(axis = var_336, x = qk_5_cast); + 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_247 = transpose(perm = var_397, x = var_396_cast); + tensor var_403_cast = matmul(transpose_x = var_403_transpose_x_0, transpose_y = var_403_transpose_y_0, x = var_401_cast, y = transpose_247); + 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_244 = transpose(perm = var_404, x = var_403_cast); + tensor x_35_cast = reshape(shape = concat_2, x = transpose_244); + tensor var_409_to_fp16 = const()[name = tensor("op_409_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102845824)))]; + tensor var_410_to_fp16 = const()[name = tensor("op_410_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106122688)))]; + tensor var_411_cast = linear(bias = var_410_to_fp16, weight = var_409_to_fp16, x = x_35_cast); + tensor x_37_cast = add(x = x_31_cast, y = var_411_cast); + 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(106125312)))]; + 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(106127936)))]; + tensor var_417_cast = 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); + tensor var_426_to_fp16 = const()[name = tensor("op_426_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106130560)))]; + tensor var_427_to_fp16 = const()[name = tensor("op_427_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119237824)))]; + tensor input_25_cast = linear(bias = var_427_to_fp16, weight = var_426_to_fp16, x = var_417_cast); + tensor x_41_mode_0 = const()[name = tensor("x_41_mode_0"), val = tensor("EXACT")]; + tensor x_41_cast = gelu(mode = x_41_mode_0, x = input_25_cast); + tensor var_432_to_fp16 = const()[name = tensor("op_432_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119248128)))]; + tensor var_433_to_fp16 = const()[name = tensor("op_433_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132355392)))]; + tensor var_434_cast = linear(bias = var_433_to_fp16, weight = var_432_to_fp16, x = x_41_cast); + tensor x_43_cast = add(x = x_37_cast, y = var_434_cast); + 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(132358016)))]; + 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(132360640)))]; + tensor var_449_to_fp16 = const()[name = tensor("op_449_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_460_cast = 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); + tensor var_471_to_fp16 = const()[name = tensor("op_471_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132363264)))]; + tensor var_472_to_fp16 = const()[name = tensor("op_472_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135640128)))]; + tensor q_13_cast = linear(bias = var_472_to_fp16, weight = var_471_to_fp16, x = var_460_cast); + tensor var_475_to_fp16 = const()[name = tensor("op_475_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135642752)))]; + tensor k_13_bias_0_to_fp16 = const()[name = tensor("k_13_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138919616)))]; + tensor k_13_cast = linear(bias = k_13_bias_0_to_fp16, weight = var_475_to_fp16, x = var_460_cast); + tensor var_479_to_fp16 = const()[name = tensor("op_479_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138922240)))]; + tensor var_480_to_fp16 = const()[name = tensor("op_480_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142199104)))]; + tensor v_13_cast = linear(bias = var_480_to_fp16, weight = var_479_to_fp16, x = var_460_cast); + tensor var_488 = const()[name = tensor("op_488"), val = tensor([1, 1500, 20, -1])]; + tensor var_489_cast = reshape(shape = var_488, x = q_13_cast); + tensor const_230_to_fp16 = const()[name = tensor("const_230_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_15_cast = mul(x = var_489_cast, y = const_230_to_fp16); + tensor var_495 = const()[name = tensor("op_495"), val = tensor([1, 1500, 20, -1])]; + tensor var_496_cast = reshape(shape = var_495, x = k_13_cast); + tensor const_231_to_fp16 = const()[name = tensor("const_231_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_15_cast = mul(x = var_496_cast, y = const_231_to_fp16); + tensor var_502 = const()[name = tensor("op_502"), val = tensor([1, 1500, 20, -1])]; + tensor var_503_cast = reshape(shape = var_502, x = v_13_cast); + 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_70_perm_0 = const()[name = tensor("transpose_70_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_71_perm_0 = const()[name = tensor("transpose_71_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_241 = transpose(perm = transpose_71_perm_0, x = k_15_cast); + tensor transpose_242 = transpose(perm = transpose_70_perm_0, x = q_15_cast); + tensor qk_7_cast = matmul(transpose_x = qk_7_transpose_x_0, transpose_y = qk_7_transpose_y_0, x = transpose_242, y = transpose_241); + tensor var_508_cast = softmax(axis = var_443, x = qk_7_cast); + 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_243 = transpose(perm = var_504, x = var_503_cast); + tensor var_510_cast = matmul(transpose_x = var_510_transpose_x_0, transpose_y = var_510_transpose_y_0, x = var_508_cast, y = transpose_243); + 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_240 = transpose(perm = var_511, x = var_510_cast); + tensor x_47_cast = reshape(shape = concat_3, x = transpose_240); + tensor var_516_to_fp16 = const()[name = tensor("op_516_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142201728)))]; + tensor var_517_to_fp16 = const()[name = tensor("op_517_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145478592)))]; + tensor var_518_cast = linear(bias = var_517_to_fp16, weight = var_516_to_fp16, x = x_47_cast); + tensor x_49_cast = add(x = x_43_cast, y = var_518_cast); + 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(145481216)))]; + 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(145483840)))]; + tensor var_524_cast = 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); + tensor var_533_to_fp16 = const()[name = tensor("op_533_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145486464)))]; + tensor var_534_to_fp16 = const()[name = tensor("op_534_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158593728)))]; + tensor input_33_cast = linear(bias = var_534_to_fp16, weight = var_533_to_fp16, x = var_524_cast); + tensor x_53_mode_0 = const()[name = tensor("x_53_mode_0"), val = tensor("EXACT")]; + tensor x_53_cast = gelu(mode = x_53_mode_0, x = input_33_cast); + tensor var_539_to_fp16 = const()[name = tensor("op_539_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158604032)))]; + tensor var_540_to_fp16 = const()[name = tensor("op_540_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171711296)))]; + tensor var_541_cast = linear(bias = var_540_to_fp16, weight = var_539_to_fp16, x = x_53_cast); + tensor x_55_cast = add(x = x_49_cast, y = var_541_cast); + 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(171713920)))]; + 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(171716544)))]; + tensor var_556_to_fp16 = const()[name = tensor("op_556_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_567_cast = 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); + tensor var_578_to_fp16 = const()[name = tensor("op_578_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171719168)))]; + tensor var_579_to_fp16 = const()[name = tensor("op_579_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174996032)))]; + tensor q_17_cast = linear(bias = var_579_to_fp16, weight = var_578_to_fp16, x = var_567_cast); + tensor var_582_to_fp16 = const()[name = tensor("op_582_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174998656)))]; + tensor k_17_bias_0_to_fp16 = const()[name = tensor("k_17_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178275520)))]; + tensor k_17_cast = linear(bias = k_17_bias_0_to_fp16, weight = var_582_to_fp16, x = var_567_cast); + tensor var_586_to_fp16 = const()[name = tensor("op_586_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178278144)))]; + tensor var_587_to_fp16 = const()[name = tensor("op_587_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(181555008)))]; + tensor v_17_cast = linear(bias = var_587_to_fp16, weight = var_586_to_fp16, x = var_567_cast); + tensor var_595 = const()[name = tensor("op_595"), val = tensor([1, 1500, 20, -1])]; + tensor var_596_cast = reshape(shape = var_595, x = q_17_cast); + tensor const_232_to_fp16 = const()[name = tensor("const_232_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_19_cast = mul(x = var_596_cast, y = const_232_to_fp16); + tensor var_602 = const()[name = tensor("op_602"), val = tensor([1, 1500, 20, -1])]; + tensor var_603_cast = reshape(shape = var_602, x = k_17_cast); + tensor const_233_to_fp16 = const()[name = tensor("const_233_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_19_cast = mul(x = var_603_cast, y = const_233_to_fp16); + tensor var_609 = const()[name = tensor("op_609"), val = tensor([1, 1500, 20, -1])]; + tensor var_610_cast = reshape(shape = var_609, x = v_17_cast); + 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_72_perm_0 = const()[name = tensor("transpose_72_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_73_perm_0 = const()[name = tensor("transpose_73_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_237 = transpose(perm = transpose_73_perm_0, x = k_19_cast); + tensor transpose_238 = transpose(perm = transpose_72_perm_0, x = q_19_cast); + tensor qk_9_cast = matmul(transpose_x = qk_9_transpose_x_0, transpose_y = qk_9_transpose_y_0, x = transpose_238, y = transpose_237); + tensor var_615_cast = softmax(axis = var_550, x = qk_9_cast); + 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_239 = transpose(perm = var_611, x = var_610_cast); + tensor var_617_cast = matmul(transpose_x = var_617_transpose_x_0, transpose_y = var_617_transpose_y_0, x = var_615_cast, y = transpose_239); + 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_236 = transpose(perm = var_618, x = var_617_cast); + tensor x_59_cast = reshape(shape = concat_4, x = transpose_236); + tensor var_623_to_fp16 = const()[name = tensor("op_623_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(181557632)))]; + tensor var_624_to_fp16 = const()[name = tensor("op_624_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184834496)))]; + tensor var_625_cast = linear(bias = var_624_to_fp16, weight = var_623_to_fp16, x = x_59_cast); + tensor x_61_cast = add(x = x_55_cast, y = var_625_cast); + 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(184837120)))]; + 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(184839744)))]; + tensor var_631_cast = 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); + tensor var_640_to_fp16 = const()[name = tensor("op_640_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184842368)))]; + tensor var_641_to_fp16 = const()[name = tensor("op_641_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(197949632)))]; + tensor input_41_cast = linear(bias = var_641_to_fp16, weight = var_640_to_fp16, x = var_631_cast); + tensor x_65_mode_0 = const()[name = tensor("x_65_mode_0"), val = tensor("EXACT")]; + tensor x_65_cast = gelu(mode = x_65_mode_0, x = input_41_cast); + tensor var_646_to_fp16 = const()[name = tensor("op_646_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(197959936)))]; + tensor var_647_to_fp16 = const()[name = tensor("op_647_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211067200)))]; + tensor var_648_cast = linear(bias = var_647_to_fp16, weight = var_646_to_fp16, x = x_65_cast); + tensor x_67_cast = add(x = x_61_cast, y = var_648_cast); + 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(211069824)))]; + 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(211072448)))]; + tensor var_663_to_fp16 = const()[name = tensor("op_663_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_674_cast = 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); + tensor var_685_to_fp16 = const()[name = tensor("op_685_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211075072)))]; + tensor var_686_to_fp16 = const()[name = tensor("op_686_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214351936)))]; + tensor q_21_cast = linear(bias = var_686_to_fp16, weight = var_685_to_fp16, x = var_674_cast); + tensor var_689_to_fp16 = const()[name = tensor("op_689_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214354560)))]; + tensor k_21_bias_0_to_fp16 = const()[name = tensor("k_21_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217631424)))]; + tensor k_21_cast = linear(bias = k_21_bias_0_to_fp16, weight = var_689_to_fp16, x = var_674_cast); + tensor var_693_to_fp16 = const()[name = tensor("op_693_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217634048)))]; + tensor var_694_to_fp16 = const()[name = tensor("op_694_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(220910912)))]; + tensor v_21_cast = linear(bias = var_694_to_fp16, weight = var_693_to_fp16, x = var_674_cast); + tensor var_702 = const()[name = tensor("op_702"), val = tensor([1, 1500, 20, -1])]; + tensor var_703_cast = reshape(shape = var_702, x = q_21_cast); + tensor const_234_to_fp16 = const()[name = tensor("const_234_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_23_cast = mul(x = var_703_cast, y = const_234_to_fp16); + tensor var_709 = const()[name = tensor("op_709"), val = tensor([1, 1500, 20, -1])]; + tensor var_710_cast = reshape(shape = var_709, x = k_21_cast); + tensor const_235_to_fp16 = const()[name = tensor("const_235_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_23_cast = mul(x = var_710_cast, y = const_235_to_fp16); + tensor var_716 = const()[name = tensor("op_716"), val = tensor([1, 1500, 20, -1])]; + tensor var_717_cast = reshape(shape = var_716, x = v_21_cast); + 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_74_perm_0 = const()[name = tensor("transpose_74_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_75_perm_0 = const()[name = tensor("transpose_75_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_233 = transpose(perm = transpose_75_perm_0, x = k_23_cast); + tensor transpose_234 = transpose(perm = transpose_74_perm_0, x = q_23_cast); + tensor qk_11_cast = matmul(transpose_x = qk_11_transpose_x_0, transpose_y = qk_11_transpose_y_0, x = transpose_234, y = transpose_233); + tensor var_722_cast = softmax(axis = var_657, x = qk_11_cast); + 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_235 = transpose(perm = var_718, x = var_717_cast); + tensor var_724_cast = matmul(transpose_x = var_724_transpose_x_0, transpose_y = var_724_transpose_y_0, x = var_722_cast, y = transpose_235); + 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_232 = transpose(perm = var_725, x = var_724_cast); + tensor x_71_cast = reshape(shape = concat_5, x = transpose_232); + tensor var_730_to_fp16 = const()[name = tensor("op_730_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(220913536)))]; + tensor var_731_to_fp16 = const()[name = tensor("op_731_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224190400)))]; + tensor var_732_cast = linear(bias = var_731_to_fp16, weight = var_730_to_fp16, x = x_71_cast); + tensor x_73_cast = add(x = x_67_cast, y = var_732_cast); + 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(224193024)))]; + 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(224195648)))]; + tensor var_738_cast = 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); + tensor var_747_to_fp16 = const()[name = tensor("op_747_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224198272)))]; + tensor var_748_to_fp16 = const()[name = tensor("op_748_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237305536)))]; + tensor input_49_cast = linear(bias = var_748_to_fp16, weight = var_747_to_fp16, x = var_738_cast); + tensor x_77_mode_0 = const()[name = tensor("x_77_mode_0"), val = tensor("EXACT")]; + tensor x_77_cast = gelu(mode = x_77_mode_0, x = input_49_cast); + tensor var_753_to_fp16 = const()[name = tensor("op_753_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237315840)))]; + tensor var_754_to_fp16 = const()[name = tensor("op_754_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250423104)))]; + tensor var_755_cast = linear(bias = var_754_to_fp16, weight = var_753_to_fp16, x = x_77_cast); + tensor x_79_cast = add(x = x_73_cast, y = var_755_cast); + 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(250425728)))]; + 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(250428352)))]; + tensor var_770_to_fp16 = const()[name = tensor("op_770_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_781_cast = 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); + tensor var_792_to_fp16 = const()[name = tensor("op_792_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250430976)))]; + tensor var_793_to_fp16 = const()[name = tensor("op_793_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(253707840)))]; + tensor q_25_cast = linear(bias = var_793_to_fp16, weight = var_792_to_fp16, x = var_781_cast); + tensor var_796_to_fp16 = const()[name = tensor("op_796_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(253710464)))]; + tensor k_25_bias_0_to_fp16 = const()[name = tensor("k_25_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(256987328)))]; + tensor k_25_cast = linear(bias = k_25_bias_0_to_fp16, weight = var_796_to_fp16, x = var_781_cast); + tensor var_800_to_fp16 = const()[name = tensor("op_800_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(256989952)))]; + tensor var_801_to_fp16 = const()[name = tensor("op_801_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260266816)))]; + tensor v_25_cast = linear(bias = var_801_to_fp16, weight = var_800_to_fp16, x = var_781_cast); + tensor var_809 = const()[name = tensor("op_809"), val = tensor([1, 1500, 20, -1])]; + tensor var_810_cast = reshape(shape = var_809, x = q_25_cast); + tensor const_236_to_fp16 = const()[name = tensor("const_236_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_27_cast = mul(x = var_810_cast, y = const_236_to_fp16); + tensor var_816 = const()[name = tensor("op_816"), val = tensor([1, 1500, 20, -1])]; + tensor var_817_cast = reshape(shape = var_816, x = k_25_cast); + tensor const_237_to_fp16 = const()[name = tensor("const_237_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_27_cast = mul(x = var_817_cast, y = const_237_to_fp16); + tensor var_823 = const()[name = tensor("op_823"), val = tensor([1, 1500, 20, -1])]; + tensor var_824_cast = reshape(shape = var_823, x = v_25_cast); + 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_76_perm_0 = const()[name = tensor("transpose_76_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_77_perm_0 = const()[name = tensor("transpose_77_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_229 = transpose(perm = transpose_77_perm_0, x = k_27_cast); + tensor transpose_230 = transpose(perm = transpose_76_perm_0, x = q_27_cast); + tensor qk_13_cast = matmul(transpose_x = qk_13_transpose_x_0, transpose_y = qk_13_transpose_y_0, x = transpose_230, y = transpose_229); + tensor var_829_cast = softmax(axis = var_764, x = qk_13_cast); + 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_231 = transpose(perm = var_825, x = var_824_cast); + tensor var_831_cast = matmul(transpose_x = var_831_transpose_x_0, transpose_y = var_831_transpose_y_0, x = var_829_cast, y = transpose_231); + 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_228 = transpose(perm = var_832, x = var_831_cast); + tensor x_83_cast = reshape(shape = concat_6, x = transpose_228); + tensor var_837_to_fp16 = const()[name = tensor("op_837_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260269440)))]; + tensor var_838_to_fp16 = const()[name = tensor("op_838_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263546304)))]; + tensor var_839_cast = linear(bias = var_838_to_fp16, weight = var_837_to_fp16, x = x_83_cast); + tensor x_85_cast = add(x = x_79_cast, y = var_839_cast); + 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(263548928)))]; + 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(263551552)))]; + tensor var_845_cast = 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); + tensor var_854_to_fp16 = const()[name = tensor("op_854_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263554176)))]; + tensor var_855_to_fp16 = const()[name = tensor("op_855_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(276661440)))]; + tensor input_57_cast = linear(bias = var_855_to_fp16, weight = var_854_to_fp16, x = var_845_cast); + tensor x_89_mode_0 = const()[name = tensor("x_89_mode_0"), val = tensor("EXACT")]; + tensor x_89_cast = gelu(mode = x_89_mode_0, x = input_57_cast); + tensor var_860_to_fp16 = const()[name = tensor("op_860_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(276671744)))]; + tensor var_861_to_fp16 = const()[name = tensor("op_861_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289779008)))]; + tensor var_862_cast = linear(bias = var_861_to_fp16, weight = var_860_to_fp16, x = x_89_cast); + tensor x_91_cast = add(x = x_85_cast, y = var_862_cast); + 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(289781632)))]; + 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(289784256)))]; + tensor var_877_to_fp16 = const()[name = tensor("op_877_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_888_cast = 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); + tensor var_899_to_fp16 = const()[name = tensor("op_899_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289786880)))]; + tensor var_900_to_fp16 = const()[name = tensor("op_900_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293063744)))]; + tensor q_29_cast = linear(bias = var_900_to_fp16, weight = var_899_to_fp16, x = var_888_cast); + tensor var_903_to_fp16 = const()[name = tensor("op_903_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293066368)))]; + tensor k_29_bias_0_to_fp16 = const()[name = tensor("k_29_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(296343232)))]; + tensor k_29_cast = linear(bias = k_29_bias_0_to_fp16, weight = var_903_to_fp16, x = var_888_cast); + tensor var_907_to_fp16 = const()[name = tensor("op_907_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(296345856)))]; + tensor var_908_to_fp16 = const()[name = tensor("op_908_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(299622720)))]; + tensor v_29_cast = linear(bias = var_908_to_fp16, weight = var_907_to_fp16, x = var_888_cast); + tensor var_916 = const()[name = tensor("op_916"), val = tensor([1, 1500, 20, -1])]; + tensor var_917_cast = reshape(shape = var_916, x = q_29_cast); + tensor const_238_to_fp16 = const()[name = tensor("const_238_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_31_cast = mul(x = var_917_cast, y = const_238_to_fp16); + tensor var_923 = const()[name = tensor("op_923"), val = tensor([1, 1500, 20, -1])]; + tensor var_924_cast = reshape(shape = var_923, x = k_29_cast); + tensor const_239_to_fp16 = const()[name = tensor("const_239_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_31_cast = mul(x = var_924_cast, y = const_239_to_fp16); + tensor var_930 = const()[name = tensor("op_930"), val = tensor([1, 1500, 20, -1])]; + tensor var_931_cast = reshape(shape = var_930, x = v_29_cast); + 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_78_perm_0 = const()[name = tensor("transpose_78_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_79_perm_0 = const()[name = tensor("transpose_79_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_225 = transpose(perm = transpose_79_perm_0, x = k_31_cast); + tensor transpose_226 = transpose(perm = transpose_78_perm_0, x = q_31_cast); + tensor qk_15_cast = matmul(transpose_x = qk_15_transpose_x_0, transpose_y = qk_15_transpose_y_0, x = transpose_226, y = transpose_225); + tensor var_936_cast = softmax(axis = var_871, x = qk_15_cast); + 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_227 = transpose(perm = var_932, x = var_931_cast); + tensor var_938_cast = matmul(transpose_x = var_938_transpose_x_0, transpose_y = var_938_transpose_y_0, x = var_936_cast, y = transpose_227); + 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_224 = transpose(perm = var_939, x = var_938_cast); + tensor x_95_cast = reshape(shape = concat_7, x = transpose_224); + tensor var_944_to_fp16 = const()[name = tensor("op_944_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(299625344)))]; + tensor var_945_to_fp16 = const()[name = tensor("op_945_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302902208)))]; + tensor var_946_cast = linear(bias = var_945_to_fp16, weight = var_944_to_fp16, x = x_95_cast); + tensor x_97_cast = add(x = x_91_cast, y = var_946_cast); + 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(302904832)))]; + 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(302907456)))]; + tensor var_952_cast = 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); + tensor var_961_to_fp16 = const()[name = tensor("op_961_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302910080)))]; + tensor var_962_to_fp16 = const()[name = tensor("op_962_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316017344)))]; + tensor input_65_cast = linear(bias = var_962_to_fp16, weight = var_961_to_fp16, x = var_952_cast); + tensor x_101_mode_0 = const()[name = tensor("x_101_mode_0"), val = tensor("EXACT")]; + tensor x_101_cast = gelu(mode = x_101_mode_0, x = input_65_cast); + tensor var_967_to_fp16 = const()[name = tensor("op_967_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316027648)))]; + tensor var_968_to_fp16 = const()[name = tensor("op_968_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329134912)))]; + tensor var_969_cast = linear(bias = var_968_to_fp16, weight = var_967_to_fp16, x = x_101_cast); + tensor x_103_cast = add(x = x_97_cast, y = var_969_cast); + 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(329137536)))]; + 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(329140160)))]; + tensor var_984_to_fp16 = const()[name = tensor("op_984_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_995_cast = 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); + tensor var_1006_to_fp16 = const()[name = tensor("op_1006_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329142784)))]; + tensor var_1007_to_fp16 = const()[name = tensor("op_1007_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(332419648)))]; + tensor q_33_cast = linear(bias = var_1007_to_fp16, weight = var_1006_to_fp16, x = var_995_cast); + tensor var_1010_to_fp16 = const()[name = tensor("op_1010_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(332422272)))]; + tensor k_33_bias_0_to_fp16 = const()[name = tensor("k_33_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335699136)))]; + tensor k_33_cast = linear(bias = k_33_bias_0_to_fp16, weight = var_1010_to_fp16, x = var_995_cast); + tensor var_1014_to_fp16 = const()[name = tensor("op_1014_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335701760)))]; + tensor var_1015_to_fp16 = const()[name = tensor("op_1015_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(338978624)))]; + tensor v_33_cast = linear(bias = var_1015_to_fp16, weight = var_1014_to_fp16, x = var_995_cast); + tensor var_1023 = const()[name = tensor("op_1023"), val = tensor([1, 1500, 20, -1])]; + tensor var_1024_cast = reshape(shape = var_1023, x = q_33_cast); + tensor const_240_to_fp16 = const()[name = tensor("const_240_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_35_cast = mul(x = var_1024_cast, y = const_240_to_fp16); + tensor var_1030 = const()[name = tensor("op_1030"), val = tensor([1, 1500, 20, -1])]; + tensor var_1031_cast = reshape(shape = var_1030, x = k_33_cast); + tensor const_241_to_fp16 = const()[name = tensor("const_241_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_35_cast = mul(x = var_1031_cast, y = const_241_to_fp16); + tensor var_1037 = const()[name = tensor("op_1037"), val = tensor([1, 1500, 20, -1])]; + tensor var_1038_cast = reshape(shape = var_1037, x = v_33_cast); + 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_80_perm_0 = const()[name = tensor("transpose_80_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_81_perm_0 = const()[name = tensor("transpose_81_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_221 = transpose(perm = transpose_81_perm_0, x = k_35_cast); + tensor transpose_222 = transpose(perm = transpose_80_perm_0, x = q_35_cast); + tensor qk_17_cast = matmul(transpose_x = qk_17_transpose_x_0, transpose_y = qk_17_transpose_y_0, x = transpose_222, y = transpose_221); + tensor var_1043_cast = softmax(axis = var_978, x = qk_17_cast); + 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_223 = transpose(perm = var_1039, x = var_1038_cast); + tensor var_1045_cast = matmul(transpose_x = var_1045_transpose_x_0, transpose_y = var_1045_transpose_y_0, x = var_1043_cast, y = transpose_223); + 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_220 = transpose(perm = var_1046, x = var_1045_cast); + tensor x_107_cast = reshape(shape = concat_8, x = transpose_220); + tensor var_1051_to_fp16 = const()[name = tensor("op_1051_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(338981248)))]; + tensor var_1052_to_fp16 = const()[name = tensor("op_1052_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342258112)))]; + tensor var_1053_cast = linear(bias = var_1052_to_fp16, weight = var_1051_to_fp16, x = x_107_cast); + tensor x_109_cast = add(x = x_103_cast, y = var_1053_cast); + 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(342260736)))]; + 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(342263360)))]; + tensor var_1059_cast = 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); + tensor var_1068_to_fp16 = const()[name = tensor("op_1068_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342265984)))]; + tensor var_1069_to_fp16 = const()[name = tensor("op_1069_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(355373248)))]; + tensor input_73_cast = linear(bias = var_1069_to_fp16, weight = var_1068_to_fp16, x = var_1059_cast); + tensor x_113_mode_0 = const()[name = tensor("x_113_mode_0"), val = tensor("EXACT")]; + tensor x_113_cast = gelu(mode = x_113_mode_0, x = input_73_cast); + tensor var_1074_to_fp16 = const()[name = tensor("op_1074_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(355383552)))]; + tensor var_1075_to_fp16 = const()[name = tensor("op_1075_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368490816)))]; + tensor var_1076_cast = linear(bias = var_1075_to_fp16, weight = var_1074_to_fp16, x = x_113_cast); + tensor x_115_cast = add(x = x_109_cast, y = var_1076_cast); + 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(368493440)))]; + 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(368496064)))]; + tensor var_1091_to_fp16 = const()[name = tensor("op_1091_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1102_cast = 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); + tensor var_1113_to_fp16 = const()[name = tensor("op_1113_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368498688)))]; + tensor var_1114_to_fp16 = const()[name = tensor("op_1114_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(371775552)))]; + tensor q_37_cast = linear(bias = var_1114_to_fp16, weight = var_1113_to_fp16, x = var_1102_cast); + tensor var_1117_to_fp16 = const()[name = tensor("op_1117_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(371778176)))]; + tensor k_37_bias_0_to_fp16 = const()[name = tensor("k_37_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(375055040)))]; + tensor k_37_cast = linear(bias = k_37_bias_0_to_fp16, weight = var_1117_to_fp16, x = var_1102_cast); + tensor var_1121_to_fp16 = const()[name = tensor("op_1121_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(375057664)))]; + tensor var_1122_to_fp16 = const()[name = tensor("op_1122_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(378334528)))]; + tensor v_37_cast = linear(bias = var_1122_to_fp16, weight = var_1121_to_fp16, x = var_1102_cast); + tensor var_1130 = const()[name = tensor("op_1130"), val = tensor([1, 1500, 20, -1])]; + tensor var_1131_cast = reshape(shape = var_1130, x = q_37_cast); + tensor const_242_to_fp16 = const()[name = tensor("const_242_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_39_cast = mul(x = var_1131_cast, y = const_242_to_fp16); + tensor var_1137 = const()[name = tensor("op_1137"), val = tensor([1, 1500, 20, -1])]; + tensor var_1138_cast = reshape(shape = var_1137, x = k_37_cast); + tensor const_243_to_fp16 = const()[name = tensor("const_243_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_39_cast = mul(x = var_1138_cast, y = const_243_to_fp16); + tensor var_1144 = const()[name = tensor("op_1144"), val = tensor([1, 1500, 20, -1])]; + tensor var_1145_cast = reshape(shape = var_1144, x = v_37_cast); + 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_82_perm_0 = const()[name = tensor("transpose_82_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_83_perm_0 = const()[name = tensor("transpose_83_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_217 = transpose(perm = transpose_83_perm_0, x = k_39_cast); + tensor transpose_218 = transpose(perm = transpose_82_perm_0, x = q_39_cast); + tensor qk_19_cast = matmul(transpose_x = qk_19_transpose_x_0, transpose_y = qk_19_transpose_y_0, x = transpose_218, y = transpose_217); + tensor var_1150_cast = softmax(axis = var_1085, x = qk_19_cast); + 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_219 = transpose(perm = var_1146, x = var_1145_cast); + tensor var_1152_cast = matmul(transpose_x = var_1152_transpose_x_0, transpose_y = var_1152_transpose_y_0, x = var_1150_cast, y = transpose_219); + 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_216 = transpose(perm = var_1153, x = var_1152_cast); + tensor x_119_cast = reshape(shape = concat_9, x = transpose_216); + tensor var_1158_to_fp16 = const()[name = tensor("op_1158_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(378337152)))]; + tensor var_1159_to_fp16 = const()[name = tensor("op_1159_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381614016)))]; + tensor var_1160_cast = linear(bias = var_1159_to_fp16, weight = var_1158_to_fp16, x = x_119_cast); + tensor x_121_cast = add(x = x_115_cast, y = var_1160_cast); + 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(381616640)))]; + 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(381619264)))]; + tensor var_1166_cast = 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); + tensor var_1175_to_fp16 = const()[name = tensor("op_1175_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381621888)))]; + tensor var_1176_to_fp16 = const()[name = tensor("op_1176_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(394729152)))]; + tensor input_81_cast = linear(bias = var_1176_to_fp16, weight = var_1175_to_fp16, x = var_1166_cast); + tensor x_125_mode_0 = const()[name = tensor("x_125_mode_0"), val = tensor("EXACT")]; + tensor x_125_cast = gelu(mode = x_125_mode_0, x = input_81_cast); + tensor var_1181_to_fp16 = const()[name = tensor("op_1181_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(394739456)))]; + tensor var_1182_to_fp16 = const()[name = tensor("op_1182_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407846720)))]; + tensor var_1183_cast = linear(bias = var_1182_to_fp16, weight = var_1181_to_fp16, x = x_125_cast); + tensor x_127_cast = add(x = x_121_cast, y = var_1183_cast); + 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(407849344)))]; + 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(407851968)))]; + tensor var_1198_to_fp16 = const()[name = tensor("op_1198_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1209_cast = 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); + tensor var_1220_to_fp16 = const()[name = tensor("op_1220_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407854592)))]; + tensor var_1221_to_fp16 = const()[name = tensor("op_1221_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411131456)))]; + tensor q_41_cast = linear(bias = var_1221_to_fp16, weight = var_1220_to_fp16, x = var_1209_cast); + tensor var_1224_to_fp16 = const()[name = tensor("op_1224_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411134080)))]; + tensor k_41_bias_0_to_fp16 = const()[name = tensor("k_41_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(414410944)))]; + tensor k_41_cast = linear(bias = k_41_bias_0_to_fp16, weight = var_1224_to_fp16, x = var_1209_cast); + tensor var_1228_to_fp16 = const()[name = tensor("op_1228_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(414413568)))]; + tensor var_1229_to_fp16 = const()[name = tensor("op_1229_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(417690432)))]; + tensor v_41_cast = linear(bias = var_1229_to_fp16, weight = var_1228_to_fp16, x = var_1209_cast); + tensor var_1237 = const()[name = tensor("op_1237"), val = tensor([1, 1500, 20, -1])]; + tensor var_1238_cast = reshape(shape = var_1237, x = q_41_cast); + tensor const_244_to_fp16 = const()[name = tensor("const_244_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_43_cast = mul(x = var_1238_cast, y = const_244_to_fp16); + tensor var_1244 = const()[name = tensor("op_1244"), val = tensor([1, 1500, 20, -1])]; + tensor var_1245_cast = reshape(shape = var_1244, x = k_41_cast); + tensor const_245_to_fp16 = const()[name = tensor("const_245_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_43_cast = mul(x = var_1245_cast, y = const_245_to_fp16); + tensor var_1251 = const()[name = tensor("op_1251"), val = tensor([1, 1500, 20, -1])]; + tensor var_1252_cast = reshape(shape = var_1251, x = v_41_cast); + 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_84_perm_0 = const()[name = tensor("transpose_84_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_85_perm_0 = const()[name = tensor("transpose_85_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_213 = transpose(perm = transpose_85_perm_0, x = k_43_cast); + tensor transpose_214 = transpose(perm = transpose_84_perm_0, x = q_43_cast); + tensor qk_21_cast = matmul(transpose_x = qk_21_transpose_x_0, transpose_y = qk_21_transpose_y_0, x = transpose_214, y = transpose_213); + tensor var_1257_cast = softmax(axis = var_1192, x = qk_21_cast); + 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_215 = transpose(perm = var_1253, x = var_1252_cast); + tensor var_1259_cast = matmul(transpose_x = var_1259_transpose_x_0, transpose_y = var_1259_transpose_y_0, x = var_1257_cast, y = transpose_215); + 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_212 = transpose(perm = var_1260, x = var_1259_cast); + tensor x_131_cast = reshape(shape = concat_10, x = transpose_212); + tensor var_1265_to_fp16 = const()[name = tensor("op_1265_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(417693056)))]; + tensor var_1266_to_fp16 = const()[name = tensor("op_1266_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(420969920)))]; + tensor var_1267_cast = linear(bias = var_1266_to_fp16, weight = var_1265_to_fp16, x = x_131_cast); + tensor x_133_cast = add(x = x_127_cast, y = var_1267_cast); + 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(420972544)))]; + 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(420975168)))]; + tensor var_1273_cast = 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); + tensor var_1282_to_fp16 = const()[name = tensor("op_1282_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(420977792)))]; + tensor var_1283_to_fp16 = const()[name = tensor("op_1283_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(434085056)))]; + tensor input_89_cast = linear(bias = var_1283_to_fp16, weight = var_1282_to_fp16, x = var_1273_cast); + tensor x_137_mode_0 = const()[name = tensor("x_137_mode_0"), val = tensor("EXACT")]; + tensor x_137_cast = gelu(mode = x_137_mode_0, x = input_89_cast); + tensor var_1288_to_fp16 = const()[name = tensor("op_1288_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(434095360)))]; + tensor var_1289_to_fp16 = const()[name = tensor("op_1289_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447202624)))]; + tensor var_1290_cast = linear(bias = var_1289_to_fp16, weight = var_1288_to_fp16, x = x_137_cast); + tensor x_139_cast = add(x = x_133_cast, y = var_1290_cast); + 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(447205248)))]; + 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(447207872)))]; + tensor var_1305_to_fp16 = const()[name = tensor("op_1305_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1316_cast = 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); + tensor var_1327_to_fp16 = const()[name = tensor("op_1327_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447210496)))]; + tensor var_1328_to_fp16 = const()[name = tensor("op_1328_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(450487360)))]; + tensor q_45_cast = linear(bias = var_1328_to_fp16, weight = var_1327_to_fp16, x = var_1316_cast); + tensor var_1331_to_fp16 = const()[name = tensor("op_1331_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(450489984)))]; + tensor k_45_bias_0_to_fp16 = const()[name = tensor("k_45_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(453766848)))]; + tensor k_45_cast = linear(bias = k_45_bias_0_to_fp16, weight = var_1331_to_fp16, x = var_1316_cast); + tensor var_1335_to_fp16 = const()[name = tensor("op_1335_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(453769472)))]; + tensor var_1336_to_fp16 = const()[name = tensor("op_1336_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(457046336)))]; + tensor v_45_cast = linear(bias = var_1336_to_fp16, weight = var_1335_to_fp16, x = var_1316_cast); + tensor var_1344 = const()[name = tensor("op_1344"), val = tensor([1, 1500, 20, -1])]; + tensor var_1345_cast = reshape(shape = var_1344, x = q_45_cast); + tensor const_246_to_fp16 = const()[name = tensor("const_246_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_47_cast = mul(x = var_1345_cast, y = const_246_to_fp16); + tensor var_1351 = const()[name = tensor("op_1351"), val = tensor([1, 1500, 20, -1])]; + tensor var_1352_cast = reshape(shape = var_1351, x = k_45_cast); + tensor const_247_to_fp16 = const()[name = tensor("const_247_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_47_cast = mul(x = var_1352_cast, y = const_247_to_fp16); + tensor var_1358 = const()[name = tensor("op_1358"), val = tensor([1, 1500, 20, -1])]; + tensor var_1359_cast = reshape(shape = var_1358, x = v_45_cast); + 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_86_perm_0 = const()[name = tensor("transpose_86_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_87_perm_0 = const()[name = tensor("transpose_87_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_209 = transpose(perm = transpose_87_perm_0, x = k_47_cast); + tensor transpose_210 = transpose(perm = transpose_86_perm_0, x = q_47_cast); + tensor qk_23_cast = matmul(transpose_x = qk_23_transpose_x_0, transpose_y = qk_23_transpose_y_0, x = transpose_210, y = transpose_209); + tensor var_1364_cast = softmax(axis = var_1299, x = qk_23_cast); + 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_211 = transpose(perm = var_1360, x = var_1359_cast); + tensor var_1366_cast = matmul(transpose_x = var_1366_transpose_x_0, transpose_y = var_1366_transpose_y_0, x = var_1364_cast, y = transpose_211); + 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_208 = transpose(perm = var_1367, x = var_1366_cast); + tensor x_143_cast = reshape(shape = concat_11, x = transpose_208); + tensor var_1372_to_fp16 = const()[name = tensor("op_1372_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(457048960)))]; + tensor var_1373_to_fp16 = const()[name = tensor("op_1373_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460325824)))]; + tensor var_1374_cast = linear(bias = var_1373_to_fp16, weight = var_1372_to_fp16, x = x_143_cast); + tensor x_145_cast = add(x = x_139_cast, y = var_1374_cast); + 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(460328448)))]; + 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(460331072)))]; + tensor var_1380_cast = 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); + tensor var_1389_to_fp16 = const()[name = tensor("op_1389_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460333696)))]; + tensor var_1390_to_fp16 = const()[name = tensor("op_1390_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473440960)))]; + tensor input_97_cast = linear(bias = var_1390_to_fp16, weight = var_1389_to_fp16, x = var_1380_cast); + tensor x_149_mode_0 = const()[name = tensor("x_149_mode_0"), val = tensor("EXACT")]; + tensor x_149_cast = gelu(mode = x_149_mode_0, x = input_97_cast); + tensor var_1395_to_fp16 = const()[name = tensor("op_1395_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473451264)))]; + tensor var_1396_to_fp16 = const()[name = tensor("op_1396_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486558528)))]; + tensor var_1397_cast = linear(bias = var_1396_to_fp16, weight = var_1395_to_fp16, x = x_149_cast); + tensor x_151_cast = add(x = x_145_cast, y = var_1397_cast); + 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(486561152)))]; + 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(486563776)))]; + tensor var_1412_to_fp16 = const()[name = tensor("op_1412_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1423_cast = 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); + tensor var_1434_to_fp16 = const()[name = tensor("op_1434_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486566400)))]; + tensor var_1435_to_fp16 = const()[name = tensor("op_1435_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(489843264)))]; + tensor q_49_cast = linear(bias = var_1435_to_fp16, weight = var_1434_to_fp16, x = var_1423_cast); + tensor var_1438_to_fp16 = const()[name = tensor("op_1438_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(489845888)))]; + tensor k_49_bias_0_to_fp16 = const()[name = tensor("k_49_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(493122752)))]; + tensor k_49_cast = linear(bias = k_49_bias_0_to_fp16, weight = var_1438_to_fp16, x = var_1423_cast); + tensor var_1442_to_fp16 = const()[name = tensor("op_1442_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(493125376)))]; + tensor var_1443_to_fp16 = const()[name = tensor("op_1443_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496402240)))]; + tensor v_49_cast = linear(bias = var_1443_to_fp16, weight = var_1442_to_fp16, x = var_1423_cast); + tensor var_1451 = const()[name = tensor("op_1451"), val = tensor([1, 1500, 20, -1])]; + tensor var_1452_cast = reshape(shape = var_1451, x = q_49_cast); + tensor const_248_to_fp16 = const()[name = tensor("const_248_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_51_cast = mul(x = var_1452_cast, y = const_248_to_fp16); + tensor var_1458 = const()[name = tensor("op_1458"), val = tensor([1, 1500, 20, -1])]; + tensor var_1459_cast = reshape(shape = var_1458, x = k_49_cast); + tensor const_249_to_fp16 = const()[name = tensor("const_249_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_51_cast = mul(x = var_1459_cast, y = const_249_to_fp16); + tensor var_1465 = const()[name = tensor("op_1465"), val = tensor([1, 1500, 20, -1])]; + tensor var_1466_cast = reshape(shape = var_1465, x = v_49_cast); + 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_88_perm_0 = const()[name = tensor("transpose_88_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_89_perm_0 = const()[name = tensor("transpose_89_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_205 = transpose(perm = transpose_89_perm_0, x = k_51_cast); + tensor transpose_206 = transpose(perm = transpose_88_perm_0, x = q_51_cast); + tensor qk_25_cast = matmul(transpose_x = qk_25_transpose_x_0, transpose_y = qk_25_transpose_y_0, x = transpose_206, y = transpose_205); + tensor var_1471_cast = softmax(axis = var_1406, x = qk_25_cast); + 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_207 = transpose(perm = var_1467, x = var_1466_cast); + tensor var_1473_cast = matmul(transpose_x = var_1473_transpose_x_0, transpose_y = var_1473_transpose_y_0, x = var_1471_cast, y = transpose_207); + 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_204 = transpose(perm = var_1474, x = var_1473_cast); + tensor x_155_cast = reshape(shape = concat_12, x = transpose_204); + tensor var_1479_to_fp16 = const()[name = tensor("op_1479_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496404864)))]; + tensor var_1480_to_fp16 = const()[name = tensor("op_1480_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(499681728)))]; + tensor var_1481_cast = linear(bias = var_1480_to_fp16, weight = var_1479_to_fp16, x = x_155_cast); + tensor x_157_cast = add(x = x_151_cast, y = var_1481_cast); + 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(499684352)))]; + 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(499686976)))]; + tensor var_1487_cast = 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); + tensor var_1496_to_fp16 = const()[name = tensor("op_1496_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(499689600)))]; + tensor var_1497_to_fp16 = const()[name = tensor("op_1497_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(512796864)))]; + tensor input_105_cast = linear(bias = var_1497_to_fp16, weight = var_1496_to_fp16, x = var_1487_cast); + tensor x_161_mode_0 = const()[name = tensor("x_161_mode_0"), val = tensor("EXACT")]; + tensor x_161_cast = gelu(mode = x_161_mode_0, x = input_105_cast); + tensor var_1502_to_fp16 = const()[name = tensor("op_1502_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(512807168)))]; + tensor var_1503_to_fp16 = const()[name = tensor("op_1503_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(525914432)))]; + tensor var_1504_cast = linear(bias = var_1503_to_fp16, weight = var_1502_to_fp16, x = x_161_cast); + tensor x_163_cast = add(x = x_157_cast, y = var_1504_cast); + 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(525917056)))]; + 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(525919680)))]; + tensor var_1519_to_fp16 = const()[name = tensor("op_1519_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1530_cast = 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); + tensor var_1541_to_fp16 = const()[name = tensor("op_1541_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(525922304)))]; + tensor var_1542_to_fp16 = const()[name = tensor("op_1542_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529199168)))]; + tensor q_53_cast = linear(bias = var_1542_to_fp16, weight = var_1541_to_fp16, x = var_1530_cast); + tensor var_1545_to_fp16 = const()[name = tensor("op_1545_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529201792)))]; + tensor k_53_bias_0_to_fp16 = const()[name = tensor("k_53_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(532478656)))]; + tensor k_53_cast = linear(bias = k_53_bias_0_to_fp16, weight = var_1545_to_fp16, x = var_1530_cast); + tensor var_1549_to_fp16 = const()[name = tensor("op_1549_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(532481280)))]; + tensor var_1550_to_fp16 = const()[name = tensor("op_1550_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(535758144)))]; + tensor v_53_cast = linear(bias = var_1550_to_fp16, weight = var_1549_to_fp16, x = var_1530_cast); + tensor var_1558 = const()[name = tensor("op_1558"), val = tensor([1, 1500, 20, -1])]; + tensor var_1559_cast = reshape(shape = var_1558, x = q_53_cast); + tensor const_250_to_fp16 = const()[name = tensor("const_250_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_55_cast = mul(x = var_1559_cast, y = const_250_to_fp16); + tensor var_1565 = const()[name = tensor("op_1565"), val = tensor([1, 1500, 20, -1])]; + tensor var_1566_cast = reshape(shape = var_1565, x = k_53_cast); + tensor const_251_to_fp16 = const()[name = tensor("const_251_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_55_cast = mul(x = var_1566_cast, y = const_251_to_fp16); + tensor var_1572 = const()[name = tensor("op_1572"), val = tensor([1, 1500, 20, -1])]; + tensor var_1573_cast = reshape(shape = var_1572, x = v_53_cast); + 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_90_perm_0 = const()[name = tensor("transpose_90_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_91_perm_0 = const()[name = tensor("transpose_91_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_201 = transpose(perm = transpose_91_perm_0, x = k_55_cast); + tensor transpose_202 = transpose(perm = transpose_90_perm_0, x = q_55_cast); + tensor qk_27_cast = matmul(transpose_x = qk_27_transpose_x_0, transpose_y = qk_27_transpose_y_0, x = transpose_202, y = transpose_201); + tensor var_1578_cast = softmax(axis = var_1513, x = qk_27_cast); + 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_203 = transpose(perm = var_1574, x = var_1573_cast); + tensor var_1580_cast = matmul(transpose_x = var_1580_transpose_x_0, transpose_y = var_1580_transpose_y_0, x = var_1578_cast, y = transpose_203); + 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_200 = transpose(perm = var_1581, x = var_1580_cast); + tensor x_167_cast = reshape(shape = concat_13, x = transpose_200); + tensor var_1586_to_fp16 = const()[name = tensor("op_1586_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(535760768)))]; + tensor var_1587_to_fp16 = const()[name = tensor("op_1587_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(539037632)))]; + tensor var_1588_cast = linear(bias = var_1587_to_fp16, weight = var_1586_to_fp16, x = x_167_cast); + tensor x_169_cast = add(x = x_163_cast, y = var_1588_cast); + 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(539040256)))]; + 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(539042880)))]; + tensor var_1594_cast = 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); + tensor var_1603_to_fp16 = const()[name = tensor("op_1603_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(539045504)))]; + tensor var_1604_to_fp16 = const()[name = tensor("op_1604_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552152768)))]; + tensor input_113_cast = linear(bias = var_1604_to_fp16, weight = var_1603_to_fp16, x = var_1594_cast); + tensor x_173_mode_0 = const()[name = tensor("x_173_mode_0"), val = tensor("EXACT")]; + tensor x_173_cast = gelu(mode = x_173_mode_0, x = input_113_cast); + tensor var_1609_to_fp16 = const()[name = tensor("op_1609_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552163072)))]; + tensor var_1610_to_fp16 = const()[name = tensor("op_1610_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565270336)))]; + tensor var_1611_cast = linear(bias = var_1610_to_fp16, weight = var_1609_to_fp16, x = x_173_cast); + tensor x_175_cast = add(x = x_169_cast, y = var_1611_cast); + 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(565272960)))]; + 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(565275584)))]; + tensor var_1626_to_fp16 = const()[name = tensor("op_1626_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1637_cast = 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); + tensor var_1648_to_fp16 = const()[name = tensor("op_1648_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565278208)))]; + tensor var_1649_to_fp16 = const()[name = tensor("op_1649_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568555072)))]; + tensor q_57_cast = linear(bias = var_1649_to_fp16, weight = var_1648_to_fp16, x = var_1637_cast); + tensor var_1652_to_fp16 = const()[name = tensor("op_1652_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568557696)))]; + tensor k_57_bias_0_to_fp16 = const()[name = tensor("k_57_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(571834560)))]; + tensor k_57_cast = linear(bias = k_57_bias_0_to_fp16, weight = var_1652_to_fp16, x = var_1637_cast); + tensor var_1656_to_fp16 = const()[name = tensor("op_1656_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(571837184)))]; + tensor var_1657_to_fp16 = const()[name = tensor("op_1657_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(575114048)))]; + tensor v_57_cast = linear(bias = var_1657_to_fp16, weight = var_1656_to_fp16, x = var_1637_cast); + tensor var_1665 = const()[name = tensor("op_1665"), val = tensor([1, 1500, 20, -1])]; + tensor var_1666_cast = reshape(shape = var_1665, x = q_57_cast); + tensor const_252_to_fp16 = const()[name = tensor("const_252_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_59_cast = mul(x = var_1666_cast, y = const_252_to_fp16); + tensor var_1672 = const()[name = tensor("op_1672"), val = tensor([1, 1500, 20, -1])]; + tensor var_1673_cast = reshape(shape = var_1672, x = k_57_cast); + tensor const_253_to_fp16 = const()[name = tensor("const_253_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_59_cast = mul(x = var_1673_cast, y = const_253_to_fp16); + tensor var_1679 = const()[name = tensor("op_1679"), val = tensor([1, 1500, 20, -1])]; + tensor var_1680_cast = reshape(shape = var_1679, x = v_57_cast); + 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_92_perm_0 = const()[name = tensor("transpose_92_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_93_perm_0 = const()[name = tensor("transpose_93_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_197 = transpose(perm = transpose_93_perm_0, x = k_59_cast); + tensor transpose_198 = transpose(perm = transpose_92_perm_0, x = q_59_cast); + tensor qk_29_cast = matmul(transpose_x = qk_29_transpose_x_0, transpose_y = qk_29_transpose_y_0, x = transpose_198, y = transpose_197); + tensor var_1685_cast = softmax(axis = var_1620, x = qk_29_cast); + 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_199 = transpose(perm = var_1681, x = var_1680_cast); + tensor var_1687_cast = matmul(transpose_x = var_1687_transpose_x_0, transpose_y = var_1687_transpose_y_0, x = var_1685_cast, y = transpose_199); + 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_196 = transpose(perm = var_1688, x = var_1687_cast); + tensor x_179_cast = reshape(shape = concat_14, x = transpose_196); + tensor var_1693_to_fp16 = const()[name = tensor("op_1693_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(575116672)))]; + tensor var_1694_to_fp16 = const()[name = tensor("op_1694_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578393536)))]; + tensor var_1695_cast = linear(bias = var_1694_to_fp16, weight = var_1693_to_fp16, x = x_179_cast); + tensor x_181_cast = add(x = x_175_cast, y = var_1695_cast); + 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(578396160)))]; + 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(578398784)))]; + tensor var_1701_cast = 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); + tensor var_1710_to_fp16 = const()[name = tensor("op_1710_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578401408)))]; + tensor var_1711_to_fp16 = const()[name = tensor("op_1711_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(591508672)))]; + tensor input_121_cast = linear(bias = var_1711_to_fp16, weight = var_1710_to_fp16, x = var_1701_cast); + tensor x_185_mode_0 = const()[name = tensor("x_185_mode_0"), val = tensor("EXACT")]; + tensor x_185_cast = gelu(mode = x_185_mode_0, x = input_121_cast); + tensor var_1716_to_fp16 = const()[name = tensor("op_1716_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(591518976)))]; + tensor var_1717_to_fp16 = const()[name = tensor("op_1717_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604626240)))]; + tensor var_1718_cast = linear(bias = var_1717_to_fp16, weight = var_1716_to_fp16, x = x_185_cast); + tensor x_187_cast = add(x = x_181_cast, y = var_1718_cast); + 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(604628864)))]; + 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(604631488)))]; + tensor var_1733_to_fp16 = const()[name = tensor("op_1733_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1744_cast = 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); + tensor var_1755_to_fp16 = const()[name = tensor("op_1755_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604634112)))]; + tensor var_1756_to_fp16 = const()[name = tensor("op_1756_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(607910976)))]; + tensor q_61_cast = linear(bias = var_1756_to_fp16, weight = var_1755_to_fp16, x = var_1744_cast); + tensor var_1759_to_fp16 = const()[name = tensor("op_1759_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(607913600)))]; + tensor k_61_bias_0_to_fp16 = const()[name = tensor("k_61_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(611190464)))]; + tensor k_61_cast = linear(bias = k_61_bias_0_to_fp16, weight = var_1759_to_fp16, x = var_1744_cast); + tensor var_1763_to_fp16 = const()[name = tensor("op_1763_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(611193088)))]; + tensor var_1764_to_fp16 = const()[name = tensor("op_1764_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614469952)))]; + tensor v_61_cast = linear(bias = var_1764_to_fp16, weight = var_1763_to_fp16, x = var_1744_cast); + tensor var_1772 = const()[name = tensor("op_1772"), val = tensor([1, 1500, 20, -1])]; + tensor var_1773_cast = reshape(shape = var_1772, x = q_61_cast); + tensor const_254_to_fp16 = const()[name = tensor("const_254_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_63_cast = mul(x = var_1773_cast, y = const_254_to_fp16); + tensor var_1779 = const()[name = tensor("op_1779"), val = tensor([1, 1500, 20, -1])]; + tensor var_1780_cast = reshape(shape = var_1779, x = k_61_cast); + tensor const_255_to_fp16 = const()[name = tensor("const_255_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_63_cast = mul(x = var_1780_cast, y = const_255_to_fp16); + tensor var_1786 = const()[name = tensor("op_1786"), val = tensor([1, 1500, 20, -1])]; + tensor var_1787_cast = reshape(shape = var_1786, x = v_61_cast); + 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_94_perm_0 = const()[name = tensor("transpose_94_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_95_perm_0 = const()[name = tensor("transpose_95_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_193 = transpose(perm = transpose_95_perm_0, x = k_63_cast); + tensor transpose_194 = transpose(perm = transpose_94_perm_0, x = q_63_cast); + tensor qk_31_cast = matmul(transpose_x = qk_31_transpose_x_0, transpose_y = qk_31_transpose_y_0, x = transpose_194, y = transpose_193); + tensor var_1792_cast = softmax(axis = var_1727, x = qk_31_cast); + 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_195 = transpose(perm = var_1788, x = var_1787_cast); + tensor var_1794_cast = matmul(transpose_x = var_1794_transpose_x_0, transpose_y = var_1794_transpose_y_0, x = var_1792_cast, y = transpose_195); + 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_192 = transpose(perm = var_1795, x = var_1794_cast); + tensor x_191_cast = reshape(shape = concat_15, x = transpose_192); + tensor var_1800_to_fp16 = const()[name = tensor("op_1800_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614472576)))]; + tensor var_1801_to_fp16 = const()[name = tensor("op_1801_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617749440)))]; + tensor var_1802_cast = linear(bias = var_1801_to_fp16, weight = var_1800_to_fp16, x = x_191_cast); + tensor x_193_cast = add(x = x_187_cast, y = var_1802_cast); + 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(617752064)))]; + 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(617754688)))]; + tensor var_1808_cast = 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); + tensor var_1817_to_fp16 = const()[name = tensor("op_1817_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617757312)))]; + tensor var_1818_to_fp16 = const()[name = tensor("op_1818_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(630864576)))]; + tensor input_129_cast = linear(bias = var_1818_to_fp16, weight = var_1817_to_fp16, x = var_1808_cast); + tensor x_197_mode_0 = const()[name = tensor("x_197_mode_0"), val = tensor("EXACT")]; + tensor x_197_cast = gelu(mode = x_197_mode_0, x = input_129_cast); + tensor var_1823_to_fp16 = const()[name = tensor("op_1823_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(630874880)))]; + tensor var_1824_to_fp16 = const()[name = tensor("op_1824_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643982144)))]; + tensor var_1825_cast = linear(bias = var_1824_to_fp16, weight = var_1823_to_fp16, x = x_197_cast); + tensor x_199_cast = add(x = x_193_cast, y = var_1825_cast); + 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(643984768)))]; + 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(643987392)))]; + tensor var_1840_to_fp16 = const()[name = tensor("op_1840_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1851_cast = 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); + tensor var_1862_to_fp16 = const()[name = tensor("op_1862_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643990016)))]; + tensor var_1863_to_fp16 = const()[name = tensor("op_1863_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647266880)))]; + tensor q_65_cast = linear(bias = var_1863_to_fp16, weight = var_1862_to_fp16, x = var_1851_cast); + tensor var_1866_to_fp16 = const()[name = tensor("op_1866_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647269504)))]; + tensor k_65_bias_0_to_fp16 = const()[name = tensor("k_65_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(650546368)))]; + tensor k_65_cast = linear(bias = k_65_bias_0_to_fp16, weight = var_1866_to_fp16, x = var_1851_cast); + tensor var_1870_to_fp16 = const()[name = tensor("op_1870_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(650548992)))]; + tensor var_1871_to_fp16 = const()[name = tensor("op_1871_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(653825856)))]; + tensor v_65_cast = linear(bias = var_1871_to_fp16, weight = var_1870_to_fp16, x = var_1851_cast); + tensor var_1879 = const()[name = tensor("op_1879"), val = tensor([1, 1500, 20, -1])]; + tensor var_1880_cast = reshape(shape = var_1879, x = q_65_cast); + tensor const_256_to_fp16 = const()[name = tensor("const_256_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_67_cast = mul(x = var_1880_cast, y = const_256_to_fp16); + tensor var_1886 = const()[name = tensor("op_1886"), val = tensor([1, 1500, 20, -1])]; + tensor var_1887_cast = reshape(shape = var_1886, x = k_65_cast); + tensor const_257_to_fp16 = const()[name = tensor("const_257_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_67_cast = mul(x = var_1887_cast, y = const_257_to_fp16); + tensor var_1893 = const()[name = tensor("op_1893"), val = tensor([1, 1500, 20, -1])]; + tensor var_1894_cast = reshape(shape = var_1893, x = v_65_cast); + 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_96_perm_0 = const()[name = tensor("transpose_96_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_97_perm_0 = const()[name = tensor("transpose_97_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_189 = transpose(perm = transpose_97_perm_0, x = k_67_cast); + tensor transpose_190 = transpose(perm = transpose_96_perm_0, x = q_67_cast); + tensor qk_33_cast = matmul(transpose_x = qk_33_transpose_x_0, transpose_y = qk_33_transpose_y_0, x = transpose_190, y = transpose_189); + tensor var_1899_cast = softmax(axis = var_1834, x = qk_33_cast); + 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_191 = transpose(perm = var_1895, x = var_1894_cast); + tensor var_1901_cast = matmul(transpose_x = var_1901_transpose_x_0, transpose_y = var_1901_transpose_y_0, x = var_1899_cast, y = transpose_191); + 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_188 = transpose(perm = var_1902, x = var_1901_cast); + tensor x_203_cast = reshape(shape = concat_16, x = transpose_188); + tensor var_1907_to_fp16 = const()[name = tensor("op_1907_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(653828480)))]; + tensor var_1908_to_fp16 = const()[name = tensor("op_1908_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657105344)))]; + tensor var_1909_cast = linear(bias = var_1908_to_fp16, weight = var_1907_to_fp16, x = x_203_cast); + tensor x_205_cast = add(x = x_199_cast, y = var_1909_cast); + 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(657107968)))]; + 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(657110592)))]; + tensor var_1915_cast = 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); + tensor var_1924_to_fp16 = const()[name = tensor("op_1924_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657113216)))]; + tensor var_1925_to_fp16 = const()[name = tensor("op_1925_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(670220480)))]; + tensor input_137_cast = linear(bias = var_1925_to_fp16, weight = var_1924_to_fp16, x = var_1915_cast); + tensor x_209_mode_0 = const()[name = tensor("x_209_mode_0"), val = tensor("EXACT")]; + tensor x_209_cast = gelu(mode = x_209_mode_0, x = input_137_cast); + tensor var_1930_to_fp16 = const()[name = tensor("op_1930_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(670230784)))]; + tensor var_1931_to_fp16 = const()[name = tensor("op_1931_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(683338048)))]; + tensor var_1932_cast = linear(bias = var_1931_to_fp16, weight = var_1930_to_fp16, x = x_209_cast); + tensor x_211_cast = add(x = x_205_cast, y = var_1932_cast); + 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(683340672)))]; + 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(683343296)))]; + tensor var_1947_to_fp16 = const()[name = tensor("op_1947_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1958_cast = 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); + tensor var_1969_to_fp16 = const()[name = tensor("op_1969_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(683345920)))]; + tensor var_1970_to_fp16 = const()[name = tensor("op_1970_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(686622784)))]; + tensor q_69_cast = linear(bias = var_1970_to_fp16, weight = var_1969_to_fp16, x = var_1958_cast); + tensor var_1973_to_fp16 = const()[name = tensor("op_1973_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(686625408)))]; + tensor k_69_bias_0_to_fp16 = const()[name = tensor("k_69_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(689902272)))]; + tensor k_69_cast = linear(bias = k_69_bias_0_to_fp16, weight = var_1973_to_fp16, x = var_1958_cast); + tensor var_1977_to_fp16 = const()[name = tensor("op_1977_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(689904896)))]; + tensor var_1978_to_fp16 = const()[name = tensor("op_1978_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(693181760)))]; + tensor v_69_cast = linear(bias = var_1978_to_fp16, weight = var_1977_to_fp16, x = var_1958_cast); + tensor var_1986 = const()[name = tensor("op_1986"), val = tensor([1, 1500, 20, -1])]; + tensor var_1987_cast = reshape(shape = var_1986, x = q_69_cast); + tensor const_258_to_fp16 = const()[name = tensor("const_258_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_71_cast = mul(x = var_1987_cast, y = const_258_to_fp16); + tensor var_1993 = const()[name = tensor("op_1993"), val = tensor([1, 1500, 20, -1])]; + tensor var_1994_cast = reshape(shape = var_1993, x = k_69_cast); + tensor const_259_to_fp16 = const()[name = tensor("const_259_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_71_cast = mul(x = var_1994_cast, y = const_259_to_fp16); + tensor var_2000 = const()[name = tensor("op_2000"), val = tensor([1, 1500, 20, -1])]; + tensor var_2001_cast = reshape(shape = var_2000, x = v_69_cast); + 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_98_perm_0 = const()[name = tensor("transpose_98_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_99_perm_0 = const()[name = tensor("transpose_99_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_185 = transpose(perm = transpose_99_perm_0, x = k_71_cast); + tensor transpose_186 = transpose(perm = transpose_98_perm_0, x = q_71_cast); + tensor qk_35_cast = matmul(transpose_x = qk_35_transpose_x_0, transpose_y = qk_35_transpose_y_0, x = transpose_186, y = transpose_185); + tensor var_2006_cast = softmax(axis = var_1941, x = qk_35_cast); + 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_187 = transpose(perm = var_2002, x = var_2001_cast); + tensor var_2008_cast = matmul(transpose_x = var_2008_transpose_x_0, transpose_y = var_2008_transpose_y_0, x = var_2006_cast, y = transpose_187); + 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_184 = transpose(perm = var_2009, x = var_2008_cast); + tensor x_215_cast = reshape(shape = concat_17, x = transpose_184); + tensor var_2014_to_fp16 = const()[name = tensor("op_2014_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(693184384)))]; + tensor var_2015_to_fp16 = const()[name = tensor("op_2015_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696461248)))]; + tensor var_2016_cast = linear(bias = var_2015_to_fp16, weight = var_2014_to_fp16, x = x_215_cast); + tensor x_217_cast = add(x = x_211_cast, y = var_2016_cast); + 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(696463872)))]; + 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(696466496)))]; + tensor var_2022_cast = 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); + tensor var_2031_to_fp16 = const()[name = tensor("op_2031_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696469120)))]; + tensor var_2032_to_fp16 = const()[name = tensor("op_2032_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(709576384)))]; + tensor input_145_cast = linear(bias = var_2032_to_fp16, weight = var_2031_to_fp16, x = var_2022_cast); + tensor x_221_mode_0 = const()[name = tensor("x_221_mode_0"), val = tensor("EXACT")]; + tensor x_221_cast = gelu(mode = x_221_mode_0, x = input_145_cast); + tensor var_2037_to_fp16 = const()[name = tensor("op_2037_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(709586688)))]; + tensor var_2038_to_fp16 = const()[name = tensor("op_2038_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(722693952)))]; + tensor var_2039_cast = linear(bias = var_2038_to_fp16, weight = var_2037_to_fp16, x = x_221_cast); + tensor x_223_cast = add(x = x_217_cast, y = var_2039_cast); + 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(722696576)))]; + 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(722699200)))]; + tensor var_2054_to_fp16 = const()[name = tensor("op_2054_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2065_cast = 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); + tensor var_2076_to_fp16 = const()[name = tensor("op_2076_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(722701824)))]; + tensor var_2077_to_fp16 = const()[name = tensor("op_2077_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(725978688)))]; + tensor q_73_cast = linear(bias = var_2077_to_fp16, weight = var_2076_to_fp16, x = var_2065_cast); + tensor var_2080_to_fp16 = const()[name = tensor("op_2080_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(725981312)))]; + tensor k_73_bias_0_to_fp16 = const()[name = tensor("k_73_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(729258176)))]; + tensor k_73_cast = linear(bias = k_73_bias_0_to_fp16, weight = var_2080_to_fp16, x = var_2065_cast); + tensor var_2084_to_fp16 = const()[name = tensor("op_2084_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(729260800)))]; + tensor var_2085_to_fp16 = const()[name = tensor("op_2085_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(732537664)))]; + tensor v_73_cast = linear(bias = var_2085_to_fp16, weight = var_2084_to_fp16, x = var_2065_cast); + tensor var_2093 = const()[name = tensor("op_2093"), val = tensor([1, 1500, 20, -1])]; + tensor var_2094_cast = reshape(shape = var_2093, x = q_73_cast); + tensor const_260_to_fp16 = const()[name = tensor("const_260_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_75_cast = mul(x = var_2094_cast, y = const_260_to_fp16); + tensor var_2100 = const()[name = tensor("op_2100"), val = tensor([1, 1500, 20, -1])]; + tensor var_2101_cast = reshape(shape = var_2100, x = k_73_cast); + tensor const_261_to_fp16 = const()[name = tensor("const_261_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_75_cast = mul(x = var_2101_cast, y = const_261_to_fp16); + tensor var_2107 = const()[name = tensor("op_2107"), val = tensor([1, 1500, 20, -1])]; + tensor var_2108_cast = reshape(shape = var_2107, x = v_73_cast); + 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_100_perm_0 = const()[name = tensor("transpose_100_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_101_perm_0 = const()[name = tensor("transpose_101_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_181 = transpose(perm = transpose_101_perm_0, x = k_75_cast); + tensor transpose_182 = transpose(perm = transpose_100_perm_0, x = q_75_cast); + tensor qk_37_cast = matmul(transpose_x = qk_37_transpose_x_0, transpose_y = qk_37_transpose_y_0, x = transpose_182, y = transpose_181); + tensor var_2113_cast = softmax(axis = var_2048, x = qk_37_cast); + 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_183 = transpose(perm = var_2109, x = var_2108_cast); + tensor var_2115_cast = matmul(transpose_x = var_2115_transpose_x_0, transpose_y = var_2115_transpose_y_0, x = var_2113_cast, y = transpose_183); + 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_180 = transpose(perm = var_2116, x = var_2115_cast); + tensor x_227_cast = reshape(shape = concat_18, x = transpose_180); + tensor var_2121_to_fp16 = const()[name = tensor("op_2121_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(732540288)))]; + tensor var_2122_to_fp16 = const()[name = tensor("op_2122_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(735817152)))]; + tensor var_2123_cast = linear(bias = var_2122_to_fp16, weight = var_2121_to_fp16, x = x_227_cast); + tensor x_229_cast = add(x = x_223_cast, y = var_2123_cast); + 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(735819776)))]; + 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(735822400)))]; + tensor var_2129_cast = 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); + tensor var_2138_to_fp16 = const()[name = tensor("op_2138_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(735825024)))]; + tensor var_2139_to_fp16 = const()[name = tensor("op_2139_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(748932288)))]; + tensor input_153_cast = linear(bias = var_2139_to_fp16, weight = var_2138_to_fp16, x = var_2129_cast); + tensor x_233_mode_0 = const()[name = tensor("x_233_mode_0"), val = tensor("EXACT")]; + tensor x_233_cast = gelu(mode = x_233_mode_0, x = input_153_cast); + tensor var_2144_to_fp16 = const()[name = tensor("op_2144_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(748942592)))]; + tensor var_2145_to_fp16 = const()[name = tensor("op_2145_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762049856)))]; + tensor var_2146_cast = linear(bias = var_2145_to_fp16, weight = var_2144_to_fp16, x = x_233_cast); + tensor x_235_cast = add(x = x_229_cast, y = var_2146_cast); + 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(762052480)))]; + 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(762055104)))]; + tensor var_2161_to_fp16 = const()[name = tensor("op_2161_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2172_cast = 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); + tensor var_2183_to_fp16 = const()[name = tensor("op_2183_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762057728)))]; + tensor var_2184_to_fp16 = const()[name = tensor("op_2184_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(765334592)))]; + tensor q_77_cast = linear(bias = var_2184_to_fp16, weight = var_2183_to_fp16, x = var_2172_cast); + tensor var_2187_to_fp16 = const()[name = tensor("op_2187_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(765337216)))]; + tensor k_77_bias_0_to_fp16 = const()[name = tensor("k_77_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(768614080)))]; + tensor k_77_cast = linear(bias = k_77_bias_0_to_fp16, weight = var_2187_to_fp16, x = var_2172_cast); + tensor var_2191_to_fp16 = const()[name = tensor("op_2191_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(768616704)))]; + tensor var_2192_to_fp16 = const()[name = tensor("op_2192_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(771893568)))]; + tensor v_77_cast = linear(bias = var_2192_to_fp16, weight = var_2191_to_fp16, x = var_2172_cast); + tensor var_2200 = const()[name = tensor("op_2200"), val = tensor([1, 1500, 20, -1])]; + tensor var_2201_cast = reshape(shape = var_2200, x = q_77_cast); + tensor const_262_to_fp16 = const()[name = tensor("const_262_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_79_cast = mul(x = var_2201_cast, y = const_262_to_fp16); + tensor var_2207 = const()[name = tensor("op_2207"), val = tensor([1, 1500, 20, -1])]; + tensor var_2208_cast = reshape(shape = var_2207, x = k_77_cast); + tensor const_263_to_fp16 = const()[name = tensor("const_263_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_79_cast = mul(x = var_2208_cast, y = const_263_to_fp16); + tensor var_2214 = const()[name = tensor("op_2214"), val = tensor([1, 1500, 20, -1])]; + tensor var_2215_cast = reshape(shape = var_2214, x = v_77_cast); + 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_102_perm_0 = const()[name = tensor("transpose_102_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_103_perm_0 = const()[name = tensor("transpose_103_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_177 = transpose(perm = transpose_103_perm_0, x = k_79_cast); + tensor transpose_178 = transpose(perm = transpose_102_perm_0, x = q_79_cast); + tensor qk_39_cast = matmul(transpose_x = qk_39_transpose_x_0, transpose_y = qk_39_transpose_y_0, x = transpose_178, y = transpose_177); + tensor var_2220_cast = softmax(axis = var_2155, x = qk_39_cast); + 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_179 = transpose(perm = var_2216, x = var_2215_cast); + tensor var_2222_cast = matmul(transpose_x = var_2222_transpose_x_0, transpose_y = var_2222_transpose_y_0, x = var_2220_cast, y = transpose_179); + 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_176 = transpose(perm = var_2223, x = var_2222_cast); + tensor x_239_cast = reshape(shape = concat_19, x = transpose_176); + tensor var_2228_to_fp16 = const()[name = tensor("op_2228_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(771896192)))]; + tensor var_2229_to_fp16 = const()[name = tensor("op_2229_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775173056)))]; + tensor var_2230_cast = linear(bias = var_2229_to_fp16, weight = var_2228_to_fp16, x = x_239_cast); + tensor x_241_cast = add(x = x_235_cast, y = var_2230_cast); + 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(775175680)))]; + 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(775178304)))]; + tensor var_2236_cast = 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); + tensor var_2245_to_fp16 = const()[name = tensor("op_2245_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775180928)))]; + tensor var_2246_to_fp16 = const()[name = tensor("op_2246_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(788288192)))]; + tensor input_161_cast = linear(bias = var_2246_to_fp16, weight = var_2245_to_fp16, x = var_2236_cast); + tensor x_245_mode_0 = const()[name = tensor("x_245_mode_0"), val = tensor("EXACT")]; + tensor x_245_cast = gelu(mode = x_245_mode_0, x = input_161_cast); + tensor var_2251_to_fp16 = const()[name = tensor("op_2251_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(788298496)))]; + tensor var_2252_to_fp16 = const()[name = tensor("op_2252_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(801405760)))]; + tensor var_2253_cast = linear(bias = var_2252_to_fp16, weight = var_2251_to_fp16, x = x_245_cast); + tensor x_247_cast = add(x = x_241_cast, y = var_2253_cast); + 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(801408384)))]; + 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(801411008)))]; + tensor var_2268_to_fp16 = const()[name = tensor("op_2268_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2279_cast = 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); + tensor var_2290_to_fp16 = const()[name = tensor("op_2290_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(801413632)))]; + tensor var_2291_to_fp16 = const()[name = tensor("op_2291_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(804690496)))]; + tensor q_81_cast = linear(bias = var_2291_to_fp16, weight = var_2290_to_fp16, x = var_2279_cast); + tensor var_2294_to_fp16 = const()[name = tensor("op_2294_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(804693120)))]; + tensor k_81_bias_0_to_fp16 = const()[name = tensor("k_81_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(807969984)))]; + tensor k_81_cast = linear(bias = k_81_bias_0_to_fp16, weight = var_2294_to_fp16, x = var_2279_cast); + tensor var_2298_to_fp16 = const()[name = tensor("op_2298_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(807972608)))]; + tensor var_2299_to_fp16 = const()[name = tensor("op_2299_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(811249472)))]; + tensor v_81_cast = linear(bias = var_2299_to_fp16, weight = var_2298_to_fp16, x = var_2279_cast); + tensor var_2307 = const()[name = tensor("op_2307"), val = tensor([1, 1500, 20, -1])]; + tensor var_2308_cast = reshape(shape = var_2307, x = q_81_cast); + tensor const_264_to_fp16 = const()[name = tensor("const_264_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_83_cast = mul(x = var_2308_cast, y = const_264_to_fp16); + tensor var_2314 = const()[name = tensor("op_2314"), val = tensor([1, 1500, 20, -1])]; + tensor var_2315_cast = reshape(shape = var_2314, x = k_81_cast); + tensor const_265_to_fp16 = const()[name = tensor("const_265_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_83_cast = mul(x = var_2315_cast, y = const_265_to_fp16); + tensor var_2321 = const()[name = tensor("op_2321"), val = tensor([1, 1500, 20, -1])]; + tensor var_2322_cast = reshape(shape = var_2321, x = v_81_cast); + 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_104_perm_0 = const()[name = tensor("transpose_104_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_105_perm_0 = const()[name = tensor("transpose_105_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_173 = transpose(perm = transpose_105_perm_0, x = k_83_cast); + tensor transpose_174 = transpose(perm = transpose_104_perm_0, x = q_83_cast); + tensor qk_41_cast = matmul(transpose_x = qk_41_transpose_x_0, transpose_y = qk_41_transpose_y_0, x = transpose_174, y = transpose_173); + tensor var_2327_cast = softmax(axis = var_2262, x = qk_41_cast); + 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_175 = transpose(perm = var_2323, x = var_2322_cast); + tensor var_2329_cast = matmul(transpose_x = var_2329_transpose_x_0, transpose_y = var_2329_transpose_y_0, x = var_2327_cast, y = transpose_175); + 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_172 = transpose(perm = var_2330, x = var_2329_cast); + tensor x_251_cast = reshape(shape = concat_20, x = transpose_172); + tensor var_2335_to_fp16 = const()[name = tensor("op_2335_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(811252096)))]; + tensor var_2336_to_fp16 = const()[name = tensor("op_2336_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814528960)))]; + tensor var_2337_cast = linear(bias = var_2336_to_fp16, weight = var_2335_to_fp16, x = x_251_cast); + tensor x_253_cast = add(x = x_247_cast, y = var_2337_cast); + 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(814531584)))]; + 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(814534208)))]; + tensor var_2343_cast = 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); + tensor var_2352_to_fp16 = const()[name = tensor("op_2352_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814536832)))]; + tensor var_2353_to_fp16 = const()[name = tensor("op_2353_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(827644096)))]; + tensor input_169_cast = linear(bias = var_2353_to_fp16, weight = var_2352_to_fp16, x = var_2343_cast); + tensor x_257_mode_0 = const()[name = tensor("x_257_mode_0"), val = tensor("EXACT")]; + tensor x_257_cast = gelu(mode = x_257_mode_0, x = input_169_cast); + tensor var_2358_to_fp16 = const()[name = tensor("op_2358_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(827654400)))]; + tensor var_2359_to_fp16 = const()[name = tensor("op_2359_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(840761664)))]; + tensor var_2360_cast = linear(bias = var_2359_to_fp16, weight = var_2358_to_fp16, x = x_257_cast); + tensor x_259_cast = add(x = x_253_cast, y = var_2360_cast); + 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(840764288)))]; + 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(840766912)))]; + tensor var_2375_to_fp16 = const()[name = tensor("op_2375_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2386_cast = 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); + tensor var_2397_to_fp16 = const()[name = tensor("op_2397_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(840769536)))]; + tensor var_2398_to_fp16 = const()[name = tensor("op_2398_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(844046400)))]; + tensor q_85_cast = linear(bias = var_2398_to_fp16, weight = var_2397_to_fp16, x = var_2386_cast); + tensor var_2401_to_fp16 = const()[name = tensor("op_2401_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(844049024)))]; + tensor k_85_bias_0_to_fp16 = const()[name = tensor("k_85_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(847325888)))]; + tensor k_85_cast = linear(bias = k_85_bias_0_to_fp16, weight = var_2401_to_fp16, x = var_2386_cast); + tensor var_2405_to_fp16 = const()[name = tensor("op_2405_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(847328512)))]; + tensor var_2406_to_fp16 = const()[name = tensor("op_2406_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(850605376)))]; + tensor v_85_cast = linear(bias = var_2406_to_fp16, weight = var_2405_to_fp16, x = var_2386_cast); + tensor var_2414 = const()[name = tensor("op_2414"), val = tensor([1, 1500, 20, -1])]; + tensor var_2415_cast = reshape(shape = var_2414, x = q_85_cast); + tensor const_266_to_fp16 = const()[name = tensor("const_266_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_87_cast = mul(x = var_2415_cast, y = const_266_to_fp16); + tensor var_2421 = const()[name = tensor("op_2421"), val = tensor([1, 1500, 20, -1])]; + tensor var_2422_cast = reshape(shape = var_2421, x = k_85_cast); + tensor const_267_to_fp16 = const()[name = tensor("const_267_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_87_cast = mul(x = var_2422_cast, y = const_267_to_fp16); + tensor var_2428 = const()[name = tensor("op_2428"), val = tensor([1, 1500, 20, -1])]; + tensor var_2429_cast = reshape(shape = var_2428, x = v_85_cast); + 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_106_perm_0 = const()[name = tensor("transpose_106_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_107_perm_0 = const()[name = tensor("transpose_107_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_169 = transpose(perm = transpose_107_perm_0, x = k_87_cast); + tensor transpose_170 = transpose(perm = transpose_106_perm_0, x = q_87_cast); + tensor qk_43_cast = matmul(transpose_x = qk_43_transpose_x_0, transpose_y = qk_43_transpose_y_0, x = transpose_170, y = transpose_169); + tensor var_2434_cast = softmax(axis = var_2369, x = qk_43_cast); + 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_171 = transpose(perm = var_2430, x = var_2429_cast); + tensor var_2436_cast = matmul(transpose_x = var_2436_transpose_x_0, transpose_y = var_2436_transpose_y_0, x = var_2434_cast, y = transpose_171); + 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_168 = transpose(perm = var_2437, x = var_2436_cast); + tensor x_263_cast = reshape(shape = concat_21, x = transpose_168); + tensor var_2442_to_fp16 = const()[name = tensor("op_2442_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(850608000)))]; + tensor var_2443_to_fp16 = const()[name = tensor("op_2443_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(853884864)))]; + tensor var_2444_cast = linear(bias = var_2443_to_fp16, weight = var_2442_to_fp16, x = x_263_cast); + tensor x_265_cast = add(x = x_259_cast, y = var_2444_cast); + 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(853887488)))]; + 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(853890112)))]; + tensor var_2450_cast = 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); + tensor var_2459_to_fp16 = const()[name = tensor("op_2459_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(853892736)))]; + tensor var_2460_to_fp16 = const()[name = tensor("op_2460_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(867000000)))]; + tensor input_177_cast = linear(bias = var_2460_to_fp16, weight = var_2459_to_fp16, x = var_2450_cast); + tensor x_269_mode_0 = const()[name = tensor("x_269_mode_0"), val = tensor("EXACT")]; + tensor x_269_cast = gelu(mode = x_269_mode_0, x = input_177_cast); + tensor var_2465_to_fp16 = const()[name = tensor("op_2465_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(867010304)))]; + tensor var_2466_to_fp16 = const()[name = tensor("op_2466_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880117568)))]; + tensor var_2467_cast = linear(bias = var_2466_to_fp16, weight = var_2465_to_fp16, x = x_269_cast); + tensor x_271_cast = add(x = x_265_cast, y = var_2467_cast); + 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(880120192)))]; + 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(880122816)))]; + tensor var_2482_to_fp16 = const()[name = tensor("op_2482_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2493_cast = 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); + tensor var_2504_to_fp16 = const()[name = tensor("op_2504_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880125440)))]; + tensor var_2505_to_fp16 = const()[name = tensor("op_2505_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(883402304)))]; + tensor q_89_cast = linear(bias = var_2505_to_fp16, weight = var_2504_to_fp16, x = var_2493_cast); + tensor var_2508_to_fp16 = const()[name = tensor("op_2508_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(883404928)))]; + tensor k_89_bias_0_to_fp16 = const()[name = tensor("k_89_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(886681792)))]; + tensor k_89_cast = linear(bias = k_89_bias_0_to_fp16, weight = var_2508_to_fp16, x = var_2493_cast); + tensor var_2512_to_fp16 = const()[name = tensor("op_2512_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(886684416)))]; + tensor var_2513_to_fp16 = const()[name = tensor("op_2513_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(889961280)))]; + tensor v_89_cast = linear(bias = var_2513_to_fp16, weight = var_2512_to_fp16, x = var_2493_cast); + tensor var_2521 = const()[name = tensor("op_2521"), val = tensor([1, 1500, 20, -1])]; + tensor var_2522_cast = reshape(shape = var_2521, x = q_89_cast); + tensor const_268_to_fp16 = const()[name = tensor("const_268_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_91_cast = mul(x = var_2522_cast, y = const_268_to_fp16); + tensor var_2528 = const()[name = tensor("op_2528"), val = tensor([1, 1500, 20, -1])]; + tensor var_2529_cast = reshape(shape = var_2528, x = k_89_cast); + tensor const_269_to_fp16 = const()[name = tensor("const_269_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_91_cast = mul(x = var_2529_cast, y = const_269_to_fp16); + tensor var_2535 = const()[name = tensor("op_2535"), val = tensor([1, 1500, 20, -1])]; + tensor var_2536_cast = reshape(shape = var_2535, x = v_89_cast); + 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_108_perm_0 = const()[name = tensor("transpose_108_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_109_perm_0 = const()[name = tensor("transpose_109_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_165 = transpose(perm = transpose_109_perm_0, x = k_91_cast); + tensor transpose_166 = transpose(perm = transpose_108_perm_0, x = q_91_cast); + tensor qk_45_cast = matmul(transpose_x = qk_45_transpose_x_0, transpose_y = qk_45_transpose_y_0, x = transpose_166, y = transpose_165); + tensor var_2541_cast = softmax(axis = var_2476, x = qk_45_cast); + 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_167 = transpose(perm = var_2537, x = var_2536_cast); + tensor var_2543_cast = matmul(transpose_x = var_2543_transpose_x_0, transpose_y = var_2543_transpose_y_0, x = var_2541_cast, y = transpose_167); + 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_164 = transpose(perm = var_2544, x = var_2543_cast); + tensor x_275_cast = reshape(shape = concat_22, x = transpose_164); + tensor var_2549_to_fp16 = const()[name = tensor("op_2549_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(889963904)))]; + tensor var_2550_to_fp16 = const()[name = tensor("op_2550_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893240768)))]; + tensor var_2551_cast = linear(bias = var_2550_to_fp16, weight = var_2549_to_fp16, x = x_275_cast); + tensor x_277_cast = add(x = x_271_cast, y = var_2551_cast); + 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(893243392)))]; + 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(893246016)))]; + tensor var_2557_cast = 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); + tensor var_2566_to_fp16 = const()[name = tensor("op_2566_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893248640)))]; + tensor var_2567_to_fp16 = const()[name = tensor("op_2567_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(906355904)))]; + tensor input_185_cast = linear(bias = var_2567_to_fp16, weight = var_2566_to_fp16, x = var_2557_cast); + tensor x_281_mode_0 = const()[name = tensor("x_281_mode_0"), val = tensor("EXACT")]; + tensor x_281_cast = gelu(mode = x_281_mode_0, x = input_185_cast); + tensor var_2572_to_fp16 = const()[name = tensor("op_2572_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(906366208)))]; + tensor var_2573_to_fp16 = const()[name = tensor("op_2573_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(919473472)))]; + tensor var_2574_cast = linear(bias = var_2573_to_fp16, weight = var_2572_to_fp16, x = x_281_cast); + tensor x_283_cast = add(x = x_277_cast, y = var_2574_cast); + 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(919476096)))]; + 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(919478720)))]; + tensor var_2589_to_fp16 = const()[name = tensor("op_2589_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2600_cast = 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); + tensor var_2611_to_fp16 = const()[name = tensor("op_2611_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(919481344)))]; + tensor var_2612_to_fp16 = const()[name = tensor("op_2612_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(922758208)))]; + tensor q_93_cast = linear(bias = var_2612_to_fp16, weight = var_2611_to_fp16, x = var_2600_cast); + tensor var_2615_to_fp16 = const()[name = tensor("op_2615_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(922760832)))]; + tensor k_93_bias_0_to_fp16 = const()[name = tensor("k_93_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(926037696)))]; + tensor k_93_cast = linear(bias = k_93_bias_0_to_fp16, weight = var_2615_to_fp16, x = var_2600_cast); + tensor var_2619_to_fp16 = const()[name = tensor("op_2619_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(926040320)))]; + tensor var_2620_to_fp16 = const()[name = tensor("op_2620_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(929317184)))]; + tensor v_93_cast = linear(bias = var_2620_to_fp16, weight = var_2619_to_fp16, x = var_2600_cast); + tensor var_2628 = const()[name = tensor("op_2628"), val = tensor([1, 1500, 20, -1])]; + tensor var_2629_cast = reshape(shape = var_2628, x = q_93_cast); + tensor const_270_to_fp16 = const()[name = tensor("const_270_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_95_cast = mul(x = var_2629_cast, y = const_270_to_fp16); + tensor var_2635 = const()[name = tensor("op_2635"), val = tensor([1, 1500, 20, -1])]; + tensor var_2636_cast = reshape(shape = var_2635, x = k_93_cast); + tensor const_271_to_fp16 = const()[name = tensor("const_271_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_95_cast = mul(x = var_2636_cast, y = const_271_to_fp16); + tensor var_2642 = const()[name = tensor("op_2642"), val = tensor([1, 1500, 20, -1])]; + tensor var_2643_cast = reshape(shape = var_2642, x = v_93_cast); + 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_110_perm_0 = const()[name = tensor("transpose_110_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_111_perm_0 = const()[name = tensor("transpose_111_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_161 = transpose(perm = transpose_111_perm_0, x = k_95_cast); + tensor transpose_162 = transpose(perm = transpose_110_perm_0, x = q_95_cast); + tensor qk_47_cast = matmul(transpose_x = qk_47_transpose_x_0, transpose_y = qk_47_transpose_y_0, x = transpose_162, y = transpose_161); + tensor var_2648_cast = softmax(axis = var_2583, x = qk_47_cast); + 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_163 = transpose(perm = var_2644, x = var_2643_cast); + tensor var_2650_cast = matmul(transpose_x = var_2650_transpose_x_0, transpose_y = var_2650_transpose_y_0, x = var_2648_cast, y = transpose_163); + 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_160 = transpose(perm = var_2651, x = var_2650_cast); + tensor x_287_cast = reshape(shape = concat_23, x = transpose_160); + tensor var_2656_to_fp16 = const()[name = tensor("op_2656_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(929319808)))]; + tensor var_2657_to_fp16 = const()[name = tensor("op_2657_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(932596672)))]; + tensor var_2658_cast = linear(bias = var_2657_to_fp16, weight = var_2656_to_fp16, x = x_287_cast); + tensor x_289_cast = add(x = x_283_cast, y = var_2658_cast); + 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(932599296)))]; + 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(932601920)))]; + tensor var_2664_cast = 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); + tensor var_2673_to_fp16 = const()[name = tensor("op_2673_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(932604544)))]; + tensor var_2674_to_fp16 = const()[name = tensor("op_2674_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(945711808)))]; + tensor input_193_cast = linear(bias = var_2674_to_fp16, weight = var_2673_to_fp16, x = var_2664_cast); + tensor x_293_mode_0 = const()[name = tensor("x_293_mode_0"), val = tensor("EXACT")]; + tensor x_293_cast = gelu(mode = x_293_mode_0, x = input_193_cast); + tensor var_2679_to_fp16 = const()[name = tensor("op_2679_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(945722112)))]; + tensor var_2680_to_fp16 = const()[name = tensor("op_2680_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(958829376)))]; + tensor var_2681_cast = linear(bias = var_2680_to_fp16, weight = var_2679_to_fp16, x = x_293_cast); + tensor x_295_cast = add(x = x_289_cast, y = var_2681_cast); + 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(958832000)))]; + 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(958834624)))]; + tensor var_2696_to_fp16 = const()[name = tensor("op_2696_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2707_cast = 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); + tensor var_2718_to_fp16 = const()[name = tensor("op_2718_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(958837248)))]; + tensor var_2719_to_fp16 = const()[name = tensor("op_2719_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(962114112)))]; + tensor q_97_cast = linear(bias = var_2719_to_fp16, weight = var_2718_to_fp16, x = var_2707_cast); + tensor var_2722_to_fp16 = const()[name = tensor("op_2722_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(962116736)))]; + tensor k_97_bias_0_to_fp16 = const()[name = tensor("k_97_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(965393600)))]; + tensor k_97_cast = linear(bias = k_97_bias_0_to_fp16, weight = var_2722_to_fp16, x = var_2707_cast); + tensor var_2726_to_fp16 = const()[name = tensor("op_2726_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(965396224)))]; + tensor var_2727_to_fp16 = const()[name = tensor("op_2727_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(968673088)))]; + tensor v_97_cast = linear(bias = var_2727_to_fp16, weight = var_2726_to_fp16, x = var_2707_cast); + tensor var_2735 = const()[name = tensor("op_2735"), val = tensor([1, 1500, 20, -1])]; + tensor var_2736_cast = reshape(shape = var_2735, x = q_97_cast); + tensor const_272_to_fp16 = const()[name = tensor("const_272_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_99_cast = mul(x = var_2736_cast, y = const_272_to_fp16); + tensor var_2742 = const()[name = tensor("op_2742"), val = tensor([1, 1500, 20, -1])]; + tensor var_2743_cast = reshape(shape = var_2742, x = k_97_cast); + tensor const_273_to_fp16 = const()[name = tensor("const_273_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_99_cast = mul(x = var_2743_cast, y = const_273_to_fp16); + tensor var_2749 = const()[name = tensor("op_2749"), val = tensor([1, 1500, 20, -1])]; + tensor var_2750_cast = reshape(shape = var_2749, x = v_97_cast); + 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_112_perm_0 = const()[name = tensor("transpose_112_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_113_perm_0 = const()[name = tensor("transpose_113_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_157 = transpose(perm = transpose_113_perm_0, x = k_99_cast); + tensor transpose_158 = transpose(perm = transpose_112_perm_0, x = q_99_cast); + tensor qk_49_cast = matmul(transpose_x = qk_49_transpose_x_0, transpose_y = qk_49_transpose_y_0, x = transpose_158, y = transpose_157); + tensor var_2755_cast = softmax(axis = var_2690, x = qk_49_cast); + 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_159 = transpose(perm = var_2751, x = var_2750_cast); + tensor var_2757_cast = matmul(transpose_x = var_2757_transpose_x_0, transpose_y = var_2757_transpose_y_0, x = var_2755_cast, y = transpose_159); + 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_156 = transpose(perm = var_2758, x = var_2757_cast); + tensor x_299_cast = reshape(shape = concat_24, x = transpose_156); + tensor var_2763_to_fp16 = const()[name = tensor("op_2763_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(968675712)))]; + tensor var_2764_to_fp16 = const()[name = tensor("op_2764_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(971952576)))]; + tensor var_2765_cast = linear(bias = var_2764_to_fp16, weight = var_2763_to_fp16, x = x_299_cast); + tensor x_301_cast = add(x = x_295_cast, y = var_2765_cast); + 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(971955200)))]; + 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(971957824)))]; + tensor var_2771_cast = 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); + tensor var_2780_to_fp16 = const()[name = tensor("op_2780_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(971960448)))]; + tensor var_2781_to_fp16 = const()[name = tensor("op_2781_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(985067712)))]; + tensor input_201_cast = linear(bias = var_2781_to_fp16, weight = var_2780_to_fp16, x = var_2771_cast); + tensor x_305_mode_0 = const()[name = tensor("x_305_mode_0"), val = tensor("EXACT")]; + tensor x_305_cast = gelu(mode = x_305_mode_0, x = input_201_cast); + tensor var_2786_to_fp16 = const()[name = tensor("op_2786_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(985078016)))]; + tensor var_2787_to_fp16 = const()[name = tensor("op_2787_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(998185280)))]; + tensor var_2788_cast = linear(bias = var_2787_to_fp16, weight = var_2786_to_fp16, x = x_305_cast); + tensor x_307_cast = add(x = x_301_cast, y = var_2788_cast); + 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(998187904)))]; + 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(998190528)))]; + tensor var_2803_to_fp16 = const()[name = tensor("op_2803_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2814_cast = 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); + tensor var_2825_to_fp16 = const()[name = tensor("op_2825_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(998193152)))]; + tensor var_2826_to_fp16 = const()[name = tensor("op_2826_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1001470016)))]; + tensor q_101_cast = linear(bias = var_2826_to_fp16, weight = var_2825_to_fp16, x = var_2814_cast); + tensor var_2829_to_fp16 = const()[name = tensor("op_2829_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1001472640)))]; + tensor k_101_bias_0_to_fp16 = const()[name = tensor("k_101_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1004749504)))]; + tensor k_101_cast = linear(bias = k_101_bias_0_to_fp16, weight = var_2829_to_fp16, x = var_2814_cast); + tensor var_2833_to_fp16 = const()[name = tensor("op_2833_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1004752128)))]; + tensor var_2834_to_fp16 = const()[name = tensor("op_2834_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1008028992)))]; + tensor v_101_cast = linear(bias = var_2834_to_fp16, weight = var_2833_to_fp16, x = var_2814_cast); + tensor var_2842 = const()[name = tensor("op_2842"), val = tensor([1, 1500, 20, -1])]; + tensor var_2843_cast = reshape(shape = var_2842, x = q_101_cast); + tensor const_274_to_fp16 = const()[name = tensor("const_274_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_103_cast = mul(x = var_2843_cast, y = const_274_to_fp16); + tensor var_2849 = const()[name = tensor("op_2849"), val = tensor([1, 1500, 20, -1])]; + tensor var_2850_cast = reshape(shape = var_2849, x = k_101_cast); + tensor const_275_to_fp16 = const()[name = tensor("const_275_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_103_cast = mul(x = var_2850_cast, y = const_275_to_fp16); + tensor var_2856 = const()[name = tensor("op_2856"), val = tensor([1, 1500, 20, -1])]; + tensor var_2857_cast = reshape(shape = var_2856, x = v_101_cast); + 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_114_perm_0 = const()[name = tensor("transpose_114_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_115_perm_0 = const()[name = tensor("transpose_115_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_153 = transpose(perm = transpose_115_perm_0, x = k_103_cast); + tensor transpose_154 = transpose(perm = transpose_114_perm_0, x = q_103_cast); + tensor qk_51_cast = matmul(transpose_x = qk_51_transpose_x_0, transpose_y = qk_51_transpose_y_0, x = transpose_154, y = transpose_153); + tensor var_2862_cast = softmax(axis = var_2797, x = qk_51_cast); + 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_155 = transpose(perm = var_2858, x = var_2857_cast); + tensor var_2864_cast = matmul(transpose_x = var_2864_transpose_x_0, transpose_y = var_2864_transpose_y_0, x = var_2862_cast, y = transpose_155); + 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_152 = transpose(perm = var_2865, x = var_2864_cast); + tensor x_311_cast = reshape(shape = concat_25, x = transpose_152); + tensor var_2870_to_fp16 = const()[name = tensor("op_2870_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1008031616)))]; + tensor var_2871_to_fp16 = const()[name = tensor("op_2871_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1011308480)))]; + tensor var_2872_cast = linear(bias = var_2871_to_fp16, weight = var_2870_to_fp16, x = x_311_cast); + tensor x_313_cast = add(x = x_307_cast, y = var_2872_cast); + 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(1011311104)))]; + 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(1011313728)))]; + tensor var_2878_cast = 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); + tensor var_2887_to_fp16 = const()[name = tensor("op_2887_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1011316352)))]; + tensor var_2888_to_fp16 = const()[name = tensor("op_2888_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1024423616)))]; + tensor input_209_cast = linear(bias = var_2888_to_fp16, weight = var_2887_to_fp16, x = var_2878_cast); + tensor x_317_mode_0 = const()[name = tensor("x_317_mode_0"), val = tensor("EXACT")]; + tensor x_317_cast = gelu(mode = x_317_mode_0, x = input_209_cast); + tensor var_2893_to_fp16 = const()[name = tensor("op_2893_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1024433920)))]; + tensor var_2894_to_fp16 = const()[name = tensor("op_2894_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1037541184)))]; + tensor var_2895_cast = linear(bias = var_2894_to_fp16, weight = var_2893_to_fp16, x = x_317_cast); + tensor x_319_cast = add(x = x_313_cast, y = var_2895_cast); + 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(1037543808)))]; + 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(1037546432)))]; + tensor var_2910_to_fp16 = const()[name = tensor("op_2910_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2921_cast = 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); + tensor var_2932_to_fp16 = const()[name = tensor("op_2932_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1037549056)))]; + tensor var_2933_to_fp16 = const()[name = tensor("op_2933_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1040825920)))]; + tensor q_105_cast = linear(bias = var_2933_to_fp16, weight = var_2932_to_fp16, x = var_2921_cast); + tensor var_2936_to_fp16 = const()[name = tensor("op_2936_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1040828544)))]; + tensor k_105_bias_0_to_fp16 = const()[name = tensor("k_105_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1044105408)))]; + tensor k_105_cast = linear(bias = k_105_bias_0_to_fp16, weight = var_2936_to_fp16, x = var_2921_cast); + tensor var_2940_to_fp16 = const()[name = tensor("op_2940_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1044108032)))]; + tensor var_2941_to_fp16 = const()[name = tensor("op_2941_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1047384896)))]; + tensor v_105_cast = linear(bias = var_2941_to_fp16, weight = var_2940_to_fp16, x = var_2921_cast); + tensor var_2949 = const()[name = tensor("op_2949"), val = tensor([1, 1500, 20, -1])]; + tensor var_2950_cast = reshape(shape = var_2949, x = q_105_cast); + tensor const_276_to_fp16 = const()[name = tensor("const_276_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_107_cast = mul(x = var_2950_cast, y = const_276_to_fp16); + tensor var_2956 = const()[name = tensor("op_2956"), val = tensor([1, 1500, 20, -1])]; + tensor var_2957_cast = reshape(shape = var_2956, x = k_105_cast); + tensor const_277_to_fp16 = const()[name = tensor("const_277_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_107_cast = mul(x = var_2957_cast, y = const_277_to_fp16); + tensor var_2963 = const()[name = tensor("op_2963"), val = tensor([1, 1500, 20, -1])]; + tensor var_2964_cast = reshape(shape = var_2963, x = v_105_cast); + 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_116_perm_0 = const()[name = tensor("transpose_116_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_117_perm_0 = const()[name = tensor("transpose_117_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_149 = transpose(perm = transpose_117_perm_0, x = k_107_cast); + tensor transpose_150 = transpose(perm = transpose_116_perm_0, x = q_107_cast); + tensor qk_53_cast = matmul(transpose_x = qk_53_transpose_x_0, transpose_y = qk_53_transpose_y_0, x = transpose_150, y = transpose_149); + tensor var_2969_cast = softmax(axis = var_2904, x = qk_53_cast); + 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_151 = transpose(perm = var_2965, x = var_2964_cast); + tensor var_2971_cast = matmul(transpose_x = var_2971_transpose_x_0, transpose_y = var_2971_transpose_y_0, x = var_2969_cast, y = transpose_151); + 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_148 = transpose(perm = var_2972, x = var_2971_cast); + tensor x_323_cast = reshape(shape = concat_26, x = transpose_148); + tensor var_2977_to_fp16 = const()[name = tensor("op_2977_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1047387520)))]; + tensor var_2978_to_fp16 = const()[name = tensor("op_2978_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1050664384)))]; + tensor var_2979_cast = linear(bias = var_2978_to_fp16, weight = var_2977_to_fp16, x = x_323_cast); + tensor x_325_cast = add(x = x_319_cast, y = var_2979_cast); + 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(1050667008)))]; + 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(1050669632)))]; + tensor var_2985_cast = 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); + tensor var_2994_to_fp16 = const()[name = tensor("op_2994_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1050672256)))]; + tensor var_2995_to_fp16 = const()[name = tensor("op_2995_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1063779520)))]; + tensor input_217_cast = linear(bias = var_2995_to_fp16, weight = var_2994_to_fp16, x = var_2985_cast); + tensor x_329_mode_0 = const()[name = tensor("x_329_mode_0"), val = tensor("EXACT")]; + tensor x_329_cast = gelu(mode = x_329_mode_0, x = input_217_cast); + tensor var_3000_to_fp16 = const()[name = tensor("op_3000_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1063789824)))]; + tensor var_3001_to_fp16 = const()[name = tensor("op_3001_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1076897088)))]; + tensor var_3002_cast = linear(bias = var_3001_to_fp16, weight = var_3000_to_fp16, x = x_329_cast); + tensor x_331_cast = add(x = x_325_cast, y = var_3002_cast); + 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(1076899712)))]; + 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(1076902336)))]; + tensor var_3017_to_fp16 = const()[name = tensor("op_3017_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3028_cast = 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); + tensor var_3039_to_fp16 = const()[name = tensor("op_3039_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1076904960)))]; + tensor var_3040_to_fp16 = const()[name = tensor("op_3040_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1080181824)))]; + tensor q_109_cast = linear(bias = var_3040_to_fp16, weight = var_3039_to_fp16, x = var_3028_cast); + tensor var_3043_to_fp16 = const()[name = tensor("op_3043_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1080184448)))]; + tensor k_109_bias_0_to_fp16 = const()[name = tensor("k_109_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1083461312)))]; + tensor k_109_cast = linear(bias = k_109_bias_0_to_fp16, weight = var_3043_to_fp16, x = var_3028_cast); + tensor var_3047_to_fp16 = const()[name = tensor("op_3047_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1083463936)))]; + tensor var_3048_to_fp16 = const()[name = tensor("op_3048_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1086740800)))]; + tensor v_109_cast = linear(bias = var_3048_to_fp16, weight = var_3047_to_fp16, x = var_3028_cast); + tensor var_3056 = const()[name = tensor("op_3056"), val = tensor([1, 1500, 20, -1])]; + tensor var_3057_cast = reshape(shape = var_3056, x = q_109_cast); + tensor const_278_to_fp16 = const()[name = tensor("const_278_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_111_cast = mul(x = var_3057_cast, y = const_278_to_fp16); + tensor var_3063 = const()[name = tensor("op_3063"), val = tensor([1, 1500, 20, -1])]; + tensor var_3064_cast = reshape(shape = var_3063, x = k_109_cast); + tensor const_279_to_fp16 = const()[name = tensor("const_279_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_111_cast = mul(x = var_3064_cast, y = const_279_to_fp16); + tensor var_3070 = const()[name = tensor("op_3070"), val = tensor([1, 1500, 20, -1])]; + tensor var_3071_cast = reshape(shape = var_3070, x = v_109_cast); + 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_118_perm_0 = const()[name = tensor("transpose_118_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_119_perm_0 = const()[name = tensor("transpose_119_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_145 = transpose(perm = transpose_119_perm_0, x = k_111_cast); + tensor transpose_146 = transpose(perm = transpose_118_perm_0, x = q_111_cast); + tensor qk_55_cast = matmul(transpose_x = qk_55_transpose_x_0, transpose_y = qk_55_transpose_y_0, x = transpose_146, y = transpose_145); + tensor var_3076_cast = softmax(axis = var_3011, x = qk_55_cast); + 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_147 = transpose(perm = var_3072, x = var_3071_cast); + tensor var_3078_cast = matmul(transpose_x = var_3078_transpose_x_0, transpose_y = var_3078_transpose_y_0, x = var_3076_cast, y = transpose_147); + 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_144 = transpose(perm = var_3079, x = var_3078_cast); + tensor x_335_cast = reshape(shape = concat_27, x = transpose_144); + tensor var_3084_to_fp16 = const()[name = tensor("op_3084_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1086743424)))]; + tensor var_3085_to_fp16 = const()[name = tensor("op_3085_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1090020288)))]; + tensor var_3086_cast = linear(bias = var_3085_to_fp16, weight = var_3084_to_fp16, x = x_335_cast); + tensor x_337_cast = add(x = x_331_cast, y = var_3086_cast); + 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(1090022912)))]; + 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(1090025536)))]; + tensor var_3092_cast = 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); + tensor var_3101_to_fp16 = const()[name = tensor("op_3101_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1090028160)))]; + tensor var_3102_to_fp16 = const()[name = tensor("op_3102_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1103135424)))]; + tensor input_225_cast = linear(bias = var_3102_to_fp16, weight = var_3101_to_fp16, x = var_3092_cast); + tensor x_341_mode_0 = const()[name = tensor("x_341_mode_0"), val = tensor("EXACT")]; + tensor x_341_cast = gelu(mode = x_341_mode_0, x = input_225_cast); + tensor var_3107_to_fp16 = const()[name = tensor("op_3107_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1103145728)))]; + tensor var_3108_to_fp16 = const()[name = tensor("op_3108_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116252992)))]; + tensor var_3109_cast = linear(bias = var_3108_to_fp16, weight = var_3107_to_fp16, x = x_341_cast); + tensor x_343_cast = add(x = x_337_cast, y = var_3109_cast); + 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(1116255616)))]; + 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(1116258240)))]; + tensor var_3124_to_fp16 = const()[name = tensor("op_3124_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3135_cast = 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); + tensor var_3146_to_fp16 = const()[name = tensor("op_3146_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116260864)))]; + tensor var_3147_to_fp16 = const()[name = tensor("op_3147_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1119537728)))]; + tensor q_113_cast = linear(bias = var_3147_to_fp16, weight = var_3146_to_fp16, x = var_3135_cast); + tensor var_3150_to_fp16 = const()[name = tensor("op_3150_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1119540352)))]; + tensor k_113_bias_0_to_fp16 = const()[name = tensor("k_113_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1122817216)))]; + tensor k_113_cast = linear(bias = k_113_bias_0_to_fp16, weight = var_3150_to_fp16, x = var_3135_cast); + tensor var_3154_to_fp16 = const()[name = tensor("op_3154_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1122819840)))]; + tensor var_3155_to_fp16 = const()[name = tensor("op_3155_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1126096704)))]; + tensor v_113_cast = linear(bias = var_3155_to_fp16, weight = var_3154_to_fp16, x = var_3135_cast); + tensor var_3163 = const()[name = tensor("op_3163"), val = tensor([1, 1500, 20, -1])]; + tensor var_3164_cast = reshape(shape = var_3163, x = q_113_cast); + tensor const_280_to_fp16 = const()[name = tensor("const_280_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_115_cast = mul(x = var_3164_cast, y = const_280_to_fp16); + tensor var_3170 = const()[name = tensor("op_3170"), val = tensor([1, 1500, 20, -1])]; + tensor var_3171_cast = reshape(shape = var_3170, x = k_113_cast); + tensor const_281_to_fp16 = const()[name = tensor("const_281_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_115_cast = mul(x = var_3171_cast, y = const_281_to_fp16); + tensor var_3177 = const()[name = tensor("op_3177"), val = tensor([1, 1500, 20, -1])]; + tensor var_3178_cast = reshape(shape = var_3177, x = v_113_cast); + 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_120_perm_0 = const()[name = tensor("transpose_120_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_121_perm_0 = const()[name = tensor("transpose_121_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_141 = transpose(perm = transpose_121_perm_0, x = k_115_cast); + tensor transpose_142 = transpose(perm = transpose_120_perm_0, x = q_115_cast); + tensor qk_57_cast = matmul(transpose_x = qk_57_transpose_x_0, transpose_y = qk_57_transpose_y_0, x = transpose_142, y = transpose_141); + tensor var_3183_cast = softmax(axis = var_3118, x = qk_57_cast); + 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_143 = transpose(perm = var_3179, x = var_3178_cast); + tensor var_3185_cast = matmul(transpose_x = var_3185_transpose_x_0, transpose_y = var_3185_transpose_y_0, x = var_3183_cast, y = transpose_143); + 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_140 = transpose(perm = var_3186, x = var_3185_cast); + tensor x_347_cast = reshape(shape = concat_28, x = transpose_140); + tensor var_3191_to_fp16 = const()[name = tensor("op_3191_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1126099328)))]; + tensor var_3192_to_fp16 = const()[name = tensor("op_3192_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1129376192)))]; + tensor var_3193_cast = linear(bias = var_3192_to_fp16, weight = var_3191_to_fp16, x = x_347_cast); + tensor x_349_cast = add(x = x_343_cast, y = var_3193_cast); + 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(1129378816)))]; + 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(1129381440)))]; + tensor var_3199_cast = 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); + tensor var_3208_to_fp16 = const()[name = tensor("op_3208_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1129384064)))]; + tensor var_3209_to_fp16 = const()[name = tensor("op_3209_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1142491328)))]; + tensor input_233_cast = linear(bias = var_3209_to_fp16, weight = var_3208_to_fp16, x = var_3199_cast); + tensor x_353_mode_0 = const()[name = tensor("x_353_mode_0"), val = tensor("EXACT")]; + tensor x_353_cast = gelu(mode = x_353_mode_0, x = input_233_cast); + tensor var_3214_to_fp16 = const()[name = tensor("op_3214_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1142501632)))]; + tensor var_3215_to_fp16 = const()[name = tensor("op_3215_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1155608896)))]; + tensor var_3216_cast = linear(bias = var_3215_to_fp16, weight = var_3214_to_fp16, x = x_353_cast); + tensor x_355_cast = add(x = x_349_cast, y = var_3216_cast); + 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(1155611520)))]; + 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(1155614144)))]; + tensor var_3231_to_fp16 = const()[name = tensor("op_3231_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3242_cast = 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); + tensor var_3253_to_fp16 = const()[name = tensor("op_3253_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1155616768)))]; + tensor var_3254_to_fp16 = const()[name = tensor("op_3254_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1158893632)))]; + tensor q_117_cast = linear(bias = var_3254_to_fp16, weight = var_3253_to_fp16, x = var_3242_cast); + tensor var_3257_to_fp16 = const()[name = tensor("op_3257_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1158896256)))]; + tensor k_117_bias_0_to_fp16 = const()[name = tensor("k_117_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1162173120)))]; + tensor k_117_cast = linear(bias = k_117_bias_0_to_fp16, weight = var_3257_to_fp16, x = var_3242_cast); + tensor var_3261_to_fp16 = const()[name = tensor("op_3261_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1162175744)))]; + tensor var_3262_to_fp16 = const()[name = tensor("op_3262_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1165452608)))]; + tensor v_117_cast = linear(bias = var_3262_to_fp16, weight = var_3261_to_fp16, x = var_3242_cast); + tensor var_3270 = const()[name = tensor("op_3270"), val = tensor([1, 1500, 20, -1])]; + tensor var_3271_cast = reshape(shape = var_3270, x = q_117_cast); + tensor const_282_to_fp16 = const()[name = tensor("const_282_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_119_cast = mul(x = var_3271_cast, y = const_282_to_fp16); + tensor var_3277 = const()[name = tensor("op_3277"), val = tensor([1, 1500, 20, -1])]; + tensor var_3278_cast = reshape(shape = var_3277, x = k_117_cast); + tensor const_283_to_fp16 = const()[name = tensor("const_283_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_119_cast = mul(x = var_3278_cast, y = const_283_to_fp16); + tensor var_3284 = const()[name = tensor("op_3284"), val = tensor([1, 1500, 20, -1])]; + tensor var_3285_cast = reshape(shape = var_3284, x = v_117_cast); + 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_122_perm_0 = const()[name = tensor("transpose_122_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_123_perm_0 = const()[name = tensor("transpose_123_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_137 = transpose(perm = transpose_123_perm_0, x = k_119_cast); + tensor transpose_138 = transpose(perm = transpose_122_perm_0, x = q_119_cast); + tensor qk_59_cast = matmul(transpose_x = qk_59_transpose_x_0, transpose_y = qk_59_transpose_y_0, x = transpose_138, y = transpose_137); + tensor var_3290_cast = softmax(axis = var_3225, x = qk_59_cast); + 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_139 = transpose(perm = var_3286, x = var_3285_cast); + tensor var_3292_cast = matmul(transpose_x = var_3292_transpose_x_0, transpose_y = var_3292_transpose_y_0, x = var_3290_cast, y = transpose_139); + 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_136 = transpose(perm = var_3293, x = var_3292_cast); + tensor x_359_cast = reshape(shape = concat_29, x = transpose_136); + tensor var_3298_to_fp16 = const()[name = tensor("op_3298_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1165455232)))]; + tensor var_3299_to_fp16 = const()[name = tensor("op_3299_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1168732096)))]; + tensor var_3300_cast = linear(bias = var_3299_to_fp16, weight = var_3298_to_fp16, x = x_359_cast); + tensor x_361_cast = add(x = x_355_cast, y = var_3300_cast); + 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(1168734720)))]; + 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(1168737344)))]; + tensor var_3306_cast = 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); + tensor var_3315_to_fp16 = const()[name = tensor("op_3315_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1168739968)))]; + tensor var_3316_to_fp16 = const()[name = tensor("op_3316_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1181847232)))]; + tensor input_241_cast = linear(bias = var_3316_to_fp16, weight = var_3315_to_fp16, x = var_3306_cast); + tensor x_365_mode_0 = const()[name = tensor("x_365_mode_0"), val = tensor("EXACT")]; + tensor x_365_cast = gelu(mode = x_365_mode_0, x = input_241_cast); + tensor var_3321_to_fp16 = const()[name = tensor("op_3321_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1181857536)))]; + tensor var_3322_to_fp16 = const()[name = tensor("op_3322_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1194964800)))]; + tensor var_3323_cast = linear(bias = var_3322_to_fp16, weight = var_3321_to_fp16, x = x_365_cast); + tensor x_367_cast = add(x = x_361_cast, y = var_3323_cast); + 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(1194967424)))]; + 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(1194970048)))]; + tensor var_3338_to_fp16 = const()[name = tensor("op_3338_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3349_cast = 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); + tensor var_3360_to_fp16 = const()[name = tensor("op_3360_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1194972672)))]; + tensor var_3361_to_fp16 = const()[name = tensor("op_3361_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1198249536)))]; + tensor q_121_cast = linear(bias = var_3361_to_fp16, weight = var_3360_to_fp16, x = var_3349_cast); + tensor var_3364_to_fp16 = const()[name = tensor("op_3364_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1198252160)))]; + tensor k_121_bias_0_to_fp16 = const()[name = tensor("k_121_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1201529024)))]; + tensor k_121_cast = linear(bias = k_121_bias_0_to_fp16, weight = var_3364_to_fp16, x = var_3349_cast); + tensor var_3368_to_fp16 = const()[name = tensor("op_3368_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1201531648)))]; + tensor var_3369_to_fp16 = const()[name = tensor("op_3369_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1204808512)))]; + tensor v_121_cast = linear(bias = var_3369_to_fp16, weight = var_3368_to_fp16, x = var_3349_cast); + tensor var_3377 = const()[name = tensor("op_3377"), val = tensor([1, 1500, 20, -1])]; + tensor var_3378_cast = reshape(shape = var_3377, x = q_121_cast); + tensor const_284_to_fp16 = const()[name = tensor("const_284_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_123_cast = mul(x = var_3378_cast, y = const_284_to_fp16); + tensor var_3384 = const()[name = tensor("op_3384"), val = tensor([1, 1500, 20, -1])]; + tensor var_3385_cast = reshape(shape = var_3384, x = k_121_cast); + tensor const_285_to_fp16 = const()[name = tensor("const_285_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_123_cast = mul(x = var_3385_cast, y = const_285_to_fp16); + tensor var_3391 = const()[name = tensor("op_3391"), val = tensor([1, 1500, 20, -1])]; + tensor var_3392_cast = reshape(shape = var_3391, x = v_121_cast); + 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_124_perm_0 = const()[name = tensor("transpose_124_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_125_perm_0 = const()[name = tensor("transpose_125_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_133 = transpose(perm = transpose_125_perm_0, x = k_123_cast); + tensor transpose_134 = transpose(perm = transpose_124_perm_0, x = q_123_cast); + tensor qk_61_cast = matmul(transpose_x = qk_61_transpose_x_0, transpose_y = qk_61_transpose_y_0, x = transpose_134, y = transpose_133); + tensor var_3397_cast = softmax(axis = var_3332, x = qk_61_cast); + 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_135 = transpose(perm = var_3393, x = var_3392_cast); + tensor var_3399_cast = matmul(transpose_x = var_3399_transpose_x_0, transpose_y = var_3399_transpose_y_0, x = var_3397_cast, y = transpose_135); + 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_132 = transpose(perm = var_3400, x = var_3399_cast); + tensor x_371_cast = reshape(shape = concat_30, x = transpose_132); + tensor var_3405_to_fp16 = const()[name = tensor("op_3405_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1204811136)))]; + tensor var_3406_to_fp16 = const()[name = tensor("op_3406_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208088000)))]; + tensor var_3407_cast = linear(bias = var_3406_to_fp16, weight = var_3405_to_fp16, x = x_371_cast); + tensor x_373_cast = add(x = x_367_cast, y = var_3407_cast); + 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(1208090624)))]; + 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(1208093248)))]; + tensor var_3413_cast = 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); + tensor var_3422_to_fp16 = const()[name = tensor("op_3422_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208095872)))]; + tensor var_3423_to_fp16 = const()[name = tensor("op_3423_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1221203136)))]; + tensor input_249_cast = linear(bias = var_3423_to_fp16, weight = var_3422_to_fp16, x = var_3413_cast); + tensor x_377_mode_0 = const()[name = tensor("x_377_mode_0"), val = tensor("EXACT")]; + tensor x_377_cast = gelu(mode = x_377_mode_0, x = input_249_cast); + tensor var_3428_to_fp16 = const()[name = tensor("op_3428_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1221213440)))]; + tensor var_3429_to_fp16 = const()[name = tensor("op_3429_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1234320704)))]; + tensor var_3430_cast = linear(bias = var_3429_to_fp16, weight = var_3428_to_fp16, x = x_377_cast); + tensor x_379_cast = add(x = x_373_cast, y = var_3430_cast); + 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(1234323328)))]; + 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(1234325952)))]; + tensor var_3445_to_fp16 = const()[name = tensor("op_3445_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3456_cast = 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); + tensor var_3467_to_fp16 = const()[name = tensor("op_3467_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1234328576)))]; + tensor var_3468_to_fp16 = const()[name = tensor("op_3468_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1237605440)))]; + tensor q_125_cast = linear(bias = var_3468_to_fp16, weight = var_3467_to_fp16, x = var_3456_cast); + tensor var_3471_to_fp16 = const()[name = tensor("op_3471_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1237608064)))]; + tensor k_125_bias_0_to_fp16 = const()[name = tensor("k_125_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1240884928)))]; + tensor k_125_cast = linear(bias = k_125_bias_0_to_fp16, weight = var_3471_to_fp16, x = var_3456_cast); + tensor var_3475_to_fp16 = const()[name = tensor("op_3475_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1240887552)))]; + tensor var_3476_to_fp16 = const()[name = tensor("op_3476_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1244164416)))]; + tensor v_125_cast = linear(bias = var_3476_to_fp16, weight = var_3475_to_fp16, x = var_3456_cast); + tensor var_3484 = const()[name = tensor("op_3484"), val = tensor([1, 1500, 20, -1])]; + tensor var_3485_cast = reshape(shape = var_3484, x = q_125_cast); + tensor const_286_to_fp16 = const()[name = tensor("const_286_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_cast = mul(x = var_3485_cast, y = const_286_to_fp16); + tensor var_3491 = const()[name = tensor("op_3491"), val = tensor([1, 1500, 20, -1])]; + tensor var_3492_cast = reshape(shape = var_3491, x = k_125_cast); + tensor const_287_to_fp16 = const()[name = tensor("const_287_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_cast = mul(x = var_3492_cast, y = const_287_to_fp16); + tensor var_3498 = const()[name = tensor("op_3498"), val = tensor([1, 1500, 20, -1])]; + tensor var_3499_cast = reshape(shape = var_3498, x = v_125_cast); + 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_126_perm_0 = const()[name = tensor("transpose_126_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_127_perm_0 = const()[name = tensor("transpose_127_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_129 = transpose(perm = transpose_127_perm_0, x = k_cast); + tensor transpose_130 = transpose(perm = transpose_126_perm_0, x = q_cast); + tensor qk_cast = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_130, y = transpose_129); + tensor var_3504_cast = softmax(axis = var_3439, x = qk_cast); + 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_131 = transpose(perm = var_3500, x = var_3499_cast); + tensor var_3506_cast = matmul(transpose_x = var_3506_transpose_x_0, transpose_y = var_3506_transpose_y_0, x = var_3504_cast, y = transpose_131); + 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_128 = transpose(perm = var_3507, x = var_3506_cast); + tensor x_383_cast = reshape(shape = concat_31, x = transpose_128); + tensor var_3512_to_fp16 = const()[name = tensor("op_3512_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1244167040)))]; + tensor var_3513_to_fp16 = const()[name = tensor("op_3513_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1247443904)))]; + tensor var_3514_cast = linear(bias = var_3513_to_fp16, weight = var_3512_to_fp16, x = x_383_cast); + tensor x_385_cast = add(x = x_379_cast, y = var_3514_cast); + 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(1247446528)))]; + 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(1247449152)))]; + tensor var_3520_cast = 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); + tensor var_3529_to_fp16 = const()[name = tensor("op_3529_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1247451776)))]; + tensor var_3530_to_fp16 = const()[name = tensor("op_3530_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1260559040)))]; + tensor input_257_cast = linear(bias = var_3530_to_fp16, weight = var_3529_to_fp16, x = var_3520_cast); + tensor x_389_mode_0 = const()[name = tensor("x_389_mode_0"), val = tensor("EXACT")]; + tensor x_389_cast = gelu(mode = x_389_mode_0, x = input_257_cast); + tensor var_3535_to_fp16 = const()[name = tensor("op_3535_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1260569344)))]; + tensor var_3536_to_fp16 = const()[name = tensor("op_3536_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1273676608)))]; + tensor var_3537_cast = linear(bias = var_3536_to_fp16, weight = var_3535_to_fp16, x = x_389_cast); + tensor x_cast = add(x = x_385_cast, y = var_3537_cast); + 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(1273679232)))]; + 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(1273681856)))]; + tensor var_3541_to_fp16 = const()[name = tensor("op_3541_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3550_cast = 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); + tensor var_3550_cast_to_fp32_dtype_0 = const()[name = tensor("op_3550_cast_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor output = cast(dtype = var_3550_cast_to_fp32_dtype_0, x = var_3550_cast); + } -> (output); +} \ No newline at end of file