diff --git "a/model.mil" "b/model.mil" new file mode 100644--- /dev/null +++ "b/model.mil" @@ -0,0 +1,1928 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "3405.2.1"}, {"coremlc-version", "3404.23.1"}, {"coremltools-component-torch", "2.4.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "7.1"}})] +{ + func main(tensor logmel_data) { + tensor var_72 = const()[name = tensor("op_72"), val = tensor(1)]; + tensor var_80 = const()[name = tensor("op_80"), val = tensor([1])]; + tensor var_82 = const()[name = tensor("op_82"), val = tensor([1])]; + tensor var_84_pad_type_0 = const()[name = tensor("op_84_pad_type_0"), val = tensor("custom")]; + tensor var_84_pad_0 = const()[name = tensor("op_84_pad_0"), val = tensor([1, 1])]; + tensor logmel_data_to_fp16_dtype_0 = const()[name = tensor("logmel_data_to_fp16_dtype_0"), val = tensor("fp16")]; + tensor weight_3_to_fp16 = const()[name = tensor("weight_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor bias_3_to_fp16 = const()[name = tensor("bias_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(983168)))]; + tensor cast_129 = cast(dtype = logmel_data_to_fp16_dtype_0, x = logmel_data)[name = tensor("cast_129")]; + tensor var_84_cast_fp16 = conv(bias = bias_3_to_fp16, dilations = var_82, groups = var_72, pad = var_84_pad_0, pad_type = var_84_pad_type_0, strides = var_80, weight = weight_3_to_fp16, x = cast_129)[name = tensor("op_84_cast_fp16")]; + tensor input_1_mode_0 = const()[name = tensor("input_1_mode_0"), val = tensor("EXACT")]; + tensor input_1_cast_fp16 = gelu(mode = input_1_mode_0, x = var_84_cast_fp16)[name = tensor("input_1_cast_fp16")]; + tensor var_89 = const()[name = tensor("op_89"), val = tensor(1)]; + tensor var_98 = const()[name = tensor("op_98"), val = tensor([2])]; + tensor var_100 = const()[name = tensor("op_100"), val = tensor([1])]; + tensor var_102_pad_type_0 = const()[name = tensor("op_102_pad_type_0"), val = tensor("custom")]; + tensor var_102_pad_0 = const()[name = tensor("op_102_pad_0"), val = tensor([1, 1])]; + tensor weight_7_to_fp16 = const()[name = tensor("weight_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(985792)))]; + tensor bias_7_to_fp16 = const()[name = tensor("bias_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10816256)))]; + tensor var_102_cast_fp16 = conv(bias = bias_7_to_fp16, dilations = var_100, groups = var_89, pad = var_102_pad_0, pad_type = var_102_pad_type_0, strides = var_98, weight = weight_7_to_fp16, x = input_1_cast_fp16)[name = tensor("op_102_cast_fp16")]; + tensor x_3_mode_0 = const()[name = tensor("x_3_mode_0"), val = tensor("EXACT")]; + tensor x_3_cast_fp16 = gelu(mode = x_3_mode_0, x = var_102_cast_fp16)[name = tensor("x_3_cast_fp16")]; + tensor var_108 = const()[name = tensor("op_108"), val = tensor([0, 2, 1])]; + tensor positional_embedding_to_fp16 = const()[name = tensor("positional_embedding_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10818880)))]; + tensor transpose_160 = transpose(perm = var_108, x = x_3_cast_fp16)[name = tensor("transpose_160")]; + tensor var_111_cast_fp16 = add(x = transpose_160, y = positional_embedding_to_fp16)[name = tensor("op_111_cast_fp16")]; + tensor var_124 = const()[name = tensor("op_124"), val = tensor(-1)]; + tensor var_141_axes_0 = const()[name = tensor("op_141_axes_0"), val = tensor([-1])]; + tensor blocks_0_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_0_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14658944)))]; + tensor blocks_0_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_0_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14661568)))]; + tensor var_131_to_fp16 = const()[name = tensor("op_131_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_141_cast_fp16 = layer_norm(axes = var_141_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_131_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_111_cast_fp16)[name = tensor("op_141_cast_fp16")]; + tensor var_152_to_fp16 = const()[name = tensor("op_152_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14664192)))]; + tensor var_153_to_fp16 = const()[name = tensor("op_153_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17941056)))]; + tensor linear_0_cast_fp16 = linear(bias = var_153_to_fp16, weight = var_152_to_fp16, x = var_141_cast_fp16)[name = tensor("linear_0_cast_fp16")]; + tensor var_156_to_fp16 = const()[name = tensor("op_156_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17943680)))]; + tensor linear_1_bias_0_to_fp16 = const()[name = tensor("linear_1_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21220544)))]; + tensor linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_156_to_fp16, x = var_141_cast_fp16)[name = tensor("linear_1_cast_fp16")]; + tensor var_160_to_fp16 = const()[name = tensor("op_160_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21223168)))]; + tensor var_161_to_fp16 = const()[name = tensor("op_161_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24500032)))]; + tensor linear_2_cast_fp16 = linear(bias = var_161_to_fp16, weight = var_160_to_fp16, x = var_141_cast_fp16)[name = tensor("linear_2_cast_fp16")]; + tensor var_169 = const()[name = tensor("op_169"), val = tensor([1, 1500, 20, -1])]; + tensor var_170_cast_fp16 = reshape(shape = var_169, x = linear_0_cast_fp16)[name = tensor("op_170_cast_fp16")]; + tensor var_171 = const()[name = tensor("op_171"), val = tensor([0, 2, 1, 3])]; + tensor var_175 = const()[name = tensor("op_175"), val = tensor([1, 1500, 20, -1])]; + tensor var_176_cast_fp16 = reshape(shape = var_175, x = linear_1_cast_fp16)[name = tensor("op_176_cast_fp16")]; + tensor var_177 = const()[name = tensor("op_177"), val = tensor([0, 2, 1, 3])]; + tensor var_181 = const()[name = tensor("op_181"), val = tensor([1, 1500, 20, -1])]; + tensor var_182_cast_fp16 = reshape(shape = var_181, x = linear_2_cast_fp16)[name = tensor("op_182_cast_fp16")]; + tensor var_183 = const()[name = tensor("op_183"), val = tensor([0, 2, 1, 3])]; + tensor scale_1_to_fp16 = const()[name = tensor("scale_1_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_159 = transpose(perm = var_171, x = var_170_cast_fp16)[name = tensor("transpose_159")]; + tensor var_185_cast_fp16 = mul(x = transpose_159, y = scale_1_to_fp16)[name = tensor("op_185_cast_fp16")]; + tensor transpose_158 = transpose(perm = var_177, x = var_176_cast_fp16)[name = tensor("transpose_158")]; + tensor var_186_cast_fp16 = mul(x = transpose_158, y = scale_1_to_fp16)[name = tensor("op_186_cast_fp16")]; + tensor var_187_perm_0 = const()[name = tensor("op_187_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_156 = transpose(perm = var_187_perm_0, x = var_186_cast_fp16)[name = tensor("transpose_156")]; + tensor qk_1_cast_fp16 = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = var_185_cast_fp16, y = transpose_156)[name = tensor("qk_1_cast_fp16")]; + tensor var_190_cast_fp16 = softmax(axis = var_124, x = qk_1_cast_fp16)[name = tensor("op_190_cast_fp16")]; + tensor var_192_transpose_x_0 = const()[name = tensor("op_192_transpose_x_0"), val = tensor(false)]; + tensor var_192_transpose_y_0 = const()[name = tensor("op_192_transpose_y_0"), val = tensor(false)]; + tensor transpose_157 = transpose(perm = var_183, x = var_182_cast_fp16)[name = tensor("transpose_157")]; + tensor var_192_cast_fp16 = matmul(transpose_x = var_192_transpose_x_0, transpose_y = var_192_transpose_y_0, x = var_190_cast_fp16, y = transpose_157)[name = tensor("op_192_cast_fp16")]; + tensor var_193 = const()[name = tensor("op_193"), val = tensor([0, 2, 1, 3])]; + tensor concat_0 = const()[name = tensor("concat_0"), val = tensor([1, 1500, 1280])]; + tensor transpose_155 = transpose(perm = var_193, x = var_192_cast_fp16)[name = tensor("transpose_155")]; + tensor x_11_cast_fp16 = reshape(shape = concat_0, x = transpose_155)[name = tensor("x_11_cast_fp16")]; + tensor var_198_to_fp16 = const()[name = tensor("op_198_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24502656)))]; + tensor var_199_to_fp16 = const()[name = tensor("op_199_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27779520)))]; + tensor linear_3_cast_fp16 = linear(bias = var_199_to_fp16, weight = var_198_to_fp16, x = x_11_cast_fp16)[name = tensor("linear_3_cast_fp16")]; + tensor x_13_cast_fp16 = add(x = var_111_cast_fp16, y = linear_3_cast_fp16)[name = tensor("x_13_cast_fp16")]; + tensor var_206_axes_0 = const()[name = tensor("op_206_axes_0"), val = tensor([-1])]; + tensor blocks_0_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27782144)))]; + tensor blocks_0_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27784768)))]; + tensor var_206_cast_fp16 = layer_norm(axes = var_206_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_131_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast_fp16)[name = tensor("op_206_cast_fp16")]; + tensor var_215_to_fp16 = const()[name = tensor("op_215_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27787392)))]; + tensor var_216_to_fp16 = const()[name = tensor("op_216_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40894656)))]; + tensor linear_4_cast_fp16 = linear(bias = var_216_to_fp16, weight = var_215_to_fp16, x = var_206_cast_fp16)[name = tensor("linear_4_cast_fp16")]; + tensor x_17_mode_0 = const()[name = tensor("x_17_mode_0"), val = tensor("EXACT")]; + tensor x_17_cast_fp16 = gelu(mode = x_17_mode_0, x = linear_4_cast_fp16)[name = tensor("x_17_cast_fp16")]; + tensor var_221_to_fp16 = const()[name = tensor("op_221_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40904960)))]; + tensor var_222_to_fp16 = const()[name = tensor("op_222_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(54012224)))]; + tensor linear_5_cast_fp16 = linear(bias = var_222_to_fp16, weight = var_221_to_fp16, x = x_17_cast_fp16)[name = tensor("linear_5_cast_fp16")]; + tensor x_19_cast_fp16 = add(x = x_13_cast_fp16, y = linear_5_cast_fp16)[name = tensor("x_19_cast_fp16")]; + tensor var_232 = const()[name = tensor("op_232"), val = tensor(-1)]; + tensor var_249_axes_0 = const()[name = tensor("op_249_axes_0"), val = tensor([-1])]; + tensor blocks_1_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_1_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(54014848)))]; + tensor blocks_1_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_1_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(54017472)))]; + tensor var_239_to_fp16 = const()[name = tensor("op_239_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_249_cast_fp16 = layer_norm(axes = var_249_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_239_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast_fp16)[name = tensor("op_249_cast_fp16")]; + tensor var_260_to_fp16 = const()[name = tensor("op_260_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(54020096)))]; + tensor var_261_to_fp16 = const()[name = tensor("op_261_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57296960)))]; + tensor linear_6_cast_fp16 = linear(bias = var_261_to_fp16, weight = var_260_to_fp16, x = var_249_cast_fp16)[name = tensor("linear_6_cast_fp16")]; + tensor var_264_to_fp16 = const()[name = tensor("op_264_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57299584)))]; + tensor linear_7_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_264_to_fp16, x = var_249_cast_fp16)[name = tensor("linear_7_cast_fp16")]; + tensor var_268_to_fp16 = const()[name = tensor("op_268_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60576448)))]; + tensor var_269_to_fp16 = const()[name = tensor("op_269_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63853312)))]; + tensor linear_8_cast_fp16 = linear(bias = var_269_to_fp16, weight = var_268_to_fp16, x = var_249_cast_fp16)[name = tensor("linear_8_cast_fp16")]; + tensor var_277 = const()[name = tensor("op_277"), val = tensor([1, 1500, 20, -1])]; + tensor var_278_cast_fp16 = reshape(shape = var_277, x = linear_6_cast_fp16)[name = tensor("op_278_cast_fp16")]; + tensor var_279 = const()[name = tensor("op_279"), val = tensor([0, 2, 1, 3])]; + tensor var_283 = const()[name = tensor("op_283"), val = tensor([1, 1500, 20, -1])]; + tensor var_284_cast_fp16 = reshape(shape = var_283, x = linear_7_cast_fp16)[name = tensor("op_284_cast_fp16")]; + tensor var_285 = const()[name = tensor("op_285"), val = tensor([0, 2, 1, 3])]; + tensor var_289 = const()[name = tensor("op_289"), val = tensor([1, 1500, 20, -1])]; + tensor var_290_cast_fp16 = reshape(shape = var_289, x = linear_8_cast_fp16)[name = tensor("op_290_cast_fp16")]; + tensor var_291 = const()[name = tensor("op_291"), val = tensor([0, 2, 1, 3])]; + tensor scale_3_to_fp16 = const()[name = tensor("scale_3_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_154 = transpose(perm = var_279, x = var_278_cast_fp16)[name = tensor("transpose_154")]; + tensor var_293_cast_fp16 = mul(x = transpose_154, y = scale_3_to_fp16)[name = tensor("op_293_cast_fp16")]; + tensor transpose_153 = transpose(perm = var_285, x = var_284_cast_fp16)[name = tensor("transpose_153")]; + tensor var_294_cast_fp16 = mul(x = transpose_153, y = scale_3_to_fp16)[name = tensor("op_294_cast_fp16")]; + tensor var_295_perm_0 = const()[name = tensor("op_295_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_151 = transpose(perm = var_295_perm_0, x = var_294_cast_fp16)[name = tensor("transpose_151")]; + tensor qk_3_cast_fp16 = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = var_293_cast_fp16, y = transpose_151)[name = tensor("qk_3_cast_fp16")]; + tensor var_298_cast_fp16 = softmax(axis = var_232, x = qk_3_cast_fp16)[name = tensor("op_298_cast_fp16")]; + tensor var_300_transpose_x_0 = const()[name = tensor("op_300_transpose_x_0"), val = tensor(false)]; + tensor var_300_transpose_y_0 = const()[name = tensor("op_300_transpose_y_0"), val = tensor(false)]; + tensor transpose_152 = transpose(perm = var_291, x = var_290_cast_fp16)[name = tensor("transpose_152")]; + tensor var_300_cast_fp16 = matmul(transpose_x = var_300_transpose_x_0, transpose_y = var_300_transpose_y_0, x = var_298_cast_fp16, y = transpose_152)[name = tensor("op_300_cast_fp16")]; + tensor var_301 = const()[name = tensor("op_301"), val = tensor([0, 2, 1, 3])]; + tensor concat_1 = const()[name = tensor("concat_1"), val = tensor([1, 1500, 1280])]; + tensor transpose_150 = transpose(perm = var_301, x = var_300_cast_fp16)[name = tensor("transpose_150")]; + tensor x_23_cast_fp16 = reshape(shape = concat_1, x = transpose_150)[name = tensor("x_23_cast_fp16")]; + tensor var_306_to_fp16 = const()[name = tensor("op_306_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63855936)))]; + tensor var_307_to_fp16 = const()[name = tensor("op_307_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67132800)))]; + tensor linear_9_cast_fp16 = linear(bias = var_307_to_fp16, weight = var_306_to_fp16, x = x_23_cast_fp16)[name = tensor("linear_9_cast_fp16")]; + tensor x_25_cast_fp16 = add(x = x_19_cast_fp16, y = linear_9_cast_fp16)[name = tensor("x_25_cast_fp16")]; + tensor var_314_axes_0 = const()[name = tensor("op_314_axes_0"), val = tensor([-1])]; + tensor blocks_1_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67135424)))]; + tensor blocks_1_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67138048)))]; + tensor var_314_cast_fp16 = layer_norm(axes = var_314_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_239_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast_fp16)[name = tensor("op_314_cast_fp16")]; + tensor var_323_to_fp16 = const()[name = tensor("op_323_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67140672)))]; + tensor var_324_to_fp16 = const()[name = tensor("op_324_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80247936)))]; + tensor linear_10_cast_fp16 = linear(bias = var_324_to_fp16, weight = var_323_to_fp16, x = var_314_cast_fp16)[name = tensor("linear_10_cast_fp16")]; + tensor x_29_mode_0 = const()[name = tensor("x_29_mode_0"), val = tensor("EXACT")]; + tensor x_29_cast_fp16 = gelu(mode = x_29_mode_0, x = linear_10_cast_fp16)[name = tensor("x_29_cast_fp16")]; + tensor var_329_to_fp16 = const()[name = tensor("op_329_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80258240)))]; + tensor var_330_to_fp16 = const()[name = tensor("op_330_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93365504)))]; + tensor linear_11_cast_fp16 = linear(bias = var_330_to_fp16, weight = var_329_to_fp16, x = x_29_cast_fp16)[name = tensor("linear_11_cast_fp16")]; + tensor x_31_cast_fp16 = add(x = x_25_cast_fp16, y = linear_11_cast_fp16)[name = tensor("x_31_cast_fp16")]; + tensor var_340 = const()[name = tensor("op_340"), val = tensor(-1)]; + tensor var_357_axes_0 = const()[name = tensor("op_357_axes_0"), val = tensor([-1])]; + tensor blocks_2_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_2_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93368128)))]; + tensor blocks_2_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_2_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93370752)))]; + tensor var_347_to_fp16 = const()[name = tensor("op_347_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_357_cast_fp16 = layer_norm(axes = var_357_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_347_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast_fp16)[name = tensor("op_357_cast_fp16")]; + tensor var_368_to_fp16 = const()[name = tensor("op_368_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93373376)))]; + tensor var_369_to_fp16 = const()[name = tensor("op_369_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96650240)))]; + tensor linear_12_cast_fp16 = linear(bias = var_369_to_fp16, weight = var_368_to_fp16, x = var_357_cast_fp16)[name = tensor("linear_12_cast_fp16")]; + tensor var_372_to_fp16 = const()[name = tensor("op_372_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96652864)))]; + tensor linear_13_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_372_to_fp16, x = var_357_cast_fp16)[name = tensor("linear_13_cast_fp16")]; + tensor var_376_to_fp16 = const()[name = tensor("op_376_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99929728)))]; + tensor var_377_to_fp16 = const()[name = tensor("op_377_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103206592)))]; + tensor linear_14_cast_fp16 = linear(bias = var_377_to_fp16, weight = var_376_to_fp16, x = var_357_cast_fp16)[name = tensor("linear_14_cast_fp16")]; + tensor var_385 = const()[name = tensor("op_385"), val = tensor([1, 1500, 20, -1])]; + tensor var_386_cast_fp16 = reshape(shape = var_385, x = linear_12_cast_fp16)[name = tensor("op_386_cast_fp16")]; + tensor var_387 = const()[name = tensor("op_387"), val = tensor([0, 2, 1, 3])]; + tensor var_391 = const()[name = tensor("op_391"), val = tensor([1, 1500, 20, -1])]; + tensor var_392_cast_fp16 = reshape(shape = var_391, x = linear_13_cast_fp16)[name = tensor("op_392_cast_fp16")]; + tensor var_393 = const()[name = tensor("op_393"), val = tensor([0, 2, 1, 3])]; + tensor var_397 = const()[name = tensor("op_397"), val = tensor([1, 1500, 20, -1])]; + tensor var_398_cast_fp16 = reshape(shape = var_397, x = linear_14_cast_fp16)[name = tensor("op_398_cast_fp16")]; + tensor var_399 = const()[name = tensor("op_399"), val = tensor([0, 2, 1, 3])]; + tensor scale_5_to_fp16 = const()[name = tensor("scale_5_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_149 = transpose(perm = var_387, x = var_386_cast_fp16)[name = tensor("transpose_149")]; + tensor var_401_cast_fp16 = mul(x = transpose_149, y = scale_5_to_fp16)[name = tensor("op_401_cast_fp16")]; + tensor transpose_148 = transpose(perm = var_393, x = var_392_cast_fp16)[name = tensor("transpose_148")]; + tensor var_402_cast_fp16 = mul(x = transpose_148, y = scale_5_to_fp16)[name = tensor("op_402_cast_fp16")]; + tensor var_403_perm_0 = const()[name = tensor("op_403_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_146 = transpose(perm = var_403_perm_0, x = var_402_cast_fp16)[name = tensor("transpose_146")]; + tensor qk_5_cast_fp16 = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = var_401_cast_fp16, y = transpose_146)[name = tensor("qk_5_cast_fp16")]; + tensor var_406_cast_fp16 = softmax(axis = var_340, x = qk_5_cast_fp16)[name = tensor("op_406_cast_fp16")]; + tensor var_408_transpose_x_0 = const()[name = tensor("op_408_transpose_x_0"), val = tensor(false)]; + tensor var_408_transpose_y_0 = const()[name = tensor("op_408_transpose_y_0"), val = tensor(false)]; + tensor transpose_147 = transpose(perm = var_399, x = var_398_cast_fp16)[name = tensor("transpose_147")]; + tensor var_408_cast_fp16 = matmul(transpose_x = var_408_transpose_x_0, transpose_y = var_408_transpose_y_0, x = var_406_cast_fp16, y = transpose_147)[name = tensor("op_408_cast_fp16")]; + tensor var_409 = const()[name = tensor("op_409"), val = tensor([0, 2, 1, 3])]; + tensor concat_2 = const()[name = tensor("concat_2"), val = tensor([1, 1500, 1280])]; + tensor transpose_145 = transpose(perm = var_409, x = var_408_cast_fp16)[name = tensor("transpose_145")]; + tensor x_35_cast_fp16 = reshape(shape = concat_2, x = transpose_145)[name = tensor("x_35_cast_fp16")]; + tensor var_414_to_fp16 = const()[name = tensor("op_414_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(103209216)))]; + tensor var_415_to_fp16 = const()[name = tensor("op_415_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106486080)))]; + tensor linear_15_cast_fp16 = linear(bias = var_415_to_fp16, weight = var_414_to_fp16, x = x_35_cast_fp16)[name = tensor("linear_15_cast_fp16")]; + tensor x_37_cast_fp16 = add(x = x_31_cast_fp16, y = linear_15_cast_fp16)[name = tensor("x_37_cast_fp16")]; + tensor var_422_axes_0 = const()[name = tensor("op_422_axes_0"), val = tensor([-1])]; + tensor blocks_2_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106488704)))]; + tensor blocks_2_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106491328)))]; + tensor var_422_cast_fp16 = layer_norm(axes = var_422_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_347_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast_fp16)[name = tensor("op_422_cast_fp16")]; + tensor var_431_to_fp16 = const()[name = tensor("op_431_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106493952)))]; + tensor var_432_to_fp16 = const()[name = tensor("op_432_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119601216)))]; + tensor linear_16_cast_fp16 = linear(bias = var_432_to_fp16, weight = var_431_to_fp16, x = var_422_cast_fp16)[name = tensor("linear_16_cast_fp16")]; + tensor x_41_mode_0 = const()[name = tensor("x_41_mode_0"), val = tensor("EXACT")]; + tensor x_41_cast_fp16 = gelu(mode = x_41_mode_0, x = linear_16_cast_fp16)[name = tensor("x_41_cast_fp16")]; + tensor var_437_to_fp16 = const()[name = tensor("op_437_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119611520)))]; + tensor var_438_to_fp16 = const()[name = tensor("op_438_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132718784)))]; + tensor linear_17_cast_fp16 = linear(bias = var_438_to_fp16, weight = var_437_to_fp16, x = x_41_cast_fp16)[name = tensor("linear_17_cast_fp16")]; + tensor x_43_cast_fp16 = add(x = x_37_cast_fp16, y = linear_17_cast_fp16)[name = tensor("x_43_cast_fp16")]; + tensor var_448 = const()[name = tensor("op_448"), val = tensor(-1)]; + tensor var_465_axes_0 = const()[name = tensor("op_465_axes_0"), val = tensor([-1])]; + tensor blocks_3_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_3_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132721408)))]; + tensor blocks_3_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_3_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132724032)))]; + tensor var_455_to_fp16 = const()[name = tensor("op_455_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_465_cast_fp16 = layer_norm(axes = var_465_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_455_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast_fp16)[name = tensor("op_465_cast_fp16")]; + tensor var_476_to_fp16 = const()[name = tensor("op_476_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132726656)))]; + tensor var_477_to_fp16 = const()[name = tensor("op_477_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136003520)))]; + tensor linear_18_cast_fp16 = linear(bias = var_477_to_fp16, weight = var_476_to_fp16, x = var_465_cast_fp16)[name = tensor("linear_18_cast_fp16")]; + tensor var_480_to_fp16 = const()[name = tensor("op_480_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136006144)))]; + tensor linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_480_to_fp16, x = var_465_cast_fp16)[name = tensor("linear_19_cast_fp16")]; + tensor var_484_to_fp16 = const()[name = tensor("op_484_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(139283008)))]; + tensor var_485_to_fp16 = const()[name = tensor("op_485_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142559872)))]; + tensor linear_20_cast_fp16 = linear(bias = var_485_to_fp16, weight = var_484_to_fp16, x = var_465_cast_fp16)[name = tensor("linear_20_cast_fp16")]; + tensor var_493 = const()[name = tensor("op_493"), val = tensor([1, 1500, 20, -1])]; + tensor var_494_cast_fp16 = reshape(shape = var_493, x = linear_18_cast_fp16)[name = tensor("op_494_cast_fp16")]; + tensor var_495 = const()[name = tensor("op_495"), val = tensor([0, 2, 1, 3])]; + tensor var_499 = const()[name = tensor("op_499"), val = tensor([1, 1500, 20, -1])]; + tensor var_500_cast_fp16 = reshape(shape = var_499, x = linear_19_cast_fp16)[name = tensor("op_500_cast_fp16")]; + tensor var_501 = const()[name = tensor("op_501"), val = tensor([0, 2, 1, 3])]; + tensor var_505 = const()[name = tensor("op_505"), val = tensor([1, 1500, 20, -1])]; + tensor var_506_cast_fp16 = reshape(shape = var_505, x = linear_20_cast_fp16)[name = tensor("op_506_cast_fp16")]; + tensor var_507 = const()[name = tensor("op_507"), val = tensor([0, 2, 1, 3])]; + tensor scale_7_to_fp16 = const()[name = tensor("scale_7_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_144 = transpose(perm = var_495, x = var_494_cast_fp16)[name = tensor("transpose_144")]; + tensor var_509_cast_fp16 = mul(x = transpose_144, y = scale_7_to_fp16)[name = tensor("op_509_cast_fp16")]; + tensor transpose_143 = transpose(perm = var_501, x = var_500_cast_fp16)[name = tensor("transpose_143")]; + tensor var_510_cast_fp16 = mul(x = transpose_143, y = scale_7_to_fp16)[name = tensor("op_510_cast_fp16")]; + tensor var_511_perm_0 = const()[name = tensor("op_511_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_141 = transpose(perm = var_511_perm_0, x = var_510_cast_fp16)[name = tensor("transpose_141")]; + tensor qk_7_cast_fp16 = matmul(transpose_x = qk_7_transpose_x_0, transpose_y = qk_7_transpose_y_0, x = var_509_cast_fp16, y = transpose_141)[name = tensor("qk_7_cast_fp16")]; + tensor var_514_cast_fp16 = softmax(axis = var_448, x = qk_7_cast_fp16)[name = tensor("op_514_cast_fp16")]; + tensor var_516_transpose_x_0 = const()[name = tensor("op_516_transpose_x_0"), val = tensor(false)]; + tensor var_516_transpose_y_0 = const()[name = tensor("op_516_transpose_y_0"), val = tensor(false)]; + tensor transpose_142 = transpose(perm = var_507, x = var_506_cast_fp16)[name = tensor("transpose_142")]; + tensor var_516_cast_fp16 = matmul(transpose_x = var_516_transpose_x_0, transpose_y = var_516_transpose_y_0, x = var_514_cast_fp16, y = transpose_142)[name = tensor("op_516_cast_fp16")]; + tensor var_517 = const()[name = tensor("op_517"), val = tensor([0, 2, 1, 3])]; + tensor concat_3 = const()[name = tensor("concat_3"), val = tensor([1, 1500, 1280])]; + tensor transpose_140 = transpose(perm = var_517, x = var_516_cast_fp16)[name = tensor("transpose_140")]; + tensor x_47_cast_fp16 = reshape(shape = concat_3, x = transpose_140)[name = tensor("x_47_cast_fp16")]; + tensor var_522_to_fp16 = const()[name = tensor("op_522_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142562496)))]; + tensor var_523_to_fp16 = const()[name = tensor("op_523_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145839360)))]; + tensor linear_21_cast_fp16 = linear(bias = var_523_to_fp16, weight = var_522_to_fp16, x = x_47_cast_fp16)[name = tensor("linear_21_cast_fp16")]; + tensor x_49_cast_fp16 = add(x = x_43_cast_fp16, y = linear_21_cast_fp16)[name = tensor("x_49_cast_fp16")]; + tensor var_530_axes_0 = const()[name = tensor("op_530_axes_0"), val = tensor([-1])]; + tensor blocks_3_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145841984)))]; + tensor blocks_3_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145844608)))]; + tensor var_530_cast_fp16 = layer_norm(axes = var_530_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_455_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast_fp16)[name = tensor("op_530_cast_fp16")]; + tensor var_539_to_fp16 = const()[name = tensor("op_539_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145847232)))]; + tensor var_540_to_fp16 = const()[name = tensor("op_540_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158954496)))]; + tensor linear_22_cast_fp16 = linear(bias = var_540_to_fp16, weight = var_539_to_fp16, x = var_530_cast_fp16)[name = tensor("linear_22_cast_fp16")]; + tensor x_53_mode_0 = const()[name = tensor("x_53_mode_0"), val = tensor("EXACT")]; + tensor x_53_cast_fp16 = gelu(mode = x_53_mode_0, x = linear_22_cast_fp16)[name = tensor("x_53_cast_fp16")]; + tensor var_545_to_fp16 = const()[name = tensor("op_545_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158964800)))]; + tensor var_546_to_fp16 = const()[name = tensor("op_546_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(172072064)))]; + tensor linear_23_cast_fp16 = linear(bias = var_546_to_fp16, weight = var_545_to_fp16, x = x_53_cast_fp16)[name = tensor("linear_23_cast_fp16")]; + tensor x_55_cast_fp16 = add(x = x_49_cast_fp16, y = linear_23_cast_fp16)[name = tensor("x_55_cast_fp16")]; + tensor var_556 = const()[name = tensor("op_556"), val = tensor(-1)]; + tensor var_573_axes_0 = const()[name = tensor("op_573_axes_0"), val = tensor([-1])]; + tensor blocks_4_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_4_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(172074688)))]; + tensor blocks_4_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_4_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(172077312)))]; + tensor var_563_to_fp16 = const()[name = tensor("op_563_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_573_cast_fp16 = layer_norm(axes = var_573_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_563_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast_fp16)[name = tensor("op_573_cast_fp16")]; + tensor var_584_to_fp16 = const()[name = tensor("op_584_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(172079936)))]; + tensor var_585_to_fp16 = const()[name = tensor("op_585_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175356800)))]; + tensor linear_24_cast_fp16 = linear(bias = var_585_to_fp16, weight = var_584_to_fp16, x = var_573_cast_fp16)[name = tensor("linear_24_cast_fp16")]; + tensor var_588_to_fp16 = const()[name = tensor("op_588_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(175359424)))]; + tensor linear_25_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_588_to_fp16, x = var_573_cast_fp16)[name = tensor("linear_25_cast_fp16")]; + tensor var_592_to_fp16 = const()[name = tensor("op_592_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178636288)))]; + tensor var_593_to_fp16 = const()[name = tensor("op_593_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(181913152)))]; + tensor linear_26_cast_fp16 = linear(bias = var_593_to_fp16, weight = var_592_to_fp16, x = var_573_cast_fp16)[name = tensor("linear_26_cast_fp16")]; + tensor var_601 = const()[name = tensor("op_601"), val = tensor([1, 1500, 20, -1])]; + tensor var_602_cast_fp16 = reshape(shape = var_601, x = linear_24_cast_fp16)[name = tensor("op_602_cast_fp16")]; + tensor var_603 = const()[name = tensor("op_603"), val = tensor([0, 2, 1, 3])]; + tensor var_607 = const()[name = tensor("op_607"), val = tensor([1, 1500, 20, -1])]; + tensor var_608_cast_fp16 = reshape(shape = var_607, x = linear_25_cast_fp16)[name = tensor("op_608_cast_fp16")]; + tensor var_609 = const()[name = tensor("op_609"), val = tensor([0, 2, 1, 3])]; + tensor var_613 = const()[name = tensor("op_613"), val = tensor([1, 1500, 20, -1])]; + tensor var_614_cast_fp16 = reshape(shape = var_613, x = linear_26_cast_fp16)[name = tensor("op_614_cast_fp16")]; + tensor var_615 = const()[name = tensor("op_615"), val = tensor([0, 2, 1, 3])]; + tensor scale_9_to_fp16 = const()[name = tensor("scale_9_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_139 = transpose(perm = var_603, x = var_602_cast_fp16)[name = tensor("transpose_139")]; + tensor var_617_cast_fp16 = mul(x = transpose_139, y = scale_9_to_fp16)[name = tensor("op_617_cast_fp16")]; + tensor transpose_138 = transpose(perm = var_609, x = var_608_cast_fp16)[name = tensor("transpose_138")]; + tensor var_618_cast_fp16 = mul(x = transpose_138, y = scale_9_to_fp16)[name = tensor("op_618_cast_fp16")]; + tensor var_619_perm_0 = const()[name = tensor("op_619_perm_0"), val = tensor([0, 1, -1, -2])]; + tensor qk_9_transpose_x_0 = const()[name = tensor("qk_9_transpose_x_0"), val = tensor(false)]; + tensor qk_9_transpose_y_0 = const()[name = tensor("qk_9_transpose_y_0"), val = tensor(false)]; + tensor transpose_136 = transpose(perm = var_619_perm_0, x = var_618_cast_fp16)[name = tensor("transpose_136")]; + tensor qk_9_cast_fp16 = matmul(transpose_x = qk_9_transpose_x_0, transpose_y = qk_9_transpose_y_0, x = var_617_cast_fp16, y = transpose_136)[name = tensor("qk_9_cast_fp16")]; + tensor var_622_cast_fp16 = softmax(axis = var_556, x = qk_9_cast_fp16)[name = tensor("op_622_cast_fp16")]; + tensor var_624_transpose_x_0 = const()[name = tensor("op_624_transpose_x_0"), val = tensor(false)]; + tensor var_624_transpose_y_0 = const()[name = tensor("op_624_transpose_y_0"), val = tensor(false)]; + tensor transpose_137 = transpose(perm = var_615, x = var_614_cast_fp16)[name = tensor("transpose_137")]; + tensor var_624_cast_fp16 = matmul(transpose_x = var_624_transpose_x_0, transpose_y = var_624_transpose_y_0, x = var_622_cast_fp16, y = transpose_137)[name = tensor("op_624_cast_fp16")]; + tensor var_625 = const()[name = tensor("op_625"), val = tensor([0, 2, 1, 3])]; + tensor concat_4 = const()[name = tensor("concat_4"), val = tensor([1, 1500, 1280])]; + tensor transpose_135 = transpose(perm = var_625, x = var_624_cast_fp16)[name = tensor("transpose_135")]; + tensor x_59_cast_fp16 = reshape(shape = concat_4, x = transpose_135)[name = tensor("x_59_cast_fp16")]; + tensor var_630_to_fp16 = const()[name = tensor("op_630_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(181915776)))]; + tensor var_631_to_fp16 = const()[name = tensor("op_631_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(185192640)))]; + tensor linear_27_cast_fp16 = linear(bias = var_631_to_fp16, weight = var_630_to_fp16, x = x_59_cast_fp16)[name = tensor("linear_27_cast_fp16")]; + tensor x_61_cast_fp16 = add(x = x_55_cast_fp16, y = linear_27_cast_fp16)[name = tensor("x_61_cast_fp16")]; + tensor var_638_axes_0 = const()[name = tensor("op_638_axes_0"), val = tensor([-1])]; + tensor blocks_4_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(185195264)))]; + tensor blocks_4_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(185197888)))]; + tensor var_638_cast_fp16 = layer_norm(axes = var_638_axes_0, beta = blocks_4_mlp_ln_bias_to_fp16, epsilon = var_563_to_fp16, gamma = blocks_4_mlp_ln_weight_to_fp16, x = x_61_cast_fp16)[name = tensor("op_638_cast_fp16")]; + tensor var_647_to_fp16 = const()[name = tensor("op_647_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(185200512)))]; + tensor var_648_to_fp16 = const()[name = tensor("op_648_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198307776)))]; + tensor linear_28_cast_fp16 = linear(bias = var_648_to_fp16, weight = var_647_to_fp16, x = var_638_cast_fp16)[name = tensor("linear_28_cast_fp16")]; + tensor x_65_mode_0 = const()[name = tensor("x_65_mode_0"), val = tensor("EXACT")]; + tensor x_65_cast_fp16 = gelu(mode = x_65_mode_0, x = linear_28_cast_fp16)[name = tensor("x_65_cast_fp16")]; + tensor var_653_to_fp16 = const()[name = tensor("op_653_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(198318080)))]; + tensor var_654_to_fp16 = const()[name = tensor("op_654_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211425344)))]; + tensor linear_29_cast_fp16 = linear(bias = var_654_to_fp16, weight = var_653_to_fp16, x = x_65_cast_fp16)[name = tensor("linear_29_cast_fp16")]; + tensor x_67_cast_fp16 = add(x = x_61_cast_fp16, y = linear_29_cast_fp16)[name = tensor("x_67_cast_fp16")]; + tensor var_664 = const()[name = tensor("op_664"), val = tensor(-1)]; + tensor var_681_axes_0 = const()[name = tensor("op_681_axes_0"), val = tensor([-1])]; + tensor blocks_5_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_5_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211427968)))]; + tensor blocks_5_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_5_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211430592)))]; + tensor var_671_to_fp16 = const()[name = tensor("op_671_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_681_cast_fp16 = layer_norm(axes = var_681_axes_0, beta = blocks_5_attn_ln_bias_to_fp16, epsilon = var_671_to_fp16, gamma = blocks_5_attn_ln_weight_to_fp16, x = x_67_cast_fp16)[name = tensor("op_681_cast_fp16")]; + tensor var_692_to_fp16 = const()[name = tensor("op_692_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211433216)))]; + tensor var_693_to_fp16 = const()[name = tensor("op_693_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214710080)))]; + tensor linear_30_cast_fp16 = linear(bias = var_693_to_fp16, weight = var_692_to_fp16, x = var_681_cast_fp16)[name = tensor("linear_30_cast_fp16")]; + tensor var_696_to_fp16 = const()[name = tensor("op_696_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214712704)))]; + tensor linear_31_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_696_to_fp16, x = var_681_cast_fp16)[name = tensor("linear_31_cast_fp16")]; + tensor var_700_to_fp16 = const()[name = tensor("op_700_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217989568)))]; + tensor var_701_to_fp16 = const()[name = tensor("op_701_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(221266432)))]; + tensor linear_32_cast_fp16 = linear(bias = var_701_to_fp16, weight = var_700_to_fp16, x = var_681_cast_fp16)[name = tensor("linear_32_cast_fp16")]; + tensor var_709 = const()[name = tensor("op_709"), val = tensor([1, 1500, 20, -1])]; + tensor var_710_cast_fp16 = reshape(shape = var_709, x = linear_30_cast_fp16)[name = tensor("op_710_cast_fp16")]; + tensor var_711 = const()[name = tensor("op_711"), val = tensor([0, 2, 1, 3])]; + tensor var_715 = const()[name = tensor("op_715"), val = tensor([1, 1500, 20, -1])]; + tensor var_716_cast_fp16 = reshape(shape = var_715, x = linear_31_cast_fp16)[name = tensor("op_716_cast_fp16")]; + tensor var_717 = const()[name = tensor("op_717"), val = tensor([0, 2, 1, 3])]; + tensor var_721 = const()[name = tensor("op_721"), val = tensor([1, 1500, 20, -1])]; + tensor var_722_cast_fp16 = reshape(shape = var_721, x = linear_32_cast_fp16)[name = tensor("op_722_cast_fp16")]; + tensor var_723 = const()[name = tensor("op_723"), val = tensor([0, 2, 1, 3])]; + tensor scale_11_to_fp16 = const()[name = tensor("scale_11_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_134 = transpose(perm = var_711, x = var_710_cast_fp16)[name = tensor("transpose_134")]; + tensor var_725_cast_fp16 = mul(x = transpose_134, y = scale_11_to_fp16)[name = tensor("op_725_cast_fp16")]; + tensor transpose_133 = transpose(perm = var_717, x = var_716_cast_fp16)[name = tensor("transpose_133")]; + tensor var_726_cast_fp16 = mul(x = transpose_133, y = scale_11_to_fp16)[name = tensor("op_726_cast_fp16")]; + tensor var_727_perm_0 = const()[name = tensor("op_727_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_131 = transpose(perm = var_727_perm_0, x = var_726_cast_fp16)[name = tensor("transpose_131")]; + tensor qk_11_cast_fp16 = matmul(transpose_x = qk_11_transpose_x_0, transpose_y = qk_11_transpose_y_0, x = var_725_cast_fp16, y = transpose_131)[name = tensor("qk_11_cast_fp16")]; + tensor var_730_cast_fp16 = softmax(axis = var_664, x = qk_11_cast_fp16)[name = tensor("op_730_cast_fp16")]; + tensor var_732_transpose_x_0 = const()[name = tensor("op_732_transpose_x_0"), val = tensor(false)]; + tensor var_732_transpose_y_0 = const()[name = tensor("op_732_transpose_y_0"), val = tensor(false)]; + tensor transpose_132 = transpose(perm = var_723, x = var_722_cast_fp16)[name = tensor("transpose_132")]; + tensor var_732_cast_fp16 = matmul(transpose_x = var_732_transpose_x_0, transpose_y = var_732_transpose_y_0, x = var_730_cast_fp16, y = transpose_132)[name = tensor("op_732_cast_fp16")]; + tensor var_733 = const()[name = tensor("op_733"), val = tensor([0, 2, 1, 3])]; + tensor concat_5 = const()[name = tensor("concat_5"), val = tensor([1, 1500, 1280])]; + tensor transpose_130 = transpose(perm = var_733, x = var_732_cast_fp16)[name = tensor("transpose_130")]; + tensor x_71_cast_fp16 = reshape(shape = concat_5, x = transpose_130)[name = tensor("x_71_cast_fp16")]; + tensor var_738_to_fp16 = const()[name = tensor("op_738_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(221269056)))]; + tensor var_739_to_fp16 = const()[name = tensor("op_739_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224545920)))]; + tensor linear_33_cast_fp16 = linear(bias = var_739_to_fp16, weight = var_738_to_fp16, x = x_71_cast_fp16)[name = tensor("linear_33_cast_fp16")]; + tensor x_73_cast_fp16 = add(x = x_67_cast_fp16, y = linear_33_cast_fp16)[name = tensor("x_73_cast_fp16")]; + tensor var_746_axes_0 = const()[name = tensor("op_746_axes_0"), val = tensor([-1])]; + tensor blocks_5_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224548544)))]; + tensor blocks_5_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224551168)))]; + tensor var_746_cast_fp16 = layer_norm(axes = var_746_axes_0, beta = blocks_5_mlp_ln_bias_to_fp16, epsilon = var_671_to_fp16, gamma = blocks_5_mlp_ln_weight_to_fp16, x = x_73_cast_fp16)[name = tensor("op_746_cast_fp16")]; + tensor var_755_to_fp16 = const()[name = tensor("op_755_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224553792)))]; + tensor var_756_to_fp16 = const()[name = tensor("op_756_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237661056)))]; + tensor linear_34_cast_fp16 = linear(bias = var_756_to_fp16, weight = var_755_to_fp16, x = var_746_cast_fp16)[name = tensor("linear_34_cast_fp16")]; + tensor x_77_mode_0 = const()[name = tensor("x_77_mode_0"), val = tensor("EXACT")]; + tensor x_77_cast_fp16 = gelu(mode = x_77_mode_0, x = linear_34_cast_fp16)[name = tensor("x_77_cast_fp16")]; + tensor var_761_to_fp16 = const()[name = tensor("op_761_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237671360)))]; + tensor var_762_to_fp16 = const()[name = tensor("op_762_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250778624)))]; + tensor linear_35_cast_fp16 = linear(bias = var_762_to_fp16, weight = var_761_to_fp16, x = x_77_cast_fp16)[name = tensor("linear_35_cast_fp16")]; + tensor x_79_cast_fp16 = add(x = x_73_cast_fp16, y = linear_35_cast_fp16)[name = tensor("x_79_cast_fp16")]; + tensor var_772 = const()[name = tensor("op_772"), val = tensor(-1)]; + tensor var_789_axes_0 = const()[name = tensor("op_789_axes_0"), val = tensor([-1])]; + tensor blocks_6_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_6_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250781248)))]; + tensor blocks_6_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_6_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250783872)))]; + tensor var_779_to_fp16 = const()[name = tensor("op_779_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_789_cast_fp16 = layer_norm(axes = var_789_axes_0, beta = blocks_6_attn_ln_bias_to_fp16, epsilon = var_779_to_fp16, gamma = blocks_6_attn_ln_weight_to_fp16, x = x_79_cast_fp16)[name = tensor("op_789_cast_fp16")]; + tensor var_800_to_fp16 = const()[name = tensor("op_800_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250786496)))]; + tensor var_801_to_fp16 = const()[name = tensor("op_801_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(254063360)))]; + tensor linear_36_cast_fp16 = linear(bias = var_801_to_fp16, weight = var_800_to_fp16, x = var_789_cast_fp16)[name = tensor("linear_36_cast_fp16")]; + tensor var_804_to_fp16 = const()[name = tensor("op_804_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(254065984)))]; + tensor linear_37_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_804_to_fp16, x = var_789_cast_fp16)[name = tensor("linear_37_cast_fp16")]; + tensor var_808_to_fp16 = const()[name = tensor("op_808_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(257342848)))]; + tensor var_809_to_fp16 = const()[name = tensor("op_809_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260619712)))]; + tensor linear_38_cast_fp16 = linear(bias = var_809_to_fp16, weight = var_808_to_fp16, x = var_789_cast_fp16)[name = tensor("linear_38_cast_fp16")]; + tensor var_817 = const()[name = tensor("op_817"), val = tensor([1, 1500, 20, -1])]; + tensor var_818_cast_fp16 = reshape(shape = var_817, x = linear_36_cast_fp16)[name = tensor("op_818_cast_fp16")]; + tensor var_819 = const()[name = tensor("op_819"), val = tensor([0, 2, 1, 3])]; + tensor var_823 = const()[name = tensor("op_823"), val = tensor([1, 1500, 20, -1])]; + tensor var_824_cast_fp16 = reshape(shape = var_823, x = linear_37_cast_fp16)[name = tensor("op_824_cast_fp16")]; + tensor var_825 = const()[name = tensor("op_825"), val = tensor([0, 2, 1, 3])]; + tensor var_829 = const()[name = tensor("op_829"), val = tensor([1, 1500, 20, -1])]; + tensor var_830_cast_fp16 = reshape(shape = var_829, x = linear_38_cast_fp16)[name = tensor("op_830_cast_fp16")]; + tensor var_831 = const()[name = tensor("op_831"), val = tensor([0, 2, 1, 3])]; + tensor scale_13_to_fp16 = const()[name = tensor("scale_13_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_129 = transpose(perm = var_819, x = var_818_cast_fp16)[name = tensor("transpose_129")]; + tensor var_833_cast_fp16 = mul(x = transpose_129, y = scale_13_to_fp16)[name = tensor("op_833_cast_fp16")]; + tensor transpose_128 = transpose(perm = var_825, x = var_824_cast_fp16)[name = tensor("transpose_128")]; + tensor var_834_cast_fp16 = mul(x = transpose_128, y = scale_13_to_fp16)[name = tensor("op_834_cast_fp16")]; + tensor var_835_perm_0 = const()[name = tensor("op_835_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_126 = transpose(perm = var_835_perm_0, x = var_834_cast_fp16)[name = tensor("transpose_126")]; + tensor qk_13_cast_fp16 = matmul(transpose_x = qk_13_transpose_x_0, transpose_y = qk_13_transpose_y_0, x = var_833_cast_fp16, y = transpose_126)[name = tensor("qk_13_cast_fp16")]; + tensor var_838_cast_fp16 = softmax(axis = var_772, x = qk_13_cast_fp16)[name = tensor("op_838_cast_fp16")]; + tensor var_840_transpose_x_0 = const()[name = tensor("op_840_transpose_x_0"), val = tensor(false)]; + tensor var_840_transpose_y_0 = const()[name = tensor("op_840_transpose_y_0"), val = tensor(false)]; + tensor transpose_127 = transpose(perm = var_831, x = var_830_cast_fp16)[name = tensor("transpose_127")]; + tensor var_840_cast_fp16 = matmul(transpose_x = var_840_transpose_x_0, transpose_y = var_840_transpose_y_0, x = var_838_cast_fp16, y = transpose_127)[name = tensor("op_840_cast_fp16")]; + tensor var_841 = const()[name = tensor("op_841"), val = tensor([0, 2, 1, 3])]; + tensor concat_6 = const()[name = tensor("concat_6"), val = tensor([1, 1500, 1280])]; + tensor transpose_125 = transpose(perm = var_841, x = var_840_cast_fp16)[name = tensor("transpose_125")]; + tensor x_83_cast_fp16 = reshape(shape = concat_6, x = transpose_125)[name = tensor("x_83_cast_fp16")]; + tensor var_846_to_fp16 = const()[name = tensor("op_846_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260622336)))]; + tensor var_847_to_fp16 = const()[name = tensor("op_847_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263899200)))]; + tensor linear_39_cast_fp16 = linear(bias = var_847_to_fp16, weight = var_846_to_fp16, x = x_83_cast_fp16)[name = tensor("linear_39_cast_fp16")]; + tensor x_85_cast_fp16 = add(x = x_79_cast_fp16, y = linear_39_cast_fp16)[name = tensor("x_85_cast_fp16")]; + tensor var_854_axes_0 = const()[name = tensor("op_854_axes_0"), val = tensor([-1])]; + tensor blocks_6_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_6_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263901824)))]; + tensor blocks_6_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_6_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263904448)))]; + tensor var_854_cast_fp16 = layer_norm(axes = var_854_axes_0, beta = blocks_6_mlp_ln_bias_to_fp16, epsilon = var_779_to_fp16, gamma = blocks_6_mlp_ln_weight_to_fp16, x = x_85_cast_fp16)[name = tensor("op_854_cast_fp16")]; + tensor var_863_to_fp16 = const()[name = tensor("op_863_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263907072)))]; + tensor var_864_to_fp16 = const()[name = tensor("op_864_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(277014336)))]; + tensor linear_40_cast_fp16 = linear(bias = var_864_to_fp16, weight = var_863_to_fp16, x = var_854_cast_fp16)[name = tensor("linear_40_cast_fp16")]; + tensor x_89_mode_0 = const()[name = tensor("x_89_mode_0"), val = tensor("EXACT")]; + tensor x_89_cast_fp16 = gelu(mode = x_89_mode_0, x = linear_40_cast_fp16)[name = tensor("x_89_cast_fp16")]; + tensor var_869_to_fp16 = const()[name = tensor("op_869_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(277024640)))]; + tensor var_870_to_fp16 = const()[name = tensor("op_870_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(290131904)))]; + tensor linear_41_cast_fp16 = linear(bias = var_870_to_fp16, weight = var_869_to_fp16, x = x_89_cast_fp16)[name = tensor("linear_41_cast_fp16")]; + tensor x_91_cast_fp16 = add(x = x_85_cast_fp16, y = linear_41_cast_fp16)[name = tensor("x_91_cast_fp16")]; + tensor var_880 = const()[name = tensor("op_880"), val = tensor(-1)]; + tensor var_897_axes_0 = const()[name = tensor("op_897_axes_0"), val = tensor([-1])]; + tensor blocks_7_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_7_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(290134528)))]; + tensor blocks_7_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_7_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(290137152)))]; + tensor var_887_to_fp16 = const()[name = tensor("op_887_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_897_cast_fp16 = layer_norm(axes = var_897_axes_0, beta = blocks_7_attn_ln_bias_to_fp16, epsilon = var_887_to_fp16, gamma = blocks_7_attn_ln_weight_to_fp16, x = x_91_cast_fp16)[name = tensor("op_897_cast_fp16")]; + tensor var_908_to_fp16 = const()[name = tensor("op_908_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(290139776)))]; + tensor var_909_to_fp16 = const()[name = tensor("op_909_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293416640)))]; + tensor linear_42_cast_fp16 = linear(bias = var_909_to_fp16, weight = var_908_to_fp16, x = var_897_cast_fp16)[name = tensor("linear_42_cast_fp16")]; + tensor var_912_to_fp16 = const()[name = tensor("op_912_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293419264)))]; + tensor linear_43_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_912_to_fp16, x = var_897_cast_fp16)[name = tensor("linear_43_cast_fp16")]; + tensor var_916_to_fp16 = const()[name = tensor("op_916_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(296696128)))]; + tensor var_917_to_fp16 = const()[name = tensor("op_917_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(299972992)))]; + tensor linear_44_cast_fp16 = linear(bias = var_917_to_fp16, weight = var_916_to_fp16, x = var_897_cast_fp16)[name = tensor("linear_44_cast_fp16")]; + tensor var_925 = const()[name = tensor("op_925"), val = tensor([1, 1500, 20, -1])]; + tensor var_926_cast_fp16 = reshape(shape = var_925, x = linear_42_cast_fp16)[name = tensor("op_926_cast_fp16")]; + tensor var_927 = const()[name = tensor("op_927"), val = tensor([0, 2, 1, 3])]; + tensor var_931 = const()[name = tensor("op_931"), val = tensor([1, 1500, 20, -1])]; + tensor var_932_cast_fp16 = reshape(shape = var_931, x = linear_43_cast_fp16)[name = tensor("op_932_cast_fp16")]; + tensor var_933 = const()[name = tensor("op_933"), val = tensor([0, 2, 1, 3])]; + tensor var_937 = const()[name = tensor("op_937"), val = tensor([1, 1500, 20, -1])]; + tensor var_938_cast_fp16 = reshape(shape = var_937, x = linear_44_cast_fp16)[name = tensor("op_938_cast_fp16")]; + tensor var_939 = const()[name = tensor("op_939"), val = tensor([0, 2, 1, 3])]; + tensor scale_15_to_fp16 = const()[name = tensor("scale_15_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_124 = transpose(perm = var_927, x = var_926_cast_fp16)[name = tensor("transpose_124")]; + tensor var_941_cast_fp16 = mul(x = transpose_124, y = scale_15_to_fp16)[name = tensor("op_941_cast_fp16")]; + tensor transpose_123 = transpose(perm = var_933, x = var_932_cast_fp16)[name = tensor("transpose_123")]; + tensor var_942_cast_fp16 = mul(x = transpose_123, y = scale_15_to_fp16)[name = tensor("op_942_cast_fp16")]; + tensor var_943_perm_0 = const()[name = tensor("op_943_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_121 = transpose(perm = var_943_perm_0, x = var_942_cast_fp16)[name = tensor("transpose_121")]; + tensor qk_15_cast_fp16 = matmul(transpose_x = qk_15_transpose_x_0, transpose_y = qk_15_transpose_y_0, x = var_941_cast_fp16, y = transpose_121)[name = tensor("qk_15_cast_fp16")]; + tensor var_946_cast_fp16 = softmax(axis = var_880, x = qk_15_cast_fp16)[name = tensor("op_946_cast_fp16")]; + tensor var_948_transpose_x_0 = const()[name = tensor("op_948_transpose_x_0"), val = tensor(false)]; + tensor var_948_transpose_y_0 = const()[name = tensor("op_948_transpose_y_0"), val = tensor(false)]; + tensor transpose_122 = transpose(perm = var_939, x = var_938_cast_fp16)[name = tensor("transpose_122")]; + tensor var_948_cast_fp16 = matmul(transpose_x = var_948_transpose_x_0, transpose_y = var_948_transpose_y_0, x = var_946_cast_fp16, y = transpose_122)[name = tensor("op_948_cast_fp16")]; + tensor var_949 = const()[name = tensor("op_949"), val = tensor([0, 2, 1, 3])]; + tensor concat_7 = const()[name = tensor("concat_7"), val = tensor([1, 1500, 1280])]; + tensor transpose_120 = transpose(perm = var_949, x = var_948_cast_fp16)[name = tensor("transpose_120")]; + tensor x_95_cast_fp16 = reshape(shape = concat_7, x = transpose_120)[name = tensor("x_95_cast_fp16")]; + tensor var_954_to_fp16 = const()[name = tensor("op_954_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(299975616)))]; + tensor var_955_to_fp16 = const()[name = tensor("op_955_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303252480)))]; + tensor linear_45_cast_fp16 = linear(bias = var_955_to_fp16, weight = var_954_to_fp16, x = x_95_cast_fp16)[name = tensor("linear_45_cast_fp16")]; + tensor x_97_cast_fp16 = add(x = x_91_cast_fp16, y = linear_45_cast_fp16)[name = tensor("x_97_cast_fp16")]; + tensor var_962_axes_0 = const()[name = tensor("op_962_axes_0"), val = tensor([-1])]; + tensor blocks_7_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_7_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303255104)))]; + tensor blocks_7_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_7_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303257728)))]; + tensor var_962_cast_fp16 = layer_norm(axes = var_962_axes_0, beta = blocks_7_mlp_ln_bias_to_fp16, epsilon = var_887_to_fp16, gamma = blocks_7_mlp_ln_weight_to_fp16, x = x_97_cast_fp16)[name = tensor("op_962_cast_fp16")]; + tensor var_971_to_fp16 = const()[name = tensor("op_971_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303260352)))]; + tensor var_972_to_fp16 = const()[name = tensor("op_972_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316367616)))]; + tensor linear_46_cast_fp16 = linear(bias = var_972_to_fp16, weight = var_971_to_fp16, x = var_962_cast_fp16)[name = tensor("linear_46_cast_fp16")]; + tensor x_101_mode_0 = const()[name = tensor("x_101_mode_0"), val = tensor("EXACT")]; + tensor x_101_cast_fp16 = gelu(mode = x_101_mode_0, x = linear_46_cast_fp16)[name = tensor("x_101_cast_fp16")]; + tensor var_977_to_fp16 = const()[name = tensor("op_977_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316377920)))]; + tensor var_978_to_fp16 = const()[name = tensor("op_978_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329485184)))]; + tensor linear_47_cast_fp16 = linear(bias = var_978_to_fp16, weight = var_977_to_fp16, x = x_101_cast_fp16)[name = tensor("linear_47_cast_fp16")]; + tensor x_103_cast_fp16 = add(x = x_97_cast_fp16, y = linear_47_cast_fp16)[name = tensor("x_103_cast_fp16")]; + tensor var_988 = const()[name = tensor("op_988"), val = tensor(-1)]; + tensor var_1005_axes_0 = const()[name = tensor("op_1005_axes_0"), val = tensor([-1])]; + tensor blocks_8_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_8_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329487808)))]; + tensor blocks_8_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_8_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329490432)))]; + tensor var_995_to_fp16 = const()[name = tensor("op_995_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1005_cast_fp16 = layer_norm(axes = var_1005_axes_0, beta = blocks_8_attn_ln_bias_to_fp16, epsilon = var_995_to_fp16, gamma = blocks_8_attn_ln_weight_to_fp16, x = x_103_cast_fp16)[name = tensor("op_1005_cast_fp16")]; + tensor var_1016_to_fp16 = const()[name = tensor("op_1016_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329493056)))]; + tensor var_1017_to_fp16 = const()[name = tensor("op_1017_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(332769920)))]; + tensor linear_48_cast_fp16 = linear(bias = var_1017_to_fp16, weight = var_1016_to_fp16, x = var_1005_cast_fp16)[name = tensor("linear_48_cast_fp16")]; + tensor var_1020_to_fp16 = const()[name = tensor("op_1020_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(332772544)))]; + tensor linear_49_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1020_to_fp16, x = var_1005_cast_fp16)[name = tensor("linear_49_cast_fp16")]; + tensor var_1024_to_fp16 = const()[name = tensor("op_1024_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(336049408)))]; + tensor var_1025_to_fp16 = const()[name = tensor("op_1025_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339326272)))]; + tensor linear_50_cast_fp16 = linear(bias = var_1025_to_fp16, weight = var_1024_to_fp16, x = var_1005_cast_fp16)[name = tensor("linear_50_cast_fp16")]; + tensor var_1033 = const()[name = tensor("op_1033"), val = tensor([1, 1500, 20, -1])]; + tensor var_1034_cast_fp16 = reshape(shape = var_1033, x = linear_48_cast_fp16)[name = tensor("op_1034_cast_fp16")]; + tensor var_1035 = const()[name = tensor("op_1035"), val = tensor([0, 2, 1, 3])]; + tensor var_1039 = const()[name = tensor("op_1039"), val = tensor([1, 1500, 20, -1])]; + tensor var_1040_cast_fp16 = reshape(shape = var_1039, x = linear_49_cast_fp16)[name = tensor("op_1040_cast_fp16")]; + tensor var_1041 = const()[name = tensor("op_1041"), val = tensor([0, 2, 1, 3])]; + tensor var_1045 = const()[name = tensor("op_1045"), val = tensor([1, 1500, 20, -1])]; + tensor var_1046_cast_fp16 = reshape(shape = var_1045, x = linear_50_cast_fp16)[name = tensor("op_1046_cast_fp16")]; + tensor var_1047 = const()[name = tensor("op_1047"), val = tensor([0, 2, 1, 3])]; + tensor scale_17_to_fp16 = const()[name = tensor("scale_17_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_119 = transpose(perm = var_1035, x = var_1034_cast_fp16)[name = tensor("transpose_119")]; + tensor var_1049_cast_fp16 = mul(x = transpose_119, y = scale_17_to_fp16)[name = tensor("op_1049_cast_fp16")]; + tensor transpose_118 = transpose(perm = var_1041, x = var_1040_cast_fp16)[name = tensor("transpose_118")]; + tensor var_1050_cast_fp16 = mul(x = transpose_118, y = scale_17_to_fp16)[name = tensor("op_1050_cast_fp16")]; + tensor var_1051_perm_0 = const()[name = tensor("op_1051_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_116 = transpose(perm = var_1051_perm_0, x = var_1050_cast_fp16)[name = tensor("transpose_116")]; + tensor qk_17_cast_fp16 = matmul(transpose_x = qk_17_transpose_x_0, transpose_y = qk_17_transpose_y_0, x = var_1049_cast_fp16, y = transpose_116)[name = tensor("qk_17_cast_fp16")]; + tensor var_1054_cast_fp16 = softmax(axis = var_988, x = qk_17_cast_fp16)[name = tensor("op_1054_cast_fp16")]; + tensor var_1056_transpose_x_0 = const()[name = tensor("op_1056_transpose_x_0"), val = tensor(false)]; + tensor var_1056_transpose_y_0 = const()[name = tensor("op_1056_transpose_y_0"), val = tensor(false)]; + tensor transpose_117 = transpose(perm = var_1047, x = var_1046_cast_fp16)[name = tensor("transpose_117")]; + tensor var_1056_cast_fp16 = matmul(transpose_x = var_1056_transpose_x_0, transpose_y = var_1056_transpose_y_0, x = var_1054_cast_fp16, y = transpose_117)[name = tensor("op_1056_cast_fp16")]; + tensor var_1057 = const()[name = tensor("op_1057"), val = tensor([0, 2, 1, 3])]; + tensor concat_8 = const()[name = tensor("concat_8"), val = tensor([1, 1500, 1280])]; + tensor transpose_115 = transpose(perm = var_1057, x = var_1056_cast_fp16)[name = tensor("transpose_115")]; + tensor x_107_cast_fp16 = reshape(shape = concat_8, x = transpose_115)[name = tensor("x_107_cast_fp16")]; + tensor var_1062_to_fp16 = const()[name = tensor("op_1062_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339328896)))]; + tensor var_1063_to_fp16 = const()[name = tensor("op_1063_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342605760)))]; + tensor linear_51_cast_fp16 = linear(bias = var_1063_to_fp16, weight = var_1062_to_fp16, x = x_107_cast_fp16)[name = tensor("linear_51_cast_fp16")]; + tensor x_109_cast_fp16 = add(x = x_103_cast_fp16, y = linear_51_cast_fp16)[name = tensor("x_109_cast_fp16")]; + tensor var_1070_axes_0 = const()[name = tensor("op_1070_axes_0"), val = tensor([-1])]; + tensor blocks_8_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_8_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342608384)))]; + tensor blocks_8_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_8_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342611008)))]; + tensor var_1070_cast_fp16 = layer_norm(axes = var_1070_axes_0, beta = blocks_8_mlp_ln_bias_to_fp16, epsilon = var_995_to_fp16, gamma = blocks_8_mlp_ln_weight_to_fp16, x = x_109_cast_fp16)[name = tensor("op_1070_cast_fp16")]; + tensor var_1079_to_fp16 = const()[name = tensor("op_1079_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342613632)))]; + tensor var_1080_to_fp16 = const()[name = tensor("op_1080_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(355720896)))]; + tensor linear_52_cast_fp16 = linear(bias = var_1080_to_fp16, weight = var_1079_to_fp16, x = var_1070_cast_fp16)[name = tensor("linear_52_cast_fp16")]; + tensor x_113_mode_0 = const()[name = tensor("x_113_mode_0"), val = tensor("EXACT")]; + tensor x_113_cast_fp16 = gelu(mode = x_113_mode_0, x = linear_52_cast_fp16)[name = tensor("x_113_cast_fp16")]; + tensor var_1085_to_fp16 = const()[name = tensor("op_1085_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(355731200)))]; + tensor var_1086_to_fp16 = const()[name = tensor("op_1086_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368838464)))]; + tensor linear_53_cast_fp16 = linear(bias = var_1086_to_fp16, weight = var_1085_to_fp16, x = x_113_cast_fp16)[name = tensor("linear_53_cast_fp16")]; + tensor x_115_cast_fp16 = add(x = x_109_cast_fp16, y = linear_53_cast_fp16)[name = tensor("x_115_cast_fp16")]; + tensor var_1096 = const()[name = tensor("op_1096"), val = tensor(-1)]; + tensor var_1113_axes_0 = const()[name = tensor("op_1113_axes_0"), val = tensor([-1])]; + tensor blocks_9_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_9_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368841088)))]; + tensor blocks_9_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_9_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368843712)))]; + tensor var_1103_to_fp16 = const()[name = tensor("op_1103_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1113_cast_fp16 = layer_norm(axes = var_1113_axes_0, beta = blocks_9_attn_ln_bias_to_fp16, epsilon = var_1103_to_fp16, gamma = blocks_9_attn_ln_weight_to_fp16, x = x_115_cast_fp16)[name = tensor("op_1113_cast_fp16")]; + tensor var_1124_to_fp16 = const()[name = tensor("op_1124_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368846336)))]; + tensor var_1125_to_fp16 = const()[name = tensor("op_1125_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(372123200)))]; + tensor linear_54_cast_fp16 = linear(bias = var_1125_to_fp16, weight = var_1124_to_fp16, x = var_1113_cast_fp16)[name = tensor("linear_54_cast_fp16")]; + tensor var_1128_to_fp16 = const()[name = tensor("op_1128_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(372125824)))]; + tensor linear_55_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1128_to_fp16, x = var_1113_cast_fp16)[name = tensor("linear_55_cast_fp16")]; + tensor var_1132_to_fp16 = const()[name = tensor("op_1132_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(375402688)))]; + tensor var_1133_to_fp16 = const()[name = tensor("op_1133_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(378679552)))]; + tensor linear_56_cast_fp16 = linear(bias = var_1133_to_fp16, weight = var_1132_to_fp16, x = var_1113_cast_fp16)[name = tensor("linear_56_cast_fp16")]; + tensor var_1141 = const()[name = tensor("op_1141"), val = tensor([1, 1500, 20, -1])]; + tensor var_1142_cast_fp16 = reshape(shape = var_1141, x = linear_54_cast_fp16)[name = tensor("op_1142_cast_fp16")]; + tensor var_1143 = const()[name = tensor("op_1143"), val = tensor([0, 2, 1, 3])]; + tensor var_1147 = const()[name = tensor("op_1147"), val = tensor([1, 1500, 20, -1])]; + tensor var_1148_cast_fp16 = reshape(shape = var_1147, x = linear_55_cast_fp16)[name = tensor("op_1148_cast_fp16")]; + tensor var_1149 = const()[name = tensor("op_1149"), val = tensor([0, 2, 1, 3])]; + tensor var_1153 = const()[name = tensor("op_1153"), val = tensor([1, 1500, 20, -1])]; + tensor var_1154_cast_fp16 = reshape(shape = var_1153, x = linear_56_cast_fp16)[name = tensor("op_1154_cast_fp16")]; + tensor var_1155 = const()[name = tensor("op_1155"), val = tensor([0, 2, 1, 3])]; + tensor scale_19_to_fp16 = const()[name = tensor("scale_19_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_114 = transpose(perm = var_1143, x = var_1142_cast_fp16)[name = tensor("transpose_114")]; + tensor var_1157_cast_fp16 = mul(x = transpose_114, y = scale_19_to_fp16)[name = tensor("op_1157_cast_fp16")]; + tensor transpose_113 = transpose(perm = var_1149, x = var_1148_cast_fp16)[name = tensor("transpose_113")]; + tensor var_1158_cast_fp16 = mul(x = transpose_113, y = scale_19_to_fp16)[name = tensor("op_1158_cast_fp16")]; + tensor var_1159_perm_0 = const()[name = tensor("op_1159_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_111 = transpose(perm = var_1159_perm_0, x = var_1158_cast_fp16)[name = tensor("transpose_111")]; + tensor qk_19_cast_fp16 = matmul(transpose_x = qk_19_transpose_x_0, transpose_y = qk_19_transpose_y_0, x = var_1157_cast_fp16, y = transpose_111)[name = tensor("qk_19_cast_fp16")]; + tensor var_1162_cast_fp16 = softmax(axis = var_1096, x = qk_19_cast_fp16)[name = tensor("op_1162_cast_fp16")]; + tensor var_1164_transpose_x_0 = const()[name = tensor("op_1164_transpose_x_0"), val = tensor(false)]; + tensor var_1164_transpose_y_0 = const()[name = tensor("op_1164_transpose_y_0"), val = tensor(false)]; + tensor transpose_112 = transpose(perm = var_1155, x = var_1154_cast_fp16)[name = tensor("transpose_112")]; + tensor var_1164_cast_fp16 = matmul(transpose_x = var_1164_transpose_x_0, transpose_y = var_1164_transpose_y_0, x = var_1162_cast_fp16, y = transpose_112)[name = tensor("op_1164_cast_fp16")]; + tensor var_1165 = const()[name = tensor("op_1165"), val = tensor([0, 2, 1, 3])]; + tensor concat_9 = const()[name = tensor("concat_9"), val = tensor([1, 1500, 1280])]; + tensor transpose_110 = transpose(perm = var_1165, x = var_1164_cast_fp16)[name = tensor("transpose_110")]; + tensor x_119_cast_fp16 = reshape(shape = concat_9, x = transpose_110)[name = tensor("x_119_cast_fp16")]; + tensor var_1170_to_fp16 = const()[name = tensor("op_1170_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(378682176)))]; + tensor var_1171_to_fp16 = const()[name = tensor("op_1171_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381959040)))]; + tensor linear_57_cast_fp16 = linear(bias = var_1171_to_fp16, weight = var_1170_to_fp16, x = x_119_cast_fp16)[name = tensor("linear_57_cast_fp16")]; + tensor x_121_cast_fp16 = add(x = x_115_cast_fp16, y = linear_57_cast_fp16)[name = tensor("x_121_cast_fp16")]; + tensor var_1178_axes_0 = const()[name = tensor("op_1178_axes_0"), val = tensor([-1])]; + tensor blocks_9_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_9_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381961664)))]; + tensor blocks_9_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_9_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381964288)))]; + tensor var_1178_cast_fp16 = layer_norm(axes = var_1178_axes_0, beta = blocks_9_mlp_ln_bias_to_fp16, epsilon = var_1103_to_fp16, gamma = blocks_9_mlp_ln_weight_to_fp16, x = x_121_cast_fp16)[name = tensor("op_1178_cast_fp16")]; + tensor var_1187_to_fp16 = const()[name = tensor("op_1187_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381966912)))]; + tensor var_1188_to_fp16 = const()[name = tensor("op_1188_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(395074176)))]; + tensor linear_58_cast_fp16 = linear(bias = var_1188_to_fp16, weight = var_1187_to_fp16, x = var_1178_cast_fp16)[name = tensor("linear_58_cast_fp16")]; + tensor x_125_mode_0 = const()[name = tensor("x_125_mode_0"), val = tensor("EXACT")]; + tensor x_125_cast_fp16 = gelu(mode = x_125_mode_0, x = linear_58_cast_fp16)[name = tensor("x_125_cast_fp16")]; + tensor var_1193_to_fp16 = const()[name = tensor("op_1193_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(395084480)))]; + tensor var_1194_to_fp16 = const()[name = tensor("op_1194_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(408191744)))]; + tensor linear_59_cast_fp16 = linear(bias = var_1194_to_fp16, weight = var_1193_to_fp16, x = x_125_cast_fp16)[name = tensor("linear_59_cast_fp16")]; + tensor x_127_cast_fp16 = add(x = x_121_cast_fp16, y = linear_59_cast_fp16)[name = tensor("x_127_cast_fp16")]; + tensor var_1204 = const()[name = tensor("op_1204"), val = tensor(-1)]; + tensor var_1221_axes_0 = const()[name = tensor("op_1221_axes_0"), val = tensor([-1])]; + tensor blocks_10_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_10_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(408194368)))]; + tensor blocks_10_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_10_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(408196992)))]; + tensor var_1211_to_fp16 = const()[name = tensor("op_1211_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1221_cast_fp16 = layer_norm(axes = var_1221_axes_0, beta = blocks_10_attn_ln_bias_to_fp16, epsilon = var_1211_to_fp16, gamma = blocks_10_attn_ln_weight_to_fp16, x = x_127_cast_fp16)[name = tensor("op_1221_cast_fp16")]; + tensor var_1232_to_fp16 = const()[name = tensor("op_1232_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(408199616)))]; + tensor var_1233_to_fp16 = const()[name = tensor("op_1233_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411476480)))]; + tensor linear_60_cast_fp16 = linear(bias = var_1233_to_fp16, weight = var_1232_to_fp16, x = var_1221_cast_fp16)[name = tensor("linear_60_cast_fp16")]; + tensor var_1236_to_fp16 = const()[name = tensor("op_1236_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411479104)))]; + tensor linear_61_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1236_to_fp16, x = var_1221_cast_fp16)[name = tensor("linear_61_cast_fp16")]; + tensor var_1240_to_fp16 = const()[name = tensor("op_1240_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(414755968)))]; + tensor var_1241_to_fp16 = const()[name = tensor("op_1241_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(418032832)))]; + tensor linear_62_cast_fp16 = linear(bias = var_1241_to_fp16, weight = var_1240_to_fp16, x = var_1221_cast_fp16)[name = tensor("linear_62_cast_fp16")]; + tensor var_1249 = const()[name = tensor("op_1249"), val = tensor([1, 1500, 20, -1])]; + tensor var_1250_cast_fp16 = reshape(shape = var_1249, x = linear_60_cast_fp16)[name = tensor("op_1250_cast_fp16")]; + tensor var_1251 = const()[name = tensor("op_1251"), val = tensor([0, 2, 1, 3])]; + tensor var_1255 = const()[name = tensor("op_1255"), val = tensor([1, 1500, 20, -1])]; + tensor var_1256_cast_fp16 = reshape(shape = var_1255, x = linear_61_cast_fp16)[name = tensor("op_1256_cast_fp16")]; + tensor var_1257 = const()[name = tensor("op_1257"), val = tensor([0, 2, 1, 3])]; + tensor var_1261 = const()[name = tensor("op_1261"), val = tensor([1, 1500, 20, -1])]; + tensor var_1262_cast_fp16 = reshape(shape = var_1261, x = linear_62_cast_fp16)[name = tensor("op_1262_cast_fp16")]; + tensor var_1263 = const()[name = tensor("op_1263"), val = tensor([0, 2, 1, 3])]; + tensor scale_21_to_fp16 = const()[name = tensor("scale_21_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_109 = transpose(perm = var_1251, x = var_1250_cast_fp16)[name = tensor("transpose_109")]; + tensor var_1265_cast_fp16 = mul(x = transpose_109, y = scale_21_to_fp16)[name = tensor("op_1265_cast_fp16")]; + tensor transpose_108 = transpose(perm = var_1257, x = var_1256_cast_fp16)[name = tensor("transpose_108")]; + tensor var_1266_cast_fp16 = mul(x = transpose_108, y = scale_21_to_fp16)[name = tensor("op_1266_cast_fp16")]; + tensor var_1267_perm_0 = const()[name = tensor("op_1267_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_106 = transpose(perm = var_1267_perm_0, x = var_1266_cast_fp16)[name = tensor("transpose_106")]; + tensor qk_21_cast_fp16 = matmul(transpose_x = qk_21_transpose_x_0, transpose_y = qk_21_transpose_y_0, x = var_1265_cast_fp16, y = transpose_106)[name = tensor("qk_21_cast_fp16")]; + tensor var_1270_cast_fp16 = softmax(axis = var_1204, x = qk_21_cast_fp16)[name = tensor("op_1270_cast_fp16")]; + tensor var_1272_transpose_x_0 = const()[name = tensor("op_1272_transpose_x_0"), val = tensor(false)]; + tensor var_1272_transpose_y_0 = const()[name = tensor("op_1272_transpose_y_0"), val = tensor(false)]; + tensor transpose_107 = transpose(perm = var_1263, x = var_1262_cast_fp16)[name = tensor("transpose_107")]; + tensor var_1272_cast_fp16 = matmul(transpose_x = var_1272_transpose_x_0, transpose_y = var_1272_transpose_y_0, x = var_1270_cast_fp16, y = transpose_107)[name = tensor("op_1272_cast_fp16")]; + tensor var_1273 = const()[name = tensor("op_1273"), val = tensor([0, 2, 1, 3])]; + tensor concat_10 = const()[name = tensor("concat_10"), val = tensor([1, 1500, 1280])]; + tensor transpose_105 = transpose(perm = var_1273, x = var_1272_cast_fp16)[name = tensor("transpose_105")]; + tensor x_131_cast_fp16 = reshape(shape = concat_10, x = transpose_105)[name = tensor("x_131_cast_fp16")]; + tensor var_1278_to_fp16 = const()[name = tensor("op_1278_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(418035456)))]; + tensor var_1279_to_fp16 = const()[name = tensor("op_1279_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(421312320)))]; + tensor linear_63_cast_fp16 = linear(bias = var_1279_to_fp16, weight = var_1278_to_fp16, x = x_131_cast_fp16)[name = tensor("linear_63_cast_fp16")]; + tensor x_133_cast_fp16 = add(x = x_127_cast_fp16, y = linear_63_cast_fp16)[name = tensor("x_133_cast_fp16")]; + tensor var_1286_axes_0 = const()[name = tensor("op_1286_axes_0"), val = tensor([-1])]; + tensor blocks_10_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_10_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(421314944)))]; + tensor blocks_10_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_10_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(421317568)))]; + tensor var_1286_cast_fp16 = layer_norm(axes = var_1286_axes_0, beta = blocks_10_mlp_ln_bias_to_fp16, epsilon = var_1211_to_fp16, gamma = blocks_10_mlp_ln_weight_to_fp16, x = x_133_cast_fp16)[name = tensor("op_1286_cast_fp16")]; + tensor var_1295_to_fp16 = const()[name = tensor("op_1295_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(421320192)))]; + tensor var_1296_to_fp16 = const()[name = tensor("op_1296_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(434427456)))]; + tensor linear_64_cast_fp16 = linear(bias = var_1296_to_fp16, weight = var_1295_to_fp16, x = var_1286_cast_fp16)[name = tensor("linear_64_cast_fp16")]; + tensor x_137_mode_0 = const()[name = tensor("x_137_mode_0"), val = tensor("EXACT")]; + tensor x_137_cast_fp16 = gelu(mode = x_137_mode_0, x = linear_64_cast_fp16)[name = tensor("x_137_cast_fp16")]; + tensor var_1301_to_fp16 = const()[name = tensor("op_1301_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(434437760)))]; + tensor var_1302_to_fp16 = const()[name = tensor("op_1302_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447545024)))]; + tensor linear_65_cast_fp16 = linear(bias = var_1302_to_fp16, weight = var_1301_to_fp16, x = x_137_cast_fp16)[name = tensor("linear_65_cast_fp16")]; + tensor x_139_cast_fp16 = add(x = x_133_cast_fp16, y = linear_65_cast_fp16)[name = tensor("x_139_cast_fp16")]; + tensor var_1312 = const()[name = tensor("op_1312"), val = tensor(-1)]; + tensor var_1329_axes_0 = const()[name = tensor("op_1329_axes_0"), val = tensor([-1])]; + tensor blocks_11_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_11_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447547648)))]; + tensor blocks_11_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_11_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447550272)))]; + tensor var_1319_to_fp16 = const()[name = tensor("op_1319_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1329_cast_fp16 = layer_norm(axes = var_1329_axes_0, beta = blocks_11_attn_ln_bias_to_fp16, epsilon = var_1319_to_fp16, gamma = blocks_11_attn_ln_weight_to_fp16, x = x_139_cast_fp16)[name = tensor("op_1329_cast_fp16")]; + tensor var_1340_to_fp16 = const()[name = tensor("op_1340_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447552896)))]; + tensor var_1341_to_fp16 = const()[name = tensor("op_1341_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(450829760)))]; + tensor linear_66_cast_fp16 = linear(bias = var_1341_to_fp16, weight = var_1340_to_fp16, x = var_1329_cast_fp16)[name = tensor("linear_66_cast_fp16")]; + tensor var_1344_to_fp16 = const()[name = tensor("op_1344_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(450832384)))]; + tensor linear_67_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1344_to_fp16, x = var_1329_cast_fp16)[name = tensor("linear_67_cast_fp16")]; + tensor var_1348_to_fp16 = const()[name = tensor("op_1348_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(454109248)))]; + tensor var_1349_to_fp16 = const()[name = tensor("op_1349_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(457386112)))]; + tensor linear_68_cast_fp16 = linear(bias = var_1349_to_fp16, weight = var_1348_to_fp16, x = var_1329_cast_fp16)[name = tensor("linear_68_cast_fp16")]; + tensor var_1357 = const()[name = tensor("op_1357"), val = tensor([1, 1500, 20, -1])]; + tensor var_1358_cast_fp16 = reshape(shape = var_1357, x = linear_66_cast_fp16)[name = tensor("op_1358_cast_fp16")]; + tensor var_1359 = const()[name = tensor("op_1359"), val = tensor([0, 2, 1, 3])]; + tensor var_1363 = const()[name = tensor("op_1363"), val = tensor([1, 1500, 20, -1])]; + tensor var_1364_cast_fp16 = reshape(shape = var_1363, x = linear_67_cast_fp16)[name = tensor("op_1364_cast_fp16")]; + tensor var_1365 = const()[name = tensor("op_1365"), val = tensor([0, 2, 1, 3])]; + tensor var_1369 = const()[name = tensor("op_1369"), val = tensor([1, 1500, 20, -1])]; + tensor var_1370_cast_fp16 = reshape(shape = var_1369, x = linear_68_cast_fp16)[name = tensor("op_1370_cast_fp16")]; + tensor var_1371 = const()[name = tensor("op_1371"), val = tensor([0, 2, 1, 3])]; + tensor scale_23_to_fp16 = const()[name = tensor("scale_23_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_104 = transpose(perm = var_1359, x = var_1358_cast_fp16)[name = tensor("transpose_104")]; + tensor var_1373_cast_fp16 = mul(x = transpose_104, y = scale_23_to_fp16)[name = tensor("op_1373_cast_fp16")]; + tensor transpose_103 = transpose(perm = var_1365, x = var_1364_cast_fp16)[name = tensor("transpose_103")]; + tensor var_1374_cast_fp16 = mul(x = transpose_103, y = scale_23_to_fp16)[name = tensor("op_1374_cast_fp16")]; + tensor var_1375_perm_0 = const()[name = tensor("op_1375_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_101 = transpose(perm = var_1375_perm_0, x = var_1374_cast_fp16)[name = tensor("transpose_101")]; + tensor qk_23_cast_fp16 = matmul(transpose_x = qk_23_transpose_x_0, transpose_y = qk_23_transpose_y_0, x = var_1373_cast_fp16, y = transpose_101)[name = tensor("qk_23_cast_fp16")]; + tensor var_1378_cast_fp16 = softmax(axis = var_1312, x = qk_23_cast_fp16)[name = tensor("op_1378_cast_fp16")]; + tensor var_1380_transpose_x_0 = const()[name = tensor("op_1380_transpose_x_0"), val = tensor(false)]; + tensor var_1380_transpose_y_0 = const()[name = tensor("op_1380_transpose_y_0"), val = tensor(false)]; + tensor transpose_102 = transpose(perm = var_1371, x = var_1370_cast_fp16)[name = tensor("transpose_102")]; + tensor var_1380_cast_fp16 = matmul(transpose_x = var_1380_transpose_x_0, transpose_y = var_1380_transpose_y_0, x = var_1378_cast_fp16, y = transpose_102)[name = tensor("op_1380_cast_fp16")]; + tensor var_1381 = const()[name = tensor("op_1381"), val = tensor([0, 2, 1, 3])]; + tensor concat_11 = const()[name = tensor("concat_11"), val = tensor([1, 1500, 1280])]; + tensor transpose_100 = transpose(perm = var_1381, x = var_1380_cast_fp16)[name = tensor("transpose_100")]; + tensor x_143_cast_fp16 = reshape(shape = concat_11, x = transpose_100)[name = tensor("x_143_cast_fp16")]; + tensor var_1386_to_fp16 = const()[name = tensor("op_1386_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(457388736)))]; + tensor var_1387_to_fp16 = const()[name = tensor("op_1387_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460665600)))]; + tensor linear_69_cast_fp16 = linear(bias = var_1387_to_fp16, weight = var_1386_to_fp16, x = x_143_cast_fp16)[name = tensor("linear_69_cast_fp16")]; + tensor x_145_cast_fp16 = add(x = x_139_cast_fp16, y = linear_69_cast_fp16)[name = tensor("x_145_cast_fp16")]; + tensor var_1394_axes_0 = const()[name = tensor("op_1394_axes_0"), val = tensor([-1])]; + tensor blocks_11_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_11_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460668224)))]; + tensor blocks_11_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_11_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460670848)))]; + tensor var_1394_cast_fp16 = layer_norm(axes = var_1394_axes_0, beta = blocks_11_mlp_ln_bias_to_fp16, epsilon = var_1319_to_fp16, gamma = blocks_11_mlp_ln_weight_to_fp16, x = x_145_cast_fp16)[name = tensor("op_1394_cast_fp16")]; + tensor var_1403_to_fp16 = const()[name = tensor("op_1403_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460673472)))]; + tensor var_1404_to_fp16 = const()[name = tensor("op_1404_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473780736)))]; + tensor linear_70_cast_fp16 = linear(bias = var_1404_to_fp16, weight = var_1403_to_fp16, x = var_1394_cast_fp16)[name = tensor("linear_70_cast_fp16")]; + tensor x_149_mode_0 = const()[name = tensor("x_149_mode_0"), val = tensor("EXACT")]; + tensor x_149_cast_fp16 = gelu(mode = x_149_mode_0, x = linear_70_cast_fp16)[name = tensor("x_149_cast_fp16")]; + tensor var_1409_to_fp16 = const()[name = tensor("op_1409_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473791040)))]; + tensor var_1410_to_fp16 = const()[name = tensor("op_1410_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486898304)))]; + tensor linear_71_cast_fp16 = linear(bias = var_1410_to_fp16, weight = var_1409_to_fp16, x = x_149_cast_fp16)[name = tensor("linear_71_cast_fp16")]; + tensor x_151_cast_fp16 = add(x = x_145_cast_fp16, y = linear_71_cast_fp16)[name = tensor("x_151_cast_fp16")]; + tensor var_1420 = const()[name = tensor("op_1420"), val = tensor(-1)]; + tensor var_1437_axes_0 = const()[name = tensor("op_1437_axes_0"), val = tensor([-1])]; + tensor blocks_12_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_12_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486900928)))]; + tensor blocks_12_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_12_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486903552)))]; + tensor var_1427_to_fp16 = const()[name = tensor("op_1427_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1437_cast_fp16 = layer_norm(axes = var_1437_axes_0, beta = blocks_12_attn_ln_bias_to_fp16, epsilon = var_1427_to_fp16, gamma = blocks_12_attn_ln_weight_to_fp16, x = x_151_cast_fp16)[name = tensor("op_1437_cast_fp16")]; + tensor var_1448_to_fp16 = const()[name = tensor("op_1448_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486906176)))]; + tensor var_1449_to_fp16 = const()[name = tensor("op_1449_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(490183040)))]; + tensor linear_72_cast_fp16 = linear(bias = var_1449_to_fp16, weight = var_1448_to_fp16, x = var_1437_cast_fp16)[name = tensor("linear_72_cast_fp16")]; + tensor var_1452_to_fp16 = const()[name = tensor("op_1452_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(490185664)))]; + tensor linear_73_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1452_to_fp16, x = var_1437_cast_fp16)[name = tensor("linear_73_cast_fp16")]; + tensor var_1456_to_fp16 = const()[name = tensor("op_1456_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(493462528)))]; + tensor var_1457_to_fp16 = const()[name = tensor("op_1457_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496739392)))]; + tensor linear_74_cast_fp16 = linear(bias = var_1457_to_fp16, weight = var_1456_to_fp16, x = var_1437_cast_fp16)[name = tensor("linear_74_cast_fp16")]; + tensor var_1465 = const()[name = tensor("op_1465"), val = tensor([1, 1500, 20, -1])]; + tensor var_1466_cast_fp16 = reshape(shape = var_1465, x = linear_72_cast_fp16)[name = tensor("op_1466_cast_fp16")]; + tensor var_1467 = const()[name = tensor("op_1467"), val = tensor([0, 2, 1, 3])]; + tensor var_1471 = const()[name = tensor("op_1471"), val = tensor([1, 1500, 20, -1])]; + tensor var_1472_cast_fp16 = reshape(shape = var_1471, x = linear_73_cast_fp16)[name = tensor("op_1472_cast_fp16")]; + tensor var_1473 = const()[name = tensor("op_1473"), val = tensor([0, 2, 1, 3])]; + tensor var_1477 = const()[name = tensor("op_1477"), val = tensor([1, 1500, 20, -1])]; + tensor var_1478_cast_fp16 = reshape(shape = var_1477, x = linear_74_cast_fp16)[name = tensor("op_1478_cast_fp16")]; + tensor var_1479 = const()[name = tensor("op_1479"), val = tensor([0, 2, 1, 3])]; + tensor scale_25_to_fp16 = const()[name = tensor("scale_25_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_99 = transpose(perm = var_1467, x = var_1466_cast_fp16)[name = tensor("transpose_99")]; + tensor var_1481_cast_fp16 = mul(x = transpose_99, y = scale_25_to_fp16)[name = tensor("op_1481_cast_fp16")]; + tensor transpose_98 = transpose(perm = var_1473, x = var_1472_cast_fp16)[name = tensor("transpose_98")]; + tensor var_1482_cast_fp16 = mul(x = transpose_98, y = scale_25_to_fp16)[name = tensor("op_1482_cast_fp16")]; + tensor var_1483_perm_0 = const()[name = tensor("op_1483_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_96 = transpose(perm = var_1483_perm_0, x = var_1482_cast_fp16)[name = tensor("transpose_96")]; + tensor qk_25_cast_fp16 = matmul(transpose_x = qk_25_transpose_x_0, transpose_y = qk_25_transpose_y_0, x = var_1481_cast_fp16, y = transpose_96)[name = tensor("qk_25_cast_fp16")]; + tensor var_1486_cast_fp16 = softmax(axis = var_1420, x = qk_25_cast_fp16)[name = tensor("op_1486_cast_fp16")]; + tensor var_1488_transpose_x_0 = const()[name = tensor("op_1488_transpose_x_0"), val = tensor(false)]; + tensor var_1488_transpose_y_0 = const()[name = tensor("op_1488_transpose_y_0"), val = tensor(false)]; + tensor transpose_97 = transpose(perm = var_1479, x = var_1478_cast_fp16)[name = tensor("transpose_97")]; + tensor var_1488_cast_fp16 = matmul(transpose_x = var_1488_transpose_x_0, transpose_y = var_1488_transpose_y_0, x = var_1486_cast_fp16, y = transpose_97)[name = tensor("op_1488_cast_fp16")]; + tensor var_1489 = const()[name = tensor("op_1489"), val = tensor([0, 2, 1, 3])]; + tensor concat_12 = const()[name = tensor("concat_12"), val = tensor([1, 1500, 1280])]; + tensor transpose_95 = transpose(perm = var_1489, x = var_1488_cast_fp16)[name = tensor("transpose_95")]; + tensor x_155_cast_fp16 = reshape(shape = concat_12, x = transpose_95)[name = tensor("x_155_cast_fp16")]; + tensor var_1494_to_fp16 = const()[name = tensor("op_1494_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496742016)))]; + tensor var_1495_to_fp16 = const()[name = tensor("op_1495_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(500018880)))]; + tensor linear_75_cast_fp16 = linear(bias = var_1495_to_fp16, weight = var_1494_to_fp16, x = x_155_cast_fp16)[name = tensor("linear_75_cast_fp16")]; + tensor x_157_cast_fp16 = add(x = x_151_cast_fp16, y = linear_75_cast_fp16)[name = tensor("x_157_cast_fp16")]; + tensor var_1502_axes_0 = const()[name = tensor("op_1502_axes_0"), val = tensor([-1])]; + tensor blocks_12_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_12_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(500021504)))]; + tensor blocks_12_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_12_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(500024128)))]; + tensor var_1502_cast_fp16 = layer_norm(axes = var_1502_axes_0, beta = blocks_12_mlp_ln_bias_to_fp16, epsilon = var_1427_to_fp16, gamma = blocks_12_mlp_ln_weight_to_fp16, x = x_157_cast_fp16)[name = tensor("op_1502_cast_fp16")]; + tensor var_1511_to_fp16 = const()[name = tensor("op_1511_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(500026752)))]; + tensor var_1512_to_fp16 = const()[name = tensor("op_1512_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(513134016)))]; + tensor linear_76_cast_fp16 = linear(bias = var_1512_to_fp16, weight = var_1511_to_fp16, x = var_1502_cast_fp16)[name = tensor("linear_76_cast_fp16")]; + tensor x_161_mode_0 = const()[name = tensor("x_161_mode_0"), val = tensor("EXACT")]; + tensor x_161_cast_fp16 = gelu(mode = x_161_mode_0, x = linear_76_cast_fp16)[name = tensor("x_161_cast_fp16")]; + tensor var_1517_to_fp16 = const()[name = tensor("op_1517_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(513144320)))]; + tensor var_1518_to_fp16 = const()[name = tensor("op_1518_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(526251584)))]; + tensor linear_77_cast_fp16 = linear(bias = var_1518_to_fp16, weight = var_1517_to_fp16, x = x_161_cast_fp16)[name = tensor("linear_77_cast_fp16")]; + tensor x_163_cast_fp16 = add(x = x_157_cast_fp16, y = linear_77_cast_fp16)[name = tensor("x_163_cast_fp16")]; + tensor var_1528 = const()[name = tensor("op_1528"), val = tensor(-1)]; + tensor var_1545_axes_0 = const()[name = tensor("op_1545_axes_0"), val = tensor([-1])]; + tensor blocks_13_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_13_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(526254208)))]; + tensor blocks_13_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_13_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(526256832)))]; + tensor var_1535_to_fp16 = const()[name = tensor("op_1535_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1545_cast_fp16 = layer_norm(axes = var_1545_axes_0, beta = blocks_13_attn_ln_bias_to_fp16, epsilon = var_1535_to_fp16, gamma = blocks_13_attn_ln_weight_to_fp16, x = x_163_cast_fp16)[name = tensor("op_1545_cast_fp16")]; + tensor var_1556_to_fp16 = const()[name = tensor("op_1556_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(526259456)))]; + tensor var_1557_to_fp16 = const()[name = tensor("op_1557_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529536320)))]; + tensor linear_78_cast_fp16 = linear(bias = var_1557_to_fp16, weight = var_1556_to_fp16, x = var_1545_cast_fp16)[name = tensor("linear_78_cast_fp16")]; + tensor var_1560_to_fp16 = const()[name = tensor("op_1560_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529538944)))]; + tensor linear_79_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1560_to_fp16, x = var_1545_cast_fp16)[name = tensor("linear_79_cast_fp16")]; + tensor var_1564_to_fp16 = const()[name = tensor("op_1564_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(532815808)))]; + tensor var_1565_to_fp16 = const()[name = tensor("op_1565_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(536092672)))]; + tensor linear_80_cast_fp16 = linear(bias = var_1565_to_fp16, weight = var_1564_to_fp16, x = var_1545_cast_fp16)[name = tensor("linear_80_cast_fp16")]; + tensor var_1573 = const()[name = tensor("op_1573"), val = tensor([1, 1500, 20, -1])]; + tensor var_1574_cast_fp16 = reshape(shape = var_1573, x = linear_78_cast_fp16)[name = tensor("op_1574_cast_fp16")]; + tensor var_1575 = const()[name = tensor("op_1575"), val = tensor([0, 2, 1, 3])]; + tensor var_1579 = const()[name = tensor("op_1579"), val = tensor([1, 1500, 20, -1])]; + tensor var_1580_cast_fp16 = reshape(shape = var_1579, x = linear_79_cast_fp16)[name = tensor("op_1580_cast_fp16")]; + tensor var_1581 = const()[name = tensor("op_1581"), val = tensor([0, 2, 1, 3])]; + tensor var_1585 = const()[name = tensor("op_1585"), val = tensor([1, 1500, 20, -1])]; + tensor var_1586_cast_fp16 = reshape(shape = var_1585, x = linear_80_cast_fp16)[name = tensor("op_1586_cast_fp16")]; + tensor var_1587 = const()[name = tensor("op_1587"), val = tensor([0, 2, 1, 3])]; + tensor scale_27_to_fp16 = const()[name = tensor("scale_27_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_94 = transpose(perm = var_1575, x = var_1574_cast_fp16)[name = tensor("transpose_94")]; + tensor var_1589_cast_fp16 = mul(x = transpose_94, y = scale_27_to_fp16)[name = tensor("op_1589_cast_fp16")]; + tensor transpose_93 = transpose(perm = var_1581, x = var_1580_cast_fp16)[name = tensor("transpose_93")]; + tensor var_1590_cast_fp16 = mul(x = transpose_93, y = scale_27_to_fp16)[name = tensor("op_1590_cast_fp16")]; + tensor var_1591_perm_0 = const()[name = tensor("op_1591_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_91 = transpose(perm = var_1591_perm_0, x = var_1590_cast_fp16)[name = tensor("transpose_91")]; + tensor qk_27_cast_fp16 = matmul(transpose_x = qk_27_transpose_x_0, transpose_y = qk_27_transpose_y_0, x = var_1589_cast_fp16, y = transpose_91)[name = tensor("qk_27_cast_fp16")]; + tensor var_1594_cast_fp16 = softmax(axis = var_1528, x = qk_27_cast_fp16)[name = tensor("op_1594_cast_fp16")]; + tensor var_1596_transpose_x_0 = const()[name = tensor("op_1596_transpose_x_0"), val = tensor(false)]; + tensor var_1596_transpose_y_0 = const()[name = tensor("op_1596_transpose_y_0"), val = tensor(false)]; + tensor transpose_92 = transpose(perm = var_1587, x = var_1586_cast_fp16)[name = tensor("transpose_92")]; + tensor var_1596_cast_fp16 = matmul(transpose_x = var_1596_transpose_x_0, transpose_y = var_1596_transpose_y_0, x = var_1594_cast_fp16, y = transpose_92)[name = tensor("op_1596_cast_fp16")]; + tensor var_1597 = const()[name = tensor("op_1597"), val = tensor([0, 2, 1, 3])]; + tensor concat_13 = const()[name = tensor("concat_13"), val = tensor([1, 1500, 1280])]; + tensor transpose_90 = transpose(perm = var_1597, x = var_1596_cast_fp16)[name = tensor("transpose_90")]; + tensor x_167_cast_fp16 = reshape(shape = concat_13, x = transpose_90)[name = tensor("x_167_cast_fp16")]; + tensor var_1602_to_fp16 = const()[name = tensor("op_1602_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(536095296)))]; + tensor var_1603_to_fp16 = const()[name = tensor("op_1603_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(539372160)))]; + tensor linear_81_cast_fp16 = linear(bias = var_1603_to_fp16, weight = var_1602_to_fp16, x = x_167_cast_fp16)[name = tensor("linear_81_cast_fp16")]; + tensor x_169_cast_fp16 = add(x = x_163_cast_fp16, y = linear_81_cast_fp16)[name = tensor("x_169_cast_fp16")]; + tensor var_1610_axes_0 = const()[name = tensor("op_1610_axes_0"), val = tensor([-1])]; + tensor blocks_13_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_13_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(539374784)))]; + tensor blocks_13_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_13_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(539377408)))]; + tensor var_1610_cast_fp16 = layer_norm(axes = var_1610_axes_0, beta = blocks_13_mlp_ln_bias_to_fp16, epsilon = var_1535_to_fp16, gamma = blocks_13_mlp_ln_weight_to_fp16, x = x_169_cast_fp16)[name = tensor("op_1610_cast_fp16")]; + tensor var_1619_to_fp16 = const()[name = tensor("op_1619_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(539380032)))]; + tensor var_1620_to_fp16 = const()[name = tensor("op_1620_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552487296)))]; + tensor linear_82_cast_fp16 = linear(bias = var_1620_to_fp16, weight = var_1619_to_fp16, x = var_1610_cast_fp16)[name = tensor("linear_82_cast_fp16")]; + tensor x_173_mode_0 = const()[name = tensor("x_173_mode_0"), val = tensor("EXACT")]; + tensor x_173_cast_fp16 = gelu(mode = x_173_mode_0, x = linear_82_cast_fp16)[name = tensor("x_173_cast_fp16")]; + tensor var_1625_to_fp16 = const()[name = tensor("op_1625_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552497600)))]; + tensor var_1626_to_fp16 = const()[name = tensor("op_1626_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565604864)))]; + tensor linear_83_cast_fp16 = linear(bias = var_1626_to_fp16, weight = var_1625_to_fp16, x = x_173_cast_fp16)[name = tensor("linear_83_cast_fp16")]; + tensor x_175_cast_fp16 = add(x = x_169_cast_fp16, y = linear_83_cast_fp16)[name = tensor("x_175_cast_fp16")]; + tensor var_1636 = const()[name = tensor("op_1636"), val = tensor(-1)]; + tensor var_1653_axes_0 = const()[name = tensor("op_1653_axes_0"), val = tensor([-1])]; + tensor blocks_14_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_14_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565607488)))]; + tensor blocks_14_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_14_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565610112)))]; + tensor var_1643_to_fp16 = const()[name = tensor("op_1643_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1653_cast_fp16 = layer_norm(axes = var_1653_axes_0, beta = blocks_14_attn_ln_bias_to_fp16, epsilon = var_1643_to_fp16, gamma = blocks_14_attn_ln_weight_to_fp16, x = x_175_cast_fp16)[name = tensor("op_1653_cast_fp16")]; + tensor var_1664_to_fp16 = const()[name = tensor("op_1664_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565612736)))]; + tensor var_1665_to_fp16 = const()[name = tensor("op_1665_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568889600)))]; + tensor linear_84_cast_fp16 = linear(bias = var_1665_to_fp16, weight = var_1664_to_fp16, x = var_1653_cast_fp16)[name = tensor("linear_84_cast_fp16")]; + tensor var_1668_to_fp16 = const()[name = tensor("op_1668_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568892224)))]; + tensor linear_85_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1668_to_fp16, x = var_1653_cast_fp16)[name = tensor("linear_85_cast_fp16")]; + tensor var_1672_to_fp16 = const()[name = tensor("op_1672_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(572169088)))]; + tensor var_1673_to_fp16 = const()[name = tensor("op_1673_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(575445952)))]; + tensor linear_86_cast_fp16 = linear(bias = var_1673_to_fp16, weight = var_1672_to_fp16, x = var_1653_cast_fp16)[name = tensor("linear_86_cast_fp16")]; + tensor var_1681 = const()[name = tensor("op_1681"), val = tensor([1, 1500, 20, -1])]; + tensor var_1682_cast_fp16 = reshape(shape = var_1681, x = linear_84_cast_fp16)[name = tensor("op_1682_cast_fp16")]; + tensor var_1683 = const()[name = tensor("op_1683"), val = tensor([0, 2, 1, 3])]; + tensor var_1687 = const()[name = tensor("op_1687"), val = tensor([1, 1500, 20, -1])]; + tensor var_1688_cast_fp16 = reshape(shape = var_1687, x = linear_85_cast_fp16)[name = tensor("op_1688_cast_fp16")]; + tensor var_1689 = const()[name = tensor("op_1689"), val = tensor([0, 2, 1, 3])]; + tensor var_1693 = const()[name = tensor("op_1693"), val = tensor([1, 1500, 20, -1])]; + tensor var_1694_cast_fp16 = reshape(shape = var_1693, x = linear_86_cast_fp16)[name = tensor("op_1694_cast_fp16")]; + tensor var_1695 = const()[name = tensor("op_1695"), val = tensor([0, 2, 1, 3])]; + tensor scale_29_to_fp16 = const()[name = tensor("scale_29_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_89 = transpose(perm = var_1683, x = var_1682_cast_fp16)[name = tensor("transpose_89")]; + tensor var_1697_cast_fp16 = mul(x = transpose_89, y = scale_29_to_fp16)[name = tensor("op_1697_cast_fp16")]; + tensor transpose_88 = transpose(perm = var_1689, x = var_1688_cast_fp16)[name = tensor("transpose_88")]; + tensor var_1698_cast_fp16 = mul(x = transpose_88, y = scale_29_to_fp16)[name = tensor("op_1698_cast_fp16")]; + tensor var_1699_perm_0 = const()[name = tensor("op_1699_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_86 = transpose(perm = var_1699_perm_0, x = var_1698_cast_fp16)[name = tensor("transpose_86")]; + tensor qk_29_cast_fp16 = matmul(transpose_x = qk_29_transpose_x_0, transpose_y = qk_29_transpose_y_0, x = var_1697_cast_fp16, y = transpose_86)[name = tensor("qk_29_cast_fp16")]; + tensor var_1702_cast_fp16 = softmax(axis = var_1636, x = qk_29_cast_fp16)[name = tensor("op_1702_cast_fp16")]; + tensor var_1704_transpose_x_0 = const()[name = tensor("op_1704_transpose_x_0"), val = tensor(false)]; + tensor var_1704_transpose_y_0 = const()[name = tensor("op_1704_transpose_y_0"), val = tensor(false)]; + tensor transpose_87 = transpose(perm = var_1695, x = var_1694_cast_fp16)[name = tensor("transpose_87")]; + tensor var_1704_cast_fp16 = matmul(transpose_x = var_1704_transpose_x_0, transpose_y = var_1704_transpose_y_0, x = var_1702_cast_fp16, y = transpose_87)[name = tensor("op_1704_cast_fp16")]; + tensor var_1705 = const()[name = tensor("op_1705"), val = tensor([0, 2, 1, 3])]; + tensor concat_14 = const()[name = tensor("concat_14"), val = tensor([1, 1500, 1280])]; + tensor transpose_85 = transpose(perm = var_1705, x = var_1704_cast_fp16)[name = tensor("transpose_85")]; + tensor x_179_cast_fp16 = reshape(shape = concat_14, x = transpose_85)[name = tensor("x_179_cast_fp16")]; + tensor var_1710_to_fp16 = const()[name = tensor("op_1710_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(575448576)))]; + tensor var_1711_to_fp16 = const()[name = tensor("op_1711_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578725440)))]; + tensor linear_87_cast_fp16 = linear(bias = var_1711_to_fp16, weight = var_1710_to_fp16, x = x_179_cast_fp16)[name = tensor("linear_87_cast_fp16")]; + tensor x_181_cast_fp16 = add(x = x_175_cast_fp16, y = linear_87_cast_fp16)[name = tensor("x_181_cast_fp16")]; + tensor var_1718_axes_0 = const()[name = tensor("op_1718_axes_0"), val = tensor([-1])]; + tensor blocks_14_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_14_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578728064)))]; + tensor blocks_14_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_14_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578730688)))]; + tensor var_1718_cast_fp16 = layer_norm(axes = var_1718_axes_0, beta = blocks_14_mlp_ln_bias_to_fp16, epsilon = var_1643_to_fp16, gamma = blocks_14_mlp_ln_weight_to_fp16, x = x_181_cast_fp16)[name = tensor("op_1718_cast_fp16")]; + tensor var_1727_to_fp16 = const()[name = tensor("op_1727_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578733312)))]; + tensor var_1728_to_fp16 = const()[name = tensor("op_1728_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(591840576)))]; + tensor linear_88_cast_fp16 = linear(bias = var_1728_to_fp16, weight = var_1727_to_fp16, x = var_1718_cast_fp16)[name = tensor("linear_88_cast_fp16")]; + tensor x_185_mode_0 = const()[name = tensor("x_185_mode_0"), val = tensor("EXACT")]; + tensor x_185_cast_fp16 = gelu(mode = x_185_mode_0, x = linear_88_cast_fp16)[name = tensor("x_185_cast_fp16")]; + tensor var_1733_to_fp16 = const()[name = tensor("op_1733_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(591850880)))]; + tensor var_1734_to_fp16 = const()[name = tensor("op_1734_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604958144)))]; + tensor linear_89_cast_fp16 = linear(bias = var_1734_to_fp16, weight = var_1733_to_fp16, x = x_185_cast_fp16)[name = tensor("linear_89_cast_fp16")]; + tensor x_187_cast_fp16 = add(x = x_181_cast_fp16, y = linear_89_cast_fp16)[name = tensor("x_187_cast_fp16")]; + tensor var_1744 = const()[name = tensor("op_1744"), val = tensor(-1)]; + tensor var_1761_axes_0 = const()[name = tensor("op_1761_axes_0"), val = tensor([-1])]; + tensor blocks_15_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_15_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604960768)))]; + tensor blocks_15_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_15_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604963392)))]; + tensor var_1751_to_fp16 = const()[name = tensor("op_1751_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1761_cast_fp16 = layer_norm(axes = var_1761_axes_0, beta = blocks_15_attn_ln_bias_to_fp16, epsilon = var_1751_to_fp16, gamma = blocks_15_attn_ln_weight_to_fp16, x = x_187_cast_fp16)[name = tensor("op_1761_cast_fp16")]; + tensor var_1772_to_fp16 = const()[name = tensor("op_1772_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604966016)))]; + tensor var_1773_to_fp16 = const()[name = tensor("op_1773_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(608242880)))]; + tensor linear_90_cast_fp16 = linear(bias = var_1773_to_fp16, weight = var_1772_to_fp16, x = var_1761_cast_fp16)[name = tensor("linear_90_cast_fp16")]; + tensor var_1776_to_fp16 = const()[name = tensor("op_1776_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(608245504)))]; + tensor linear_91_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1776_to_fp16, x = var_1761_cast_fp16)[name = tensor("linear_91_cast_fp16")]; + tensor var_1780_to_fp16 = const()[name = tensor("op_1780_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(611522368)))]; + tensor var_1781_to_fp16 = const()[name = tensor("op_1781_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614799232)))]; + tensor linear_92_cast_fp16 = linear(bias = var_1781_to_fp16, weight = var_1780_to_fp16, x = var_1761_cast_fp16)[name = tensor("linear_92_cast_fp16")]; + tensor var_1789 = const()[name = tensor("op_1789"), val = tensor([1, 1500, 20, -1])]; + tensor var_1790_cast_fp16 = reshape(shape = var_1789, x = linear_90_cast_fp16)[name = tensor("op_1790_cast_fp16")]; + tensor var_1791 = const()[name = tensor("op_1791"), val = tensor([0, 2, 1, 3])]; + tensor var_1795 = const()[name = tensor("op_1795"), val = tensor([1, 1500, 20, -1])]; + tensor var_1796_cast_fp16 = reshape(shape = var_1795, x = linear_91_cast_fp16)[name = tensor("op_1796_cast_fp16")]; + tensor var_1797 = const()[name = tensor("op_1797"), val = tensor([0, 2, 1, 3])]; + tensor var_1801 = const()[name = tensor("op_1801"), val = tensor([1, 1500, 20, -1])]; + tensor var_1802_cast_fp16 = reshape(shape = var_1801, x = linear_92_cast_fp16)[name = tensor("op_1802_cast_fp16")]; + tensor var_1803 = const()[name = tensor("op_1803"), val = tensor([0, 2, 1, 3])]; + tensor scale_31_to_fp16 = const()[name = tensor("scale_31_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_84 = transpose(perm = var_1791, x = var_1790_cast_fp16)[name = tensor("transpose_84")]; + tensor var_1805_cast_fp16 = mul(x = transpose_84, y = scale_31_to_fp16)[name = tensor("op_1805_cast_fp16")]; + tensor transpose_83 = transpose(perm = var_1797, x = var_1796_cast_fp16)[name = tensor("transpose_83")]; + tensor var_1806_cast_fp16 = mul(x = transpose_83, y = scale_31_to_fp16)[name = tensor("op_1806_cast_fp16")]; + tensor var_1807_perm_0 = const()[name = tensor("op_1807_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_81 = transpose(perm = var_1807_perm_0, x = var_1806_cast_fp16)[name = tensor("transpose_81")]; + tensor qk_31_cast_fp16 = matmul(transpose_x = qk_31_transpose_x_0, transpose_y = qk_31_transpose_y_0, x = var_1805_cast_fp16, y = transpose_81)[name = tensor("qk_31_cast_fp16")]; + tensor var_1810_cast_fp16 = softmax(axis = var_1744, x = qk_31_cast_fp16)[name = tensor("op_1810_cast_fp16")]; + tensor var_1812_transpose_x_0 = const()[name = tensor("op_1812_transpose_x_0"), val = tensor(false)]; + tensor var_1812_transpose_y_0 = const()[name = tensor("op_1812_transpose_y_0"), val = tensor(false)]; + tensor transpose_82 = transpose(perm = var_1803, x = var_1802_cast_fp16)[name = tensor("transpose_82")]; + tensor var_1812_cast_fp16 = matmul(transpose_x = var_1812_transpose_x_0, transpose_y = var_1812_transpose_y_0, x = var_1810_cast_fp16, y = transpose_82)[name = tensor("op_1812_cast_fp16")]; + tensor var_1813 = const()[name = tensor("op_1813"), val = tensor([0, 2, 1, 3])]; + tensor concat_15 = const()[name = tensor("concat_15"), val = tensor([1, 1500, 1280])]; + tensor transpose_80 = transpose(perm = var_1813, x = var_1812_cast_fp16)[name = tensor("transpose_80")]; + tensor x_191_cast_fp16 = reshape(shape = concat_15, x = transpose_80)[name = tensor("x_191_cast_fp16")]; + tensor var_1818_to_fp16 = const()[name = tensor("op_1818_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614801856)))]; + tensor var_1819_to_fp16 = const()[name = tensor("op_1819_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(618078720)))]; + tensor linear_93_cast_fp16 = linear(bias = var_1819_to_fp16, weight = var_1818_to_fp16, x = x_191_cast_fp16)[name = tensor("linear_93_cast_fp16")]; + tensor x_193_cast_fp16 = add(x = x_187_cast_fp16, y = linear_93_cast_fp16)[name = tensor("x_193_cast_fp16")]; + tensor var_1826_axes_0 = const()[name = tensor("op_1826_axes_0"), val = tensor([-1])]; + tensor blocks_15_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_15_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(618081344)))]; + tensor blocks_15_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_15_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(618083968)))]; + tensor var_1826_cast_fp16 = layer_norm(axes = var_1826_axes_0, beta = blocks_15_mlp_ln_bias_to_fp16, epsilon = var_1751_to_fp16, gamma = blocks_15_mlp_ln_weight_to_fp16, x = x_193_cast_fp16)[name = tensor("op_1826_cast_fp16")]; + tensor var_1835_to_fp16 = const()[name = tensor("op_1835_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(618086592)))]; + tensor var_1836_to_fp16 = const()[name = tensor("op_1836_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(631193856)))]; + tensor linear_94_cast_fp16 = linear(bias = var_1836_to_fp16, weight = var_1835_to_fp16, x = var_1826_cast_fp16)[name = tensor("linear_94_cast_fp16")]; + tensor x_197_mode_0 = const()[name = tensor("x_197_mode_0"), val = tensor("EXACT")]; + tensor x_197_cast_fp16 = gelu(mode = x_197_mode_0, x = linear_94_cast_fp16)[name = tensor("x_197_cast_fp16")]; + tensor var_1841_to_fp16 = const()[name = tensor("op_1841_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(631204160)))]; + tensor var_1842_to_fp16 = const()[name = tensor("op_1842_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(644311424)))]; + tensor linear_95_cast_fp16 = linear(bias = var_1842_to_fp16, weight = var_1841_to_fp16, x = x_197_cast_fp16)[name = tensor("linear_95_cast_fp16")]; + tensor x_199_cast_fp16 = add(x = x_193_cast_fp16, y = linear_95_cast_fp16)[name = tensor("x_199_cast_fp16")]; + tensor var_1852 = const()[name = tensor("op_1852"), val = tensor(-1)]; + tensor var_1869_axes_0 = const()[name = tensor("op_1869_axes_0"), val = tensor([-1])]; + tensor blocks_16_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_16_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(644314048)))]; + tensor blocks_16_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_16_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(644316672)))]; + tensor var_1859_to_fp16 = const()[name = tensor("op_1859_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1869_cast_fp16 = layer_norm(axes = var_1869_axes_0, beta = blocks_16_attn_ln_bias_to_fp16, epsilon = var_1859_to_fp16, gamma = blocks_16_attn_ln_weight_to_fp16, x = x_199_cast_fp16)[name = tensor("op_1869_cast_fp16")]; + tensor var_1880_to_fp16 = const()[name = tensor("op_1880_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(644319296)))]; + tensor var_1881_to_fp16 = const()[name = tensor("op_1881_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647596160)))]; + tensor linear_96_cast_fp16 = linear(bias = var_1881_to_fp16, weight = var_1880_to_fp16, x = var_1869_cast_fp16)[name = tensor("linear_96_cast_fp16")]; + tensor var_1884_to_fp16 = const()[name = tensor("op_1884_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647598784)))]; + tensor linear_97_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1884_to_fp16, x = var_1869_cast_fp16)[name = tensor("linear_97_cast_fp16")]; + tensor var_1888_to_fp16 = const()[name = tensor("op_1888_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(650875648)))]; + tensor var_1889_to_fp16 = const()[name = tensor("op_1889_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(654152512)))]; + tensor linear_98_cast_fp16 = linear(bias = var_1889_to_fp16, weight = var_1888_to_fp16, x = var_1869_cast_fp16)[name = tensor("linear_98_cast_fp16")]; + tensor var_1897 = const()[name = tensor("op_1897"), val = tensor([1, 1500, 20, -1])]; + tensor var_1898_cast_fp16 = reshape(shape = var_1897, x = linear_96_cast_fp16)[name = tensor("op_1898_cast_fp16")]; + tensor var_1899 = const()[name = tensor("op_1899"), val = tensor([0, 2, 1, 3])]; + tensor var_1903 = const()[name = tensor("op_1903"), val = tensor([1, 1500, 20, -1])]; + tensor var_1904_cast_fp16 = reshape(shape = var_1903, x = linear_97_cast_fp16)[name = tensor("op_1904_cast_fp16")]; + tensor var_1905 = const()[name = tensor("op_1905"), val = tensor([0, 2, 1, 3])]; + tensor var_1909 = const()[name = tensor("op_1909"), val = tensor([1, 1500, 20, -1])]; + tensor var_1910_cast_fp16 = reshape(shape = var_1909, x = linear_98_cast_fp16)[name = tensor("op_1910_cast_fp16")]; + tensor var_1911 = const()[name = tensor("op_1911"), val = tensor([0, 2, 1, 3])]; + tensor scale_33_to_fp16 = const()[name = tensor("scale_33_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_79 = transpose(perm = var_1899, x = var_1898_cast_fp16)[name = tensor("transpose_79")]; + tensor var_1913_cast_fp16 = mul(x = transpose_79, y = scale_33_to_fp16)[name = tensor("op_1913_cast_fp16")]; + tensor transpose_78 = transpose(perm = var_1905, x = var_1904_cast_fp16)[name = tensor("transpose_78")]; + tensor var_1914_cast_fp16 = mul(x = transpose_78, y = scale_33_to_fp16)[name = tensor("op_1914_cast_fp16")]; + tensor var_1915_perm_0 = const()[name = tensor("op_1915_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_76 = transpose(perm = var_1915_perm_0, x = var_1914_cast_fp16)[name = tensor("transpose_76")]; + tensor qk_33_cast_fp16 = matmul(transpose_x = qk_33_transpose_x_0, transpose_y = qk_33_transpose_y_0, x = var_1913_cast_fp16, y = transpose_76)[name = tensor("qk_33_cast_fp16")]; + tensor var_1918_cast_fp16 = softmax(axis = var_1852, x = qk_33_cast_fp16)[name = tensor("op_1918_cast_fp16")]; + tensor var_1920_transpose_x_0 = const()[name = tensor("op_1920_transpose_x_0"), val = tensor(false)]; + tensor var_1920_transpose_y_0 = const()[name = tensor("op_1920_transpose_y_0"), val = tensor(false)]; + tensor transpose_77 = transpose(perm = var_1911, x = var_1910_cast_fp16)[name = tensor("transpose_77")]; + tensor var_1920_cast_fp16 = matmul(transpose_x = var_1920_transpose_x_0, transpose_y = var_1920_transpose_y_0, x = var_1918_cast_fp16, y = transpose_77)[name = tensor("op_1920_cast_fp16")]; + tensor var_1921 = const()[name = tensor("op_1921"), val = tensor([0, 2, 1, 3])]; + tensor concat_16 = const()[name = tensor("concat_16"), val = tensor([1, 1500, 1280])]; + tensor transpose_75 = transpose(perm = var_1921, x = var_1920_cast_fp16)[name = tensor("transpose_75")]; + tensor x_203_cast_fp16 = reshape(shape = concat_16, x = transpose_75)[name = tensor("x_203_cast_fp16")]; + tensor var_1926_to_fp16 = const()[name = tensor("op_1926_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(654155136)))]; + tensor var_1927_to_fp16 = const()[name = tensor("op_1927_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657432000)))]; + tensor linear_99_cast_fp16 = linear(bias = var_1927_to_fp16, weight = var_1926_to_fp16, x = x_203_cast_fp16)[name = tensor("linear_99_cast_fp16")]; + tensor x_205_cast_fp16 = add(x = x_199_cast_fp16, y = linear_99_cast_fp16)[name = tensor("x_205_cast_fp16")]; + tensor var_1934_axes_0 = const()[name = tensor("op_1934_axes_0"), val = tensor([-1])]; + tensor blocks_16_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_16_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657434624)))]; + tensor blocks_16_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_16_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657437248)))]; + tensor var_1934_cast_fp16 = layer_norm(axes = var_1934_axes_0, beta = blocks_16_mlp_ln_bias_to_fp16, epsilon = var_1859_to_fp16, gamma = blocks_16_mlp_ln_weight_to_fp16, x = x_205_cast_fp16)[name = tensor("op_1934_cast_fp16")]; + tensor var_1943_to_fp16 = const()[name = tensor("op_1943_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657439872)))]; + tensor var_1944_to_fp16 = const()[name = tensor("op_1944_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(670547136)))]; + tensor linear_100_cast_fp16 = linear(bias = var_1944_to_fp16, weight = var_1943_to_fp16, x = var_1934_cast_fp16)[name = tensor("linear_100_cast_fp16")]; + tensor x_209_mode_0 = const()[name = tensor("x_209_mode_0"), val = tensor("EXACT")]; + tensor x_209_cast_fp16 = gelu(mode = x_209_mode_0, x = linear_100_cast_fp16)[name = tensor("x_209_cast_fp16")]; + tensor var_1949_to_fp16 = const()[name = tensor("op_1949_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(670557440)))]; + tensor var_1950_to_fp16 = const()[name = tensor("op_1950_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(683664704)))]; + tensor linear_101_cast_fp16 = linear(bias = var_1950_to_fp16, weight = var_1949_to_fp16, x = x_209_cast_fp16)[name = tensor("linear_101_cast_fp16")]; + tensor x_211_cast_fp16 = add(x = x_205_cast_fp16, y = linear_101_cast_fp16)[name = tensor("x_211_cast_fp16")]; + tensor var_1960 = const()[name = tensor("op_1960"), val = tensor(-1)]; + tensor var_1977_axes_0 = const()[name = tensor("op_1977_axes_0"), val = tensor([-1])]; + tensor blocks_17_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_17_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(683667328)))]; + tensor blocks_17_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_17_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(683669952)))]; + tensor var_1967_to_fp16 = const()[name = tensor("op_1967_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1977_cast_fp16 = layer_norm(axes = var_1977_axes_0, beta = blocks_17_attn_ln_bias_to_fp16, epsilon = var_1967_to_fp16, gamma = blocks_17_attn_ln_weight_to_fp16, x = x_211_cast_fp16)[name = tensor("op_1977_cast_fp16")]; + tensor var_1988_to_fp16 = const()[name = tensor("op_1988_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(683672576)))]; + tensor var_1989_to_fp16 = const()[name = tensor("op_1989_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(686949440)))]; + tensor linear_102_cast_fp16 = linear(bias = var_1989_to_fp16, weight = var_1988_to_fp16, x = var_1977_cast_fp16)[name = tensor("linear_102_cast_fp16")]; + tensor var_1992_to_fp16 = const()[name = tensor("op_1992_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(686952064)))]; + tensor linear_103_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1992_to_fp16, x = var_1977_cast_fp16)[name = tensor("linear_103_cast_fp16")]; + tensor var_1996_to_fp16 = const()[name = tensor("op_1996_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(690228928)))]; + tensor var_1997_to_fp16 = const()[name = tensor("op_1997_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(693505792)))]; + tensor linear_104_cast_fp16 = linear(bias = var_1997_to_fp16, weight = var_1996_to_fp16, x = var_1977_cast_fp16)[name = tensor("linear_104_cast_fp16")]; + tensor var_2005 = const()[name = tensor("op_2005"), val = tensor([1, 1500, 20, -1])]; + tensor var_2006_cast_fp16 = reshape(shape = var_2005, x = linear_102_cast_fp16)[name = tensor("op_2006_cast_fp16")]; + tensor var_2007 = const()[name = tensor("op_2007"), val = tensor([0, 2, 1, 3])]; + tensor var_2011 = const()[name = tensor("op_2011"), val = tensor([1, 1500, 20, -1])]; + tensor var_2012_cast_fp16 = reshape(shape = var_2011, x = linear_103_cast_fp16)[name = tensor("op_2012_cast_fp16")]; + tensor var_2013 = const()[name = tensor("op_2013"), val = tensor([0, 2, 1, 3])]; + tensor var_2017 = const()[name = tensor("op_2017"), val = tensor([1, 1500, 20, -1])]; + tensor var_2018_cast_fp16 = reshape(shape = var_2017, x = linear_104_cast_fp16)[name = tensor("op_2018_cast_fp16")]; + tensor var_2019 = const()[name = tensor("op_2019"), val = tensor([0, 2, 1, 3])]; + tensor scale_35_to_fp16 = const()[name = tensor("scale_35_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_74 = transpose(perm = var_2007, x = var_2006_cast_fp16)[name = tensor("transpose_74")]; + tensor var_2021_cast_fp16 = mul(x = transpose_74, y = scale_35_to_fp16)[name = tensor("op_2021_cast_fp16")]; + tensor transpose_73 = transpose(perm = var_2013, x = var_2012_cast_fp16)[name = tensor("transpose_73")]; + tensor var_2022_cast_fp16 = mul(x = transpose_73, y = scale_35_to_fp16)[name = tensor("op_2022_cast_fp16")]; + tensor var_2023_perm_0 = const()[name = tensor("op_2023_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_71 = transpose(perm = var_2023_perm_0, x = var_2022_cast_fp16)[name = tensor("transpose_71")]; + tensor qk_35_cast_fp16 = matmul(transpose_x = qk_35_transpose_x_0, transpose_y = qk_35_transpose_y_0, x = var_2021_cast_fp16, y = transpose_71)[name = tensor("qk_35_cast_fp16")]; + tensor var_2026_cast_fp16 = softmax(axis = var_1960, x = qk_35_cast_fp16)[name = tensor("op_2026_cast_fp16")]; + tensor var_2028_transpose_x_0 = const()[name = tensor("op_2028_transpose_x_0"), val = tensor(false)]; + tensor var_2028_transpose_y_0 = const()[name = tensor("op_2028_transpose_y_0"), val = tensor(false)]; + tensor transpose_72 = transpose(perm = var_2019, x = var_2018_cast_fp16)[name = tensor("transpose_72")]; + tensor var_2028_cast_fp16 = matmul(transpose_x = var_2028_transpose_x_0, transpose_y = var_2028_transpose_y_0, x = var_2026_cast_fp16, y = transpose_72)[name = tensor("op_2028_cast_fp16")]; + tensor var_2029 = const()[name = tensor("op_2029"), val = tensor([0, 2, 1, 3])]; + tensor concat_17 = const()[name = tensor("concat_17"), val = tensor([1, 1500, 1280])]; + tensor transpose_70 = transpose(perm = var_2029, x = var_2028_cast_fp16)[name = tensor("transpose_70")]; + tensor x_215_cast_fp16 = reshape(shape = concat_17, x = transpose_70)[name = tensor("x_215_cast_fp16")]; + tensor var_2034_to_fp16 = const()[name = tensor("op_2034_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(693508416)))]; + tensor var_2035_to_fp16 = const()[name = tensor("op_2035_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696785280)))]; + tensor linear_105_cast_fp16 = linear(bias = var_2035_to_fp16, weight = var_2034_to_fp16, x = x_215_cast_fp16)[name = tensor("linear_105_cast_fp16")]; + tensor x_217_cast_fp16 = add(x = x_211_cast_fp16, y = linear_105_cast_fp16)[name = tensor("x_217_cast_fp16")]; + tensor var_2042_axes_0 = const()[name = tensor("op_2042_axes_0"), val = tensor([-1])]; + tensor blocks_17_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_17_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696787904)))]; + tensor blocks_17_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_17_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696790528)))]; + tensor var_2042_cast_fp16 = layer_norm(axes = var_2042_axes_0, beta = blocks_17_mlp_ln_bias_to_fp16, epsilon = var_1967_to_fp16, gamma = blocks_17_mlp_ln_weight_to_fp16, x = x_217_cast_fp16)[name = tensor("op_2042_cast_fp16")]; + tensor var_2051_to_fp16 = const()[name = tensor("op_2051_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696793152)))]; + tensor var_2052_to_fp16 = const()[name = tensor("op_2052_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(709900416)))]; + tensor linear_106_cast_fp16 = linear(bias = var_2052_to_fp16, weight = var_2051_to_fp16, x = var_2042_cast_fp16)[name = tensor("linear_106_cast_fp16")]; + tensor x_221_mode_0 = const()[name = tensor("x_221_mode_0"), val = tensor("EXACT")]; + tensor x_221_cast_fp16 = gelu(mode = x_221_mode_0, x = linear_106_cast_fp16)[name = tensor("x_221_cast_fp16")]; + tensor var_2057_to_fp16 = const()[name = tensor("op_2057_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(709910720)))]; + tensor var_2058_to_fp16 = const()[name = tensor("op_2058_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(723017984)))]; + tensor linear_107_cast_fp16 = linear(bias = var_2058_to_fp16, weight = var_2057_to_fp16, x = x_221_cast_fp16)[name = tensor("linear_107_cast_fp16")]; + tensor x_223_cast_fp16 = add(x = x_217_cast_fp16, y = linear_107_cast_fp16)[name = tensor("x_223_cast_fp16")]; + tensor var_2068 = const()[name = tensor("op_2068"), val = tensor(-1)]; + tensor var_2085_axes_0 = const()[name = tensor("op_2085_axes_0"), val = tensor([-1])]; + tensor blocks_18_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_18_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(723020608)))]; + tensor blocks_18_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_18_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(723023232)))]; + tensor var_2075_to_fp16 = const()[name = tensor("op_2075_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2085_cast_fp16 = layer_norm(axes = var_2085_axes_0, beta = blocks_18_attn_ln_bias_to_fp16, epsilon = var_2075_to_fp16, gamma = blocks_18_attn_ln_weight_to_fp16, x = x_223_cast_fp16)[name = tensor("op_2085_cast_fp16")]; + tensor var_2096_to_fp16 = const()[name = tensor("op_2096_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(723025856)))]; + tensor var_2097_to_fp16 = const()[name = tensor("op_2097_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(726302720)))]; + tensor linear_108_cast_fp16 = linear(bias = var_2097_to_fp16, weight = var_2096_to_fp16, x = var_2085_cast_fp16)[name = tensor("linear_108_cast_fp16")]; + tensor var_2100_to_fp16 = const()[name = tensor("op_2100_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(726305344)))]; + tensor linear_109_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2100_to_fp16, x = var_2085_cast_fp16)[name = tensor("linear_109_cast_fp16")]; + tensor var_2104_to_fp16 = const()[name = tensor("op_2104_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(729582208)))]; + tensor var_2105_to_fp16 = const()[name = tensor("op_2105_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(732859072)))]; + tensor linear_110_cast_fp16 = linear(bias = var_2105_to_fp16, weight = var_2104_to_fp16, x = var_2085_cast_fp16)[name = tensor("linear_110_cast_fp16")]; + tensor var_2113 = const()[name = tensor("op_2113"), val = tensor([1, 1500, 20, -1])]; + tensor var_2114_cast_fp16 = reshape(shape = var_2113, x = linear_108_cast_fp16)[name = tensor("op_2114_cast_fp16")]; + tensor var_2115 = const()[name = tensor("op_2115"), val = tensor([0, 2, 1, 3])]; + tensor var_2119 = const()[name = tensor("op_2119"), val = tensor([1, 1500, 20, -1])]; + tensor var_2120_cast_fp16 = reshape(shape = var_2119, x = linear_109_cast_fp16)[name = tensor("op_2120_cast_fp16")]; + tensor var_2121 = const()[name = tensor("op_2121"), val = tensor([0, 2, 1, 3])]; + tensor var_2125 = const()[name = tensor("op_2125"), val = tensor([1, 1500, 20, -1])]; + tensor var_2126_cast_fp16 = reshape(shape = var_2125, x = linear_110_cast_fp16)[name = tensor("op_2126_cast_fp16")]; + tensor var_2127 = const()[name = tensor("op_2127"), val = tensor([0, 2, 1, 3])]; + tensor scale_37_to_fp16 = const()[name = tensor("scale_37_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_69 = transpose(perm = var_2115, x = var_2114_cast_fp16)[name = tensor("transpose_69")]; + tensor var_2129_cast_fp16 = mul(x = transpose_69, y = scale_37_to_fp16)[name = tensor("op_2129_cast_fp16")]; + tensor transpose_68 = transpose(perm = var_2121, x = var_2120_cast_fp16)[name = tensor("transpose_68")]; + tensor var_2130_cast_fp16 = mul(x = transpose_68, y = scale_37_to_fp16)[name = tensor("op_2130_cast_fp16")]; + tensor var_2131_perm_0 = const()[name = tensor("op_2131_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_66 = transpose(perm = var_2131_perm_0, x = var_2130_cast_fp16)[name = tensor("transpose_66")]; + tensor qk_37_cast_fp16 = matmul(transpose_x = qk_37_transpose_x_0, transpose_y = qk_37_transpose_y_0, x = var_2129_cast_fp16, y = transpose_66)[name = tensor("qk_37_cast_fp16")]; + tensor var_2134_cast_fp16 = softmax(axis = var_2068, x = qk_37_cast_fp16)[name = tensor("op_2134_cast_fp16")]; + tensor var_2136_transpose_x_0 = const()[name = tensor("op_2136_transpose_x_0"), val = tensor(false)]; + tensor var_2136_transpose_y_0 = const()[name = tensor("op_2136_transpose_y_0"), val = tensor(false)]; + tensor transpose_67 = transpose(perm = var_2127, x = var_2126_cast_fp16)[name = tensor("transpose_67")]; + tensor var_2136_cast_fp16 = matmul(transpose_x = var_2136_transpose_x_0, transpose_y = var_2136_transpose_y_0, x = var_2134_cast_fp16, y = transpose_67)[name = tensor("op_2136_cast_fp16")]; + tensor var_2137 = const()[name = tensor("op_2137"), val = tensor([0, 2, 1, 3])]; + tensor concat_18 = const()[name = tensor("concat_18"), val = tensor([1, 1500, 1280])]; + tensor transpose_65 = transpose(perm = var_2137, x = var_2136_cast_fp16)[name = tensor("transpose_65")]; + tensor x_227_cast_fp16 = reshape(shape = concat_18, x = transpose_65)[name = tensor("x_227_cast_fp16")]; + tensor var_2142_to_fp16 = const()[name = tensor("op_2142_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(732861696)))]; + tensor var_2143_to_fp16 = const()[name = tensor("op_2143_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(736138560)))]; + tensor linear_111_cast_fp16 = linear(bias = var_2143_to_fp16, weight = var_2142_to_fp16, x = x_227_cast_fp16)[name = tensor("linear_111_cast_fp16")]; + tensor x_229_cast_fp16 = add(x = x_223_cast_fp16, y = linear_111_cast_fp16)[name = tensor("x_229_cast_fp16")]; + tensor var_2150_axes_0 = const()[name = tensor("op_2150_axes_0"), val = tensor([-1])]; + tensor blocks_18_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_18_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(736141184)))]; + tensor blocks_18_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_18_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(736143808)))]; + tensor var_2150_cast_fp16 = layer_norm(axes = var_2150_axes_0, beta = blocks_18_mlp_ln_bias_to_fp16, epsilon = var_2075_to_fp16, gamma = blocks_18_mlp_ln_weight_to_fp16, x = x_229_cast_fp16)[name = tensor("op_2150_cast_fp16")]; + tensor var_2159_to_fp16 = const()[name = tensor("op_2159_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(736146432)))]; + tensor var_2160_to_fp16 = const()[name = tensor("op_2160_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(749253696)))]; + tensor linear_112_cast_fp16 = linear(bias = var_2160_to_fp16, weight = var_2159_to_fp16, x = var_2150_cast_fp16)[name = tensor("linear_112_cast_fp16")]; + tensor x_233_mode_0 = const()[name = tensor("x_233_mode_0"), val = tensor("EXACT")]; + tensor x_233_cast_fp16 = gelu(mode = x_233_mode_0, x = linear_112_cast_fp16)[name = tensor("x_233_cast_fp16")]; + tensor var_2165_to_fp16 = const()[name = tensor("op_2165_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(749264000)))]; + tensor var_2166_to_fp16 = const()[name = tensor("op_2166_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762371264)))]; + tensor linear_113_cast_fp16 = linear(bias = var_2166_to_fp16, weight = var_2165_to_fp16, x = x_233_cast_fp16)[name = tensor("linear_113_cast_fp16")]; + tensor x_235_cast_fp16 = add(x = x_229_cast_fp16, y = linear_113_cast_fp16)[name = tensor("x_235_cast_fp16")]; + tensor var_2176 = const()[name = tensor("op_2176"), val = tensor(-1)]; + tensor var_2193_axes_0 = const()[name = tensor("op_2193_axes_0"), val = tensor([-1])]; + tensor blocks_19_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_19_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762373888)))]; + tensor blocks_19_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_19_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762376512)))]; + tensor var_2183_to_fp16 = const()[name = tensor("op_2183_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2193_cast_fp16 = layer_norm(axes = var_2193_axes_0, beta = blocks_19_attn_ln_bias_to_fp16, epsilon = var_2183_to_fp16, gamma = blocks_19_attn_ln_weight_to_fp16, x = x_235_cast_fp16)[name = tensor("op_2193_cast_fp16")]; + tensor var_2204_to_fp16 = const()[name = tensor("op_2204_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762379136)))]; + tensor var_2205_to_fp16 = const()[name = tensor("op_2205_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(765656000)))]; + tensor linear_114_cast_fp16 = linear(bias = var_2205_to_fp16, weight = var_2204_to_fp16, x = var_2193_cast_fp16)[name = tensor("linear_114_cast_fp16")]; + tensor var_2208_to_fp16 = const()[name = tensor("op_2208_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(765658624)))]; + tensor linear_115_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2208_to_fp16, x = var_2193_cast_fp16)[name = tensor("linear_115_cast_fp16")]; + tensor var_2212_to_fp16 = const()[name = tensor("op_2212_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(768935488)))]; + tensor var_2213_to_fp16 = const()[name = tensor("op_2213_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(772212352)))]; + tensor linear_116_cast_fp16 = linear(bias = var_2213_to_fp16, weight = var_2212_to_fp16, x = var_2193_cast_fp16)[name = tensor("linear_116_cast_fp16")]; + tensor var_2221 = const()[name = tensor("op_2221"), val = tensor([1, 1500, 20, -1])]; + tensor var_2222_cast_fp16 = reshape(shape = var_2221, x = linear_114_cast_fp16)[name = tensor("op_2222_cast_fp16")]; + tensor var_2223 = const()[name = tensor("op_2223"), val = tensor([0, 2, 1, 3])]; + tensor var_2227 = const()[name = tensor("op_2227"), val = tensor([1, 1500, 20, -1])]; + tensor var_2228_cast_fp16 = reshape(shape = var_2227, x = linear_115_cast_fp16)[name = tensor("op_2228_cast_fp16")]; + tensor var_2229 = const()[name = tensor("op_2229"), val = tensor([0, 2, 1, 3])]; + tensor var_2233 = const()[name = tensor("op_2233"), val = tensor([1, 1500, 20, -1])]; + tensor var_2234_cast_fp16 = reshape(shape = var_2233, x = linear_116_cast_fp16)[name = tensor("op_2234_cast_fp16")]; + tensor var_2235 = const()[name = tensor("op_2235"), val = tensor([0, 2, 1, 3])]; + tensor scale_39_to_fp16 = const()[name = tensor("scale_39_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_64 = transpose(perm = var_2223, x = var_2222_cast_fp16)[name = tensor("transpose_64")]; + tensor var_2237_cast_fp16 = mul(x = transpose_64, y = scale_39_to_fp16)[name = tensor("op_2237_cast_fp16")]; + tensor transpose_63 = transpose(perm = var_2229, x = var_2228_cast_fp16)[name = tensor("transpose_63")]; + tensor var_2238_cast_fp16 = mul(x = transpose_63, y = scale_39_to_fp16)[name = tensor("op_2238_cast_fp16")]; + tensor var_2239_perm_0 = const()[name = tensor("op_2239_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_61 = transpose(perm = var_2239_perm_0, x = var_2238_cast_fp16)[name = tensor("transpose_61")]; + tensor qk_39_cast_fp16 = matmul(transpose_x = qk_39_transpose_x_0, transpose_y = qk_39_transpose_y_0, x = var_2237_cast_fp16, y = transpose_61)[name = tensor("qk_39_cast_fp16")]; + tensor var_2242_cast_fp16 = softmax(axis = var_2176, x = qk_39_cast_fp16)[name = tensor("op_2242_cast_fp16")]; + tensor var_2244_transpose_x_0 = const()[name = tensor("op_2244_transpose_x_0"), val = tensor(false)]; + tensor var_2244_transpose_y_0 = const()[name = tensor("op_2244_transpose_y_0"), val = tensor(false)]; + tensor transpose_62 = transpose(perm = var_2235, x = var_2234_cast_fp16)[name = tensor("transpose_62")]; + tensor var_2244_cast_fp16 = matmul(transpose_x = var_2244_transpose_x_0, transpose_y = var_2244_transpose_y_0, x = var_2242_cast_fp16, y = transpose_62)[name = tensor("op_2244_cast_fp16")]; + tensor var_2245 = const()[name = tensor("op_2245"), val = tensor([0, 2, 1, 3])]; + tensor concat_19 = const()[name = tensor("concat_19"), val = tensor([1, 1500, 1280])]; + tensor transpose_60 = transpose(perm = var_2245, x = var_2244_cast_fp16)[name = tensor("transpose_60")]; + tensor x_239_cast_fp16 = reshape(shape = concat_19, x = transpose_60)[name = tensor("x_239_cast_fp16")]; + tensor var_2250_to_fp16 = const()[name = tensor("op_2250_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(772214976)))]; + tensor var_2251_to_fp16 = const()[name = tensor("op_2251_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775491840)))]; + tensor linear_117_cast_fp16 = linear(bias = var_2251_to_fp16, weight = var_2250_to_fp16, x = x_239_cast_fp16)[name = tensor("linear_117_cast_fp16")]; + tensor x_241_cast_fp16 = add(x = x_235_cast_fp16, y = linear_117_cast_fp16)[name = tensor("x_241_cast_fp16")]; + tensor var_2258_axes_0 = const()[name = tensor("op_2258_axes_0"), val = tensor([-1])]; + tensor blocks_19_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_19_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775494464)))]; + tensor blocks_19_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_19_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775497088)))]; + tensor var_2258_cast_fp16 = layer_norm(axes = var_2258_axes_0, beta = blocks_19_mlp_ln_bias_to_fp16, epsilon = var_2183_to_fp16, gamma = blocks_19_mlp_ln_weight_to_fp16, x = x_241_cast_fp16)[name = tensor("op_2258_cast_fp16")]; + tensor var_2267_to_fp16 = const()[name = tensor("op_2267_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775499712)))]; + tensor var_2268_to_fp16 = const()[name = tensor("op_2268_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(788606976)))]; + tensor linear_118_cast_fp16 = linear(bias = var_2268_to_fp16, weight = var_2267_to_fp16, x = var_2258_cast_fp16)[name = tensor("linear_118_cast_fp16")]; + tensor x_245_mode_0 = const()[name = tensor("x_245_mode_0"), val = tensor("EXACT")]; + tensor x_245_cast_fp16 = gelu(mode = x_245_mode_0, x = linear_118_cast_fp16)[name = tensor("x_245_cast_fp16")]; + tensor var_2273_to_fp16 = const()[name = tensor("op_2273_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(788617280)))]; + tensor var_2274_to_fp16 = const()[name = tensor("op_2274_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(801724544)))]; + tensor linear_119_cast_fp16 = linear(bias = var_2274_to_fp16, weight = var_2273_to_fp16, x = x_245_cast_fp16)[name = tensor("linear_119_cast_fp16")]; + tensor x_247_cast_fp16 = add(x = x_241_cast_fp16, y = linear_119_cast_fp16)[name = tensor("x_247_cast_fp16")]; + tensor var_2284 = const()[name = tensor("op_2284"), val = tensor(-1)]; + tensor var_2301_axes_0 = const()[name = tensor("op_2301_axes_0"), val = tensor([-1])]; + tensor blocks_20_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_20_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(801727168)))]; + tensor blocks_20_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_20_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(801729792)))]; + tensor var_2291_to_fp16 = const()[name = tensor("op_2291_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2301_cast_fp16 = layer_norm(axes = var_2301_axes_0, beta = blocks_20_attn_ln_bias_to_fp16, epsilon = var_2291_to_fp16, gamma = blocks_20_attn_ln_weight_to_fp16, x = x_247_cast_fp16)[name = tensor("op_2301_cast_fp16")]; + tensor var_2312_to_fp16 = const()[name = tensor("op_2312_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(801732416)))]; + tensor var_2313_to_fp16 = const()[name = tensor("op_2313_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(805009280)))]; + tensor linear_120_cast_fp16 = linear(bias = var_2313_to_fp16, weight = var_2312_to_fp16, x = var_2301_cast_fp16)[name = tensor("linear_120_cast_fp16")]; + tensor var_2316_to_fp16 = const()[name = tensor("op_2316_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(805011904)))]; + tensor linear_121_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2316_to_fp16, x = var_2301_cast_fp16)[name = tensor("linear_121_cast_fp16")]; + tensor var_2320_to_fp16 = const()[name = tensor("op_2320_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(808288768)))]; + tensor var_2321_to_fp16 = const()[name = tensor("op_2321_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(811565632)))]; + tensor linear_122_cast_fp16 = linear(bias = var_2321_to_fp16, weight = var_2320_to_fp16, x = var_2301_cast_fp16)[name = tensor("linear_122_cast_fp16")]; + tensor var_2329 = const()[name = tensor("op_2329"), val = tensor([1, 1500, 20, -1])]; + tensor var_2330_cast_fp16 = reshape(shape = var_2329, x = linear_120_cast_fp16)[name = tensor("op_2330_cast_fp16")]; + tensor var_2331 = const()[name = tensor("op_2331"), val = tensor([0, 2, 1, 3])]; + tensor var_2335 = const()[name = tensor("op_2335"), val = tensor([1, 1500, 20, -1])]; + tensor var_2336_cast_fp16 = reshape(shape = var_2335, x = linear_121_cast_fp16)[name = tensor("op_2336_cast_fp16")]; + tensor var_2337 = const()[name = tensor("op_2337"), val = tensor([0, 2, 1, 3])]; + tensor var_2341 = const()[name = tensor("op_2341"), val = tensor([1, 1500, 20, -1])]; + tensor var_2342_cast_fp16 = reshape(shape = var_2341, x = linear_122_cast_fp16)[name = tensor("op_2342_cast_fp16")]; + tensor var_2343 = const()[name = tensor("op_2343"), val = tensor([0, 2, 1, 3])]; + tensor scale_41_to_fp16 = const()[name = tensor("scale_41_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_59 = transpose(perm = var_2331, x = var_2330_cast_fp16)[name = tensor("transpose_59")]; + tensor var_2345_cast_fp16 = mul(x = transpose_59, y = scale_41_to_fp16)[name = tensor("op_2345_cast_fp16")]; + tensor transpose_58 = transpose(perm = var_2337, x = var_2336_cast_fp16)[name = tensor("transpose_58")]; + tensor var_2346_cast_fp16 = mul(x = transpose_58, y = scale_41_to_fp16)[name = tensor("op_2346_cast_fp16")]; + tensor var_2347_perm_0 = const()[name = tensor("op_2347_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_56 = transpose(perm = var_2347_perm_0, x = var_2346_cast_fp16)[name = tensor("transpose_56")]; + tensor qk_41_cast_fp16 = matmul(transpose_x = qk_41_transpose_x_0, transpose_y = qk_41_transpose_y_0, x = var_2345_cast_fp16, y = transpose_56)[name = tensor("qk_41_cast_fp16")]; + tensor var_2350_cast_fp16 = softmax(axis = var_2284, x = qk_41_cast_fp16)[name = tensor("op_2350_cast_fp16")]; + tensor var_2352_transpose_x_0 = const()[name = tensor("op_2352_transpose_x_0"), val = tensor(false)]; + tensor var_2352_transpose_y_0 = const()[name = tensor("op_2352_transpose_y_0"), val = tensor(false)]; + tensor transpose_57 = transpose(perm = var_2343, x = var_2342_cast_fp16)[name = tensor("transpose_57")]; + tensor var_2352_cast_fp16 = matmul(transpose_x = var_2352_transpose_x_0, transpose_y = var_2352_transpose_y_0, x = var_2350_cast_fp16, y = transpose_57)[name = tensor("op_2352_cast_fp16")]; + tensor var_2353 = const()[name = tensor("op_2353"), val = tensor([0, 2, 1, 3])]; + tensor concat_20 = const()[name = tensor("concat_20"), val = tensor([1, 1500, 1280])]; + tensor transpose_55 = transpose(perm = var_2353, x = var_2352_cast_fp16)[name = tensor("transpose_55")]; + tensor x_251_cast_fp16 = reshape(shape = concat_20, x = transpose_55)[name = tensor("x_251_cast_fp16")]; + tensor var_2358_to_fp16 = const()[name = tensor("op_2358_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(811568256)))]; + tensor var_2359_to_fp16 = const()[name = tensor("op_2359_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814845120)))]; + tensor linear_123_cast_fp16 = linear(bias = var_2359_to_fp16, weight = var_2358_to_fp16, x = x_251_cast_fp16)[name = tensor("linear_123_cast_fp16")]; + tensor x_253_cast_fp16 = add(x = x_247_cast_fp16, y = linear_123_cast_fp16)[name = tensor("x_253_cast_fp16")]; + tensor var_2366_axes_0 = const()[name = tensor("op_2366_axes_0"), val = tensor([-1])]; + tensor blocks_20_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_20_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814847744)))]; + tensor blocks_20_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_20_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814850368)))]; + tensor var_2366_cast_fp16 = layer_norm(axes = var_2366_axes_0, beta = blocks_20_mlp_ln_bias_to_fp16, epsilon = var_2291_to_fp16, gamma = blocks_20_mlp_ln_weight_to_fp16, x = x_253_cast_fp16)[name = tensor("op_2366_cast_fp16")]; + tensor var_2375_to_fp16 = const()[name = tensor("op_2375_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814852992)))]; + tensor var_2376_to_fp16 = const()[name = tensor("op_2376_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(827960256)))]; + tensor linear_124_cast_fp16 = linear(bias = var_2376_to_fp16, weight = var_2375_to_fp16, x = var_2366_cast_fp16)[name = tensor("linear_124_cast_fp16")]; + tensor x_257_mode_0 = const()[name = tensor("x_257_mode_0"), val = tensor("EXACT")]; + tensor x_257_cast_fp16 = gelu(mode = x_257_mode_0, x = linear_124_cast_fp16)[name = tensor("x_257_cast_fp16")]; + tensor var_2381_to_fp16 = const()[name = tensor("op_2381_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(827970560)))]; + tensor var_2382_to_fp16 = const()[name = tensor("op_2382_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(841077824)))]; + tensor linear_125_cast_fp16 = linear(bias = var_2382_to_fp16, weight = var_2381_to_fp16, x = x_257_cast_fp16)[name = tensor("linear_125_cast_fp16")]; + tensor x_259_cast_fp16 = add(x = x_253_cast_fp16, y = linear_125_cast_fp16)[name = tensor("x_259_cast_fp16")]; + tensor var_2392 = const()[name = tensor("op_2392"), val = tensor(-1)]; + tensor var_2409_axes_0 = const()[name = tensor("op_2409_axes_0"), val = tensor([-1])]; + tensor blocks_21_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_21_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(841080448)))]; + tensor blocks_21_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_21_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(841083072)))]; + tensor var_2399_to_fp16 = const()[name = tensor("op_2399_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2409_cast_fp16 = layer_norm(axes = var_2409_axes_0, beta = blocks_21_attn_ln_bias_to_fp16, epsilon = var_2399_to_fp16, gamma = blocks_21_attn_ln_weight_to_fp16, x = x_259_cast_fp16)[name = tensor("op_2409_cast_fp16")]; + tensor var_2420_to_fp16 = const()[name = tensor("op_2420_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(841085696)))]; + tensor var_2421_to_fp16 = const()[name = tensor("op_2421_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(844362560)))]; + tensor linear_126_cast_fp16 = linear(bias = var_2421_to_fp16, weight = var_2420_to_fp16, x = var_2409_cast_fp16)[name = tensor("linear_126_cast_fp16")]; + tensor var_2424_to_fp16 = const()[name = tensor("op_2424_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(844365184)))]; + tensor linear_127_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2424_to_fp16, x = var_2409_cast_fp16)[name = tensor("linear_127_cast_fp16")]; + tensor var_2428_to_fp16 = const()[name = tensor("op_2428_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(847642048)))]; + tensor var_2429_to_fp16 = const()[name = tensor("op_2429_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(850918912)))]; + tensor linear_128_cast_fp16 = linear(bias = var_2429_to_fp16, weight = var_2428_to_fp16, x = var_2409_cast_fp16)[name = tensor("linear_128_cast_fp16")]; + tensor var_2437 = const()[name = tensor("op_2437"), val = tensor([1, 1500, 20, -1])]; + tensor var_2438_cast_fp16 = reshape(shape = var_2437, x = linear_126_cast_fp16)[name = tensor("op_2438_cast_fp16")]; + tensor var_2439 = const()[name = tensor("op_2439"), val = tensor([0, 2, 1, 3])]; + tensor var_2443 = const()[name = tensor("op_2443"), val = tensor([1, 1500, 20, -1])]; + tensor var_2444_cast_fp16 = reshape(shape = var_2443, x = linear_127_cast_fp16)[name = tensor("op_2444_cast_fp16")]; + tensor var_2445 = const()[name = tensor("op_2445"), val = tensor([0, 2, 1, 3])]; + tensor var_2449 = const()[name = tensor("op_2449"), val = tensor([1, 1500, 20, -1])]; + tensor var_2450_cast_fp16 = reshape(shape = var_2449, x = linear_128_cast_fp16)[name = tensor("op_2450_cast_fp16")]; + tensor var_2451 = const()[name = tensor("op_2451"), val = tensor([0, 2, 1, 3])]; + tensor scale_43_to_fp16 = const()[name = tensor("scale_43_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_54 = transpose(perm = var_2439, x = var_2438_cast_fp16)[name = tensor("transpose_54")]; + tensor var_2453_cast_fp16 = mul(x = transpose_54, y = scale_43_to_fp16)[name = tensor("op_2453_cast_fp16")]; + tensor transpose_53 = transpose(perm = var_2445, x = var_2444_cast_fp16)[name = tensor("transpose_53")]; + tensor var_2454_cast_fp16 = mul(x = transpose_53, y = scale_43_to_fp16)[name = tensor("op_2454_cast_fp16")]; + tensor var_2455_perm_0 = const()[name = tensor("op_2455_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_51 = transpose(perm = var_2455_perm_0, x = var_2454_cast_fp16)[name = tensor("transpose_51")]; + tensor qk_43_cast_fp16 = matmul(transpose_x = qk_43_transpose_x_0, transpose_y = qk_43_transpose_y_0, x = var_2453_cast_fp16, y = transpose_51)[name = tensor("qk_43_cast_fp16")]; + tensor var_2458_cast_fp16 = softmax(axis = var_2392, x = qk_43_cast_fp16)[name = tensor("op_2458_cast_fp16")]; + tensor var_2460_transpose_x_0 = const()[name = tensor("op_2460_transpose_x_0"), val = tensor(false)]; + tensor var_2460_transpose_y_0 = const()[name = tensor("op_2460_transpose_y_0"), val = tensor(false)]; + tensor transpose_52 = transpose(perm = var_2451, x = var_2450_cast_fp16)[name = tensor("transpose_52")]; + tensor var_2460_cast_fp16 = matmul(transpose_x = var_2460_transpose_x_0, transpose_y = var_2460_transpose_y_0, x = var_2458_cast_fp16, y = transpose_52)[name = tensor("op_2460_cast_fp16")]; + tensor var_2461 = const()[name = tensor("op_2461"), val = tensor([0, 2, 1, 3])]; + tensor concat_21 = const()[name = tensor("concat_21"), val = tensor([1, 1500, 1280])]; + tensor transpose_50 = transpose(perm = var_2461, x = var_2460_cast_fp16)[name = tensor("transpose_50")]; + tensor x_263_cast_fp16 = reshape(shape = concat_21, x = transpose_50)[name = tensor("x_263_cast_fp16")]; + tensor var_2466_to_fp16 = const()[name = tensor("op_2466_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(850921536)))]; + tensor var_2467_to_fp16 = const()[name = tensor("op_2467_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(854198400)))]; + tensor linear_129_cast_fp16 = linear(bias = var_2467_to_fp16, weight = var_2466_to_fp16, x = x_263_cast_fp16)[name = tensor("linear_129_cast_fp16")]; + tensor x_265_cast_fp16 = add(x = x_259_cast_fp16, y = linear_129_cast_fp16)[name = tensor("x_265_cast_fp16")]; + tensor var_2474_axes_0 = const()[name = tensor("op_2474_axes_0"), val = tensor([-1])]; + tensor blocks_21_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_21_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(854201024)))]; + tensor blocks_21_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_21_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(854203648)))]; + tensor var_2474_cast_fp16 = layer_norm(axes = var_2474_axes_0, beta = blocks_21_mlp_ln_bias_to_fp16, epsilon = var_2399_to_fp16, gamma = blocks_21_mlp_ln_weight_to_fp16, x = x_265_cast_fp16)[name = tensor("op_2474_cast_fp16")]; + tensor var_2483_to_fp16 = const()[name = tensor("op_2483_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(854206272)))]; + tensor var_2484_to_fp16 = const()[name = tensor("op_2484_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(867313536)))]; + tensor linear_130_cast_fp16 = linear(bias = var_2484_to_fp16, weight = var_2483_to_fp16, x = var_2474_cast_fp16)[name = tensor("linear_130_cast_fp16")]; + tensor x_269_mode_0 = const()[name = tensor("x_269_mode_0"), val = tensor("EXACT")]; + tensor x_269_cast_fp16 = gelu(mode = x_269_mode_0, x = linear_130_cast_fp16)[name = tensor("x_269_cast_fp16")]; + tensor var_2489_to_fp16 = const()[name = tensor("op_2489_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(867323840)))]; + tensor var_2490_to_fp16 = const()[name = tensor("op_2490_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880431104)))]; + tensor linear_131_cast_fp16 = linear(bias = var_2490_to_fp16, weight = var_2489_to_fp16, x = x_269_cast_fp16)[name = tensor("linear_131_cast_fp16")]; + tensor x_271_cast_fp16 = add(x = x_265_cast_fp16, y = linear_131_cast_fp16)[name = tensor("x_271_cast_fp16")]; + tensor var_2500 = const()[name = tensor("op_2500"), val = tensor(-1)]; + tensor var_2517_axes_0 = const()[name = tensor("op_2517_axes_0"), val = tensor([-1])]; + tensor blocks_22_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_22_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880433728)))]; + tensor blocks_22_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_22_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880436352)))]; + tensor var_2507_to_fp16 = const()[name = tensor("op_2507_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2517_cast_fp16 = layer_norm(axes = var_2517_axes_0, beta = blocks_22_attn_ln_bias_to_fp16, epsilon = var_2507_to_fp16, gamma = blocks_22_attn_ln_weight_to_fp16, x = x_271_cast_fp16)[name = tensor("op_2517_cast_fp16")]; + tensor var_2528_to_fp16 = const()[name = tensor("op_2528_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880438976)))]; + tensor var_2529_to_fp16 = const()[name = tensor("op_2529_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(883715840)))]; + tensor linear_132_cast_fp16 = linear(bias = var_2529_to_fp16, weight = var_2528_to_fp16, x = var_2517_cast_fp16)[name = tensor("linear_132_cast_fp16")]; + tensor var_2532_to_fp16 = const()[name = tensor("op_2532_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(883718464)))]; + tensor linear_133_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2532_to_fp16, x = var_2517_cast_fp16)[name = tensor("linear_133_cast_fp16")]; + tensor var_2536_to_fp16 = const()[name = tensor("op_2536_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(886995328)))]; + tensor var_2537_to_fp16 = const()[name = tensor("op_2537_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(890272192)))]; + tensor linear_134_cast_fp16 = linear(bias = var_2537_to_fp16, weight = var_2536_to_fp16, x = var_2517_cast_fp16)[name = tensor("linear_134_cast_fp16")]; + tensor var_2545 = const()[name = tensor("op_2545"), val = tensor([1, 1500, 20, -1])]; + tensor var_2546_cast_fp16 = reshape(shape = var_2545, x = linear_132_cast_fp16)[name = tensor("op_2546_cast_fp16")]; + tensor var_2547 = const()[name = tensor("op_2547"), val = tensor([0, 2, 1, 3])]; + tensor var_2551 = const()[name = tensor("op_2551"), val = tensor([1, 1500, 20, -1])]; + tensor var_2552_cast_fp16 = reshape(shape = var_2551, x = linear_133_cast_fp16)[name = tensor("op_2552_cast_fp16")]; + tensor var_2553 = const()[name = tensor("op_2553"), val = tensor([0, 2, 1, 3])]; + tensor var_2557 = const()[name = tensor("op_2557"), val = tensor([1, 1500, 20, -1])]; + tensor var_2558_cast_fp16 = reshape(shape = var_2557, x = linear_134_cast_fp16)[name = tensor("op_2558_cast_fp16")]; + tensor var_2559 = const()[name = tensor("op_2559"), val = tensor([0, 2, 1, 3])]; + tensor scale_45_to_fp16 = const()[name = tensor("scale_45_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_49 = transpose(perm = var_2547, x = var_2546_cast_fp16)[name = tensor("transpose_49")]; + tensor var_2561_cast_fp16 = mul(x = transpose_49, y = scale_45_to_fp16)[name = tensor("op_2561_cast_fp16")]; + tensor transpose_48 = transpose(perm = var_2553, x = var_2552_cast_fp16)[name = tensor("transpose_48")]; + tensor var_2562_cast_fp16 = mul(x = transpose_48, y = scale_45_to_fp16)[name = tensor("op_2562_cast_fp16")]; + tensor var_2563_perm_0 = const()[name = tensor("op_2563_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_46 = transpose(perm = var_2563_perm_0, x = var_2562_cast_fp16)[name = tensor("transpose_46")]; + tensor qk_45_cast_fp16 = matmul(transpose_x = qk_45_transpose_x_0, transpose_y = qk_45_transpose_y_0, x = var_2561_cast_fp16, y = transpose_46)[name = tensor("qk_45_cast_fp16")]; + tensor var_2566_cast_fp16 = softmax(axis = var_2500, x = qk_45_cast_fp16)[name = tensor("op_2566_cast_fp16")]; + tensor var_2568_transpose_x_0 = const()[name = tensor("op_2568_transpose_x_0"), val = tensor(false)]; + tensor var_2568_transpose_y_0 = const()[name = tensor("op_2568_transpose_y_0"), val = tensor(false)]; + tensor transpose_47 = transpose(perm = var_2559, x = var_2558_cast_fp16)[name = tensor("transpose_47")]; + tensor var_2568_cast_fp16 = matmul(transpose_x = var_2568_transpose_x_0, transpose_y = var_2568_transpose_y_0, x = var_2566_cast_fp16, y = transpose_47)[name = tensor("op_2568_cast_fp16")]; + tensor var_2569 = const()[name = tensor("op_2569"), val = tensor([0, 2, 1, 3])]; + tensor concat_22 = const()[name = tensor("concat_22"), val = tensor([1, 1500, 1280])]; + tensor transpose_45 = transpose(perm = var_2569, x = var_2568_cast_fp16)[name = tensor("transpose_45")]; + tensor x_275_cast_fp16 = reshape(shape = concat_22, x = transpose_45)[name = tensor("x_275_cast_fp16")]; + tensor var_2574_to_fp16 = const()[name = tensor("op_2574_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(890274816)))]; + tensor var_2575_to_fp16 = const()[name = tensor("op_2575_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893551680)))]; + tensor linear_135_cast_fp16 = linear(bias = var_2575_to_fp16, weight = var_2574_to_fp16, x = x_275_cast_fp16)[name = tensor("linear_135_cast_fp16")]; + tensor x_277_cast_fp16 = add(x = x_271_cast_fp16, y = linear_135_cast_fp16)[name = tensor("x_277_cast_fp16")]; + tensor var_2582_axes_0 = const()[name = tensor("op_2582_axes_0"), val = tensor([-1])]; + tensor blocks_22_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_22_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893554304)))]; + tensor blocks_22_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_22_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893556928)))]; + tensor var_2582_cast_fp16 = layer_norm(axes = var_2582_axes_0, beta = blocks_22_mlp_ln_bias_to_fp16, epsilon = var_2507_to_fp16, gamma = blocks_22_mlp_ln_weight_to_fp16, x = x_277_cast_fp16)[name = tensor("op_2582_cast_fp16")]; + tensor var_2591_to_fp16 = const()[name = tensor("op_2591_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893559552)))]; + tensor var_2592_to_fp16 = const()[name = tensor("op_2592_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(906666816)))]; + tensor linear_136_cast_fp16 = linear(bias = var_2592_to_fp16, weight = var_2591_to_fp16, x = var_2582_cast_fp16)[name = tensor("linear_136_cast_fp16")]; + tensor x_281_mode_0 = const()[name = tensor("x_281_mode_0"), val = tensor("EXACT")]; + tensor x_281_cast_fp16 = gelu(mode = x_281_mode_0, x = linear_136_cast_fp16)[name = tensor("x_281_cast_fp16")]; + tensor var_2597_to_fp16 = const()[name = tensor("op_2597_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(906677120)))]; + tensor var_2598_to_fp16 = const()[name = tensor("op_2598_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(919784384)))]; + tensor linear_137_cast_fp16 = linear(bias = var_2598_to_fp16, weight = var_2597_to_fp16, x = x_281_cast_fp16)[name = tensor("linear_137_cast_fp16")]; + tensor x_283_cast_fp16 = add(x = x_277_cast_fp16, y = linear_137_cast_fp16)[name = tensor("x_283_cast_fp16")]; + tensor var_2608 = const()[name = tensor("op_2608"), val = tensor(-1)]; + tensor var_2625_axes_0 = const()[name = tensor("op_2625_axes_0"), val = tensor([-1])]; + tensor blocks_23_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_23_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(919787008)))]; + tensor blocks_23_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_23_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(919789632)))]; + tensor var_2615_to_fp16 = const()[name = tensor("op_2615_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2625_cast_fp16 = layer_norm(axes = var_2625_axes_0, beta = blocks_23_attn_ln_bias_to_fp16, epsilon = var_2615_to_fp16, gamma = blocks_23_attn_ln_weight_to_fp16, x = x_283_cast_fp16)[name = tensor("op_2625_cast_fp16")]; + tensor var_2636_to_fp16 = const()[name = tensor("op_2636_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(919792256)))]; + tensor var_2637_to_fp16 = const()[name = tensor("op_2637_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(923069120)))]; + tensor linear_138_cast_fp16 = linear(bias = var_2637_to_fp16, weight = var_2636_to_fp16, x = var_2625_cast_fp16)[name = tensor("linear_138_cast_fp16")]; + tensor var_2640_to_fp16 = const()[name = tensor("op_2640_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(923071744)))]; + tensor linear_139_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2640_to_fp16, x = var_2625_cast_fp16)[name = tensor("linear_139_cast_fp16")]; + tensor var_2644_to_fp16 = const()[name = tensor("op_2644_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(926348608)))]; + tensor var_2645_to_fp16 = const()[name = tensor("op_2645_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(929625472)))]; + tensor linear_140_cast_fp16 = linear(bias = var_2645_to_fp16, weight = var_2644_to_fp16, x = var_2625_cast_fp16)[name = tensor("linear_140_cast_fp16")]; + tensor var_2653 = const()[name = tensor("op_2653"), val = tensor([1, 1500, 20, -1])]; + tensor var_2654_cast_fp16 = reshape(shape = var_2653, x = linear_138_cast_fp16)[name = tensor("op_2654_cast_fp16")]; + tensor var_2655 = const()[name = tensor("op_2655"), val = tensor([0, 2, 1, 3])]; + tensor var_2659 = const()[name = tensor("op_2659"), val = tensor([1, 1500, 20, -1])]; + tensor var_2660_cast_fp16 = reshape(shape = var_2659, x = linear_139_cast_fp16)[name = tensor("op_2660_cast_fp16")]; + tensor var_2661 = const()[name = tensor("op_2661"), val = tensor([0, 2, 1, 3])]; + tensor var_2665 = const()[name = tensor("op_2665"), val = tensor([1, 1500, 20, -1])]; + tensor var_2666_cast_fp16 = reshape(shape = var_2665, x = linear_140_cast_fp16)[name = tensor("op_2666_cast_fp16")]; + tensor var_2667 = const()[name = tensor("op_2667"), val = tensor([0, 2, 1, 3])]; + tensor scale_47_to_fp16 = const()[name = tensor("scale_47_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_44 = transpose(perm = var_2655, x = var_2654_cast_fp16)[name = tensor("transpose_44")]; + tensor var_2669_cast_fp16 = mul(x = transpose_44, y = scale_47_to_fp16)[name = tensor("op_2669_cast_fp16")]; + tensor transpose_43 = transpose(perm = var_2661, x = var_2660_cast_fp16)[name = tensor("transpose_43")]; + tensor var_2670_cast_fp16 = mul(x = transpose_43, y = scale_47_to_fp16)[name = tensor("op_2670_cast_fp16")]; + tensor var_2671_perm_0 = const()[name = tensor("op_2671_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_41 = transpose(perm = var_2671_perm_0, x = var_2670_cast_fp16)[name = tensor("transpose_41")]; + tensor qk_47_cast_fp16 = matmul(transpose_x = qk_47_transpose_x_0, transpose_y = qk_47_transpose_y_0, x = var_2669_cast_fp16, y = transpose_41)[name = tensor("qk_47_cast_fp16")]; + tensor var_2674_cast_fp16 = softmax(axis = var_2608, x = qk_47_cast_fp16)[name = tensor("op_2674_cast_fp16")]; + tensor var_2676_transpose_x_0 = const()[name = tensor("op_2676_transpose_x_0"), val = tensor(false)]; + tensor var_2676_transpose_y_0 = const()[name = tensor("op_2676_transpose_y_0"), val = tensor(false)]; + tensor transpose_42 = transpose(perm = var_2667, x = var_2666_cast_fp16)[name = tensor("transpose_42")]; + tensor var_2676_cast_fp16 = matmul(transpose_x = var_2676_transpose_x_0, transpose_y = var_2676_transpose_y_0, x = var_2674_cast_fp16, y = transpose_42)[name = tensor("op_2676_cast_fp16")]; + tensor var_2677 = const()[name = tensor("op_2677"), val = tensor([0, 2, 1, 3])]; + tensor concat_23 = const()[name = tensor("concat_23"), val = tensor([1, 1500, 1280])]; + tensor transpose_40 = transpose(perm = var_2677, x = var_2676_cast_fp16)[name = tensor("transpose_40")]; + tensor x_287_cast_fp16 = reshape(shape = concat_23, x = transpose_40)[name = tensor("x_287_cast_fp16")]; + tensor var_2682_to_fp16 = const()[name = tensor("op_2682_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(929628096)))]; + tensor var_2683_to_fp16 = const()[name = tensor("op_2683_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(932904960)))]; + tensor linear_141_cast_fp16 = linear(bias = var_2683_to_fp16, weight = var_2682_to_fp16, x = x_287_cast_fp16)[name = tensor("linear_141_cast_fp16")]; + tensor x_289_cast_fp16 = add(x = x_283_cast_fp16, y = linear_141_cast_fp16)[name = tensor("x_289_cast_fp16")]; + tensor var_2690_axes_0 = const()[name = tensor("op_2690_axes_0"), val = tensor([-1])]; + tensor blocks_23_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_23_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(932907584)))]; + tensor blocks_23_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_23_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(932910208)))]; + tensor var_2690_cast_fp16 = layer_norm(axes = var_2690_axes_0, beta = blocks_23_mlp_ln_bias_to_fp16, epsilon = var_2615_to_fp16, gamma = blocks_23_mlp_ln_weight_to_fp16, x = x_289_cast_fp16)[name = tensor("op_2690_cast_fp16")]; + tensor var_2699_to_fp16 = const()[name = tensor("op_2699_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(932912832)))]; + tensor var_2700_to_fp16 = const()[name = tensor("op_2700_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(946020096)))]; + tensor linear_142_cast_fp16 = linear(bias = var_2700_to_fp16, weight = var_2699_to_fp16, x = var_2690_cast_fp16)[name = tensor("linear_142_cast_fp16")]; + tensor x_293_mode_0 = const()[name = tensor("x_293_mode_0"), val = tensor("EXACT")]; + tensor x_293_cast_fp16 = gelu(mode = x_293_mode_0, x = linear_142_cast_fp16)[name = tensor("x_293_cast_fp16")]; + tensor var_2705_to_fp16 = const()[name = tensor("op_2705_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(946030400)))]; + tensor var_2706_to_fp16 = const()[name = tensor("op_2706_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(959137664)))]; + tensor linear_143_cast_fp16 = linear(bias = var_2706_to_fp16, weight = var_2705_to_fp16, x = x_293_cast_fp16)[name = tensor("linear_143_cast_fp16")]; + tensor x_295_cast_fp16 = add(x = x_289_cast_fp16, y = linear_143_cast_fp16)[name = tensor("x_295_cast_fp16")]; + tensor var_2716 = const()[name = tensor("op_2716"), val = tensor(-1)]; + tensor var_2733_axes_0 = const()[name = tensor("op_2733_axes_0"), val = tensor([-1])]; + tensor blocks_24_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_24_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(959140288)))]; + tensor blocks_24_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_24_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(959142912)))]; + tensor var_2723_to_fp16 = const()[name = tensor("op_2723_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2733_cast_fp16 = layer_norm(axes = var_2733_axes_0, beta = blocks_24_attn_ln_bias_to_fp16, epsilon = var_2723_to_fp16, gamma = blocks_24_attn_ln_weight_to_fp16, x = x_295_cast_fp16)[name = tensor("op_2733_cast_fp16")]; + tensor var_2744_to_fp16 = const()[name = tensor("op_2744_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(959145536)))]; + tensor var_2745_to_fp16 = const()[name = tensor("op_2745_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(962422400)))]; + tensor linear_144_cast_fp16 = linear(bias = var_2745_to_fp16, weight = var_2744_to_fp16, x = var_2733_cast_fp16)[name = tensor("linear_144_cast_fp16")]; + tensor var_2748_to_fp16 = const()[name = tensor("op_2748_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(962425024)))]; + tensor linear_145_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2748_to_fp16, x = var_2733_cast_fp16)[name = tensor("linear_145_cast_fp16")]; + tensor var_2752_to_fp16 = const()[name = tensor("op_2752_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(965701888)))]; + tensor var_2753_to_fp16 = const()[name = tensor("op_2753_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(968978752)))]; + tensor linear_146_cast_fp16 = linear(bias = var_2753_to_fp16, weight = var_2752_to_fp16, x = var_2733_cast_fp16)[name = tensor("linear_146_cast_fp16")]; + tensor var_2761 = const()[name = tensor("op_2761"), val = tensor([1, 1500, 20, -1])]; + tensor var_2762_cast_fp16 = reshape(shape = var_2761, x = linear_144_cast_fp16)[name = tensor("op_2762_cast_fp16")]; + tensor var_2763 = const()[name = tensor("op_2763"), val = tensor([0, 2, 1, 3])]; + tensor var_2767 = const()[name = tensor("op_2767"), val = tensor([1, 1500, 20, -1])]; + tensor var_2768_cast_fp16 = reshape(shape = var_2767, x = linear_145_cast_fp16)[name = tensor("op_2768_cast_fp16")]; + tensor var_2769 = const()[name = tensor("op_2769"), val = tensor([0, 2, 1, 3])]; + tensor var_2773 = const()[name = tensor("op_2773"), val = tensor([1, 1500, 20, -1])]; + tensor var_2774_cast_fp16 = reshape(shape = var_2773, x = linear_146_cast_fp16)[name = tensor("op_2774_cast_fp16")]; + tensor var_2775 = const()[name = tensor("op_2775"), val = tensor([0, 2, 1, 3])]; + tensor scale_49_to_fp16 = const()[name = tensor("scale_49_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_39 = transpose(perm = var_2763, x = var_2762_cast_fp16)[name = tensor("transpose_39")]; + tensor var_2777_cast_fp16 = mul(x = transpose_39, y = scale_49_to_fp16)[name = tensor("op_2777_cast_fp16")]; + tensor transpose_38 = transpose(perm = var_2769, x = var_2768_cast_fp16)[name = tensor("transpose_38")]; + tensor var_2778_cast_fp16 = mul(x = transpose_38, y = scale_49_to_fp16)[name = tensor("op_2778_cast_fp16")]; + tensor var_2779_perm_0 = const()[name = tensor("op_2779_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_36 = transpose(perm = var_2779_perm_0, x = var_2778_cast_fp16)[name = tensor("transpose_36")]; + tensor qk_49_cast_fp16 = matmul(transpose_x = qk_49_transpose_x_0, transpose_y = qk_49_transpose_y_0, x = var_2777_cast_fp16, y = transpose_36)[name = tensor("qk_49_cast_fp16")]; + tensor var_2782_cast_fp16 = softmax(axis = var_2716, x = qk_49_cast_fp16)[name = tensor("op_2782_cast_fp16")]; + tensor var_2784_transpose_x_0 = const()[name = tensor("op_2784_transpose_x_0"), val = tensor(false)]; + tensor var_2784_transpose_y_0 = const()[name = tensor("op_2784_transpose_y_0"), val = tensor(false)]; + tensor transpose_37 = transpose(perm = var_2775, x = var_2774_cast_fp16)[name = tensor("transpose_37")]; + tensor var_2784_cast_fp16 = matmul(transpose_x = var_2784_transpose_x_0, transpose_y = var_2784_transpose_y_0, x = var_2782_cast_fp16, y = transpose_37)[name = tensor("op_2784_cast_fp16")]; + tensor var_2785 = const()[name = tensor("op_2785"), val = tensor([0, 2, 1, 3])]; + tensor concat_24 = const()[name = tensor("concat_24"), val = tensor([1, 1500, 1280])]; + tensor transpose_35 = transpose(perm = var_2785, x = var_2784_cast_fp16)[name = tensor("transpose_35")]; + tensor x_299_cast_fp16 = reshape(shape = concat_24, x = transpose_35)[name = tensor("x_299_cast_fp16")]; + tensor var_2790_to_fp16 = const()[name = tensor("op_2790_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(968981376)))]; + tensor var_2791_to_fp16 = const()[name = tensor("op_2791_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(972258240)))]; + tensor linear_147_cast_fp16 = linear(bias = var_2791_to_fp16, weight = var_2790_to_fp16, x = x_299_cast_fp16)[name = tensor("linear_147_cast_fp16")]; + tensor x_301_cast_fp16 = add(x = x_295_cast_fp16, y = linear_147_cast_fp16)[name = tensor("x_301_cast_fp16")]; + tensor var_2798_axes_0 = const()[name = tensor("op_2798_axes_0"), val = tensor([-1])]; + tensor blocks_24_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_24_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(972260864)))]; + tensor blocks_24_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_24_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(972263488)))]; + tensor var_2798_cast_fp16 = layer_norm(axes = var_2798_axes_0, beta = blocks_24_mlp_ln_bias_to_fp16, epsilon = var_2723_to_fp16, gamma = blocks_24_mlp_ln_weight_to_fp16, x = x_301_cast_fp16)[name = tensor("op_2798_cast_fp16")]; + tensor var_2807_to_fp16 = const()[name = tensor("op_2807_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(972266112)))]; + tensor var_2808_to_fp16 = const()[name = tensor("op_2808_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(985373376)))]; + tensor linear_148_cast_fp16 = linear(bias = var_2808_to_fp16, weight = var_2807_to_fp16, x = var_2798_cast_fp16)[name = tensor("linear_148_cast_fp16")]; + tensor x_305_mode_0 = const()[name = tensor("x_305_mode_0"), val = tensor("EXACT")]; + tensor x_305_cast_fp16 = gelu(mode = x_305_mode_0, x = linear_148_cast_fp16)[name = tensor("x_305_cast_fp16")]; + tensor var_2813_to_fp16 = const()[name = tensor("op_2813_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(985383680)))]; + tensor var_2814_to_fp16 = const()[name = tensor("op_2814_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(998490944)))]; + tensor linear_149_cast_fp16 = linear(bias = var_2814_to_fp16, weight = var_2813_to_fp16, x = x_305_cast_fp16)[name = tensor("linear_149_cast_fp16")]; + tensor x_307_cast_fp16 = add(x = x_301_cast_fp16, y = linear_149_cast_fp16)[name = tensor("x_307_cast_fp16")]; + tensor var_2824 = const()[name = tensor("op_2824"), val = tensor(-1)]; + tensor var_2841_axes_0 = const()[name = tensor("op_2841_axes_0"), val = tensor([-1])]; + tensor blocks_25_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_25_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(998493568)))]; + tensor blocks_25_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_25_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(998496192)))]; + tensor var_2831_to_fp16 = const()[name = tensor("op_2831_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2841_cast_fp16 = layer_norm(axes = var_2841_axes_0, beta = blocks_25_attn_ln_bias_to_fp16, epsilon = var_2831_to_fp16, gamma = blocks_25_attn_ln_weight_to_fp16, x = x_307_cast_fp16)[name = tensor("op_2841_cast_fp16")]; + tensor var_2852_to_fp16 = const()[name = tensor("op_2852_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(998498816)))]; + tensor var_2853_to_fp16 = const()[name = tensor("op_2853_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1001775680)))]; + tensor linear_150_cast_fp16 = linear(bias = var_2853_to_fp16, weight = var_2852_to_fp16, x = var_2841_cast_fp16)[name = tensor("linear_150_cast_fp16")]; + tensor var_2856_to_fp16 = const()[name = tensor("op_2856_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1001778304)))]; + tensor linear_151_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2856_to_fp16, x = var_2841_cast_fp16)[name = tensor("linear_151_cast_fp16")]; + tensor var_2860_to_fp16 = const()[name = tensor("op_2860_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1005055168)))]; + tensor var_2861_to_fp16 = const()[name = tensor("op_2861_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1008332032)))]; + tensor linear_152_cast_fp16 = linear(bias = var_2861_to_fp16, weight = var_2860_to_fp16, x = var_2841_cast_fp16)[name = tensor("linear_152_cast_fp16")]; + tensor var_2869 = const()[name = tensor("op_2869"), val = tensor([1, 1500, 20, -1])]; + tensor var_2870_cast_fp16 = reshape(shape = var_2869, x = linear_150_cast_fp16)[name = tensor("op_2870_cast_fp16")]; + tensor var_2871 = const()[name = tensor("op_2871"), val = tensor([0, 2, 1, 3])]; + tensor var_2875 = const()[name = tensor("op_2875"), val = tensor([1, 1500, 20, -1])]; + tensor var_2876_cast_fp16 = reshape(shape = var_2875, x = linear_151_cast_fp16)[name = tensor("op_2876_cast_fp16")]; + tensor var_2877 = const()[name = tensor("op_2877"), val = tensor([0, 2, 1, 3])]; + tensor var_2881 = const()[name = tensor("op_2881"), val = tensor([1, 1500, 20, -1])]; + tensor var_2882_cast_fp16 = reshape(shape = var_2881, x = linear_152_cast_fp16)[name = tensor("op_2882_cast_fp16")]; + tensor var_2883 = const()[name = tensor("op_2883"), val = tensor([0, 2, 1, 3])]; + tensor scale_51_to_fp16 = const()[name = tensor("scale_51_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_34 = transpose(perm = var_2871, x = var_2870_cast_fp16)[name = tensor("transpose_34")]; + tensor var_2885_cast_fp16 = mul(x = transpose_34, y = scale_51_to_fp16)[name = tensor("op_2885_cast_fp16")]; + tensor transpose_33 = transpose(perm = var_2877, x = var_2876_cast_fp16)[name = tensor("transpose_33")]; + tensor var_2886_cast_fp16 = mul(x = transpose_33, y = scale_51_to_fp16)[name = tensor("op_2886_cast_fp16")]; + tensor var_2887_perm_0 = const()[name = tensor("op_2887_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_31 = transpose(perm = var_2887_perm_0, x = var_2886_cast_fp16)[name = tensor("transpose_31")]; + tensor qk_51_cast_fp16 = matmul(transpose_x = qk_51_transpose_x_0, transpose_y = qk_51_transpose_y_0, x = var_2885_cast_fp16, y = transpose_31)[name = tensor("qk_51_cast_fp16")]; + tensor var_2890_cast_fp16 = softmax(axis = var_2824, x = qk_51_cast_fp16)[name = tensor("op_2890_cast_fp16")]; + tensor var_2892_transpose_x_0 = const()[name = tensor("op_2892_transpose_x_0"), val = tensor(false)]; + tensor var_2892_transpose_y_0 = const()[name = tensor("op_2892_transpose_y_0"), val = tensor(false)]; + tensor transpose_32 = transpose(perm = var_2883, x = var_2882_cast_fp16)[name = tensor("transpose_32")]; + tensor var_2892_cast_fp16 = matmul(transpose_x = var_2892_transpose_x_0, transpose_y = var_2892_transpose_y_0, x = var_2890_cast_fp16, y = transpose_32)[name = tensor("op_2892_cast_fp16")]; + tensor var_2893 = const()[name = tensor("op_2893"), val = tensor([0, 2, 1, 3])]; + tensor concat_25 = const()[name = tensor("concat_25"), val = tensor([1, 1500, 1280])]; + tensor transpose_30 = transpose(perm = var_2893, x = var_2892_cast_fp16)[name = tensor("transpose_30")]; + tensor x_311_cast_fp16 = reshape(shape = concat_25, x = transpose_30)[name = tensor("x_311_cast_fp16")]; + tensor var_2898_to_fp16 = const()[name = tensor("op_2898_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1008334656)))]; + tensor var_2899_to_fp16 = const()[name = tensor("op_2899_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1011611520)))]; + tensor linear_153_cast_fp16 = linear(bias = var_2899_to_fp16, weight = var_2898_to_fp16, x = x_311_cast_fp16)[name = tensor("linear_153_cast_fp16")]; + tensor x_313_cast_fp16 = add(x = x_307_cast_fp16, y = linear_153_cast_fp16)[name = tensor("x_313_cast_fp16")]; + tensor var_2906_axes_0 = const()[name = tensor("op_2906_axes_0"), val = tensor([-1])]; + tensor blocks_25_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_25_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1011614144)))]; + tensor blocks_25_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_25_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1011616768)))]; + tensor var_2906_cast_fp16 = layer_norm(axes = var_2906_axes_0, beta = blocks_25_mlp_ln_bias_to_fp16, epsilon = var_2831_to_fp16, gamma = blocks_25_mlp_ln_weight_to_fp16, x = x_313_cast_fp16)[name = tensor("op_2906_cast_fp16")]; + tensor var_2915_to_fp16 = const()[name = tensor("op_2915_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1011619392)))]; + tensor var_2916_to_fp16 = const()[name = tensor("op_2916_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1024726656)))]; + tensor linear_154_cast_fp16 = linear(bias = var_2916_to_fp16, weight = var_2915_to_fp16, x = var_2906_cast_fp16)[name = tensor("linear_154_cast_fp16")]; + tensor x_317_mode_0 = const()[name = tensor("x_317_mode_0"), val = tensor("EXACT")]; + tensor x_317_cast_fp16 = gelu(mode = x_317_mode_0, x = linear_154_cast_fp16)[name = tensor("x_317_cast_fp16")]; + tensor var_2921_to_fp16 = const()[name = tensor("op_2921_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1024736960)))]; + tensor var_2922_to_fp16 = const()[name = tensor("op_2922_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1037844224)))]; + tensor linear_155_cast_fp16 = linear(bias = var_2922_to_fp16, weight = var_2921_to_fp16, x = x_317_cast_fp16)[name = tensor("linear_155_cast_fp16")]; + tensor x_319_cast_fp16 = add(x = x_313_cast_fp16, y = linear_155_cast_fp16)[name = tensor("x_319_cast_fp16")]; + tensor var_2932 = const()[name = tensor("op_2932"), val = tensor(-1)]; + tensor var_2949_axes_0 = const()[name = tensor("op_2949_axes_0"), val = tensor([-1])]; + tensor blocks_26_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_26_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1037846848)))]; + tensor blocks_26_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_26_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1037849472)))]; + tensor var_2939_to_fp16 = const()[name = tensor("op_2939_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2949_cast_fp16 = layer_norm(axes = var_2949_axes_0, beta = blocks_26_attn_ln_bias_to_fp16, epsilon = var_2939_to_fp16, gamma = blocks_26_attn_ln_weight_to_fp16, x = x_319_cast_fp16)[name = tensor("op_2949_cast_fp16")]; + tensor var_2960_to_fp16 = const()[name = tensor("op_2960_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1037852096)))]; + tensor var_2961_to_fp16 = const()[name = tensor("op_2961_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1041128960)))]; + tensor linear_156_cast_fp16 = linear(bias = var_2961_to_fp16, weight = var_2960_to_fp16, x = var_2949_cast_fp16)[name = tensor("linear_156_cast_fp16")]; + tensor var_2964_to_fp16 = const()[name = tensor("op_2964_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1041131584)))]; + tensor linear_157_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2964_to_fp16, x = var_2949_cast_fp16)[name = tensor("linear_157_cast_fp16")]; + tensor var_2968_to_fp16 = const()[name = tensor("op_2968_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1044408448)))]; + tensor var_2969_to_fp16 = const()[name = tensor("op_2969_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1047685312)))]; + tensor linear_158_cast_fp16 = linear(bias = var_2969_to_fp16, weight = var_2968_to_fp16, x = var_2949_cast_fp16)[name = tensor("linear_158_cast_fp16")]; + tensor var_2977 = const()[name = tensor("op_2977"), val = tensor([1, 1500, 20, -1])]; + tensor var_2978_cast_fp16 = reshape(shape = var_2977, x = linear_156_cast_fp16)[name = tensor("op_2978_cast_fp16")]; + tensor var_2979 = const()[name = tensor("op_2979"), val = tensor([0, 2, 1, 3])]; + tensor var_2983 = const()[name = tensor("op_2983"), val = tensor([1, 1500, 20, -1])]; + tensor var_2984_cast_fp16 = reshape(shape = var_2983, x = linear_157_cast_fp16)[name = tensor("op_2984_cast_fp16")]; + tensor var_2985 = const()[name = tensor("op_2985"), val = tensor([0, 2, 1, 3])]; + tensor var_2989 = const()[name = tensor("op_2989"), val = tensor([1, 1500, 20, -1])]; + tensor var_2990_cast_fp16 = reshape(shape = var_2989, x = linear_158_cast_fp16)[name = tensor("op_2990_cast_fp16")]; + tensor var_2991 = const()[name = tensor("op_2991"), val = tensor([0, 2, 1, 3])]; + tensor scale_53_to_fp16 = const()[name = tensor("scale_53_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_29 = transpose(perm = var_2979, x = var_2978_cast_fp16)[name = tensor("transpose_29")]; + tensor var_2993_cast_fp16 = mul(x = transpose_29, y = scale_53_to_fp16)[name = tensor("op_2993_cast_fp16")]; + tensor transpose_28 = transpose(perm = var_2985, x = var_2984_cast_fp16)[name = tensor("transpose_28")]; + tensor var_2994_cast_fp16 = mul(x = transpose_28, y = scale_53_to_fp16)[name = tensor("op_2994_cast_fp16")]; + tensor var_2995_perm_0 = const()[name = tensor("op_2995_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_26 = transpose(perm = var_2995_perm_0, x = var_2994_cast_fp16)[name = tensor("transpose_26")]; + tensor qk_53_cast_fp16 = matmul(transpose_x = qk_53_transpose_x_0, transpose_y = qk_53_transpose_y_0, x = var_2993_cast_fp16, y = transpose_26)[name = tensor("qk_53_cast_fp16")]; + tensor var_2998_cast_fp16 = softmax(axis = var_2932, x = qk_53_cast_fp16)[name = tensor("op_2998_cast_fp16")]; + tensor var_3000_transpose_x_0 = const()[name = tensor("op_3000_transpose_x_0"), val = tensor(false)]; + tensor var_3000_transpose_y_0 = const()[name = tensor("op_3000_transpose_y_0"), val = tensor(false)]; + tensor transpose_27 = transpose(perm = var_2991, x = var_2990_cast_fp16)[name = tensor("transpose_27")]; + tensor var_3000_cast_fp16 = matmul(transpose_x = var_3000_transpose_x_0, transpose_y = var_3000_transpose_y_0, x = var_2998_cast_fp16, y = transpose_27)[name = tensor("op_3000_cast_fp16")]; + tensor var_3001 = const()[name = tensor("op_3001"), val = tensor([0, 2, 1, 3])]; + tensor concat_26 = const()[name = tensor("concat_26"), val = tensor([1, 1500, 1280])]; + tensor transpose_25 = transpose(perm = var_3001, x = var_3000_cast_fp16)[name = tensor("transpose_25")]; + tensor x_323_cast_fp16 = reshape(shape = concat_26, x = transpose_25)[name = tensor("x_323_cast_fp16")]; + tensor var_3006_to_fp16 = const()[name = tensor("op_3006_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1047687936)))]; + tensor var_3007_to_fp16 = const()[name = tensor("op_3007_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1050964800)))]; + tensor linear_159_cast_fp16 = linear(bias = var_3007_to_fp16, weight = var_3006_to_fp16, x = x_323_cast_fp16)[name = tensor("linear_159_cast_fp16")]; + tensor x_325_cast_fp16 = add(x = x_319_cast_fp16, y = linear_159_cast_fp16)[name = tensor("x_325_cast_fp16")]; + tensor var_3014_axes_0 = const()[name = tensor("op_3014_axes_0"), val = tensor([-1])]; + tensor blocks_26_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_26_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1050967424)))]; + tensor blocks_26_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_26_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1050970048)))]; + tensor var_3014_cast_fp16 = layer_norm(axes = var_3014_axes_0, beta = blocks_26_mlp_ln_bias_to_fp16, epsilon = var_2939_to_fp16, gamma = blocks_26_mlp_ln_weight_to_fp16, x = x_325_cast_fp16)[name = tensor("op_3014_cast_fp16")]; + tensor var_3023_to_fp16 = const()[name = tensor("op_3023_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1050972672)))]; + tensor var_3024_to_fp16 = const()[name = tensor("op_3024_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1064079936)))]; + tensor linear_160_cast_fp16 = linear(bias = var_3024_to_fp16, weight = var_3023_to_fp16, x = var_3014_cast_fp16)[name = tensor("linear_160_cast_fp16")]; + tensor x_329_mode_0 = const()[name = tensor("x_329_mode_0"), val = tensor("EXACT")]; + tensor x_329_cast_fp16 = gelu(mode = x_329_mode_0, x = linear_160_cast_fp16)[name = tensor("x_329_cast_fp16")]; + tensor var_3029_to_fp16 = const()[name = tensor("op_3029_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1064090240)))]; + tensor var_3030_to_fp16 = const()[name = tensor("op_3030_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1077197504)))]; + tensor linear_161_cast_fp16 = linear(bias = var_3030_to_fp16, weight = var_3029_to_fp16, x = x_329_cast_fp16)[name = tensor("linear_161_cast_fp16")]; + tensor x_331_cast_fp16 = add(x = x_325_cast_fp16, y = linear_161_cast_fp16)[name = tensor("x_331_cast_fp16")]; + tensor var_3040 = const()[name = tensor("op_3040"), val = tensor(-1)]; + tensor var_3057_axes_0 = const()[name = tensor("op_3057_axes_0"), val = tensor([-1])]; + tensor blocks_27_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_27_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1077200128)))]; + tensor blocks_27_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_27_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1077202752)))]; + tensor var_3047_to_fp16 = const()[name = tensor("op_3047_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3057_cast_fp16 = layer_norm(axes = var_3057_axes_0, beta = blocks_27_attn_ln_bias_to_fp16, epsilon = var_3047_to_fp16, gamma = blocks_27_attn_ln_weight_to_fp16, x = x_331_cast_fp16)[name = tensor("op_3057_cast_fp16")]; + tensor var_3068_to_fp16 = const()[name = tensor("op_3068_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1077205376)))]; + tensor var_3069_to_fp16 = const()[name = tensor("op_3069_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1080482240)))]; + tensor linear_162_cast_fp16 = linear(bias = var_3069_to_fp16, weight = var_3068_to_fp16, x = var_3057_cast_fp16)[name = tensor("linear_162_cast_fp16")]; + tensor var_3072_to_fp16 = const()[name = tensor("op_3072_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1080484864)))]; + tensor linear_163_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3072_to_fp16, x = var_3057_cast_fp16)[name = tensor("linear_163_cast_fp16")]; + tensor var_3076_to_fp16 = const()[name = tensor("op_3076_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1083761728)))]; + tensor var_3077_to_fp16 = const()[name = tensor("op_3077_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1087038592)))]; + tensor linear_164_cast_fp16 = linear(bias = var_3077_to_fp16, weight = var_3076_to_fp16, x = var_3057_cast_fp16)[name = tensor("linear_164_cast_fp16")]; + tensor var_3085 = const()[name = tensor("op_3085"), val = tensor([1, 1500, 20, -1])]; + tensor var_3086_cast_fp16 = reshape(shape = var_3085, x = linear_162_cast_fp16)[name = tensor("op_3086_cast_fp16")]; + tensor var_3087 = const()[name = tensor("op_3087"), val = tensor([0, 2, 1, 3])]; + tensor var_3091 = const()[name = tensor("op_3091"), val = tensor([1, 1500, 20, -1])]; + tensor var_3092_cast_fp16 = reshape(shape = var_3091, x = linear_163_cast_fp16)[name = tensor("op_3092_cast_fp16")]; + tensor var_3093 = const()[name = tensor("op_3093"), val = tensor([0, 2, 1, 3])]; + tensor var_3097 = const()[name = tensor("op_3097"), val = tensor([1, 1500, 20, -1])]; + tensor var_3098_cast_fp16 = reshape(shape = var_3097, x = linear_164_cast_fp16)[name = tensor("op_3098_cast_fp16")]; + tensor var_3099 = const()[name = tensor("op_3099"), val = tensor([0, 2, 1, 3])]; + tensor scale_55_to_fp16 = const()[name = tensor("scale_55_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_24 = transpose(perm = var_3087, x = var_3086_cast_fp16)[name = tensor("transpose_24")]; + tensor var_3101_cast_fp16 = mul(x = transpose_24, y = scale_55_to_fp16)[name = tensor("op_3101_cast_fp16")]; + tensor transpose_23 = transpose(perm = var_3093, x = var_3092_cast_fp16)[name = tensor("transpose_23")]; + tensor var_3102_cast_fp16 = mul(x = transpose_23, y = scale_55_to_fp16)[name = tensor("op_3102_cast_fp16")]; + tensor var_3103_perm_0 = const()[name = tensor("op_3103_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_21 = transpose(perm = var_3103_perm_0, x = var_3102_cast_fp16)[name = tensor("transpose_21")]; + tensor qk_55_cast_fp16 = matmul(transpose_x = qk_55_transpose_x_0, transpose_y = qk_55_transpose_y_0, x = var_3101_cast_fp16, y = transpose_21)[name = tensor("qk_55_cast_fp16")]; + tensor var_3106_cast_fp16 = softmax(axis = var_3040, x = qk_55_cast_fp16)[name = tensor("op_3106_cast_fp16")]; + tensor var_3108_transpose_x_0 = const()[name = tensor("op_3108_transpose_x_0"), val = tensor(false)]; + tensor var_3108_transpose_y_0 = const()[name = tensor("op_3108_transpose_y_0"), val = tensor(false)]; + tensor transpose_22 = transpose(perm = var_3099, x = var_3098_cast_fp16)[name = tensor("transpose_22")]; + tensor var_3108_cast_fp16 = matmul(transpose_x = var_3108_transpose_x_0, transpose_y = var_3108_transpose_y_0, x = var_3106_cast_fp16, y = transpose_22)[name = tensor("op_3108_cast_fp16")]; + tensor var_3109 = const()[name = tensor("op_3109"), val = tensor([0, 2, 1, 3])]; + tensor concat_27 = const()[name = tensor("concat_27"), val = tensor([1, 1500, 1280])]; + tensor transpose_20 = transpose(perm = var_3109, x = var_3108_cast_fp16)[name = tensor("transpose_20")]; + tensor x_335_cast_fp16 = reshape(shape = concat_27, x = transpose_20)[name = tensor("x_335_cast_fp16")]; + tensor var_3114_to_fp16 = const()[name = tensor("op_3114_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1087041216)))]; + tensor var_3115_to_fp16 = const()[name = tensor("op_3115_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1090318080)))]; + tensor linear_165_cast_fp16 = linear(bias = var_3115_to_fp16, weight = var_3114_to_fp16, x = x_335_cast_fp16)[name = tensor("linear_165_cast_fp16")]; + tensor x_337_cast_fp16 = add(x = x_331_cast_fp16, y = linear_165_cast_fp16)[name = tensor("x_337_cast_fp16")]; + tensor var_3122_axes_0 = const()[name = tensor("op_3122_axes_0"), val = tensor([-1])]; + tensor blocks_27_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_27_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1090320704)))]; + tensor blocks_27_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_27_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1090323328)))]; + tensor var_3122_cast_fp16 = layer_norm(axes = var_3122_axes_0, beta = blocks_27_mlp_ln_bias_to_fp16, epsilon = var_3047_to_fp16, gamma = blocks_27_mlp_ln_weight_to_fp16, x = x_337_cast_fp16)[name = tensor("op_3122_cast_fp16")]; + tensor var_3131_to_fp16 = const()[name = tensor("op_3131_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1090325952)))]; + tensor var_3132_to_fp16 = const()[name = tensor("op_3132_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1103433216)))]; + tensor linear_166_cast_fp16 = linear(bias = var_3132_to_fp16, weight = var_3131_to_fp16, x = var_3122_cast_fp16)[name = tensor("linear_166_cast_fp16")]; + tensor x_341_mode_0 = const()[name = tensor("x_341_mode_0"), val = tensor("EXACT")]; + tensor x_341_cast_fp16 = gelu(mode = x_341_mode_0, x = linear_166_cast_fp16)[name = tensor("x_341_cast_fp16")]; + tensor var_3137_to_fp16 = const()[name = tensor("op_3137_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1103443520)))]; + tensor var_3138_to_fp16 = const()[name = tensor("op_3138_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116550784)))]; + tensor linear_167_cast_fp16 = linear(bias = var_3138_to_fp16, weight = var_3137_to_fp16, x = x_341_cast_fp16)[name = tensor("linear_167_cast_fp16")]; + tensor x_343_cast_fp16 = add(x = x_337_cast_fp16, y = linear_167_cast_fp16)[name = tensor("x_343_cast_fp16")]; + tensor var_3148 = const()[name = tensor("op_3148"), val = tensor(-1)]; + tensor var_3165_axes_0 = const()[name = tensor("op_3165_axes_0"), val = tensor([-1])]; + tensor blocks_28_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_28_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116553408)))]; + tensor blocks_28_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_28_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116556032)))]; + tensor var_3155_to_fp16 = const()[name = tensor("op_3155_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3165_cast_fp16 = layer_norm(axes = var_3165_axes_0, beta = blocks_28_attn_ln_bias_to_fp16, epsilon = var_3155_to_fp16, gamma = blocks_28_attn_ln_weight_to_fp16, x = x_343_cast_fp16)[name = tensor("op_3165_cast_fp16")]; + tensor var_3176_to_fp16 = const()[name = tensor("op_3176_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116558656)))]; + tensor var_3177_to_fp16 = const()[name = tensor("op_3177_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1119835520)))]; + tensor linear_168_cast_fp16 = linear(bias = var_3177_to_fp16, weight = var_3176_to_fp16, x = var_3165_cast_fp16)[name = tensor("linear_168_cast_fp16")]; + tensor var_3180_to_fp16 = const()[name = tensor("op_3180_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1119838144)))]; + tensor linear_169_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3180_to_fp16, x = var_3165_cast_fp16)[name = tensor("linear_169_cast_fp16")]; + tensor var_3184_to_fp16 = const()[name = tensor("op_3184_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1123115008)))]; + tensor var_3185_to_fp16 = const()[name = tensor("op_3185_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1126391872)))]; + tensor linear_170_cast_fp16 = linear(bias = var_3185_to_fp16, weight = var_3184_to_fp16, x = var_3165_cast_fp16)[name = tensor("linear_170_cast_fp16")]; + tensor var_3193 = const()[name = tensor("op_3193"), val = tensor([1, 1500, 20, -1])]; + tensor var_3194_cast_fp16 = reshape(shape = var_3193, x = linear_168_cast_fp16)[name = tensor("op_3194_cast_fp16")]; + tensor var_3195 = const()[name = tensor("op_3195"), val = tensor([0, 2, 1, 3])]; + tensor var_3199 = const()[name = tensor("op_3199"), val = tensor([1, 1500, 20, -1])]; + tensor var_3200_cast_fp16 = reshape(shape = var_3199, x = linear_169_cast_fp16)[name = tensor("op_3200_cast_fp16")]; + tensor var_3201 = const()[name = tensor("op_3201"), val = tensor([0, 2, 1, 3])]; + tensor var_3205 = const()[name = tensor("op_3205"), val = tensor([1, 1500, 20, -1])]; + tensor var_3206_cast_fp16 = reshape(shape = var_3205, x = linear_170_cast_fp16)[name = tensor("op_3206_cast_fp16")]; + tensor var_3207 = const()[name = tensor("op_3207"), val = tensor([0, 2, 1, 3])]; + tensor scale_57_to_fp16 = const()[name = tensor("scale_57_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_19 = transpose(perm = var_3195, x = var_3194_cast_fp16)[name = tensor("transpose_19")]; + tensor var_3209_cast_fp16 = mul(x = transpose_19, y = scale_57_to_fp16)[name = tensor("op_3209_cast_fp16")]; + tensor transpose_18 = transpose(perm = var_3201, x = var_3200_cast_fp16)[name = tensor("transpose_18")]; + tensor var_3210_cast_fp16 = mul(x = transpose_18, y = scale_57_to_fp16)[name = tensor("op_3210_cast_fp16")]; + tensor var_3211_perm_0 = const()[name = tensor("op_3211_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_16 = transpose(perm = var_3211_perm_0, x = var_3210_cast_fp16)[name = tensor("transpose_16")]; + tensor qk_57_cast_fp16 = matmul(transpose_x = qk_57_transpose_x_0, transpose_y = qk_57_transpose_y_0, x = var_3209_cast_fp16, y = transpose_16)[name = tensor("qk_57_cast_fp16")]; + tensor var_3214_cast_fp16 = softmax(axis = var_3148, x = qk_57_cast_fp16)[name = tensor("op_3214_cast_fp16")]; + tensor var_3216_transpose_x_0 = const()[name = tensor("op_3216_transpose_x_0"), val = tensor(false)]; + tensor var_3216_transpose_y_0 = const()[name = tensor("op_3216_transpose_y_0"), val = tensor(false)]; + tensor transpose_17 = transpose(perm = var_3207, x = var_3206_cast_fp16)[name = tensor("transpose_17")]; + tensor var_3216_cast_fp16 = matmul(transpose_x = var_3216_transpose_x_0, transpose_y = var_3216_transpose_y_0, x = var_3214_cast_fp16, y = transpose_17)[name = tensor("op_3216_cast_fp16")]; + tensor var_3217 = const()[name = tensor("op_3217"), val = tensor([0, 2, 1, 3])]; + tensor concat_28 = const()[name = tensor("concat_28"), val = tensor([1, 1500, 1280])]; + tensor transpose_15 = transpose(perm = var_3217, x = var_3216_cast_fp16)[name = tensor("transpose_15")]; + tensor x_347_cast_fp16 = reshape(shape = concat_28, x = transpose_15)[name = tensor("x_347_cast_fp16")]; + tensor var_3222_to_fp16 = const()[name = tensor("op_3222_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1126394496)))]; + tensor var_3223_to_fp16 = const()[name = tensor("op_3223_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1129671360)))]; + tensor linear_171_cast_fp16 = linear(bias = var_3223_to_fp16, weight = var_3222_to_fp16, x = x_347_cast_fp16)[name = tensor("linear_171_cast_fp16")]; + tensor x_349_cast_fp16 = add(x = x_343_cast_fp16, y = linear_171_cast_fp16)[name = tensor("x_349_cast_fp16")]; + tensor var_3230_axes_0 = const()[name = tensor("op_3230_axes_0"), val = tensor([-1])]; + tensor blocks_28_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_28_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1129673984)))]; + tensor blocks_28_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_28_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1129676608)))]; + tensor var_3230_cast_fp16 = layer_norm(axes = var_3230_axes_0, beta = blocks_28_mlp_ln_bias_to_fp16, epsilon = var_3155_to_fp16, gamma = blocks_28_mlp_ln_weight_to_fp16, x = x_349_cast_fp16)[name = tensor("op_3230_cast_fp16")]; + tensor var_3239_to_fp16 = const()[name = tensor("op_3239_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1129679232)))]; + tensor var_3240_to_fp16 = const()[name = tensor("op_3240_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1142786496)))]; + tensor linear_172_cast_fp16 = linear(bias = var_3240_to_fp16, weight = var_3239_to_fp16, x = var_3230_cast_fp16)[name = tensor("linear_172_cast_fp16")]; + tensor x_353_mode_0 = const()[name = tensor("x_353_mode_0"), val = tensor("EXACT")]; + tensor x_353_cast_fp16 = gelu(mode = x_353_mode_0, x = linear_172_cast_fp16)[name = tensor("x_353_cast_fp16")]; + tensor var_3245_to_fp16 = const()[name = tensor("op_3245_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1142796800)))]; + tensor var_3246_to_fp16 = const()[name = tensor("op_3246_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1155904064)))]; + tensor linear_173_cast_fp16 = linear(bias = var_3246_to_fp16, weight = var_3245_to_fp16, x = x_353_cast_fp16)[name = tensor("linear_173_cast_fp16")]; + tensor x_355_cast_fp16 = add(x = x_349_cast_fp16, y = linear_173_cast_fp16)[name = tensor("x_355_cast_fp16")]; + tensor var_3256 = const()[name = tensor("op_3256"), val = tensor(-1)]; + tensor var_3273_axes_0 = const()[name = tensor("op_3273_axes_0"), val = tensor([-1])]; + tensor blocks_29_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_29_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1155906688)))]; + tensor blocks_29_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_29_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1155909312)))]; + tensor var_3263_to_fp16 = const()[name = tensor("op_3263_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3273_cast_fp16 = layer_norm(axes = var_3273_axes_0, beta = blocks_29_attn_ln_bias_to_fp16, epsilon = var_3263_to_fp16, gamma = blocks_29_attn_ln_weight_to_fp16, x = x_355_cast_fp16)[name = tensor("op_3273_cast_fp16")]; + tensor var_3284_to_fp16 = const()[name = tensor("op_3284_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1155911936)))]; + tensor var_3285_to_fp16 = const()[name = tensor("op_3285_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1159188800)))]; + tensor linear_174_cast_fp16 = linear(bias = var_3285_to_fp16, weight = var_3284_to_fp16, x = var_3273_cast_fp16)[name = tensor("linear_174_cast_fp16")]; + tensor var_3288_to_fp16 = const()[name = tensor("op_3288_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1159191424)))]; + tensor linear_175_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3288_to_fp16, x = var_3273_cast_fp16)[name = tensor("linear_175_cast_fp16")]; + tensor var_3292_to_fp16 = const()[name = tensor("op_3292_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1162468288)))]; + tensor var_3293_to_fp16 = const()[name = tensor("op_3293_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1165745152)))]; + tensor linear_176_cast_fp16 = linear(bias = var_3293_to_fp16, weight = var_3292_to_fp16, x = var_3273_cast_fp16)[name = tensor("linear_176_cast_fp16")]; + tensor var_3301 = const()[name = tensor("op_3301"), val = tensor([1, 1500, 20, -1])]; + tensor var_3302_cast_fp16 = reshape(shape = var_3301, x = linear_174_cast_fp16)[name = tensor("op_3302_cast_fp16")]; + tensor var_3303 = const()[name = tensor("op_3303"), val = tensor([0, 2, 1, 3])]; + tensor var_3307 = const()[name = tensor("op_3307"), val = tensor([1, 1500, 20, -1])]; + tensor var_3308_cast_fp16 = reshape(shape = var_3307, x = linear_175_cast_fp16)[name = tensor("op_3308_cast_fp16")]; + tensor var_3309 = const()[name = tensor("op_3309"), val = tensor([0, 2, 1, 3])]; + tensor var_3313 = const()[name = tensor("op_3313"), val = tensor([1, 1500, 20, -1])]; + tensor var_3314_cast_fp16 = reshape(shape = var_3313, x = linear_176_cast_fp16)[name = tensor("op_3314_cast_fp16")]; + tensor var_3315 = const()[name = tensor("op_3315"), val = tensor([0, 2, 1, 3])]; + tensor scale_59_to_fp16 = const()[name = tensor("scale_59_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_14 = transpose(perm = var_3303, x = var_3302_cast_fp16)[name = tensor("transpose_14")]; + tensor var_3317_cast_fp16 = mul(x = transpose_14, y = scale_59_to_fp16)[name = tensor("op_3317_cast_fp16")]; + tensor transpose_13 = transpose(perm = var_3309, x = var_3308_cast_fp16)[name = tensor("transpose_13")]; + tensor var_3318_cast_fp16 = mul(x = transpose_13, y = scale_59_to_fp16)[name = tensor("op_3318_cast_fp16")]; + tensor var_3319_perm_0 = const()[name = tensor("op_3319_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_11 = transpose(perm = var_3319_perm_0, x = var_3318_cast_fp16)[name = tensor("transpose_11")]; + tensor qk_59_cast_fp16 = matmul(transpose_x = qk_59_transpose_x_0, transpose_y = qk_59_transpose_y_0, x = var_3317_cast_fp16, y = transpose_11)[name = tensor("qk_59_cast_fp16")]; + tensor var_3322_cast_fp16 = softmax(axis = var_3256, x = qk_59_cast_fp16)[name = tensor("op_3322_cast_fp16")]; + tensor var_3324_transpose_x_0 = const()[name = tensor("op_3324_transpose_x_0"), val = tensor(false)]; + tensor var_3324_transpose_y_0 = const()[name = tensor("op_3324_transpose_y_0"), val = tensor(false)]; + tensor transpose_12 = transpose(perm = var_3315, x = var_3314_cast_fp16)[name = tensor("transpose_12")]; + tensor var_3324_cast_fp16 = matmul(transpose_x = var_3324_transpose_x_0, transpose_y = var_3324_transpose_y_0, x = var_3322_cast_fp16, y = transpose_12)[name = tensor("op_3324_cast_fp16")]; + tensor var_3325 = const()[name = tensor("op_3325"), val = tensor([0, 2, 1, 3])]; + tensor concat_29 = const()[name = tensor("concat_29"), val = tensor([1, 1500, 1280])]; + tensor transpose_10 = transpose(perm = var_3325, x = var_3324_cast_fp16)[name = tensor("transpose_10")]; + tensor x_359_cast_fp16 = reshape(shape = concat_29, x = transpose_10)[name = tensor("x_359_cast_fp16")]; + tensor var_3330_to_fp16 = const()[name = tensor("op_3330_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1165747776)))]; + tensor var_3331_to_fp16 = const()[name = tensor("op_3331_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1169024640)))]; + tensor linear_177_cast_fp16 = linear(bias = var_3331_to_fp16, weight = var_3330_to_fp16, x = x_359_cast_fp16)[name = tensor("linear_177_cast_fp16")]; + tensor x_361_cast_fp16 = add(x = x_355_cast_fp16, y = linear_177_cast_fp16)[name = tensor("x_361_cast_fp16")]; + tensor var_3338_axes_0 = const()[name = tensor("op_3338_axes_0"), val = tensor([-1])]; + tensor blocks_29_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_29_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1169027264)))]; + tensor blocks_29_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_29_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1169029888)))]; + tensor var_3338_cast_fp16 = layer_norm(axes = var_3338_axes_0, beta = blocks_29_mlp_ln_bias_to_fp16, epsilon = var_3263_to_fp16, gamma = blocks_29_mlp_ln_weight_to_fp16, x = x_361_cast_fp16)[name = tensor("op_3338_cast_fp16")]; + tensor var_3347_to_fp16 = const()[name = tensor("op_3347_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1169032512)))]; + tensor var_3348_to_fp16 = const()[name = tensor("op_3348_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1182139776)))]; + tensor linear_178_cast_fp16 = linear(bias = var_3348_to_fp16, weight = var_3347_to_fp16, x = var_3338_cast_fp16)[name = tensor("linear_178_cast_fp16")]; + tensor x_365_mode_0 = const()[name = tensor("x_365_mode_0"), val = tensor("EXACT")]; + tensor x_365_cast_fp16 = gelu(mode = x_365_mode_0, x = linear_178_cast_fp16)[name = tensor("x_365_cast_fp16")]; + tensor var_3353_to_fp16 = const()[name = tensor("op_3353_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1182150080)))]; + tensor var_3354_to_fp16 = const()[name = tensor("op_3354_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1195257344)))]; + tensor linear_179_cast_fp16 = linear(bias = var_3354_to_fp16, weight = var_3353_to_fp16, x = x_365_cast_fp16)[name = tensor("linear_179_cast_fp16")]; + tensor x_367_cast_fp16 = add(x = x_361_cast_fp16, y = linear_179_cast_fp16)[name = tensor("x_367_cast_fp16")]; + tensor var_3364 = const()[name = tensor("op_3364"), val = tensor(-1)]; + tensor var_3381_axes_0 = const()[name = tensor("op_3381_axes_0"), val = tensor([-1])]; + tensor blocks_30_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_30_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1195259968)))]; + tensor blocks_30_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_30_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1195262592)))]; + tensor var_3371_to_fp16 = const()[name = tensor("op_3371_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3381_cast_fp16 = layer_norm(axes = var_3381_axes_0, beta = blocks_30_attn_ln_bias_to_fp16, epsilon = var_3371_to_fp16, gamma = blocks_30_attn_ln_weight_to_fp16, x = x_367_cast_fp16)[name = tensor("op_3381_cast_fp16")]; + tensor var_3392_to_fp16 = const()[name = tensor("op_3392_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1195265216)))]; + tensor var_3393_to_fp16 = const()[name = tensor("op_3393_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1198542080)))]; + tensor linear_180_cast_fp16 = linear(bias = var_3393_to_fp16, weight = var_3392_to_fp16, x = var_3381_cast_fp16)[name = tensor("linear_180_cast_fp16")]; + tensor var_3396_to_fp16 = const()[name = tensor("op_3396_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1198544704)))]; + tensor linear_181_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3396_to_fp16, x = var_3381_cast_fp16)[name = tensor("linear_181_cast_fp16")]; + tensor var_3400_to_fp16 = const()[name = tensor("op_3400_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1201821568)))]; + tensor var_3401_to_fp16 = const()[name = tensor("op_3401_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1205098432)))]; + tensor linear_182_cast_fp16 = linear(bias = var_3401_to_fp16, weight = var_3400_to_fp16, x = var_3381_cast_fp16)[name = tensor("linear_182_cast_fp16")]; + tensor var_3409 = const()[name = tensor("op_3409"), val = tensor([1, 1500, 20, -1])]; + tensor var_3410_cast_fp16 = reshape(shape = var_3409, x = linear_180_cast_fp16)[name = tensor("op_3410_cast_fp16")]; + tensor var_3411 = const()[name = tensor("op_3411"), val = tensor([0, 2, 1, 3])]; + tensor var_3415 = const()[name = tensor("op_3415"), val = tensor([1, 1500, 20, -1])]; + tensor var_3416_cast_fp16 = reshape(shape = var_3415, x = linear_181_cast_fp16)[name = tensor("op_3416_cast_fp16")]; + tensor var_3417 = const()[name = tensor("op_3417"), val = tensor([0, 2, 1, 3])]; + tensor var_3421 = const()[name = tensor("op_3421"), val = tensor([1, 1500, 20, -1])]; + tensor var_3422_cast_fp16 = reshape(shape = var_3421, x = linear_182_cast_fp16)[name = tensor("op_3422_cast_fp16")]; + tensor var_3423 = const()[name = tensor("op_3423"), val = tensor([0, 2, 1, 3])]; + tensor scale_61_to_fp16 = const()[name = tensor("scale_61_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_9 = transpose(perm = var_3411, x = var_3410_cast_fp16)[name = tensor("transpose_9")]; + tensor var_3425_cast_fp16 = mul(x = transpose_9, y = scale_61_to_fp16)[name = tensor("op_3425_cast_fp16")]; + tensor transpose_8 = transpose(perm = var_3417, x = var_3416_cast_fp16)[name = tensor("transpose_8")]; + tensor var_3426_cast_fp16 = mul(x = transpose_8, y = scale_61_to_fp16)[name = tensor("op_3426_cast_fp16")]; + tensor var_3427_perm_0 = const()[name = tensor("op_3427_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_6 = transpose(perm = var_3427_perm_0, x = var_3426_cast_fp16)[name = tensor("transpose_6")]; + tensor qk_61_cast_fp16 = matmul(transpose_x = qk_61_transpose_x_0, transpose_y = qk_61_transpose_y_0, x = var_3425_cast_fp16, y = transpose_6)[name = tensor("qk_61_cast_fp16")]; + tensor var_3430_cast_fp16 = softmax(axis = var_3364, x = qk_61_cast_fp16)[name = tensor("op_3430_cast_fp16")]; + tensor var_3432_transpose_x_0 = const()[name = tensor("op_3432_transpose_x_0"), val = tensor(false)]; + tensor var_3432_transpose_y_0 = const()[name = tensor("op_3432_transpose_y_0"), val = tensor(false)]; + tensor transpose_7 = transpose(perm = var_3423, x = var_3422_cast_fp16)[name = tensor("transpose_7")]; + tensor var_3432_cast_fp16 = matmul(transpose_x = var_3432_transpose_x_0, transpose_y = var_3432_transpose_y_0, x = var_3430_cast_fp16, y = transpose_7)[name = tensor("op_3432_cast_fp16")]; + tensor var_3433 = const()[name = tensor("op_3433"), val = tensor([0, 2, 1, 3])]; + tensor concat_30 = const()[name = tensor("concat_30"), val = tensor([1, 1500, 1280])]; + tensor transpose_5 = transpose(perm = var_3433, x = var_3432_cast_fp16)[name = tensor("transpose_5")]; + tensor x_371_cast_fp16 = reshape(shape = concat_30, x = transpose_5)[name = tensor("x_371_cast_fp16")]; + tensor var_3438_to_fp16 = const()[name = tensor("op_3438_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1205101056)))]; + tensor var_3439_to_fp16 = const()[name = tensor("op_3439_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208377920)))]; + tensor linear_183_cast_fp16 = linear(bias = var_3439_to_fp16, weight = var_3438_to_fp16, x = x_371_cast_fp16)[name = tensor("linear_183_cast_fp16")]; + tensor x_373_cast_fp16 = add(x = x_367_cast_fp16, y = linear_183_cast_fp16)[name = tensor("x_373_cast_fp16")]; + tensor var_3446_axes_0 = const()[name = tensor("op_3446_axes_0"), val = tensor([-1])]; + tensor blocks_30_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_30_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208380544)))]; + tensor blocks_30_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_30_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208383168)))]; + tensor var_3446_cast_fp16 = layer_norm(axes = var_3446_axes_0, beta = blocks_30_mlp_ln_bias_to_fp16, epsilon = var_3371_to_fp16, gamma = blocks_30_mlp_ln_weight_to_fp16, x = x_373_cast_fp16)[name = tensor("op_3446_cast_fp16")]; + tensor var_3455_to_fp16 = const()[name = tensor("op_3455_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208385792)))]; + tensor var_3456_to_fp16 = const()[name = tensor("op_3456_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1221493056)))]; + tensor linear_184_cast_fp16 = linear(bias = var_3456_to_fp16, weight = var_3455_to_fp16, x = var_3446_cast_fp16)[name = tensor("linear_184_cast_fp16")]; + tensor x_377_mode_0 = const()[name = tensor("x_377_mode_0"), val = tensor("EXACT")]; + tensor x_377_cast_fp16 = gelu(mode = x_377_mode_0, x = linear_184_cast_fp16)[name = tensor("x_377_cast_fp16")]; + tensor var_3461_to_fp16 = const()[name = tensor("op_3461_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1221503360)))]; + tensor var_3462_to_fp16 = const()[name = tensor("op_3462_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1234610624)))]; + tensor linear_185_cast_fp16 = linear(bias = var_3462_to_fp16, weight = var_3461_to_fp16, x = x_377_cast_fp16)[name = tensor("linear_185_cast_fp16")]; + tensor x_379_cast_fp16 = add(x = x_373_cast_fp16, y = linear_185_cast_fp16)[name = tensor("x_379_cast_fp16")]; + tensor var_3472 = const()[name = tensor("op_3472"), val = tensor(-1)]; + tensor var_3489_axes_0 = const()[name = tensor("op_3489_axes_0"), val = tensor([-1])]; + tensor blocks_31_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_31_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1234613248)))]; + tensor blocks_31_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_31_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1234615872)))]; + tensor var_3479_to_fp16 = const()[name = tensor("op_3479_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3489_cast_fp16 = layer_norm(axes = var_3489_axes_0, beta = blocks_31_attn_ln_bias_to_fp16, epsilon = var_3479_to_fp16, gamma = blocks_31_attn_ln_weight_to_fp16, x = x_379_cast_fp16)[name = tensor("op_3489_cast_fp16")]; + tensor var_3500_to_fp16 = const()[name = tensor("op_3500_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1234618496)))]; + tensor var_3501_to_fp16 = const()[name = tensor("op_3501_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1237895360)))]; + tensor linear_186_cast_fp16 = linear(bias = var_3501_to_fp16, weight = var_3500_to_fp16, x = var_3489_cast_fp16)[name = tensor("linear_186_cast_fp16")]; + tensor var_3504_to_fp16 = const()[name = tensor("op_3504_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1237897984)))]; + tensor linear_187_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3504_to_fp16, x = var_3489_cast_fp16)[name = tensor("linear_187_cast_fp16")]; + tensor var_3508_to_fp16 = const()[name = tensor("op_3508_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1241174848)))]; + tensor var_3509_to_fp16 = const()[name = tensor("op_3509_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1244451712)))]; + tensor linear_188_cast_fp16 = linear(bias = var_3509_to_fp16, weight = var_3508_to_fp16, x = var_3489_cast_fp16)[name = tensor("linear_188_cast_fp16")]; + tensor var_3517 = const()[name = tensor("op_3517"), val = tensor([1, 1500, 20, -1])]; + tensor var_3518_cast_fp16 = reshape(shape = var_3517, x = linear_186_cast_fp16)[name = tensor("op_3518_cast_fp16")]; + tensor var_3519 = const()[name = tensor("op_3519"), val = tensor([0, 2, 1, 3])]; + tensor var_3523 = const()[name = tensor("op_3523"), val = tensor([1, 1500, 20, -1])]; + tensor var_3524_cast_fp16 = reshape(shape = var_3523, x = linear_187_cast_fp16)[name = tensor("op_3524_cast_fp16")]; + tensor var_3525 = const()[name = tensor("op_3525"), val = tensor([0, 2, 1, 3])]; + tensor var_3529 = const()[name = tensor("op_3529"), val = tensor([1, 1500, 20, -1])]; + tensor var_3530_cast_fp16 = reshape(shape = var_3529, x = linear_188_cast_fp16)[name = tensor("op_3530_cast_fp16")]; + tensor var_3531 = const()[name = tensor("op_3531"), val = tensor([0, 2, 1, 3])]; + tensor scale_to_fp16 = const()[name = tensor("scale_to_fp16"), val = tensor([0x1.6ap-2])]; + tensor transpose_4 = transpose(perm = var_3519, x = var_3518_cast_fp16)[name = tensor("transpose_4")]; + tensor var_3533_cast_fp16 = mul(x = transpose_4, y = scale_to_fp16)[name = tensor("op_3533_cast_fp16")]; + tensor transpose_3 = transpose(perm = var_3525, x = var_3524_cast_fp16)[name = tensor("transpose_3")]; + tensor var_3534_cast_fp16 = mul(x = transpose_3, y = scale_to_fp16)[name = tensor("op_3534_cast_fp16")]; + tensor var_3535_perm_0 = const()[name = tensor("op_3535_perm_0"), val = tensor([0, 1, -1, -2])]; + 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_1 = transpose(perm = var_3535_perm_0, x = var_3534_cast_fp16)[name = tensor("transpose_1")]; + tensor qk_cast_fp16 = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = var_3533_cast_fp16, y = transpose_1)[name = tensor("qk_cast_fp16")]; + tensor var_3538_cast_fp16 = softmax(axis = var_3472, x = qk_cast_fp16)[name = tensor("op_3538_cast_fp16")]; + tensor var_3540_transpose_x_0 = const()[name = tensor("op_3540_transpose_x_0"), val = tensor(false)]; + tensor var_3540_transpose_y_0 = const()[name = tensor("op_3540_transpose_y_0"), val = tensor(false)]; + tensor transpose_2 = transpose(perm = var_3531, x = var_3530_cast_fp16)[name = tensor("transpose_2")]; + tensor var_3540_cast_fp16 = matmul(transpose_x = var_3540_transpose_x_0, transpose_y = var_3540_transpose_y_0, x = var_3538_cast_fp16, y = transpose_2)[name = tensor("op_3540_cast_fp16")]; + tensor var_3541 = const()[name = tensor("op_3541"), val = tensor([0, 2, 1, 3])]; + tensor concat_31 = const()[name = tensor("concat_31"), val = tensor([1, 1500, 1280])]; + tensor transpose_0 = transpose(perm = var_3541, x = var_3540_cast_fp16)[name = tensor("transpose_0")]; + tensor x_383_cast_fp16 = reshape(shape = concat_31, x = transpose_0)[name = tensor("x_383_cast_fp16")]; + tensor var_3546_to_fp16 = const()[name = tensor("op_3546_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1244454336)))]; + tensor var_3547_to_fp16 = const()[name = tensor("op_3547_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1247731200)))]; + tensor linear_189_cast_fp16 = linear(bias = var_3547_to_fp16, weight = var_3546_to_fp16, x = x_383_cast_fp16)[name = tensor("linear_189_cast_fp16")]; + tensor x_385_cast_fp16 = add(x = x_379_cast_fp16, y = linear_189_cast_fp16)[name = tensor("x_385_cast_fp16")]; + tensor var_3554_axes_0 = const()[name = tensor("op_3554_axes_0"), val = tensor([-1])]; + tensor blocks_31_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_31_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1247733824)))]; + tensor blocks_31_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_31_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1247736448)))]; + tensor var_3554_cast_fp16 = layer_norm(axes = var_3554_axes_0, beta = blocks_31_mlp_ln_bias_to_fp16, epsilon = var_3479_to_fp16, gamma = blocks_31_mlp_ln_weight_to_fp16, x = x_385_cast_fp16)[name = tensor("op_3554_cast_fp16")]; + tensor var_3563_to_fp16 = const()[name = tensor("op_3563_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1247739072)))]; + tensor var_3564_to_fp16 = const()[name = tensor("op_3564_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1260846336)))]; + tensor linear_190_cast_fp16 = linear(bias = var_3564_to_fp16, weight = var_3563_to_fp16, x = var_3554_cast_fp16)[name = tensor("linear_190_cast_fp16")]; + tensor x_389_mode_0 = const()[name = tensor("x_389_mode_0"), val = tensor("EXACT")]; + tensor x_389_cast_fp16 = gelu(mode = x_389_mode_0, x = linear_190_cast_fp16)[name = tensor("x_389_cast_fp16")]; + tensor var_3569_to_fp16 = const()[name = tensor("op_3569_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1260856640)))]; + tensor var_3570_to_fp16 = const()[name = tensor("op_3570_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1273963904)))]; + tensor linear_191_cast_fp16 = linear(bias = var_3570_to_fp16, weight = var_3569_to_fp16, x = x_389_cast_fp16)[name = tensor("linear_191_cast_fp16")]; + tensor x_cast_fp16 = add(x = x_385_cast_fp16, y = linear_191_cast_fp16)[name = tensor("x_cast_fp16")]; + tensor var_3583_axes_0 = const()[name = tensor("op_3583_axes_0"), val = tensor([-1])]; + tensor ln_post_weight_to_fp16 = const()[name = tensor("ln_post_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1273966528)))]; + tensor ln_post_bias_to_fp16 = const()[name = tensor("ln_post_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1273969152)))]; + tensor var_3574_to_fp16 = const()[name = tensor("op_3574_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3583_cast_fp16 = layer_norm(axes = var_3583_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_3574_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast_fp16)[name = tensor("op_3583_cast_fp16")]; + tensor var_3583_cast_fp16_to_fp32_dtype_0 = const()[name = tensor("op_3583_cast_fp16_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor output = cast(dtype = var_3583_cast_fp16_to_fp32_dtype_0, x = var_3583_cast_fp16)[name = tensor("cast_128")]; + } -> (output); +} \ No newline at end of file