diff --git "a/large-v2/encoder.mlmodelc/model0/model.mil" "b/large-v2/encoder.mlmodelc/model0/model.mil" new file mode 100644--- /dev/null +++ "b/large-v2/encoder.mlmodelc/model0/model.mil" @@ -0,0 +1,962 @@ +program(1.3) +[buildInfo = dict({{"coremlc-component-MIL", "3400.43.1"}, {"coremlc-version", "3400.58.2"}})] +{ + func main(tensor logmel_data) { + string var_84_pad_type_0 = const()[name = string("op_84_pad_type_0"), val = string("custom")]; + tensor var_84_pad_0 = const()[name = string("op_84_pad_0"), val = tensor([1, 1])]; + tensor var_84_strides_0 = const()[name = string("op_84_strides_0"), val = tensor([1])]; + tensor var_84_dilations_0 = const()[name = string("op_84_dilations_0"), val = tensor([1])]; + int32 var_84_groups_0 = const()[name = string("op_84_groups_0"), val = int32(1)]; + tensor weight_3_to_fp16 = const()[name = string("weight_3_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(64)))]; + tensor bias_3_to_fp16 = const()[name = string("bias_3_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(614528)))]; + tensor var_84_cast_fp16 = conv(bias = bias_3_to_fp16, dilations = var_84_dilations_0, groups = var_84_groups_0, pad = var_84_pad_0, pad_type = var_84_pad_type_0, strides = var_84_strides_0, weight = weight_3_to_fp16, x = logmel_data)[name = string("op_84_cast_fp16")]; + string input_1_mode_0 = const()[name = string("input_1_mode_0"), val = string("EXACT")]; + tensor input_1_cast_fp16 = gelu(mode = input_1_mode_0, x = var_84_cast_fp16)[name = string("input_1_cast_fp16")]; + string var_102_pad_type_0 = const()[name = string("op_102_pad_type_0"), val = string("custom")]; + tensor var_102_pad_0 = const()[name = string("op_102_pad_0"), val = tensor([1, 1])]; + tensor var_102_strides_0 = const()[name = string("op_102_strides_0"), val = tensor([2])]; + tensor var_102_dilations_0 = const()[name = string("op_102_dilations_0"), val = tensor([1])]; + int32 var_102_groups_0 = const()[name = string("op_102_groups_0"), val = int32(1)]; + tensor weight_7_to_fp16 = const()[name = string("weight_7_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(617152)))]; + tensor bias_7_to_fp16 = const()[name = string("bias_7_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(10447616)))]; + tensor var_102_cast_fp16 = conv(bias = bias_7_to_fp16, dilations = var_102_dilations_0, groups = var_102_groups_0, pad = var_102_pad_0, pad_type = var_102_pad_type_0, strides = var_102_strides_0, weight = weight_7_to_fp16, x = input_1_cast_fp16)[name = string("op_102_cast_fp16")]; + string x_3_mode_0 = const()[name = string("x_3_mode_0"), val = string("EXACT")]; + tensor x_3_cast_fp16 = gelu(mode = x_3_mode_0, x = var_102_cast_fp16)[name = string("x_3_cast_fp16")]; + tensor var_108 = const()[name = string("op_108"), val = tensor([0, 2, 1])]; + tensor positional_embedding_to_fp16 = const()[name = string("positional_embedding_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(10450240)))]; + tensor x_5_cast_fp16 = transpose(perm = var_108, x = x_3_cast_fp16)[name = string("transpose_160")]; + tensor var_111_cast_fp16 = add(x = x_5_cast_fp16, y = positional_embedding_to_fp16)[name = string("op_111_cast_fp16")]; + int32 var_124 = const()[name = string("op_124"), val = int32(-1)]; + tensor var_140_axes_0 = const()[name = string("op_140_axes_0"), val = tensor([-1])]; + tensor blocks_0_attn_ln_weight_to_fp16 = const()[name = string("blocks_0_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(14290304)))]; + tensor blocks_0_attn_ln_bias_to_fp16 = const()[name = string("blocks_0_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(14292928)))]; + fp16 var_130_to_fp16 = const()[name = string("op_130_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_140_cast_fp16 = layer_norm(axes = var_140_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_130_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_111_cast_fp16)[name = string("op_140_cast_fp16")]; + tensor var_151_to_fp16 = const()[name = string("op_151_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(14295552)))]; + tensor var_152_to_fp16 = const()[name = string("op_152_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(17572416)))]; + tensor linear_0_cast_fp16 = linear(bias = var_152_to_fp16, weight = var_151_to_fp16, x = var_140_cast_fp16)[name = string("linear_0_cast_fp16")]; + tensor var_155_to_fp16 = const()[name = string("op_155_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(17575040)))]; + tensor linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(20851904)))]; + tensor linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_155_to_fp16, x = var_140_cast_fp16)[name = string("linear_1_cast_fp16")]; + tensor var_159_to_fp16 = const()[name = string("op_159_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(20854528)))]; + tensor var_160_to_fp16 = const()[name = string("op_160_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(24131392)))]; + tensor linear_2_cast_fp16 = linear(bias = var_160_to_fp16, weight = var_159_to_fp16, x = var_140_cast_fp16)[name = string("linear_2_cast_fp16")]; + tensor var_168 = const()[name = string("op_168"), val = tensor([1, 1500, 20, -1])]; + tensor var_169_cast_fp16 = reshape(shape = var_168, x = linear_0_cast_fp16)[name = string("op_169_cast_fp16")]; + tensor const_224_to_fp16 = const()[name = string("const_224_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_3_cast_fp16 = mul(x = var_169_cast_fp16, y = const_224_to_fp16)[name = string("q_3_cast_fp16")]; + tensor var_175 = const()[name = string("op_175"), val = tensor([1, 1500, 20, -1])]; + tensor var_176_cast_fp16 = reshape(shape = var_175, x = linear_1_cast_fp16)[name = string("op_176_cast_fp16")]; + tensor const_225_to_fp16 = const()[name = string("const_225_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_3_cast_fp16 = mul(x = var_176_cast_fp16, y = const_225_to_fp16)[name = string("k_3_cast_fp16")]; + tensor var_182 = const()[name = string("op_182"), val = tensor([1, 1500, 20, -1])]; + tensor var_183_cast_fp16 = reshape(shape = var_182, x = linear_2_cast_fp16)[name = string("op_183_cast_fp16")]; + tensor var_184 = const()[name = string("op_184"), val = tensor([0, 2, -3, -1])]; + bool qk_1_transpose_x_0 = const()[name = string("qk_1_transpose_x_0"), val = bool(false)]; + bool qk_1_transpose_y_0 = const()[name = string("qk_1_transpose_y_0"), val = bool(false)]; + tensor transpose_64_perm_0 = const()[name = string("transpose_64_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_65_perm_0 = const()[name = string("transpose_65_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_65 = transpose(perm = transpose_65_perm_0, x = k_3_cast_fp16)[name = string("transpose_158")]; + tensor transpose_64 = transpose(perm = transpose_64_perm_0, x = q_3_cast_fp16)[name = string("transpose_159")]; + tensor qk_1_cast_fp16 = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_64, y = transpose_65)[name = string("qk_1_cast_fp16")]; + tensor var_188_cast_fp16 = softmax(axis = var_124, x = qk_1_cast_fp16)[name = string("op_188_cast_fp16")]; + bool var_190_transpose_x_0 = const()[name = string("op_190_transpose_x_0"), val = bool(false)]; + bool var_190_transpose_y_0 = const()[name = string("op_190_transpose_y_0"), val = bool(false)]; + tensor v_3_cast_fp16 = transpose(perm = var_184, x = var_183_cast_fp16)[name = string("transpose_157")]; + tensor var_190_cast_fp16 = matmul(transpose_x = var_190_transpose_x_0, transpose_y = var_190_transpose_y_0, x = var_188_cast_fp16, y = v_3_cast_fp16)[name = string("op_190_cast_fp16")]; + tensor var_191 = const()[name = string("op_191"), val = tensor([0, 2, 1, 3])]; + tensor concat_0 = const()[name = string("concat_0"), val = tensor([1, 1500, 1280])]; + tensor var_192_cast_fp16 = transpose(perm = var_191, x = var_190_cast_fp16)[name = string("transpose_156")]; + tensor x_11_cast_fp16 = reshape(shape = concat_0, x = var_192_cast_fp16)[name = string("x_11_cast_fp16")]; + tensor var_196_to_fp16 = const()[name = string("op_196_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(24134016)))]; + tensor var_197_to_fp16 = const()[name = string("op_197_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(27410880)))]; + tensor linear_3_cast_fp16 = linear(bias = var_197_to_fp16, weight = var_196_to_fp16, x = x_11_cast_fp16)[name = string("linear_3_cast_fp16")]; + tensor x_13_cast_fp16 = add(x = var_111_cast_fp16, y = linear_3_cast_fp16)[name = string("x_13_cast_fp16")]; + tensor var_204_axes_0 = const()[name = string("op_204_axes_0"), val = tensor([-1])]; + tensor blocks_0_mlp_ln_weight_to_fp16 = const()[name = string("blocks_0_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(27413504)))]; + tensor blocks_0_mlp_ln_bias_to_fp16 = const()[name = string("blocks_0_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(27416128)))]; + tensor var_204_cast_fp16 = layer_norm(axes = var_204_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_130_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast_fp16)[name = string("op_204_cast_fp16")]; + tensor var_213_to_fp16 = const()[name = string("op_213_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(27418752)))]; + tensor var_214_to_fp16 = const()[name = string("op_214_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(40526016)))]; + tensor linear_4_cast_fp16 = linear(bias = var_214_to_fp16, weight = var_213_to_fp16, x = var_204_cast_fp16)[name = string("linear_4_cast_fp16")]; + string x_17_mode_0 = const()[name = string("x_17_mode_0"), val = string("EXACT")]; + tensor x_17_cast_fp16 = gelu(mode = x_17_mode_0, x = linear_4_cast_fp16)[name = string("x_17_cast_fp16")]; + tensor var_219_to_fp16 = const()[name = string("op_219_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(40536320)))]; + tensor var_220_to_fp16 = const()[name = string("op_220_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(53643584)))]; + tensor linear_5_cast_fp16 = linear(bias = var_220_to_fp16, weight = var_219_to_fp16, x = x_17_cast_fp16)[name = string("linear_5_cast_fp16")]; + tensor x_19_cast_fp16 = add(x = x_13_cast_fp16, y = linear_5_cast_fp16)[name = string("x_19_cast_fp16")]; + int32 var_230 = const()[name = string("op_230"), val = int32(-1)]; + tensor var_246_axes_0 = const()[name = string("op_246_axes_0"), val = tensor([-1])]; + tensor blocks_1_attn_ln_weight_to_fp16 = const()[name = string("blocks_1_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(53646208)))]; + tensor blocks_1_attn_ln_bias_to_fp16 = const()[name = string("blocks_1_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(53648832)))]; + fp16 var_236_to_fp16 = const()[name = string("op_236_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_246_cast_fp16 = layer_norm(axes = var_246_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_236_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast_fp16)[name = string("op_246_cast_fp16")]; + tensor var_257_to_fp16 = const()[name = string("op_257_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(53651456)))]; + tensor var_258_to_fp16 = const()[name = string("op_258_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(56928320)))]; + tensor linear_6_cast_fp16 = linear(bias = var_258_to_fp16, weight = var_257_to_fp16, x = var_246_cast_fp16)[name = string("linear_6_cast_fp16")]; + tensor var_261_to_fp16 = const()[name = string("op_261_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(56930944)))]; + tensor linear_7_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_261_to_fp16, x = var_246_cast_fp16)[name = string("linear_7_cast_fp16")]; + tensor var_265_to_fp16 = const()[name = string("op_265_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(60207808)))]; + tensor var_266_to_fp16 = const()[name = string("op_266_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(63484672)))]; + tensor linear_8_cast_fp16 = linear(bias = var_266_to_fp16, weight = var_265_to_fp16, x = var_246_cast_fp16)[name = string("linear_8_cast_fp16")]; + tensor var_274 = const()[name = string("op_274"), val = tensor([1, 1500, 20, -1])]; + tensor var_275_cast_fp16 = reshape(shape = var_274, x = linear_6_cast_fp16)[name = string("op_275_cast_fp16")]; + tensor const_226_to_fp16 = const()[name = string("const_226_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_7_cast_fp16 = mul(x = var_275_cast_fp16, y = const_226_to_fp16)[name = string("q_7_cast_fp16")]; + tensor var_281 = const()[name = string("op_281"), val = tensor([1, 1500, 20, -1])]; + tensor var_282_cast_fp16 = reshape(shape = var_281, x = linear_7_cast_fp16)[name = string("op_282_cast_fp16")]; + tensor const_227_to_fp16 = const()[name = string("const_227_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_7_cast_fp16 = mul(x = var_282_cast_fp16, y = const_227_to_fp16)[name = string("k_7_cast_fp16")]; + tensor var_288 = const()[name = string("op_288"), val = tensor([1, 1500, 20, -1])]; + tensor var_289_cast_fp16 = reshape(shape = var_288, x = linear_8_cast_fp16)[name = string("op_289_cast_fp16")]; + tensor var_290 = const()[name = string("op_290"), val = tensor([0, 2, -3, -1])]; + bool qk_3_transpose_x_0 = const()[name = string("qk_3_transpose_x_0"), val = bool(false)]; + bool qk_3_transpose_y_0 = const()[name = string("qk_3_transpose_y_0"), val = bool(false)]; + tensor transpose_66_perm_0 = const()[name = string("transpose_66_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_67_perm_0 = const()[name = string("transpose_67_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_67 = transpose(perm = transpose_67_perm_0, x = k_7_cast_fp16)[name = string("transpose_154")]; + tensor transpose_66 = transpose(perm = transpose_66_perm_0, x = q_7_cast_fp16)[name = string("transpose_155")]; + tensor qk_3_cast_fp16 = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_66, y = transpose_67)[name = string("qk_3_cast_fp16")]; + tensor var_294_cast_fp16 = softmax(axis = var_230, x = qk_3_cast_fp16)[name = string("op_294_cast_fp16")]; + bool var_296_transpose_x_0 = const()[name = string("op_296_transpose_x_0"), val = bool(false)]; + bool var_296_transpose_y_0 = const()[name = string("op_296_transpose_y_0"), val = bool(false)]; + tensor v_7_cast_fp16 = transpose(perm = var_290, x = var_289_cast_fp16)[name = string("transpose_153")]; + tensor var_296_cast_fp16 = matmul(transpose_x = var_296_transpose_x_0, transpose_y = var_296_transpose_y_0, x = var_294_cast_fp16, y = v_7_cast_fp16)[name = string("op_296_cast_fp16")]; + tensor var_297 = const()[name = string("op_297"), val = tensor([0, 2, 1, 3])]; + tensor concat_1 = const()[name = string("concat_1"), val = tensor([1, 1500, 1280])]; + tensor var_298_cast_fp16 = transpose(perm = var_297, x = var_296_cast_fp16)[name = string("transpose_152")]; + tensor x_23_cast_fp16 = reshape(shape = concat_1, x = var_298_cast_fp16)[name = string("x_23_cast_fp16")]; + tensor var_302_to_fp16 = const()[name = string("op_302_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(63487296)))]; + tensor var_303_to_fp16 = const()[name = string("op_303_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(66764160)))]; + tensor linear_9_cast_fp16 = linear(bias = var_303_to_fp16, weight = var_302_to_fp16, x = x_23_cast_fp16)[name = string("linear_9_cast_fp16")]; + tensor x_25_cast_fp16 = add(x = x_19_cast_fp16, y = linear_9_cast_fp16)[name = string("x_25_cast_fp16")]; + tensor var_310_axes_0 = const()[name = string("op_310_axes_0"), val = tensor([-1])]; + tensor blocks_1_mlp_ln_weight_to_fp16 = const()[name = string("blocks_1_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(66766784)))]; + tensor blocks_1_mlp_ln_bias_to_fp16 = const()[name = string("blocks_1_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(66769408)))]; + tensor var_310_cast_fp16 = layer_norm(axes = var_310_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_236_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast_fp16)[name = string("op_310_cast_fp16")]; + tensor var_319_to_fp16 = const()[name = string("op_319_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(66772032)))]; + tensor var_320_to_fp16 = const()[name = string("op_320_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(79879296)))]; + tensor linear_10_cast_fp16 = linear(bias = var_320_to_fp16, weight = var_319_to_fp16, x = var_310_cast_fp16)[name = string("linear_10_cast_fp16")]; + string x_29_mode_0 = const()[name = string("x_29_mode_0"), val = string("EXACT")]; + tensor x_29_cast_fp16 = gelu(mode = x_29_mode_0, x = linear_10_cast_fp16)[name = string("x_29_cast_fp16")]; + tensor var_325_to_fp16 = const()[name = string("op_325_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(79889600)))]; + tensor var_326_to_fp16 = const()[name = string("op_326_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(92996864)))]; + tensor linear_11_cast_fp16 = linear(bias = var_326_to_fp16, weight = var_325_to_fp16, x = x_29_cast_fp16)[name = string("linear_11_cast_fp16")]; + tensor x_31_cast_fp16 = add(x = x_25_cast_fp16, y = linear_11_cast_fp16)[name = string("x_31_cast_fp16")]; + int32 var_336 = const()[name = string("op_336"), val = int32(-1)]; + tensor var_352_axes_0 = const()[name = string("op_352_axes_0"), val = tensor([-1])]; + tensor blocks_2_attn_ln_weight_to_fp16 = const()[name = string("blocks_2_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(92999488)))]; + tensor blocks_2_attn_ln_bias_to_fp16 = const()[name = string("blocks_2_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(93002112)))]; + fp16 var_342_to_fp16 = const()[name = string("op_342_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_352_cast_fp16 = layer_norm(axes = var_352_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_342_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast_fp16)[name = string("op_352_cast_fp16")]; + tensor var_363_to_fp16 = const()[name = string("op_363_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(93004736)))]; + tensor var_364_to_fp16 = const()[name = string("op_364_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(96281600)))]; + tensor linear_12_cast_fp16 = linear(bias = var_364_to_fp16, weight = var_363_to_fp16, x = var_352_cast_fp16)[name = string("linear_12_cast_fp16")]; + tensor var_367_to_fp16 = const()[name = string("op_367_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(96284224)))]; + tensor linear_13_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_367_to_fp16, x = var_352_cast_fp16)[name = string("linear_13_cast_fp16")]; + tensor var_371_to_fp16 = const()[name = string("op_371_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(99561088)))]; + tensor var_372_to_fp16 = const()[name = string("op_372_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(102837952)))]; + tensor linear_14_cast_fp16 = linear(bias = var_372_to_fp16, weight = var_371_to_fp16, x = var_352_cast_fp16)[name = string("linear_14_cast_fp16")]; + tensor var_380 = const()[name = string("op_380"), val = tensor([1, 1500, 20, -1])]; + tensor var_381_cast_fp16 = reshape(shape = var_380, x = linear_12_cast_fp16)[name = string("op_381_cast_fp16")]; + tensor const_228_to_fp16 = const()[name = string("const_228_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_11_cast_fp16 = mul(x = var_381_cast_fp16, y = const_228_to_fp16)[name = string("q_11_cast_fp16")]; + tensor var_387 = const()[name = string("op_387"), val = tensor([1, 1500, 20, -1])]; + tensor var_388_cast_fp16 = reshape(shape = var_387, x = linear_13_cast_fp16)[name = string("op_388_cast_fp16")]; + tensor const_229_to_fp16 = const()[name = string("const_229_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_11_cast_fp16 = mul(x = var_388_cast_fp16, y = const_229_to_fp16)[name = string("k_11_cast_fp16")]; + tensor var_394 = const()[name = string("op_394"), val = tensor([1, 1500, 20, -1])]; + tensor var_395_cast_fp16 = reshape(shape = var_394, x = linear_14_cast_fp16)[name = string("op_395_cast_fp16")]; + tensor var_396 = const()[name = string("op_396"), val = tensor([0, 2, -3, -1])]; + bool qk_5_transpose_x_0 = const()[name = string("qk_5_transpose_x_0"), val = bool(false)]; + bool qk_5_transpose_y_0 = const()[name = string("qk_5_transpose_y_0"), val = bool(false)]; + tensor transpose_68_perm_0 = const()[name = string("transpose_68_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_69_perm_0 = const()[name = string("transpose_69_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_69 = transpose(perm = transpose_69_perm_0, x = k_11_cast_fp16)[name = string("transpose_150")]; + tensor transpose_68 = transpose(perm = transpose_68_perm_0, x = q_11_cast_fp16)[name = string("transpose_151")]; + tensor qk_5_cast_fp16 = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_68, y = transpose_69)[name = string("qk_5_cast_fp16")]; + tensor var_400_cast_fp16 = softmax(axis = var_336, x = qk_5_cast_fp16)[name = string("op_400_cast_fp16")]; + bool var_402_transpose_x_0 = const()[name = string("op_402_transpose_x_0"), val = bool(false)]; + bool var_402_transpose_y_0 = const()[name = string("op_402_transpose_y_0"), val = bool(false)]; + tensor v_11_cast_fp16 = transpose(perm = var_396, x = var_395_cast_fp16)[name = string("transpose_149")]; + tensor var_402_cast_fp16 = matmul(transpose_x = var_402_transpose_x_0, transpose_y = var_402_transpose_y_0, x = var_400_cast_fp16, y = v_11_cast_fp16)[name = string("op_402_cast_fp16")]; + tensor var_403 = const()[name = string("op_403"), val = tensor([0, 2, 1, 3])]; + tensor concat_2 = const()[name = string("concat_2"), val = tensor([1, 1500, 1280])]; + tensor var_404_cast_fp16 = transpose(perm = var_403, x = var_402_cast_fp16)[name = string("transpose_148")]; + tensor x_35_cast_fp16 = reshape(shape = concat_2, x = var_404_cast_fp16)[name = string("x_35_cast_fp16")]; + tensor var_408_to_fp16 = const()[name = string("op_408_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(102840576)))]; + tensor var_409_to_fp16 = const()[name = string("op_409_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(106117440)))]; + tensor linear_15_cast_fp16 = linear(bias = var_409_to_fp16, weight = var_408_to_fp16, x = x_35_cast_fp16)[name = string("linear_15_cast_fp16")]; + tensor x_37_cast_fp16 = add(x = x_31_cast_fp16, y = linear_15_cast_fp16)[name = string("x_37_cast_fp16")]; + tensor var_416_axes_0 = const()[name = string("op_416_axes_0"), val = tensor([-1])]; + tensor blocks_2_mlp_ln_weight_to_fp16 = const()[name = string("blocks_2_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(106120064)))]; + tensor blocks_2_mlp_ln_bias_to_fp16 = const()[name = string("blocks_2_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(106122688)))]; + tensor var_416_cast_fp16 = layer_norm(axes = var_416_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_342_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast_fp16)[name = string("op_416_cast_fp16")]; + tensor var_425_to_fp16 = const()[name = string("op_425_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(106125312)))]; + tensor var_426_to_fp16 = const()[name = string("op_426_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(119232576)))]; + tensor linear_16_cast_fp16 = linear(bias = var_426_to_fp16, weight = var_425_to_fp16, x = var_416_cast_fp16)[name = string("linear_16_cast_fp16")]; + string x_41_mode_0 = const()[name = string("x_41_mode_0"), val = string("EXACT")]; + tensor x_41_cast_fp16 = gelu(mode = x_41_mode_0, x = linear_16_cast_fp16)[name = string("x_41_cast_fp16")]; + tensor var_431_to_fp16 = const()[name = string("op_431_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(119242880)))]; + tensor var_432_to_fp16 = const()[name = string("op_432_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(132350144)))]; + tensor linear_17_cast_fp16 = linear(bias = var_432_to_fp16, weight = var_431_to_fp16, x = x_41_cast_fp16)[name = string("linear_17_cast_fp16")]; + tensor x_43_cast_fp16 = add(x = x_37_cast_fp16, y = linear_17_cast_fp16)[name = string("x_43_cast_fp16")]; + int32 var_442 = const()[name = string("op_442"), val = int32(-1)]; + tensor var_458_axes_0 = const()[name = string("op_458_axes_0"), val = tensor([-1])]; + tensor blocks_3_attn_ln_weight_to_fp16 = const()[name = string("blocks_3_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(132352768)))]; + tensor blocks_3_attn_ln_bias_to_fp16 = const()[name = string("blocks_3_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(132355392)))]; + fp16 var_448_to_fp16 = const()[name = string("op_448_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_458_cast_fp16 = layer_norm(axes = var_458_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_448_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast_fp16)[name = string("op_458_cast_fp16")]; + tensor var_469_to_fp16 = const()[name = string("op_469_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(132358016)))]; + tensor var_470_to_fp16 = const()[name = string("op_470_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(135634880)))]; + tensor linear_18_cast_fp16 = linear(bias = var_470_to_fp16, weight = var_469_to_fp16, x = var_458_cast_fp16)[name = string("linear_18_cast_fp16")]; + tensor var_473_to_fp16 = const()[name = string("op_473_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(135637504)))]; + tensor linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_473_to_fp16, x = var_458_cast_fp16)[name = string("linear_19_cast_fp16")]; + tensor var_477_to_fp16 = const()[name = string("op_477_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(138914368)))]; + tensor var_478_to_fp16 = const()[name = string("op_478_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(142191232)))]; + tensor linear_20_cast_fp16 = linear(bias = var_478_to_fp16, weight = var_477_to_fp16, x = var_458_cast_fp16)[name = string("linear_20_cast_fp16")]; + tensor var_486 = const()[name = string("op_486"), val = tensor([1, 1500, 20, -1])]; + tensor var_487_cast_fp16 = reshape(shape = var_486, x = linear_18_cast_fp16)[name = string("op_487_cast_fp16")]; + tensor const_230_to_fp16 = const()[name = string("const_230_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_15_cast_fp16 = mul(x = var_487_cast_fp16, y = const_230_to_fp16)[name = string("q_15_cast_fp16")]; + tensor var_493 = const()[name = string("op_493"), val = tensor([1, 1500, 20, -1])]; + tensor var_494_cast_fp16 = reshape(shape = var_493, x = linear_19_cast_fp16)[name = string("op_494_cast_fp16")]; + tensor const_231_to_fp16 = const()[name = string("const_231_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_15_cast_fp16 = mul(x = var_494_cast_fp16, y = const_231_to_fp16)[name = string("k_15_cast_fp16")]; + tensor var_500 = const()[name = string("op_500"), val = tensor([1, 1500, 20, -1])]; + tensor var_501_cast_fp16 = reshape(shape = var_500, x = linear_20_cast_fp16)[name = string("op_501_cast_fp16")]; + tensor var_502 = const()[name = string("op_502"), val = tensor([0, 2, -3, -1])]; + bool qk_7_transpose_x_0 = const()[name = string("qk_7_transpose_x_0"), val = bool(false)]; + bool qk_7_transpose_y_0 = const()[name = string("qk_7_transpose_y_0"), val = bool(false)]; + tensor transpose_70_perm_0 = const()[name = string("transpose_70_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_71_perm_0 = const()[name = string("transpose_71_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_71 = transpose(perm = transpose_71_perm_0, x = k_15_cast_fp16)[name = string("transpose_146")]; + tensor transpose_70 = transpose(perm = transpose_70_perm_0, x = q_15_cast_fp16)[name = string("transpose_147")]; + tensor qk_7_cast_fp16 = matmul(transpose_x = qk_7_transpose_x_0, transpose_y = qk_7_transpose_y_0, x = transpose_70, y = transpose_71)[name = string("qk_7_cast_fp16")]; + tensor var_506_cast_fp16 = softmax(axis = var_442, x = qk_7_cast_fp16)[name = string("op_506_cast_fp16")]; + bool var_508_transpose_x_0 = const()[name = string("op_508_transpose_x_0"), val = bool(false)]; + bool var_508_transpose_y_0 = const()[name = string("op_508_transpose_y_0"), val = bool(false)]; + tensor v_15_cast_fp16 = transpose(perm = var_502, x = var_501_cast_fp16)[name = string("transpose_145")]; + tensor var_508_cast_fp16 = matmul(transpose_x = var_508_transpose_x_0, transpose_y = var_508_transpose_y_0, x = var_506_cast_fp16, y = v_15_cast_fp16)[name = string("op_508_cast_fp16")]; + tensor var_509 = const()[name = string("op_509"), val = tensor([0, 2, 1, 3])]; + tensor concat_3 = const()[name = string("concat_3"), val = tensor([1, 1500, 1280])]; + tensor var_510_cast_fp16 = transpose(perm = var_509, x = var_508_cast_fp16)[name = string("transpose_144")]; + tensor x_47_cast_fp16 = reshape(shape = concat_3, x = var_510_cast_fp16)[name = string("x_47_cast_fp16")]; + tensor var_514_to_fp16 = const()[name = string("op_514_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(142193856)))]; + tensor var_515_to_fp16 = const()[name = string("op_515_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(145470720)))]; + tensor linear_21_cast_fp16 = linear(bias = var_515_to_fp16, weight = var_514_to_fp16, x = x_47_cast_fp16)[name = string("linear_21_cast_fp16")]; + tensor x_49_cast_fp16 = add(x = x_43_cast_fp16, y = linear_21_cast_fp16)[name = string("x_49_cast_fp16")]; + tensor var_522_axes_0 = const()[name = string("op_522_axes_0"), val = tensor([-1])]; + tensor blocks_3_mlp_ln_weight_to_fp16 = const()[name = string("blocks_3_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(145473344)))]; + tensor blocks_3_mlp_ln_bias_to_fp16 = const()[name = string("blocks_3_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(145475968)))]; + tensor var_522_cast_fp16 = layer_norm(axes = var_522_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_448_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast_fp16)[name = string("op_522_cast_fp16")]; + tensor var_531_to_fp16 = const()[name = string("op_531_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(145478592)))]; + tensor var_532_to_fp16 = const()[name = string("op_532_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(158585856)))]; + tensor linear_22_cast_fp16 = linear(bias = var_532_to_fp16, weight = var_531_to_fp16, x = var_522_cast_fp16)[name = string("linear_22_cast_fp16")]; + string x_53_mode_0 = const()[name = string("x_53_mode_0"), val = string("EXACT")]; + tensor x_53_cast_fp16 = gelu(mode = x_53_mode_0, x = linear_22_cast_fp16)[name = string("x_53_cast_fp16")]; + tensor var_537_to_fp16 = const()[name = string("op_537_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(158596160)))]; + tensor var_538_to_fp16 = const()[name = string("op_538_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(171703424)))]; + tensor linear_23_cast_fp16 = linear(bias = var_538_to_fp16, weight = var_537_to_fp16, x = x_53_cast_fp16)[name = string("linear_23_cast_fp16")]; + tensor x_55_cast_fp16 = add(x = x_49_cast_fp16, y = linear_23_cast_fp16)[name = string("x_55_cast_fp16")]; + int32 var_548 = const()[name = string("op_548"), val = int32(-1)]; + tensor var_564_axes_0 = const()[name = string("op_564_axes_0"), val = tensor([-1])]; + tensor blocks_4_attn_ln_weight_to_fp16 = const()[name = string("blocks_4_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(171706048)))]; + tensor blocks_4_attn_ln_bias_to_fp16 = const()[name = string("blocks_4_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(171708672)))]; + fp16 var_554_to_fp16 = const()[name = string("op_554_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_564_cast_fp16 = layer_norm(axes = var_564_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_554_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast_fp16)[name = string("op_564_cast_fp16")]; + tensor var_575_to_fp16 = const()[name = string("op_575_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(171711296)))]; + tensor var_576_to_fp16 = const()[name = string("op_576_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(174988160)))]; + tensor linear_24_cast_fp16 = linear(bias = var_576_to_fp16, weight = var_575_to_fp16, x = var_564_cast_fp16)[name = string("linear_24_cast_fp16")]; + tensor var_579_to_fp16 = const()[name = string("op_579_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(174990784)))]; + tensor linear_25_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_579_to_fp16, x = var_564_cast_fp16)[name = string("linear_25_cast_fp16")]; + tensor var_583_to_fp16 = const()[name = string("op_583_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(178267648)))]; + tensor var_584_to_fp16 = const()[name = string("op_584_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(181544512)))]; + tensor linear_26_cast_fp16 = linear(bias = var_584_to_fp16, weight = var_583_to_fp16, x = var_564_cast_fp16)[name = string("linear_26_cast_fp16")]; + tensor var_592 = const()[name = string("op_592"), val = tensor([1, 1500, 20, -1])]; + tensor var_593_cast_fp16 = reshape(shape = var_592, x = linear_24_cast_fp16)[name = string("op_593_cast_fp16")]; + tensor const_232_to_fp16 = const()[name = string("const_232_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_19_cast_fp16 = mul(x = var_593_cast_fp16, y = const_232_to_fp16)[name = string("q_19_cast_fp16")]; + tensor var_599 = const()[name = string("op_599"), val = tensor([1, 1500, 20, -1])]; + tensor var_600_cast_fp16 = reshape(shape = var_599, x = linear_25_cast_fp16)[name = string("op_600_cast_fp16")]; + tensor const_233_to_fp16 = const()[name = string("const_233_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_19_cast_fp16 = mul(x = var_600_cast_fp16, y = const_233_to_fp16)[name = string("k_19_cast_fp16")]; + tensor var_606 = const()[name = string("op_606"), val = tensor([1, 1500, 20, -1])]; + tensor var_607_cast_fp16 = reshape(shape = var_606, x = linear_26_cast_fp16)[name = string("op_607_cast_fp16")]; + tensor var_608 = const()[name = string("op_608"), val = tensor([0, 2, -3, -1])]; + bool qk_9_transpose_x_0 = const()[name = string("qk_9_transpose_x_0"), val = bool(false)]; + bool qk_9_transpose_y_0 = const()[name = string("qk_9_transpose_y_0"), val = bool(false)]; + tensor transpose_72_perm_0 = const()[name = string("transpose_72_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_73_perm_0 = const()[name = string("transpose_73_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_73 = transpose(perm = transpose_73_perm_0, x = k_19_cast_fp16)[name = string("transpose_142")]; + tensor transpose_72 = transpose(perm = transpose_72_perm_0, x = q_19_cast_fp16)[name = string("transpose_143")]; + tensor qk_9_cast_fp16 = matmul(transpose_x = qk_9_transpose_x_0, transpose_y = qk_9_transpose_y_0, x = transpose_72, y = transpose_73)[name = string("qk_9_cast_fp16")]; + tensor var_612_cast_fp16 = softmax(axis = var_548, x = qk_9_cast_fp16)[name = string("op_612_cast_fp16")]; + bool var_614_transpose_x_0 = const()[name = string("op_614_transpose_x_0"), val = bool(false)]; + bool var_614_transpose_y_0 = const()[name = string("op_614_transpose_y_0"), val = bool(false)]; + tensor v_19_cast_fp16 = transpose(perm = var_608, x = var_607_cast_fp16)[name = string("transpose_141")]; + tensor var_614_cast_fp16 = matmul(transpose_x = var_614_transpose_x_0, transpose_y = var_614_transpose_y_0, x = var_612_cast_fp16, y = v_19_cast_fp16)[name = string("op_614_cast_fp16")]; + tensor var_615 = const()[name = string("op_615"), val = tensor([0, 2, 1, 3])]; + tensor concat_4 = const()[name = string("concat_4"), val = tensor([1, 1500, 1280])]; + tensor var_616_cast_fp16 = transpose(perm = var_615, x = var_614_cast_fp16)[name = string("transpose_140")]; + tensor x_59_cast_fp16 = reshape(shape = concat_4, x = var_616_cast_fp16)[name = string("x_59_cast_fp16")]; + tensor var_620_to_fp16 = const()[name = string("op_620_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(181547136)))]; + tensor var_621_to_fp16 = const()[name = string("op_621_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(184824000)))]; + tensor linear_27_cast_fp16 = linear(bias = var_621_to_fp16, weight = var_620_to_fp16, x = x_59_cast_fp16)[name = string("linear_27_cast_fp16")]; + tensor x_61_cast_fp16 = add(x = x_55_cast_fp16, y = linear_27_cast_fp16)[name = string("x_61_cast_fp16")]; + tensor var_628_axes_0 = const()[name = string("op_628_axes_0"), val = tensor([-1])]; + tensor blocks_4_mlp_ln_weight_to_fp16 = const()[name = string("blocks_4_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(184826624)))]; + tensor blocks_4_mlp_ln_bias_to_fp16 = const()[name = string("blocks_4_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(184829248)))]; + tensor var_628_cast_fp16 = layer_norm(axes = var_628_axes_0, beta = blocks_4_mlp_ln_bias_to_fp16, epsilon = var_554_to_fp16, gamma = blocks_4_mlp_ln_weight_to_fp16, x = x_61_cast_fp16)[name = string("op_628_cast_fp16")]; + tensor var_637_to_fp16 = const()[name = string("op_637_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(184831872)))]; + tensor var_638_to_fp16 = const()[name = string("op_638_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(197939136)))]; + tensor linear_28_cast_fp16 = linear(bias = var_638_to_fp16, weight = var_637_to_fp16, x = var_628_cast_fp16)[name = string("linear_28_cast_fp16")]; + string x_65_mode_0 = const()[name = string("x_65_mode_0"), val = string("EXACT")]; + tensor x_65_cast_fp16 = gelu(mode = x_65_mode_0, x = linear_28_cast_fp16)[name = string("x_65_cast_fp16")]; + tensor var_643_to_fp16 = const()[name = string("op_643_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(197949440)))]; + tensor var_644_to_fp16 = const()[name = string("op_644_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(211056704)))]; + tensor linear_29_cast_fp16 = linear(bias = var_644_to_fp16, weight = var_643_to_fp16, x = x_65_cast_fp16)[name = string("linear_29_cast_fp16")]; + tensor x_67_cast_fp16 = add(x = x_61_cast_fp16, y = linear_29_cast_fp16)[name = string("x_67_cast_fp16")]; + int32 var_654 = const()[name = string("op_654"), val = int32(-1)]; + tensor var_670_axes_0 = const()[name = string("op_670_axes_0"), val = tensor([-1])]; + tensor blocks_5_attn_ln_weight_to_fp16 = const()[name = string("blocks_5_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(211059328)))]; + tensor blocks_5_attn_ln_bias_to_fp16 = const()[name = string("blocks_5_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(211061952)))]; + fp16 var_660_to_fp16 = const()[name = string("op_660_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_670_cast_fp16 = layer_norm(axes = var_670_axes_0, beta = blocks_5_attn_ln_bias_to_fp16, epsilon = var_660_to_fp16, gamma = blocks_5_attn_ln_weight_to_fp16, x = x_67_cast_fp16)[name = string("op_670_cast_fp16")]; + tensor var_681_to_fp16 = const()[name = string("op_681_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(211064576)))]; + tensor var_682_to_fp16 = const()[name = string("op_682_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(214341440)))]; + tensor linear_30_cast_fp16 = linear(bias = var_682_to_fp16, weight = var_681_to_fp16, x = var_670_cast_fp16)[name = string("linear_30_cast_fp16")]; + tensor var_685_to_fp16 = const()[name = string("op_685_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(214344064)))]; + tensor linear_31_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_685_to_fp16, x = var_670_cast_fp16)[name = string("linear_31_cast_fp16")]; + tensor var_689_to_fp16 = const()[name = string("op_689_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(217620928)))]; + tensor var_690_to_fp16 = const()[name = string("op_690_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(220897792)))]; + tensor linear_32_cast_fp16 = linear(bias = var_690_to_fp16, weight = var_689_to_fp16, x = var_670_cast_fp16)[name = string("linear_32_cast_fp16")]; + tensor var_698 = const()[name = string("op_698"), val = tensor([1, 1500, 20, -1])]; + tensor var_699_cast_fp16 = reshape(shape = var_698, x = linear_30_cast_fp16)[name = string("op_699_cast_fp16")]; + tensor const_234_to_fp16 = const()[name = string("const_234_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_23_cast_fp16 = mul(x = var_699_cast_fp16, y = const_234_to_fp16)[name = string("q_23_cast_fp16")]; + tensor var_705 = const()[name = string("op_705"), val = tensor([1, 1500, 20, -1])]; + tensor var_706_cast_fp16 = reshape(shape = var_705, x = linear_31_cast_fp16)[name = string("op_706_cast_fp16")]; + tensor const_235_to_fp16 = const()[name = string("const_235_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_23_cast_fp16 = mul(x = var_706_cast_fp16, y = const_235_to_fp16)[name = string("k_23_cast_fp16")]; + tensor var_712 = const()[name = string("op_712"), val = tensor([1, 1500, 20, -1])]; + tensor var_713_cast_fp16 = reshape(shape = var_712, x = linear_32_cast_fp16)[name = string("op_713_cast_fp16")]; + tensor var_714 = const()[name = string("op_714"), val = tensor([0, 2, -3, -1])]; + bool qk_11_transpose_x_0 = const()[name = string("qk_11_transpose_x_0"), val = bool(false)]; + bool qk_11_transpose_y_0 = const()[name = string("qk_11_transpose_y_0"), val = bool(false)]; + tensor transpose_74_perm_0 = const()[name = string("transpose_74_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_75_perm_0 = const()[name = string("transpose_75_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_75 = transpose(perm = transpose_75_perm_0, x = k_23_cast_fp16)[name = string("transpose_138")]; + tensor transpose_74 = transpose(perm = transpose_74_perm_0, x = q_23_cast_fp16)[name = string("transpose_139")]; + tensor qk_11_cast_fp16 = matmul(transpose_x = qk_11_transpose_x_0, transpose_y = qk_11_transpose_y_0, x = transpose_74, y = transpose_75)[name = string("qk_11_cast_fp16")]; + tensor var_718_cast_fp16 = softmax(axis = var_654, x = qk_11_cast_fp16)[name = string("op_718_cast_fp16")]; + bool var_720_transpose_x_0 = const()[name = string("op_720_transpose_x_0"), val = bool(false)]; + bool var_720_transpose_y_0 = const()[name = string("op_720_transpose_y_0"), val = bool(false)]; + tensor v_23_cast_fp16 = transpose(perm = var_714, x = var_713_cast_fp16)[name = string("transpose_137")]; + tensor var_720_cast_fp16 = matmul(transpose_x = var_720_transpose_x_0, transpose_y = var_720_transpose_y_0, x = var_718_cast_fp16, y = v_23_cast_fp16)[name = string("op_720_cast_fp16")]; + tensor var_721 = const()[name = string("op_721"), val = tensor([0, 2, 1, 3])]; + tensor concat_5 = const()[name = string("concat_5"), val = tensor([1, 1500, 1280])]; + tensor var_722_cast_fp16 = transpose(perm = var_721, x = var_720_cast_fp16)[name = string("transpose_136")]; + tensor x_71_cast_fp16 = reshape(shape = concat_5, x = var_722_cast_fp16)[name = string("x_71_cast_fp16")]; + tensor var_726_to_fp16 = const()[name = string("op_726_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(220900416)))]; + tensor var_727_to_fp16 = const()[name = string("op_727_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(224177280)))]; + tensor linear_33_cast_fp16 = linear(bias = var_727_to_fp16, weight = var_726_to_fp16, x = x_71_cast_fp16)[name = string("linear_33_cast_fp16")]; + tensor x_73_cast_fp16 = add(x = x_67_cast_fp16, y = linear_33_cast_fp16)[name = string("x_73_cast_fp16")]; + tensor var_734_axes_0 = const()[name = string("op_734_axes_0"), val = tensor([-1])]; + tensor blocks_5_mlp_ln_weight_to_fp16 = const()[name = string("blocks_5_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(224179904)))]; + tensor blocks_5_mlp_ln_bias_to_fp16 = const()[name = string("blocks_5_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(224182528)))]; + tensor var_734_cast_fp16 = layer_norm(axes = var_734_axes_0, beta = blocks_5_mlp_ln_bias_to_fp16, epsilon = var_660_to_fp16, gamma = blocks_5_mlp_ln_weight_to_fp16, x = x_73_cast_fp16)[name = string("op_734_cast_fp16")]; + tensor var_743_to_fp16 = const()[name = string("op_743_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(224185152)))]; + tensor var_744_to_fp16 = const()[name = string("op_744_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(237292416)))]; + tensor linear_34_cast_fp16 = linear(bias = var_744_to_fp16, weight = var_743_to_fp16, x = var_734_cast_fp16)[name = string("linear_34_cast_fp16")]; + string x_77_mode_0 = const()[name = string("x_77_mode_0"), val = string("EXACT")]; + tensor x_77_cast_fp16 = gelu(mode = x_77_mode_0, x = linear_34_cast_fp16)[name = string("x_77_cast_fp16")]; + tensor var_749_to_fp16 = const()[name = string("op_749_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(237302720)))]; + tensor var_750_to_fp16 = const()[name = string("op_750_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(250409984)))]; + tensor linear_35_cast_fp16 = linear(bias = var_750_to_fp16, weight = var_749_to_fp16, x = x_77_cast_fp16)[name = string("linear_35_cast_fp16")]; + tensor x_79_cast_fp16 = add(x = x_73_cast_fp16, y = linear_35_cast_fp16)[name = string("x_79_cast_fp16")]; + int32 var_760 = const()[name = string("op_760"), val = int32(-1)]; + tensor var_776_axes_0 = const()[name = string("op_776_axes_0"), val = tensor([-1])]; + tensor blocks_6_attn_ln_weight_to_fp16 = const()[name = string("blocks_6_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(250412608)))]; + tensor blocks_6_attn_ln_bias_to_fp16 = const()[name = string("blocks_6_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(250415232)))]; + fp16 var_766_to_fp16 = const()[name = string("op_766_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_776_cast_fp16 = layer_norm(axes = var_776_axes_0, beta = blocks_6_attn_ln_bias_to_fp16, epsilon = var_766_to_fp16, gamma = blocks_6_attn_ln_weight_to_fp16, x = x_79_cast_fp16)[name = string("op_776_cast_fp16")]; + tensor var_787_to_fp16 = const()[name = string("op_787_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(250417856)))]; + tensor var_788_to_fp16 = const()[name = string("op_788_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(253694720)))]; + tensor linear_36_cast_fp16 = linear(bias = var_788_to_fp16, weight = var_787_to_fp16, x = var_776_cast_fp16)[name = string("linear_36_cast_fp16")]; + tensor var_791_to_fp16 = const()[name = string("op_791_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(253697344)))]; + tensor linear_37_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_791_to_fp16, x = var_776_cast_fp16)[name = string("linear_37_cast_fp16")]; + tensor var_795_to_fp16 = const()[name = string("op_795_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(256974208)))]; + tensor var_796_to_fp16 = const()[name = string("op_796_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(260251072)))]; + tensor linear_38_cast_fp16 = linear(bias = var_796_to_fp16, weight = var_795_to_fp16, x = var_776_cast_fp16)[name = string("linear_38_cast_fp16")]; + tensor var_804 = const()[name = string("op_804"), val = tensor([1, 1500, 20, -1])]; + tensor var_805_cast_fp16 = reshape(shape = var_804, x = linear_36_cast_fp16)[name = string("op_805_cast_fp16")]; + tensor const_236_to_fp16 = const()[name = string("const_236_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_27_cast_fp16 = mul(x = var_805_cast_fp16, y = const_236_to_fp16)[name = string("q_27_cast_fp16")]; + tensor var_811 = const()[name = string("op_811"), val = tensor([1, 1500, 20, -1])]; + tensor var_812_cast_fp16 = reshape(shape = var_811, x = linear_37_cast_fp16)[name = string("op_812_cast_fp16")]; + tensor const_237_to_fp16 = const()[name = string("const_237_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_27_cast_fp16 = mul(x = var_812_cast_fp16, y = const_237_to_fp16)[name = string("k_27_cast_fp16")]; + tensor var_818 = const()[name = string("op_818"), val = tensor([1, 1500, 20, -1])]; + tensor var_819_cast_fp16 = reshape(shape = var_818, x = linear_38_cast_fp16)[name = string("op_819_cast_fp16")]; + tensor var_820 = const()[name = string("op_820"), val = tensor([0, 2, -3, -1])]; + bool qk_13_transpose_x_0 = const()[name = string("qk_13_transpose_x_0"), val = bool(false)]; + bool qk_13_transpose_y_0 = const()[name = string("qk_13_transpose_y_0"), val = bool(false)]; + tensor transpose_76_perm_0 = const()[name = string("transpose_76_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_77_perm_0 = const()[name = string("transpose_77_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_77 = transpose(perm = transpose_77_perm_0, x = k_27_cast_fp16)[name = string("transpose_134")]; + tensor transpose_76 = transpose(perm = transpose_76_perm_0, x = q_27_cast_fp16)[name = string("transpose_135")]; + tensor qk_13_cast_fp16 = matmul(transpose_x = qk_13_transpose_x_0, transpose_y = qk_13_transpose_y_0, x = transpose_76, y = transpose_77)[name = string("qk_13_cast_fp16")]; + tensor var_824_cast_fp16 = softmax(axis = var_760, x = qk_13_cast_fp16)[name = string("op_824_cast_fp16")]; + bool var_826_transpose_x_0 = const()[name = string("op_826_transpose_x_0"), val = bool(false)]; + bool var_826_transpose_y_0 = const()[name = string("op_826_transpose_y_0"), val = bool(false)]; + tensor v_27_cast_fp16 = transpose(perm = var_820, x = var_819_cast_fp16)[name = string("transpose_133")]; + tensor var_826_cast_fp16 = matmul(transpose_x = var_826_transpose_x_0, transpose_y = var_826_transpose_y_0, x = var_824_cast_fp16, y = v_27_cast_fp16)[name = string("op_826_cast_fp16")]; + tensor var_827 = const()[name = string("op_827"), val = tensor([0, 2, 1, 3])]; + tensor concat_6 = const()[name = string("concat_6"), val = tensor([1, 1500, 1280])]; + tensor var_828_cast_fp16 = transpose(perm = var_827, x = var_826_cast_fp16)[name = string("transpose_132")]; + tensor x_83_cast_fp16 = reshape(shape = concat_6, x = var_828_cast_fp16)[name = string("x_83_cast_fp16")]; + tensor var_832_to_fp16 = const()[name = string("op_832_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(260253696)))]; + tensor var_833_to_fp16 = const()[name = string("op_833_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(263530560)))]; + tensor linear_39_cast_fp16 = linear(bias = var_833_to_fp16, weight = var_832_to_fp16, x = x_83_cast_fp16)[name = string("linear_39_cast_fp16")]; + tensor x_85_cast_fp16 = add(x = x_79_cast_fp16, y = linear_39_cast_fp16)[name = string("x_85_cast_fp16")]; + tensor var_840_axes_0 = const()[name = string("op_840_axes_0"), val = tensor([-1])]; + tensor blocks_6_mlp_ln_weight_to_fp16 = const()[name = string("blocks_6_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(263533184)))]; + tensor blocks_6_mlp_ln_bias_to_fp16 = const()[name = string("blocks_6_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(263535808)))]; + tensor var_840_cast_fp16 = layer_norm(axes = var_840_axes_0, beta = blocks_6_mlp_ln_bias_to_fp16, epsilon = var_766_to_fp16, gamma = blocks_6_mlp_ln_weight_to_fp16, x = x_85_cast_fp16)[name = string("op_840_cast_fp16")]; + tensor var_849_to_fp16 = const()[name = string("op_849_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(263538432)))]; + tensor var_850_to_fp16 = const()[name = string("op_850_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(276645696)))]; + tensor linear_40_cast_fp16 = linear(bias = var_850_to_fp16, weight = var_849_to_fp16, x = var_840_cast_fp16)[name = string("linear_40_cast_fp16")]; + string x_89_mode_0 = const()[name = string("x_89_mode_0"), val = string("EXACT")]; + tensor x_89_cast_fp16 = gelu(mode = x_89_mode_0, x = linear_40_cast_fp16)[name = string("x_89_cast_fp16")]; + tensor var_855_to_fp16 = const()[name = string("op_855_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(276656000)))]; + tensor var_856_to_fp16 = const()[name = string("op_856_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(289763264)))]; + tensor linear_41_cast_fp16 = linear(bias = var_856_to_fp16, weight = var_855_to_fp16, x = x_89_cast_fp16)[name = string("linear_41_cast_fp16")]; + tensor x_91_cast_fp16 = add(x = x_85_cast_fp16, y = linear_41_cast_fp16)[name = string("x_91_cast_fp16")]; + int32 var_866 = const()[name = string("op_866"), val = int32(-1)]; + tensor var_882_axes_0 = const()[name = string("op_882_axes_0"), val = tensor([-1])]; + tensor blocks_7_attn_ln_weight_to_fp16 = const()[name = string("blocks_7_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(289765888)))]; + tensor blocks_7_attn_ln_bias_to_fp16 = const()[name = string("blocks_7_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(289768512)))]; + fp16 var_872_to_fp16 = const()[name = string("op_872_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_882_cast_fp16 = layer_norm(axes = var_882_axes_0, beta = blocks_7_attn_ln_bias_to_fp16, epsilon = var_872_to_fp16, gamma = blocks_7_attn_ln_weight_to_fp16, x = x_91_cast_fp16)[name = string("op_882_cast_fp16")]; + tensor var_893_to_fp16 = const()[name = string("op_893_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(289771136)))]; + tensor var_894_to_fp16 = const()[name = string("op_894_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(293048000)))]; + tensor linear_42_cast_fp16 = linear(bias = var_894_to_fp16, weight = var_893_to_fp16, x = var_882_cast_fp16)[name = string("linear_42_cast_fp16")]; + tensor var_897_to_fp16 = const()[name = string("op_897_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(293050624)))]; + tensor linear_43_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_897_to_fp16, x = var_882_cast_fp16)[name = string("linear_43_cast_fp16")]; + tensor var_901_to_fp16 = const()[name = string("op_901_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(296327488)))]; + tensor var_902_to_fp16 = const()[name = string("op_902_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(299604352)))]; + tensor linear_44_cast_fp16 = linear(bias = var_902_to_fp16, weight = var_901_to_fp16, x = var_882_cast_fp16)[name = string("linear_44_cast_fp16")]; + tensor var_910 = const()[name = string("op_910"), val = tensor([1, 1500, 20, -1])]; + tensor var_911_cast_fp16 = reshape(shape = var_910, x = linear_42_cast_fp16)[name = string("op_911_cast_fp16")]; + tensor const_238_to_fp16 = const()[name = string("const_238_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_31_cast_fp16 = mul(x = var_911_cast_fp16, y = const_238_to_fp16)[name = string("q_31_cast_fp16")]; + tensor var_917 = const()[name = string("op_917"), val = tensor([1, 1500, 20, -1])]; + tensor var_918_cast_fp16 = reshape(shape = var_917, x = linear_43_cast_fp16)[name = string("op_918_cast_fp16")]; + tensor const_239_to_fp16 = const()[name = string("const_239_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_31_cast_fp16 = mul(x = var_918_cast_fp16, y = const_239_to_fp16)[name = string("k_31_cast_fp16")]; + tensor var_924 = const()[name = string("op_924"), val = tensor([1, 1500, 20, -1])]; + tensor var_925_cast_fp16 = reshape(shape = var_924, x = linear_44_cast_fp16)[name = string("op_925_cast_fp16")]; + tensor var_926 = const()[name = string("op_926"), val = tensor([0, 2, -3, -1])]; + bool qk_15_transpose_x_0 = const()[name = string("qk_15_transpose_x_0"), val = bool(false)]; + bool qk_15_transpose_y_0 = const()[name = string("qk_15_transpose_y_0"), val = bool(false)]; + tensor transpose_78_perm_0 = const()[name = string("transpose_78_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_79_perm_0 = const()[name = string("transpose_79_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_79 = transpose(perm = transpose_79_perm_0, x = k_31_cast_fp16)[name = string("transpose_130")]; + tensor transpose_78 = transpose(perm = transpose_78_perm_0, x = q_31_cast_fp16)[name = string("transpose_131")]; + tensor qk_15_cast_fp16 = matmul(transpose_x = qk_15_transpose_x_0, transpose_y = qk_15_transpose_y_0, x = transpose_78, y = transpose_79)[name = string("qk_15_cast_fp16")]; + tensor var_930_cast_fp16 = softmax(axis = var_866, x = qk_15_cast_fp16)[name = string("op_930_cast_fp16")]; + bool var_932_transpose_x_0 = const()[name = string("op_932_transpose_x_0"), val = bool(false)]; + bool var_932_transpose_y_0 = const()[name = string("op_932_transpose_y_0"), val = bool(false)]; + tensor v_31_cast_fp16 = transpose(perm = var_926, x = var_925_cast_fp16)[name = string("transpose_129")]; + tensor var_932_cast_fp16 = matmul(transpose_x = var_932_transpose_x_0, transpose_y = var_932_transpose_y_0, x = var_930_cast_fp16, y = v_31_cast_fp16)[name = string("op_932_cast_fp16")]; + tensor var_933 = const()[name = string("op_933"), val = tensor([0, 2, 1, 3])]; + tensor concat_7 = const()[name = string("concat_7"), val = tensor([1, 1500, 1280])]; + tensor var_934_cast_fp16 = transpose(perm = var_933, x = var_932_cast_fp16)[name = string("transpose_128")]; + tensor x_95_cast_fp16 = reshape(shape = concat_7, x = var_934_cast_fp16)[name = string("x_95_cast_fp16")]; + tensor var_938_to_fp16 = const()[name = string("op_938_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(299606976)))]; + tensor var_939_to_fp16 = const()[name = string("op_939_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(302883840)))]; + tensor linear_45_cast_fp16 = linear(bias = var_939_to_fp16, weight = var_938_to_fp16, x = x_95_cast_fp16)[name = string("linear_45_cast_fp16")]; + tensor x_97_cast_fp16 = add(x = x_91_cast_fp16, y = linear_45_cast_fp16)[name = string("x_97_cast_fp16")]; + tensor var_946_axes_0 = const()[name = string("op_946_axes_0"), val = tensor([-1])]; + tensor blocks_7_mlp_ln_weight_to_fp16 = const()[name = string("blocks_7_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(302886464)))]; + tensor blocks_7_mlp_ln_bias_to_fp16 = const()[name = string("blocks_7_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(302889088)))]; + tensor var_946_cast_fp16 = layer_norm(axes = var_946_axes_0, beta = blocks_7_mlp_ln_bias_to_fp16, epsilon = var_872_to_fp16, gamma = blocks_7_mlp_ln_weight_to_fp16, x = x_97_cast_fp16)[name = string("op_946_cast_fp16")]; + tensor var_955_to_fp16 = const()[name = string("op_955_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(302891712)))]; + tensor var_956_to_fp16 = const()[name = string("op_956_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(315998976)))]; + tensor linear_46_cast_fp16 = linear(bias = var_956_to_fp16, weight = var_955_to_fp16, x = var_946_cast_fp16)[name = string("linear_46_cast_fp16")]; + string x_101_mode_0 = const()[name = string("x_101_mode_0"), val = string("EXACT")]; + tensor x_101_cast_fp16 = gelu(mode = x_101_mode_0, x = linear_46_cast_fp16)[name = string("x_101_cast_fp16")]; + tensor var_961_to_fp16 = const()[name = string("op_961_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(316009280)))]; + tensor var_962_to_fp16 = const()[name = string("op_962_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(329116544)))]; + tensor linear_47_cast_fp16 = linear(bias = var_962_to_fp16, weight = var_961_to_fp16, x = x_101_cast_fp16)[name = string("linear_47_cast_fp16")]; + tensor x_103_cast_fp16 = add(x = x_97_cast_fp16, y = linear_47_cast_fp16)[name = string("x_103_cast_fp16")]; + int32 var_972 = const()[name = string("op_972"), val = int32(-1)]; + tensor var_988_axes_0 = const()[name = string("op_988_axes_0"), val = tensor([-1])]; + tensor blocks_8_attn_ln_weight_to_fp16 = const()[name = string("blocks_8_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(329119168)))]; + tensor blocks_8_attn_ln_bias_to_fp16 = const()[name = string("blocks_8_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(329121792)))]; + fp16 var_978_to_fp16 = const()[name = string("op_978_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_988_cast_fp16 = layer_norm(axes = var_988_axes_0, beta = blocks_8_attn_ln_bias_to_fp16, epsilon = var_978_to_fp16, gamma = blocks_8_attn_ln_weight_to_fp16, x = x_103_cast_fp16)[name = string("op_988_cast_fp16")]; + tensor var_999_to_fp16 = const()[name = string("op_999_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(329124416)))]; + tensor var_1000_to_fp16 = const()[name = string("op_1000_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(332401280)))]; + tensor linear_48_cast_fp16 = linear(bias = var_1000_to_fp16, weight = var_999_to_fp16, x = var_988_cast_fp16)[name = string("linear_48_cast_fp16")]; + tensor var_1003_to_fp16 = const()[name = string("op_1003_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(332403904)))]; + tensor linear_49_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1003_to_fp16, x = var_988_cast_fp16)[name = string("linear_49_cast_fp16")]; + tensor var_1007_to_fp16 = const()[name = string("op_1007_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(335680768)))]; + tensor var_1008_to_fp16 = const()[name = string("op_1008_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(338957632)))]; + tensor linear_50_cast_fp16 = linear(bias = var_1008_to_fp16, weight = var_1007_to_fp16, x = var_988_cast_fp16)[name = string("linear_50_cast_fp16")]; + tensor var_1016 = const()[name = string("op_1016"), val = tensor([1, 1500, 20, -1])]; + tensor var_1017_cast_fp16 = reshape(shape = var_1016, x = linear_48_cast_fp16)[name = string("op_1017_cast_fp16")]; + tensor const_240_to_fp16 = const()[name = string("const_240_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_35_cast_fp16 = mul(x = var_1017_cast_fp16, y = const_240_to_fp16)[name = string("q_35_cast_fp16")]; + tensor var_1023 = const()[name = string("op_1023"), val = tensor([1, 1500, 20, -1])]; + tensor var_1024_cast_fp16 = reshape(shape = var_1023, x = linear_49_cast_fp16)[name = string("op_1024_cast_fp16")]; + tensor const_241_to_fp16 = const()[name = string("const_241_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_35_cast_fp16 = mul(x = var_1024_cast_fp16, y = const_241_to_fp16)[name = string("k_35_cast_fp16")]; + tensor var_1030 = const()[name = string("op_1030"), val = tensor([1, 1500, 20, -1])]; + tensor var_1031_cast_fp16 = reshape(shape = var_1030, x = linear_50_cast_fp16)[name = string("op_1031_cast_fp16")]; + tensor var_1032 = const()[name = string("op_1032"), val = tensor([0, 2, -3, -1])]; + bool qk_17_transpose_x_0 = const()[name = string("qk_17_transpose_x_0"), val = bool(false)]; + bool qk_17_transpose_y_0 = const()[name = string("qk_17_transpose_y_0"), val = bool(false)]; + tensor transpose_80_perm_0 = const()[name = string("transpose_80_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_81_perm_0 = const()[name = string("transpose_81_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_81 = transpose(perm = transpose_81_perm_0, x = k_35_cast_fp16)[name = string("transpose_126")]; + tensor transpose_80 = transpose(perm = transpose_80_perm_0, x = q_35_cast_fp16)[name = string("transpose_127")]; + tensor qk_17_cast_fp16 = matmul(transpose_x = qk_17_transpose_x_0, transpose_y = qk_17_transpose_y_0, x = transpose_80, y = transpose_81)[name = string("qk_17_cast_fp16")]; + tensor var_1036_cast_fp16 = softmax(axis = var_972, x = qk_17_cast_fp16)[name = string("op_1036_cast_fp16")]; + bool var_1038_transpose_x_0 = const()[name = string("op_1038_transpose_x_0"), val = bool(false)]; + bool var_1038_transpose_y_0 = const()[name = string("op_1038_transpose_y_0"), val = bool(false)]; + tensor v_35_cast_fp16 = transpose(perm = var_1032, x = var_1031_cast_fp16)[name = string("transpose_125")]; + tensor var_1038_cast_fp16 = matmul(transpose_x = var_1038_transpose_x_0, transpose_y = var_1038_transpose_y_0, x = var_1036_cast_fp16, y = v_35_cast_fp16)[name = string("op_1038_cast_fp16")]; + tensor var_1039 = const()[name = string("op_1039"), val = tensor([0, 2, 1, 3])]; + tensor concat_8 = const()[name = string("concat_8"), val = tensor([1, 1500, 1280])]; + tensor var_1040_cast_fp16 = transpose(perm = var_1039, x = var_1038_cast_fp16)[name = string("transpose_124")]; + tensor x_107_cast_fp16 = reshape(shape = concat_8, x = var_1040_cast_fp16)[name = string("x_107_cast_fp16")]; + tensor var_1044_to_fp16 = const()[name = string("op_1044_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(338960256)))]; + tensor var_1045_to_fp16 = const()[name = string("op_1045_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(342237120)))]; + tensor linear_51_cast_fp16 = linear(bias = var_1045_to_fp16, weight = var_1044_to_fp16, x = x_107_cast_fp16)[name = string("linear_51_cast_fp16")]; + tensor x_109_cast_fp16 = add(x = x_103_cast_fp16, y = linear_51_cast_fp16)[name = string("x_109_cast_fp16")]; + tensor var_1052_axes_0 = const()[name = string("op_1052_axes_0"), val = tensor([-1])]; + tensor blocks_8_mlp_ln_weight_to_fp16 = const()[name = string("blocks_8_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(342239744)))]; + tensor blocks_8_mlp_ln_bias_to_fp16 = const()[name = string("blocks_8_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(342242368)))]; + tensor var_1052_cast_fp16 = layer_norm(axes = var_1052_axes_0, beta = blocks_8_mlp_ln_bias_to_fp16, epsilon = var_978_to_fp16, gamma = blocks_8_mlp_ln_weight_to_fp16, x = x_109_cast_fp16)[name = string("op_1052_cast_fp16")]; + tensor var_1061_to_fp16 = const()[name = string("op_1061_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(342244992)))]; + tensor var_1062_to_fp16 = const()[name = string("op_1062_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(355352256)))]; + tensor linear_52_cast_fp16 = linear(bias = var_1062_to_fp16, weight = var_1061_to_fp16, x = var_1052_cast_fp16)[name = string("linear_52_cast_fp16")]; + string x_113_mode_0 = const()[name = string("x_113_mode_0"), val = string("EXACT")]; + tensor x_113_cast_fp16 = gelu(mode = x_113_mode_0, x = linear_52_cast_fp16)[name = string("x_113_cast_fp16")]; + tensor var_1067_to_fp16 = const()[name = string("op_1067_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(355362560)))]; + tensor var_1068_to_fp16 = const()[name = string("op_1068_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(368469824)))]; + tensor linear_53_cast_fp16 = linear(bias = var_1068_to_fp16, weight = var_1067_to_fp16, x = x_113_cast_fp16)[name = string("linear_53_cast_fp16")]; + tensor x_115_cast_fp16 = add(x = x_109_cast_fp16, y = linear_53_cast_fp16)[name = string("x_115_cast_fp16")]; + int32 var_1078 = const()[name = string("op_1078"), val = int32(-1)]; + tensor var_1094_axes_0 = const()[name = string("op_1094_axes_0"), val = tensor([-1])]; + tensor blocks_9_attn_ln_weight_to_fp16 = const()[name = string("blocks_9_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(368472448)))]; + tensor blocks_9_attn_ln_bias_to_fp16 = const()[name = string("blocks_9_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(368475072)))]; + fp16 var_1084_to_fp16 = const()[name = string("op_1084_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1094_cast_fp16 = layer_norm(axes = var_1094_axes_0, beta = blocks_9_attn_ln_bias_to_fp16, epsilon = var_1084_to_fp16, gamma = blocks_9_attn_ln_weight_to_fp16, x = x_115_cast_fp16)[name = string("op_1094_cast_fp16")]; + tensor var_1105_to_fp16 = const()[name = string("op_1105_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(368477696)))]; + tensor var_1106_to_fp16 = const()[name = string("op_1106_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(371754560)))]; + tensor linear_54_cast_fp16 = linear(bias = var_1106_to_fp16, weight = var_1105_to_fp16, x = var_1094_cast_fp16)[name = string("linear_54_cast_fp16")]; + tensor var_1109_to_fp16 = const()[name = string("op_1109_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(371757184)))]; + tensor linear_55_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1109_to_fp16, x = var_1094_cast_fp16)[name = string("linear_55_cast_fp16")]; + tensor var_1113_to_fp16 = const()[name = string("op_1113_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(375034048)))]; + tensor var_1114_to_fp16 = const()[name = string("op_1114_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(378310912)))]; + tensor linear_56_cast_fp16 = linear(bias = var_1114_to_fp16, weight = var_1113_to_fp16, x = var_1094_cast_fp16)[name = string("linear_56_cast_fp16")]; + tensor var_1122 = const()[name = string("op_1122"), val = tensor([1, 1500, 20, -1])]; + tensor var_1123_cast_fp16 = reshape(shape = var_1122, x = linear_54_cast_fp16)[name = string("op_1123_cast_fp16")]; + tensor const_242_to_fp16 = const()[name = string("const_242_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_39_cast_fp16 = mul(x = var_1123_cast_fp16, y = const_242_to_fp16)[name = string("q_39_cast_fp16")]; + tensor var_1129 = const()[name = string("op_1129"), val = tensor([1, 1500, 20, -1])]; + tensor var_1130_cast_fp16 = reshape(shape = var_1129, x = linear_55_cast_fp16)[name = string("op_1130_cast_fp16")]; + tensor const_243_to_fp16 = const()[name = string("const_243_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_39_cast_fp16 = mul(x = var_1130_cast_fp16, y = const_243_to_fp16)[name = string("k_39_cast_fp16")]; + tensor var_1136 = const()[name = string("op_1136"), val = tensor([1, 1500, 20, -1])]; + tensor var_1137_cast_fp16 = reshape(shape = var_1136, x = linear_56_cast_fp16)[name = string("op_1137_cast_fp16")]; + tensor var_1138 = const()[name = string("op_1138"), val = tensor([0, 2, -3, -1])]; + bool qk_19_transpose_x_0 = const()[name = string("qk_19_transpose_x_0"), val = bool(false)]; + bool qk_19_transpose_y_0 = const()[name = string("qk_19_transpose_y_0"), val = bool(false)]; + tensor transpose_82_perm_0 = const()[name = string("transpose_82_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_83_perm_0 = const()[name = string("transpose_83_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_83 = transpose(perm = transpose_83_perm_0, x = k_39_cast_fp16)[name = string("transpose_122")]; + tensor transpose_82 = transpose(perm = transpose_82_perm_0, x = q_39_cast_fp16)[name = string("transpose_123")]; + tensor qk_19_cast_fp16 = matmul(transpose_x = qk_19_transpose_x_0, transpose_y = qk_19_transpose_y_0, x = transpose_82, y = transpose_83)[name = string("qk_19_cast_fp16")]; + tensor var_1142_cast_fp16 = softmax(axis = var_1078, x = qk_19_cast_fp16)[name = string("op_1142_cast_fp16")]; + bool var_1144_transpose_x_0 = const()[name = string("op_1144_transpose_x_0"), val = bool(false)]; + bool var_1144_transpose_y_0 = const()[name = string("op_1144_transpose_y_0"), val = bool(false)]; + tensor v_39_cast_fp16 = transpose(perm = var_1138, x = var_1137_cast_fp16)[name = string("transpose_121")]; + tensor var_1144_cast_fp16 = matmul(transpose_x = var_1144_transpose_x_0, transpose_y = var_1144_transpose_y_0, x = var_1142_cast_fp16, y = v_39_cast_fp16)[name = string("op_1144_cast_fp16")]; + tensor var_1145 = const()[name = string("op_1145"), val = tensor([0, 2, 1, 3])]; + tensor concat_9 = const()[name = string("concat_9"), val = tensor([1, 1500, 1280])]; + tensor var_1146_cast_fp16 = transpose(perm = var_1145, x = var_1144_cast_fp16)[name = string("transpose_120")]; + tensor x_119_cast_fp16 = reshape(shape = concat_9, x = var_1146_cast_fp16)[name = string("x_119_cast_fp16")]; + tensor var_1150_to_fp16 = const()[name = string("op_1150_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(378313536)))]; + tensor var_1151_to_fp16 = const()[name = string("op_1151_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(381590400)))]; + tensor linear_57_cast_fp16 = linear(bias = var_1151_to_fp16, weight = var_1150_to_fp16, x = x_119_cast_fp16)[name = string("linear_57_cast_fp16")]; + tensor x_121_cast_fp16 = add(x = x_115_cast_fp16, y = linear_57_cast_fp16)[name = string("x_121_cast_fp16")]; + tensor var_1158_axes_0 = const()[name = string("op_1158_axes_0"), val = tensor([-1])]; + tensor blocks_9_mlp_ln_weight_to_fp16 = const()[name = string("blocks_9_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(381593024)))]; + tensor blocks_9_mlp_ln_bias_to_fp16 = const()[name = string("blocks_9_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(381595648)))]; + tensor var_1158_cast_fp16 = layer_norm(axes = var_1158_axes_0, beta = blocks_9_mlp_ln_bias_to_fp16, epsilon = var_1084_to_fp16, gamma = blocks_9_mlp_ln_weight_to_fp16, x = x_121_cast_fp16)[name = string("op_1158_cast_fp16")]; + tensor var_1167_to_fp16 = const()[name = string("op_1167_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(381598272)))]; + tensor var_1168_to_fp16 = const()[name = string("op_1168_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(394705536)))]; + tensor linear_58_cast_fp16 = linear(bias = var_1168_to_fp16, weight = var_1167_to_fp16, x = var_1158_cast_fp16)[name = string("linear_58_cast_fp16")]; + string x_125_mode_0 = const()[name = string("x_125_mode_0"), val = string("EXACT")]; + tensor x_125_cast_fp16 = gelu(mode = x_125_mode_0, x = linear_58_cast_fp16)[name = string("x_125_cast_fp16")]; + tensor var_1173_to_fp16 = const()[name = string("op_1173_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(394715840)))]; + tensor var_1174_to_fp16 = const()[name = string("op_1174_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(407823104)))]; + tensor linear_59_cast_fp16 = linear(bias = var_1174_to_fp16, weight = var_1173_to_fp16, x = x_125_cast_fp16)[name = string("linear_59_cast_fp16")]; + tensor x_127_cast_fp16 = add(x = x_121_cast_fp16, y = linear_59_cast_fp16)[name = string("x_127_cast_fp16")]; + int32 var_1184 = const()[name = string("op_1184"), val = int32(-1)]; + tensor var_1200_axes_0 = const()[name = string("op_1200_axes_0"), val = tensor([-1])]; + tensor blocks_10_attn_ln_weight_to_fp16 = const()[name = string("blocks_10_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(407825728)))]; + tensor blocks_10_attn_ln_bias_to_fp16 = const()[name = string("blocks_10_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(407828352)))]; + fp16 var_1190_to_fp16 = const()[name = string("op_1190_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1200_cast_fp16 = layer_norm(axes = var_1200_axes_0, beta = blocks_10_attn_ln_bias_to_fp16, epsilon = var_1190_to_fp16, gamma = blocks_10_attn_ln_weight_to_fp16, x = x_127_cast_fp16)[name = string("op_1200_cast_fp16")]; + tensor var_1211_to_fp16 = const()[name = string("op_1211_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(407830976)))]; + tensor var_1212_to_fp16 = const()[name = string("op_1212_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(411107840)))]; + tensor linear_60_cast_fp16 = linear(bias = var_1212_to_fp16, weight = var_1211_to_fp16, x = var_1200_cast_fp16)[name = string("linear_60_cast_fp16")]; + tensor var_1215_to_fp16 = const()[name = string("op_1215_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(411110464)))]; + tensor linear_61_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1215_to_fp16, x = var_1200_cast_fp16)[name = string("linear_61_cast_fp16")]; + tensor var_1219_to_fp16 = const()[name = string("op_1219_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(414387328)))]; + tensor var_1220_to_fp16 = const()[name = string("op_1220_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(417664192)))]; + tensor linear_62_cast_fp16 = linear(bias = var_1220_to_fp16, weight = var_1219_to_fp16, x = var_1200_cast_fp16)[name = string("linear_62_cast_fp16")]; + tensor var_1228 = const()[name = string("op_1228"), val = tensor([1, 1500, 20, -1])]; + tensor var_1229_cast_fp16 = reshape(shape = var_1228, x = linear_60_cast_fp16)[name = string("op_1229_cast_fp16")]; + tensor const_244_to_fp16 = const()[name = string("const_244_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_43_cast_fp16 = mul(x = var_1229_cast_fp16, y = const_244_to_fp16)[name = string("q_43_cast_fp16")]; + tensor var_1235 = const()[name = string("op_1235"), val = tensor([1, 1500, 20, -1])]; + tensor var_1236_cast_fp16 = reshape(shape = var_1235, x = linear_61_cast_fp16)[name = string("op_1236_cast_fp16")]; + tensor const_245_to_fp16 = const()[name = string("const_245_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_43_cast_fp16 = mul(x = var_1236_cast_fp16, y = const_245_to_fp16)[name = string("k_43_cast_fp16")]; + tensor var_1242 = const()[name = string("op_1242"), val = tensor([1, 1500, 20, -1])]; + tensor var_1243_cast_fp16 = reshape(shape = var_1242, x = linear_62_cast_fp16)[name = string("op_1243_cast_fp16")]; + tensor var_1244 = const()[name = string("op_1244"), val = tensor([0, 2, -3, -1])]; + bool qk_21_transpose_x_0 = const()[name = string("qk_21_transpose_x_0"), val = bool(false)]; + bool qk_21_transpose_y_0 = const()[name = string("qk_21_transpose_y_0"), val = bool(false)]; + tensor transpose_84_perm_0 = const()[name = string("transpose_84_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_85_perm_0 = const()[name = string("transpose_85_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_85 = transpose(perm = transpose_85_perm_0, x = k_43_cast_fp16)[name = string("transpose_118")]; + tensor transpose_84 = transpose(perm = transpose_84_perm_0, x = q_43_cast_fp16)[name = string("transpose_119")]; + tensor qk_21_cast_fp16 = matmul(transpose_x = qk_21_transpose_x_0, transpose_y = qk_21_transpose_y_0, x = transpose_84, y = transpose_85)[name = string("qk_21_cast_fp16")]; + tensor var_1248_cast_fp16 = softmax(axis = var_1184, x = qk_21_cast_fp16)[name = string("op_1248_cast_fp16")]; + bool var_1250_transpose_x_0 = const()[name = string("op_1250_transpose_x_0"), val = bool(false)]; + bool var_1250_transpose_y_0 = const()[name = string("op_1250_transpose_y_0"), val = bool(false)]; + tensor v_43_cast_fp16 = transpose(perm = var_1244, x = var_1243_cast_fp16)[name = string("transpose_117")]; + tensor var_1250_cast_fp16 = matmul(transpose_x = var_1250_transpose_x_0, transpose_y = var_1250_transpose_y_0, x = var_1248_cast_fp16, y = v_43_cast_fp16)[name = string("op_1250_cast_fp16")]; + tensor var_1251 = const()[name = string("op_1251"), val = tensor([0, 2, 1, 3])]; + tensor concat_10 = const()[name = string("concat_10"), val = tensor([1, 1500, 1280])]; + tensor var_1252_cast_fp16 = transpose(perm = var_1251, x = var_1250_cast_fp16)[name = string("transpose_116")]; + tensor x_131_cast_fp16 = reshape(shape = concat_10, x = var_1252_cast_fp16)[name = string("x_131_cast_fp16")]; + tensor var_1256_to_fp16 = const()[name = string("op_1256_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(417666816)))]; + tensor var_1257_to_fp16 = const()[name = string("op_1257_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(420943680)))]; + tensor linear_63_cast_fp16 = linear(bias = var_1257_to_fp16, weight = var_1256_to_fp16, x = x_131_cast_fp16)[name = string("linear_63_cast_fp16")]; + tensor x_133_cast_fp16 = add(x = x_127_cast_fp16, y = linear_63_cast_fp16)[name = string("x_133_cast_fp16")]; + tensor var_1264_axes_0 = const()[name = string("op_1264_axes_0"), val = tensor([-1])]; + tensor blocks_10_mlp_ln_weight_to_fp16 = const()[name = string("blocks_10_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(420946304)))]; + tensor blocks_10_mlp_ln_bias_to_fp16 = const()[name = string("blocks_10_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(420948928)))]; + tensor var_1264_cast_fp16 = layer_norm(axes = var_1264_axes_0, beta = blocks_10_mlp_ln_bias_to_fp16, epsilon = var_1190_to_fp16, gamma = blocks_10_mlp_ln_weight_to_fp16, x = x_133_cast_fp16)[name = string("op_1264_cast_fp16")]; + tensor var_1273_to_fp16 = const()[name = string("op_1273_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(420951552)))]; + tensor var_1274_to_fp16 = const()[name = string("op_1274_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(434058816)))]; + tensor linear_64_cast_fp16 = linear(bias = var_1274_to_fp16, weight = var_1273_to_fp16, x = var_1264_cast_fp16)[name = string("linear_64_cast_fp16")]; + string x_137_mode_0 = const()[name = string("x_137_mode_0"), val = string("EXACT")]; + tensor x_137_cast_fp16 = gelu(mode = x_137_mode_0, x = linear_64_cast_fp16)[name = string("x_137_cast_fp16")]; + tensor var_1279_to_fp16 = const()[name = string("op_1279_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(434069120)))]; + tensor var_1280_to_fp16 = const()[name = string("op_1280_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(447176384)))]; + tensor linear_65_cast_fp16 = linear(bias = var_1280_to_fp16, weight = var_1279_to_fp16, x = x_137_cast_fp16)[name = string("linear_65_cast_fp16")]; + tensor x_139_cast_fp16 = add(x = x_133_cast_fp16, y = linear_65_cast_fp16)[name = string("x_139_cast_fp16")]; + int32 var_1290 = const()[name = string("op_1290"), val = int32(-1)]; + tensor var_1306_axes_0 = const()[name = string("op_1306_axes_0"), val = tensor([-1])]; + tensor blocks_11_attn_ln_weight_to_fp16 = const()[name = string("blocks_11_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(447179008)))]; + tensor blocks_11_attn_ln_bias_to_fp16 = const()[name = string("blocks_11_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(447181632)))]; + fp16 var_1296_to_fp16 = const()[name = string("op_1296_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1306_cast_fp16 = layer_norm(axes = var_1306_axes_0, beta = blocks_11_attn_ln_bias_to_fp16, epsilon = var_1296_to_fp16, gamma = blocks_11_attn_ln_weight_to_fp16, x = x_139_cast_fp16)[name = string("op_1306_cast_fp16")]; + tensor var_1317_to_fp16 = const()[name = string("op_1317_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(447184256)))]; + tensor var_1318_to_fp16 = const()[name = string("op_1318_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(450461120)))]; + tensor linear_66_cast_fp16 = linear(bias = var_1318_to_fp16, weight = var_1317_to_fp16, x = var_1306_cast_fp16)[name = string("linear_66_cast_fp16")]; + tensor var_1321_to_fp16 = const()[name = string("op_1321_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(450463744)))]; + tensor linear_67_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1321_to_fp16, x = var_1306_cast_fp16)[name = string("linear_67_cast_fp16")]; + tensor var_1325_to_fp16 = const()[name = string("op_1325_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(453740608)))]; + tensor var_1326_to_fp16 = const()[name = string("op_1326_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(457017472)))]; + tensor linear_68_cast_fp16 = linear(bias = var_1326_to_fp16, weight = var_1325_to_fp16, x = var_1306_cast_fp16)[name = string("linear_68_cast_fp16")]; + tensor var_1334 = const()[name = string("op_1334"), val = tensor([1, 1500, 20, -1])]; + tensor var_1335_cast_fp16 = reshape(shape = var_1334, x = linear_66_cast_fp16)[name = string("op_1335_cast_fp16")]; + tensor const_246_to_fp16 = const()[name = string("const_246_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_47_cast_fp16 = mul(x = var_1335_cast_fp16, y = const_246_to_fp16)[name = string("q_47_cast_fp16")]; + tensor var_1341 = const()[name = string("op_1341"), val = tensor([1, 1500, 20, -1])]; + tensor var_1342_cast_fp16 = reshape(shape = var_1341, x = linear_67_cast_fp16)[name = string("op_1342_cast_fp16")]; + tensor const_247_to_fp16 = const()[name = string("const_247_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_47_cast_fp16 = mul(x = var_1342_cast_fp16, y = const_247_to_fp16)[name = string("k_47_cast_fp16")]; + tensor var_1348 = const()[name = string("op_1348"), val = tensor([1, 1500, 20, -1])]; + tensor var_1349_cast_fp16 = reshape(shape = var_1348, x = linear_68_cast_fp16)[name = string("op_1349_cast_fp16")]; + tensor var_1350 = const()[name = string("op_1350"), val = tensor([0, 2, -3, -1])]; + bool qk_23_transpose_x_0 = const()[name = string("qk_23_transpose_x_0"), val = bool(false)]; + bool qk_23_transpose_y_0 = const()[name = string("qk_23_transpose_y_0"), val = bool(false)]; + tensor transpose_86_perm_0 = const()[name = string("transpose_86_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_87_perm_0 = const()[name = string("transpose_87_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_87 = transpose(perm = transpose_87_perm_0, x = k_47_cast_fp16)[name = string("transpose_114")]; + tensor transpose_86 = transpose(perm = transpose_86_perm_0, x = q_47_cast_fp16)[name = string("transpose_115")]; + tensor qk_23_cast_fp16 = matmul(transpose_x = qk_23_transpose_x_0, transpose_y = qk_23_transpose_y_0, x = transpose_86, y = transpose_87)[name = string("qk_23_cast_fp16")]; + tensor var_1354_cast_fp16 = softmax(axis = var_1290, x = qk_23_cast_fp16)[name = string("op_1354_cast_fp16")]; + bool var_1356_transpose_x_0 = const()[name = string("op_1356_transpose_x_0"), val = bool(false)]; + bool var_1356_transpose_y_0 = const()[name = string("op_1356_transpose_y_0"), val = bool(false)]; + tensor v_47_cast_fp16 = transpose(perm = var_1350, x = var_1349_cast_fp16)[name = string("transpose_113")]; + tensor var_1356_cast_fp16 = matmul(transpose_x = var_1356_transpose_x_0, transpose_y = var_1356_transpose_y_0, x = var_1354_cast_fp16, y = v_47_cast_fp16)[name = string("op_1356_cast_fp16")]; + tensor var_1357 = const()[name = string("op_1357"), val = tensor([0, 2, 1, 3])]; + tensor concat_11 = const()[name = string("concat_11"), val = tensor([1, 1500, 1280])]; + tensor var_1358_cast_fp16 = transpose(perm = var_1357, x = var_1356_cast_fp16)[name = string("transpose_112")]; + tensor x_143_cast_fp16 = reshape(shape = concat_11, x = var_1358_cast_fp16)[name = string("x_143_cast_fp16")]; + tensor var_1362_to_fp16 = const()[name = string("op_1362_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(457020096)))]; + tensor var_1363_to_fp16 = const()[name = string("op_1363_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(460296960)))]; + tensor linear_69_cast_fp16 = linear(bias = var_1363_to_fp16, weight = var_1362_to_fp16, x = x_143_cast_fp16)[name = string("linear_69_cast_fp16")]; + tensor x_145_cast_fp16 = add(x = x_139_cast_fp16, y = linear_69_cast_fp16)[name = string("x_145_cast_fp16")]; + tensor var_1370_axes_0 = const()[name = string("op_1370_axes_0"), val = tensor([-1])]; + tensor blocks_11_mlp_ln_weight_to_fp16 = const()[name = string("blocks_11_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(460299584)))]; + tensor blocks_11_mlp_ln_bias_to_fp16 = const()[name = string("blocks_11_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(460302208)))]; + tensor var_1370_cast_fp16 = layer_norm(axes = var_1370_axes_0, beta = blocks_11_mlp_ln_bias_to_fp16, epsilon = var_1296_to_fp16, gamma = blocks_11_mlp_ln_weight_to_fp16, x = x_145_cast_fp16)[name = string("op_1370_cast_fp16")]; + tensor var_1379_to_fp16 = const()[name = string("op_1379_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(460304832)))]; + tensor var_1380_to_fp16 = const()[name = string("op_1380_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(473412096)))]; + tensor linear_70_cast_fp16 = linear(bias = var_1380_to_fp16, weight = var_1379_to_fp16, x = var_1370_cast_fp16)[name = string("linear_70_cast_fp16")]; + string x_149_mode_0 = const()[name = string("x_149_mode_0"), val = string("EXACT")]; + tensor x_149_cast_fp16 = gelu(mode = x_149_mode_0, x = linear_70_cast_fp16)[name = string("x_149_cast_fp16")]; + tensor var_1385_to_fp16 = const()[name = string("op_1385_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(473422400)))]; + tensor var_1386_to_fp16 = const()[name = string("op_1386_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(486529664)))]; + tensor linear_71_cast_fp16 = linear(bias = var_1386_to_fp16, weight = var_1385_to_fp16, x = x_149_cast_fp16)[name = string("linear_71_cast_fp16")]; + tensor x_151_cast_fp16 = add(x = x_145_cast_fp16, y = linear_71_cast_fp16)[name = string("x_151_cast_fp16")]; + int32 var_1396 = const()[name = string("op_1396"), val = int32(-1)]; + tensor var_1412_axes_0 = const()[name = string("op_1412_axes_0"), val = tensor([-1])]; + tensor blocks_12_attn_ln_weight_to_fp16 = const()[name = string("blocks_12_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(486532288)))]; + tensor blocks_12_attn_ln_bias_to_fp16 = const()[name = string("blocks_12_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(486534912)))]; + fp16 var_1402_to_fp16 = const()[name = string("op_1402_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1412_cast_fp16 = layer_norm(axes = var_1412_axes_0, beta = blocks_12_attn_ln_bias_to_fp16, epsilon = var_1402_to_fp16, gamma = blocks_12_attn_ln_weight_to_fp16, x = x_151_cast_fp16)[name = string("op_1412_cast_fp16")]; + tensor var_1423_to_fp16 = const()[name = string("op_1423_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(486537536)))]; + tensor var_1424_to_fp16 = const()[name = string("op_1424_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(489814400)))]; + tensor linear_72_cast_fp16 = linear(bias = var_1424_to_fp16, weight = var_1423_to_fp16, x = var_1412_cast_fp16)[name = string("linear_72_cast_fp16")]; + tensor var_1427_to_fp16 = const()[name = string("op_1427_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(489817024)))]; + tensor linear_73_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1427_to_fp16, x = var_1412_cast_fp16)[name = string("linear_73_cast_fp16")]; + tensor var_1431_to_fp16 = const()[name = string("op_1431_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(493093888)))]; + tensor var_1432_to_fp16 = const()[name = string("op_1432_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(496370752)))]; + tensor linear_74_cast_fp16 = linear(bias = var_1432_to_fp16, weight = var_1431_to_fp16, x = var_1412_cast_fp16)[name = string("linear_74_cast_fp16")]; + tensor var_1440 = const()[name = string("op_1440"), val = tensor([1, 1500, 20, -1])]; + tensor var_1441_cast_fp16 = reshape(shape = var_1440, x = linear_72_cast_fp16)[name = string("op_1441_cast_fp16")]; + tensor const_248_to_fp16 = const()[name = string("const_248_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_51_cast_fp16 = mul(x = var_1441_cast_fp16, y = const_248_to_fp16)[name = string("q_51_cast_fp16")]; + tensor var_1447 = const()[name = string("op_1447"), val = tensor([1, 1500, 20, -1])]; + tensor var_1448_cast_fp16 = reshape(shape = var_1447, x = linear_73_cast_fp16)[name = string("op_1448_cast_fp16")]; + tensor const_249_to_fp16 = const()[name = string("const_249_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_51_cast_fp16 = mul(x = var_1448_cast_fp16, y = const_249_to_fp16)[name = string("k_51_cast_fp16")]; + tensor var_1454 = const()[name = string("op_1454"), val = tensor([1, 1500, 20, -1])]; + tensor var_1455_cast_fp16 = reshape(shape = var_1454, x = linear_74_cast_fp16)[name = string("op_1455_cast_fp16")]; + tensor var_1456 = const()[name = string("op_1456"), val = tensor([0, 2, -3, -1])]; + bool qk_25_transpose_x_0 = const()[name = string("qk_25_transpose_x_0"), val = bool(false)]; + bool qk_25_transpose_y_0 = const()[name = string("qk_25_transpose_y_0"), val = bool(false)]; + tensor transpose_88_perm_0 = const()[name = string("transpose_88_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_89_perm_0 = const()[name = string("transpose_89_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_89 = transpose(perm = transpose_89_perm_0, x = k_51_cast_fp16)[name = string("transpose_110")]; + tensor transpose_88 = transpose(perm = transpose_88_perm_0, x = q_51_cast_fp16)[name = string("transpose_111")]; + tensor qk_25_cast_fp16 = matmul(transpose_x = qk_25_transpose_x_0, transpose_y = qk_25_transpose_y_0, x = transpose_88, y = transpose_89)[name = string("qk_25_cast_fp16")]; + tensor var_1460_cast_fp16 = softmax(axis = var_1396, x = qk_25_cast_fp16)[name = string("op_1460_cast_fp16")]; + bool var_1462_transpose_x_0 = const()[name = string("op_1462_transpose_x_0"), val = bool(false)]; + bool var_1462_transpose_y_0 = const()[name = string("op_1462_transpose_y_0"), val = bool(false)]; + tensor v_51_cast_fp16 = transpose(perm = var_1456, x = var_1455_cast_fp16)[name = string("transpose_109")]; + tensor var_1462_cast_fp16 = matmul(transpose_x = var_1462_transpose_x_0, transpose_y = var_1462_transpose_y_0, x = var_1460_cast_fp16, y = v_51_cast_fp16)[name = string("op_1462_cast_fp16")]; + tensor var_1463 = const()[name = string("op_1463"), val = tensor([0, 2, 1, 3])]; + tensor concat_12 = const()[name = string("concat_12"), val = tensor([1, 1500, 1280])]; + tensor var_1464_cast_fp16 = transpose(perm = var_1463, x = var_1462_cast_fp16)[name = string("transpose_108")]; + tensor x_155_cast_fp16 = reshape(shape = concat_12, x = var_1464_cast_fp16)[name = string("x_155_cast_fp16")]; + tensor var_1468_to_fp16 = const()[name = string("op_1468_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(496373376)))]; + tensor var_1469_to_fp16 = const()[name = string("op_1469_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(499650240)))]; + tensor linear_75_cast_fp16 = linear(bias = var_1469_to_fp16, weight = var_1468_to_fp16, x = x_155_cast_fp16)[name = string("linear_75_cast_fp16")]; + tensor x_157_cast_fp16 = add(x = x_151_cast_fp16, y = linear_75_cast_fp16)[name = string("x_157_cast_fp16")]; + tensor var_1476_axes_0 = const()[name = string("op_1476_axes_0"), val = tensor([-1])]; + tensor blocks_12_mlp_ln_weight_to_fp16 = const()[name = string("blocks_12_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(499652864)))]; + tensor blocks_12_mlp_ln_bias_to_fp16 = const()[name = string("blocks_12_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(499655488)))]; + tensor var_1476_cast_fp16 = layer_norm(axes = var_1476_axes_0, beta = blocks_12_mlp_ln_bias_to_fp16, epsilon = var_1402_to_fp16, gamma = blocks_12_mlp_ln_weight_to_fp16, x = x_157_cast_fp16)[name = string("op_1476_cast_fp16")]; + tensor var_1485_to_fp16 = const()[name = string("op_1485_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(499658112)))]; + tensor var_1486_to_fp16 = const()[name = string("op_1486_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(512765376)))]; + tensor linear_76_cast_fp16 = linear(bias = var_1486_to_fp16, weight = var_1485_to_fp16, x = var_1476_cast_fp16)[name = string("linear_76_cast_fp16")]; + string x_161_mode_0 = const()[name = string("x_161_mode_0"), val = string("EXACT")]; + tensor x_161_cast_fp16 = gelu(mode = x_161_mode_0, x = linear_76_cast_fp16)[name = string("x_161_cast_fp16")]; + tensor var_1491_to_fp16 = const()[name = string("op_1491_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(512775680)))]; + tensor var_1492_to_fp16 = const()[name = string("op_1492_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(525882944)))]; + tensor linear_77_cast_fp16 = linear(bias = var_1492_to_fp16, weight = var_1491_to_fp16, x = x_161_cast_fp16)[name = string("linear_77_cast_fp16")]; + tensor x_163_cast_fp16 = add(x = x_157_cast_fp16, y = linear_77_cast_fp16)[name = string("x_163_cast_fp16")]; + int32 var_1502 = const()[name = string("op_1502"), val = int32(-1)]; + tensor var_1518_axes_0 = const()[name = string("op_1518_axes_0"), val = tensor([-1])]; + tensor blocks_13_attn_ln_weight_to_fp16 = const()[name = string("blocks_13_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(525885568)))]; + tensor blocks_13_attn_ln_bias_to_fp16 = const()[name = string("blocks_13_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(525888192)))]; + fp16 var_1508_to_fp16 = const()[name = string("op_1508_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1518_cast_fp16 = layer_norm(axes = var_1518_axes_0, beta = blocks_13_attn_ln_bias_to_fp16, epsilon = var_1508_to_fp16, gamma = blocks_13_attn_ln_weight_to_fp16, x = x_163_cast_fp16)[name = string("op_1518_cast_fp16")]; + tensor var_1529_to_fp16 = const()[name = string("op_1529_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(525890816)))]; + tensor var_1530_to_fp16 = const()[name = string("op_1530_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(529167680)))]; + tensor linear_78_cast_fp16 = linear(bias = var_1530_to_fp16, weight = var_1529_to_fp16, x = var_1518_cast_fp16)[name = string("linear_78_cast_fp16")]; + tensor var_1533_to_fp16 = const()[name = string("op_1533_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(529170304)))]; + tensor linear_79_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1533_to_fp16, x = var_1518_cast_fp16)[name = string("linear_79_cast_fp16")]; + tensor var_1537_to_fp16 = const()[name = string("op_1537_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(532447168)))]; + tensor var_1538_to_fp16 = const()[name = string("op_1538_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(535724032)))]; + tensor linear_80_cast_fp16 = linear(bias = var_1538_to_fp16, weight = var_1537_to_fp16, x = var_1518_cast_fp16)[name = string("linear_80_cast_fp16")]; + tensor var_1546 = const()[name = string("op_1546"), val = tensor([1, 1500, 20, -1])]; + tensor var_1547_cast_fp16 = reshape(shape = var_1546, x = linear_78_cast_fp16)[name = string("op_1547_cast_fp16")]; + tensor const_250_to_fp16 = const()[name = string("const_250_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_55_cast_fp16 = mul(x = var_1547_cast_fp16, y = const_250_to_fp16)[name = string("q_55_cast_fp16")]; + tensor var_1553 = const()[name = string("op_1553"), val = tensor([1, 1500, 20, -1])]; + tensor var_1554_cast_fp16 = reshape(shape = var_1553, x = linear_79_cast_fp16)[name = string("op_1554_cast_fp16")]; + tensor const_251_to_fp16 = const()[name = string("const_251_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_55_cast_fp16 = mul(x = var_1554_cast_fp16, y = const_251_to_fp16)[name = string("k_55_cast_fp16")]; + tensor var_1560 = const()[name = string("op_1560"), val = tensor([1, 1500, 20, -1])]; + tensor var_1561_cast_fp16 = reshape(shape = var_1560, x = linear_80_cast_fp16)[name = string("op_1561_cast_fp16")]; + tensor var_1562 = const()[name = string("op_1562"), val = tensor([0, 2, -3, -1])]; + bool qk_27_transpose_x_0 = const()[name = string("qk_27_transpose_x_0"), val = bool(false)]; + bool qk_27_transpose_y_0 = const()[name = string("qk_27_transpose_y_0"), val = bool(false)]; + tensor transpose_90_perm_0 = const()[name = string("transpose_90_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_91_perm_0 = const()[name = string("transpose_91_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_91 = transpose(perm = transpose_91_perm_0, x = k_55_cast_fp16)[name = string("transpose_106")]; + tensor transpose_90 = transpose(perm = transpose_90_perm_0, x = q_55_cast_fp16)[name = string("transpose_107")]; + tensor qk_27_cast_fp16 = matmul(transpose_x = qk_27_transpose_x_0, transpose_y = qk_27_transpose_y_0, x = transpose_90, y = transpose_91)[name = string("qk_27_cast_fp16")]; + tensor var_1566_cast_fp16 = softmax(axis = var_1502, x = qk_27_cast_fp16)[name = string("op_1566_cast_fp16")]; + bool var_1568_transpose_x_0 = const()[name = string("op_1568_transpose_x_0"), val = bool(false)]; + bool var_1568_transpose_y_0 = const()[name = string("op_1568_transpose_y_0"), val = bool(false)]; + tensor v_55_cast_fp16 = transpose(perm = var_1562, x = var_1561_cast_fp16)[name = string("transpose_105")]; + tensor var_1568_cast_fp16 = matmul(transpose_x = var_1568_transpose_x_0, transpose_y = var_1568_transpose_y_0, x = var_1566_cast_fp16, y = v_55_cast_fp16)[name = string("op_1568_cast_fp16")]; + tensor var_1569 = const()[name = string("op_1569"), val = tensor([0, 2, 1, 3])]; + tensor concat_13 = const()[name = string("concat_13"), val = tensor([1, 1500, 1280])]; + tensor var_1570_cast_fp16 = transpose(perm = var_1569, x = var_1568_cast_fp16)[name = string("transpose_104")]; + tensor x_167_cast_fp16 = reshape(shape = concat_13, x = var_1570_cast_fp16)[name = string("x_167_cast_fp16")]; + tensor var_1574_to_fp16 = const()[name = string("op_1574_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(535726656)))]; + tensor var_1575_to_fp16 = const()[name = string("op_1575_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(539003520)))]; + tensor linear_81_cast_fp16 = linear(bias = var_1575_to_fp16, weight = var_1574_to_fp16, x = x_167_cast_fp16)[name = string("linear_81_cast_fp16")]; + tensor x_169_cast_fp16 = add(x = x_163_cast_fp16, y = linear_81_cast_fp16)[name = string("x_169_cast_fp16")]; + tensor var_1582_axes_0 = const()[name = string("op_1582_axes_0"), val = tensor([-1])]; + tensor blocks_13_mlp_ln_weight_to_fp16 = const()[name = string("blocks_13_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(539006144)))]; + tensor blocks_13_mlp_ln_bias_to_fp16 = const()[name = string("blocks_13_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(539008768)))]; + tensor var_1582_cast_fp16 = layer_norm(axes = var_1582_axes_0, beta = blocks_13_mlp_ln_bias_to_fp16, epsilon = var_1508_to_fp16, gamma = blocks_13_mlp_ln_weight_to_fp16, x = x_169_cast_fp16)[name = string("op_1582_cast_fp16")]; + tensor var_1591_to_fp16 = const()[name = string("op_1591_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(539011392)))]; + tensor var_1592_to_fp16 = const()[name = string("op_1592_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(552118656)))]; + tensor linear_82_cast_fp16 = linear(bias = var_1592_to_fp16, weight = var_1591_to_fp16, x = var_1582_cast_fp16)[name = string("linear_82_cast_fp16")]; + string x_173_mode_0 = const()[name = string("x_173_mode_0"), val = string("EXACT")]; + tensor x_173_cast_fp16 = gelu(mode = x_173_mode_0, x = linear_82_cast_fp16)[name = string("x_173_cast_fp16")]; + tensor var_1597_to_fp16 = const()[name = string("op_1597_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(552128960)))]; + tensor var_1598_to_fp16 = const()[name = string("op_1598_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(565236224)))]; + tensor linear_83_cast_fp16 = linear(bias = var_1598_to_fp16, weight = var_1597_to_fp16, x = x_173_cast_fp16)[name = string("linear_83_cast_fp16")]; + tensor x_175_cast_fp16 = add(x = x_169_cast_fp16, y = linear_83_cast_fp16)[name = string("x_175_cast_fp16")]; + int32 var_1608 = const()[name = string("op_1608"), val = int32(-1)]; + tensor var_1624_axes_0 = const()[name = string("op_1624_axes_0"), val = tensor([-1])]; + tensor blocks_14_attn_ln_weight_to_fp16 = const()[name = string("blocks_14_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(565238848)))]; + tensor blocks_14_attn_ln_bias_to_fp16 = const()[name = string("blocks_14_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(565241472)))]; + fp16 var_1614_to_fp16 = const()[name = string("op_1614_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1624_cast_fp16 = layer_norm(axes = var_1624_axes_0, beta = blocks_14_attn_ln_bias_to_fp16, epsilon = var_1614_to_fp16, gamma = blocks_14_attn_ln_weight_to_fp16, x = x_175_cast_fp16)[name = string("op_1624_cast_fp16")]; + tensor var_1635_to_fp16 = const()[name = string("op_1635_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(565244096)))]; + tensor var_1636_to_fp16 = const()[name = string("op_1636_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(568520960)))]; + tensor linear_84_cast_fp16 = linear(bias = var_1636_to_fp16, weight = var_1635_to_fp16, x = var_1624_cast_fp16)[name = string("linear_84_cast_fp16")]; + tensor var_1639_to_fp16 = const()[name = string("op_1639_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(568523584)))]; + tensor linear_85_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1639_to_fp16, x = var_1624_cast_fp16)[name = string("linear_85_cast_fp16")]; + tensor var_1643_to_fp16 = const()[name = string("op_1643_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(571800448)))]; + tensor var_1644_to_fp16 = const()[name = string("op_1644_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(575077312)))]; + tensor linear_86_cast_fp16 = linear(bias = var_1644_to_fp16, weight = var_1643_to_fp16, x = var_1624_cast_fp16)[name = string("linear_86_cast_fp16")]; + tensor var_1652 = const()[name = string("op_1652"), val = tensor([1, 1500, 20, -1])]; + tensor var_1653_cast_fp16 = reshape(shape = var_1652, x = linear_84_cast_fp16)[name = string("op_1653_cast_fp16")]; + tensor const_252_to_fp16 = const()[name = string("const_252_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_59_cast_fp16 = mul(x = var_1653_cast_fp16, y = const_252_to_fp16)[name = string("q_59_cast_fp16")]; + tensor var_1659 = const()[name = string("op_1659"), val = tensor([1, 1500, 20, -1])]; + tensor var_1660_cast_fp16 = reshape(shape = var_1659, x = linear_85_cast_fp16)[name = string("op_1660_cast_fp16")]; + tensor const_253_to_fp16 = const()[name = string("const_253_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_59_cast_fp16 = mul(x = var_1660_cast_fp16, y = const_253_to_fp16)[name = string("k_59_cast_fp16")]; + tensor var_1666 = const()[name = string("op_1666"), val = tensor([1, 1500, 20, -1])]; + tensor var_1667_cast_fp16 = reshape(shape = var_1666, x = linear_86_cast_fp16)[name = string("op_1667_cast_fp16")]; + tensor var_1668 = const()[name = string("op_1668"), val = tensor([0, 2, -3, -1])]; + bool qk_29_transpose_x_0 = const()[name = string("qk_29_transpose_x_0"), val = bool(false)]; + bool qk_29_transpose_y_0 = const()[name = string("qk_29_transpose_y_0"), val = bool(false)]; + tensor transpose_92_perm_0 = const()[name = string("transpose_92_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_93_perm_0 = const()[name = string("transpose_93_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_93 = transpose(perm = transpose_93_perm_0, x = k_59_cast_fp16)[name = string("transpose_102")]; + tensor transpose_92 = transpose(perm = transpose_92_perm_0, x = q_59_cast_fp16)[name = string("transpose_103")]; + tensor qk_29_cast_fp16 = matmul(transpose_x = qk_29_transpose_x_0, transpose_y = qk_29_transpose_y_0, x = transpose_92, y = transpose_93)[name = string("qk_29_cast_fp16")]; + tensor var_1672_cast_fp16 = softmax(axis = var_1608, x = qk_29_cast_fp16)[name = string("op_1672_cast_fp16")]; + bool var_1674_transpose_x_0 = const()[name = string("op_1674_transpose_x_0"), val = bool(false)]; + bool var_1674_transpose_y_0 = const()[name = string("op_1674_transpose_y_0"), val = bool(false)]; + tensor v_59_cast_fp16 = transpose(perm = var_1668, x = var_1667_cast_fp16)[name = string("transpose_101")]; + tensor var_1674_cast_fp16 = matmul(transpose_x = var_1674_transpose_x_0, transpose_y = var_1674_transpose_y_0, x = var_1672_cast_fp16, y = v_59_cast_fp16)[name = string("op_1674_cast_fp16")]; + tensor var_1675 = const()[name = string("op_1675"), val = tensor([0, 2, 1, 3])]; + tensor concat_14 = const()[name = string("concat_14"), val = tensor([1, 1500, 1280])]; + tensor var_1676_cast_fp16 = transpose(perm = var_1675, x = var_1674_cast_fp16)[name = string("transpose_100")]; + tensor x_179_cast_fp16 = reshape(shape = concat_14, x = var_1676_cast_fp16)[name = string("x_179_cast_fp16")]; + tensor var_1680_to_fp16 = const()[name = string("op_1680_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(575079936)))]; + tensor var_1681_to_fp16 = const()[name = string("op_1681_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(578356800)))]; + tensor linear_87_cast_fp16 = linear(bias = var_1681_to_fp16, weight = var_1680_to_fp16, x = x_179_cast_fp16)[name = string("linear_87_cast_fp16")]; + tensor x_181_cast_fp16 = add(x = x_175_cast_fp16, y = linear_87_cast_fp16)[name = string("x_181_cast_fp16")]; + tensor var_1688_axes_0 = const()[name = string("op_1688_axes_0"), val = tensor([-1])]; + tensor blocks_14_mlp_ln_weight_to_fp16 = const()[name = string("blocks_14_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(578359424)))]; + tensor blocks_14_mlp_ln_bias_to_fp16 = const()[name = string("blocks_14_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(578362048)))]; + tensor var_1688_cast_fp16 = layer_norm(axes = var_1688_axes_0, beta = blocks_14_mlp_ln_bias_to_fp16, epsilon = var_1614_to_fp16, gamma = blocks_14_mlp_ln_weight_to_fp16, x = x_181_cast_fp16)[name = string("op_1688_cast_fp16")]; + tensor var_1697_to_fp16 = const()[name = string("op_1697_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(578364672)))]; + tensor var_1698_to_fp16 = const()[name = string("op_1698_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(591471936)))]; + tensor linear_88_cast_fp16 = linear(bias = var_1698_to_fp16, weight = var_1697_to_fp16, x = var_1688_cast_fp16)[name = string("linear_88_cast_fp16")]; + string x_185_mode_0 = const()[name = string("x_185_mode_0"), val = string("EXACT")]; + tensor x_185_cast_fp16 = gelu(mode = x_185_mode_0, x = linear_88_cast_fp16)[name = string("x_185_cast_fp16")]; + tensor var_1703_to_fp16 = const()[name = string("op_1703_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(591482240)))]; + tensor var_1704_to_fp16 = const()[name = string("op_1704_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(604589504)))]; + tensor linear_89_cast_fp16 = linear(bias = var_1704_to_fp16, weight = var_1703_to_fp16, x = x_185_cast_fp16)[name = string("linear_89_cast_fp16")]; + tensor x_187_cast_fp16 = add(x = x_181_cast_fp16, y = linear_89_cast_fp16)[name = string("x_187_cast_fp16")]; + int32 var_1714 = const()[name = string("op_1714"), val = int32(-1)]; + tensor var_1730_axes_0 = const()[name = string("op_1730_axes_0"), val = tensor([-1])]; + tensor blocks_15_attn_ln_weight_to_fp16 = const()[name = string("blocks_15_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(604592128)))]; + tensor blocks_15_attn_ln_bias_to_fp16 = const()[name = string("blocks_15_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(604594752)))]; + fp16 var_1720_to_fp16 = const()[name = string("op_1720_to_fp16"), val = fp16(0x1.5p-17)]; + tensor var_1730_cast_fp16 = layer_norm(axes = var_1730_axes_0, beta = blocks_15_attn_ln_bias_to_fp16, epsilon = var_1720_to_fp16, gamma = blocks_15_attn_ln_weight_to_fp16, x = x_187_cast_fp16)[name = string("op_1730_cast_fp16")]; + tensor var_1741_to_fp16 = const()[name = string("op_1741_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(604597376)))]; + tensor var_1742_to_fp16 = const()[name = string("op_1742_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(607874240)))]; + tensor linear_90_cast_fp16 = linear(bias = var_1742_to_fp16, weight = var_1741_to_fp16, x = var_1730_cast_fp16)[name = string("linear_90_cast_fp16")]; + tensor var_1745_to_fp16 = const()[name = string("op_1745_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(607876864)))]; + tensor linear_91_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1745_to_fp16, x = var_1730_cast_fp16)[name = string("linear_91_cast_fp16")]; + tensor var_1749_to_fp16 = const()[name = string("op_1749_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(611153728)))]; + tensor var_1750_to_fp16 = const()[name = string("op_1750_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(614430592)))]; + tensor linear_92_cast_fp16 = linear(bias = var_1750_to_fp16, weight = var_1749_to_fp16, x = var_1730_cast_fp16)[name = string("linear_92_cast_fp16")]; + tensor var_1758 = const()[name = string("op_1758"), val = tensor([1, 1500, 20, -1])]; + tensor var_1759_cast_fp16 = reshape(shape = var_1758, x = linear_90_cast_fp16)[name = string("op_1759_cast_fp16")]; + tensor const_254_to_fp16 = const()[name = string("const_254_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_63_cast_fp16 = mul(x = var_1759_cast_fp16, y = const_254_to_fp16)[name = string("q_63_cast_fp16")]; + tensor var_1765 = const()[name = string("op_1765"), val = tensor([1, 1500, 20, -1])]; + tensor var_1766_cast_fp16 = reshape(shape = var_1765, x = linear_91_cast_fp16)[name = string("op_1766_cast_fp16")]; + tensor const_255_to_fp16 = const()[name = string("const_255_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_63_cast_fp16 = mul(x = var_1766_cast_fp16, y = const_255_to_fp16)[name = string("k_63_cast_fp16")]; + tensor var_1772 = const()[name = string("op_1772"), val = tensor([1, 1500, 20, -1])]; + tensor var_1773_cast_fp16 = reshape(shape = var_1772, x = linear_92_cast_fp16)[name = string("op_1773_cast_fp16")]; + tensor var_1774 = const()[name = string("op_1774"), val = tensor([0, 2, -3, -1])]; + bool qk_31_transpose_x_0 = const()[name = string("qk_31_transpose_x_0"), val = bool(false)]; + bool qk_31_transpose_y_0 = const()[name = string("qk_31_transpose_y_0"), val = bool(false)]; + tensor transpose_94_perm_0 = const()[name = string("transpose_94_perm_0"), val = tensor([0, 2, -3, -1])]; + tensor transpose_95_perm_0 = const()[name = string("transpose_95_perm_0"), val = tensor([0, 2, -1, -3])]; + tensor transpose_95 = transpose(perm = transpose_95_perm_0, x = k_63_cast_fp16)[name = string("transpose_98")]; + tensor transpose_94 = transpose(perm = transpose_94_perm_0, x = q_63_cast_fp16)[name = string("transpose_99")]; + tensor qk_31_cast_fp16 = matmul(transpose_x = qk_31_transpose_x_0, transpose_y = qk_31_transpose_y_0, x = transpose_94, y = transpose_95)[name = string("qk_31_cast_fp16")]; + tensor var_1778_cast_fp16 = softmax(axis = var_1714, x = qk_31_cast_fp16)[name = string("op_1778_cast_fp16")]; + bool var_1780_transpose_x_0 = const()[name = string("op_1780_transpose_x_0"), val = bool(false)]; + bool var_1780_transpose_y_0 = const()[name = string("op_1780_transpose_y_0"), val = bool(false)]; + tensor v_63_cast_fp16 = transpose(perm = var_1774, x = var_1773_cast_fp16)[name = string("transpose_97")]; + tensor var_1780_cast_fp16 = matmul(transpose_x = var_1780_transpose_x_0, transpose_y = var_1780_transpose_y_0, x = var_1778_cast_fp16, y = v_63_cast_fp16)[name = string("op_1780_cast_fp16")]; + tensor var_1781 = const()[name = string("op_1781"), val = tensor([0, 2, 1, 3])]; + tensor concat_15 = const()[name = string("concat_15"), val = tensor([1, 1500, 1280])]; + tensor var_1782_cast_fp16 = transpose(perm = var_1781, x = var_1780_cast_fp16)[name = string("transpose_96")]; + tensor x_191_cast_fp16 = reshape(shape = concat_15, x = var_1782_cast_fp16)[name = string("x_191_cast_fp16")]; + tensor var_1786_to_fp16 = const()[name = string("op_1786_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(614433216)))]; + tensor var_1787_to_fp16 = const()[name = string("op_1787_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(617710080)))]; + tensor linear_93_cast_fp16 = linear(bias = var_1787_to_fp16, weight = var_1786_to_fp16, x = x_191_cast_fp16)[name = string("linear_93_cast_fp16")]; + tensor x_193_cast_fp16_1 = add(x = x_187_cast_fp16, y = linear_93_cast_fp16)[name = string("x_193_cast_fp16")]; + tensor var_1794_axes_0 = const()[name = string("op_1794_axes_0"), val = tensor([-1])]; + tensor blocks_15_mlp_ln_weight_to_fp16 = const()[name = string("blocks_15_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(617712704)))]; + tensor blocks_15_mlp_ln_bias_to_fp16 = const()[name = string("blocks_15_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(617715328)))]; + tensor var_1794_cast_fp16 = layer_norm(axes = var_1794_axes_0, beta = blocks_15_mlp_ln_bias_to_fp16, epsilon = var_1720_to_fp16, gamma = blocks_15_mlp_ln_weight_to_fp16, x = x_193_cast_fp16_1)[name = string("op_1794_cast_fp16")]; + tensor var_1803_to_fp16 = const()[name = string("op_1803_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(617717952)))]; + tensor var_1804_to_fp16 = const()[name = string("op_1804_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(630825216)))]; + tensor linear_94_cast_fp16 = linear(bias = var_1804_to_fp16, weight = var_1803_to_fp16, x = var_1794_cast_fp16)[name = string("linear_94_cast_fp16")]; + string x_197_mode_0 = const()[name = string("x_197_mode_0"), val = string("EXACT")]; + tensor x_197_cast_fp16 = gelu(mode = x_197_mode_0, x = linear_94_cast_fp16)[name = string("x_197_cast_fp16")]; + tensor var_1809_to_fp16 = const()[name = string("op_1809_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(630835520)))]; + tensor var_1810_to_fp16 = const()[name = string("op_1810_to_fp16"), val = tensor(BLOBFILE(path = string("@model_path/weights/0-weight.bin"), offset = uint64(643942784)))]; + tensor linear_95_cast_fp16_1 = linear(bias = var_1810_to_fp16, weight = var_1809_to_fp16, x = x_197_cast_fp16)[name = string("linear_95_cast_fp16")]; + string x_193_cast_fp16_dtype_0 = const()[name = string("x_193_cast_fp16_dtype_0"), val = string("fp32")]; + string linear_95_cast_fp16_dtype_0 = const()[name = string("linear_95_cast_fp16_dtype_0"), val = string("fp32")]; + tensor linear_95_cast_fp16 = cast(dtype = linear_95_cast_fp16_dtype_0, x = linear_95_cast_fp16_1)[name = string("cast_2")]; + tensor x_193_cast_fp16 = cast(dtype = x_193_cast_fp16_dtype_0, x = x_193_cast_fp16_1)[name = string("cast_3")]; + } -> (x_193_cast_fp16, linear_95_cast_fp16); +} \ No newline at end of file