diff --git a/.DS_Store b/.DS_Store new file mode 100644 index 0000000000000000000000000000000000000000..a11cd843ef3e9b919d5a3f92053de8419d55159e Binary files /dev/null and b/.DS_Store differ diff --git a/ggml-base-encoder.mlmodelc.zip b/ggml-base-encoder.mlmodelc.zip index 57caaacf2b3f622ff9cd7056b7be66410525437f..54c4e596a04c7891eabb671f9e110c76082859dc 100644 --- a/ggml-base-encoder.mlmodelc.zip +++ b/ggml-base-encoder.mlmodelc.zip @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:7e6ab77041942572f239b5b602f8aaa1c3ed29d73e3d8f20abea03a773541089 +oid sha256:9cf9278264a9e0f56a6649ca10ab2cde81f6bfe86133a1f08de15006c269cd39 size 37922638 diff --git a/ggml-base-encoder.mlmodelc/analytics/coremldata.bin b/ggml-base-encoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..3f5eb083dfd6b2e7283b81a31687c1b9ea8a3dc3 --- /dev/null +++ b/ggml-base-encoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2441ae34fc7d12946dba7b63379063e856ffc7c3e11ba5f7533efb1450562ca6 +size 207 diff --git a/ggml-base-encoder.mlmodelc/coremldata.bin b/ggml-base-encoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..c732ed7b8caeac3c15acda1c2836f3e1f525a5a8 --- /dev/null +++ b/ggml-base-encoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:05fe28591b40616fa0c34ad7b853133623f5300923ec812acb11459c411acf3b +size 149 diff --git a/ggml-base-encoder.mlmodelc/metadata.json b/ggml-base-encoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..0a32df6173b94e0faa31dcc1c960e260bfc9adfc --- /dev/null +++ b/ggml-base-encoder.mlmodelc/metadata.json @@ -0,0 +1,64 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float32", + "formattedType" : "MultiArray (Float32)", + "shortDescription" : "", + "shape" : "[]", + "name" : "output", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 6, + "mlProgramOperationTypeHistogram" : { + "Linear" : 36, + "Matmul" : 12, + "Cast" : 2, + "Conv" : 2, + "Softmax" : 6, + "Add" : 13, + "LayerNorm" : 13, + "Mul" : 12, + "Transpose" : 25, + "Gelu" : 8, + "Reshape" : 24 + }, + "computePrecision" : "Mixed (Float16, Float32, Int32)", + "isUpdatable" : "0", + "availability" : { + "macOS" : "12.0", + "tvOS" : "15.0", + "watchOS" : "8.0", + "iOS" : "15.0", + "macCatalyst" : "15.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float32", + "formattedType" : "MultiArray (Float32 1 × 80 × 3000)", + "shortDescription" : "", + "shape" : "[1, 80, 3000]", + "name" : "logmel_data", + "type" : "MultiArray" + } + ], + "generatedClassName" : "coreml_encoder_base", + "method" : "predict" + } +] \ No newline at end of file diff --git a/ggml-base-encoder.mlmodelc/model.mil b/ggml-base-encoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..1a556286ea11b368906c6d7a5613793051843109 --- /dev/null +++ b/ggml-base-encoder.mlmodelc/model.mil @@ -0,0 +1,393 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "4.28.4"}, {"coremlc-version", "1436.100.10"}})] +{ + func main(tensor logmel_data) { + tensor var_20 = const()[name = tensor("op_20"), val = tensor(1)]; + tensor var_28 = const()[name = tensor("op_28"), val = tensor([1])]; + tensor var_30 = const()[name = tensor("op_30"), val = tensor([1])]; + tensor var_32_pad_type_0 = const()[name = tensor("op_32_pad_type_0"), val = tensor("custom")]; + tensor var_32_pad_0 = const()[name = tensor("op_32_pad_0"), val = tensor([1, 1])]; + tensor logmel_data_to_fp16_dtype_0 = const()[name = tensor("logmel_data_to_fp16_dtype_0"), val = tensor("fp16")]; + tensor weight_3_to_fp16 = const()[name = tensor("weight_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor bias_3_to_fp16 = const()[name = tensor("bias_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(245888)))]; + tensor cast_187 = cast(dtype = logmel_data_to_fp16_dtype_0, x = logmel_data); + tensor var_32_cast = conv(bias = bias_3_to_fp16, dilations = var_30, groups = var_20, pad = var_32_pad_0, pad_type = var_32_pad_type_0, strides = var_28, weight = weight_3_to_fp16, x = cast_187); + tensor input_1_mode_0 = const()[name = tensor("input_1_mode_0"), val = tensor("EXACT")]; + tensor input_1_cast = gelu(mode = input_1_mode_0, x = var_32_cast); + tensor var_36 = const()[name = tensor("op_36"), val = tensor(1)]; + tensor var_45 = const()[name = tensor("op_45"), val = tensor([2])]; + tensor var_47 = const()[name = tensor("op_47"), val = tensor([1])]; + tensor var_49_pad_type_0 = const()[name = tensor("op_49_pad_type_0"), val = tensor("custom")]; + tensor var_49_pad_0 = const()[name = tensor("op_49_pad_0"), val = tensor([1, 1])]; + tensor weight_7_to_fp16 = const()[name = tensor("weight_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(246976)))]; + tensor bias_7_to_fp16 = const()[name = tensor("bias_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1819904)))]; + tensor var_49_cast = conv(bias = bias_7_to_fp16, dilations = var_47, groups = var_36, pad = var_49_pad_0, pad_type = var_49_pad_type_0, strides = var_45, weight = weight_7_to_fp16, x = input_1_cast); + tensor x_3_mode_0 = const()[name = tensor("x_3_mode_0"), val = tensor("EXACT")]; + tensor x_3_cast = gelu(mode = x_3_mode_0, x = var_49_cast); + tensor var_54 = const()[name = tensor("op_54"), val = tensor([0, 2, 1])]; + tensor positional_embedding_to_fp16 = const()[name = tensor("positional_embedding_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1820992)))]; + tensor transpose_48 = transpose(perm = var_54, x = x_3_cast); + tensor var_57_cast = add(x = transpose_48, y = positional_embedding_to_fp16); + tensor var_70 = const()[name = tensor("op_70"), val = tensor(-1)]; + tensor var_87_axes_0 = const()[name = tensor("op_87_axes_0"), val = tensor([-1])]; + tensor blocks_0_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_0_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3357056)))]; + tensor blocks_0_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_0_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3358144)))]; + tensor var_76_to_fp16 = const()[name = tensor("op_76_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_87_cast = layer_norm(axes = var_87_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_76_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_57_cast); + tensor var_98_to_fp16 = const()[name = tensor("op_98_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3359232)))]; + tensor var_99_to_fp16 = const()[name = tensor("op_99_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3883584)))]; + tensor q_1_cast = linear(bias = var_99_to_fp16, weight = var_98_to_fp16, x = var_87_cast); + tensor var_102_to_fp16 = const()[name = tensor("op_102_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3884672)))]; + tensor k_1_bias_0_to_fp16 = const()[name = tensor("k_1_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4409024)))]; + tensor k_1_cast = linear(bias = k_1_bias_0_to_fp16, weight = var_102_to_fp16, x = var_87_cast); + tensor var_106_to_fp16 = const()[name = tensor("op_106_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4410112)))]; + tensor var_107_to_fp16 = const()[name = tensor("op_107_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4934464)))]; + tensor v_1_cast = linear(bias = var_107_to_fp16, weight = var_106_to_fp16, x = var_87_cast); + tensor var_115 = const()[name = tensor("op_115"), val = tensor([1, 1500, 8, -1])]; + tensor var_116_cast = reshape(shape = var_115, x = q_1_cast); + tensor const_42_to_fp16 = const()[name = tensor("const_42_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_3_cast = mul(x = var_116_cast, y = const_42_to_fp16); + tensor var_122 = const()[name = tensor("op_122"), val = tensor([1, 1500, 8, -1])]; + tensor var_123_cast = reshape(shape = var_122, x = k_1_cast); + tensor const_43_to_fp16 = const()[name = tensor("const_43_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_3_cast = mul(x = var_123_cast, y = const_43_to_fp16); + tensor var_129 = const()[name = tensor("op_129"), val = tensor([1, 1500, 8, -1])]; + tensor var_130_cast = reshape(shape = var_129, x = v_1_cast); + tensor var_131 = const()[name = tensor("op_131"), val = tensor([0, 2, 1, 3])]; + tensor qk_1_transpose_x_0 = const()[name = tensor("qk_1_transpose_x_0"), val = tensor(false)]; + tensor qk_1_transpose_y_0 = const()[name = tensor("qk_1_transpose_y_0"), val = tensor(false)]; + tensor transpose_12_perm_0 = const()[name = tensor("transpose_12_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_13_perm_0 = const()[name = tensor("transpose_13_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_45 = transpose(perm = transpose_13_perm_0, x = k_3_cast); + tensor transpose_46 = transpose(perm = transpose_12_perm_0, x = q_3_cast); + tensor qk_1_cast = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_46, y = transpose_45); + tensor var_135_cast = softmax(axis = var_70, x = qk_1_cast); + tensor var_137_transpose_x_0 = const()[name = tensor("op_137_transpose_x_0"), val = tensor(false)]; + tensor var_137_transpose_y_0 = const()[name = tensor("op_137_transpose_y_0"), val = tensor(false)]; + tensor transpose_47 = transpose(perm = var_131, x = var_130_cast); + tensor var_137_cast = matmul(transpose_x = var_137_transpose_x_0, transpose_y = var_137_transpose_y_0, x = var_135_cast, y = transpose_47); + tensor var_138 = const()[name = tensor("op_138"), val = tensor([0, 2, 1, 3])]; + tensor concat_0 = const()[name = tensor("concat_0"), val = tensor([1, 1500, 512])]; + tensor transpose_44 = transpose(perm = var_138, x = var_137_cast); + tensor x_11_cast = reshape(shape = concat_0, x = transpose_44); + tensor var_143_to_fp16 = const()[name = tensor("op_143_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4935552)))]; + tensor var_144_to_fp16 = const()[name = tensor("op_144_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5459904)))]; + tensor var_145_cast = linear(bias = var_144_to_fp16, weight = var_143_to_fp16, x = x_11_cast); + tensor x_13_cast = add(x = var_57_cast, y = var_145_cast); + tensor var_151_axes_0 = const()[name = tensor("op_151_axes_0"), val = tensor([-1])]; + tensor blocks_0_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5460992)))]; + tensor blocks_0_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5462080)))]; + tensor var_151_cast = layer_norm(axes = var_151_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_76_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast); + tensor var_160_to_fp16 = const()[name = tensor("op_160_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5463168)))]; + tensor var_161_to_fp16 = const()[name = tensor("op_161_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7560384)))]; + tensor input_9_cast = linear(bias = var_161_to_fp16, weight = var_160_to_fp16, x = var_151_cast); + tensor x_17_mode_0 = const()[name = tensor("x_17_mode_0"), val = tensor("EXACT")]; + tensor x_17_cast = gelu(mode = x_17_mode_0, x = input_9_cast); + tensor var_166_to_fp16 = const()[name = tensor("op_166_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7564544)))]; + tensor var_167_to_fp16 = const()[name = tensor("op_167_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9661760)))]; + tensor var_168_cast = linear(bias = var_167_to_fp16, weight = var_166_to_fp16, x = x_17_cast); + tensor x_19_cast = add(x = x_13_cast, y = var_168_cast); + tensor var_177 = const()[name = tensor("op_177"), val = tensor(-1)]; + tensor var_194_axes_0 = const()[name = tensor("op_194_axes_0"), val = tensor([-1])]; + tensor blocks_1_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_1_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9662848)))]; + tensor blocks_1_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_1_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9663936)))]; + tensor var_183_to_fp16 = const()[name = tensor("op_183_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_194_cast = layer_norm(axes = var_194_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_183_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast); + tensor var_205_to_fp16 = const()[name = tensor("op_205_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9665024)))]; + tensor var_206_to_fp16 = const()[name = tensor("op_206_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10189376)))]; + tensor q_5_cast = linear(bias = var_206_to_fp16, weight = var_205_to_fp16, x = var_194_cast); + tensor var_209_to_fp16 = const()[name = tensor("op_209_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10190464)))]; + tensor k_5_bias_0_to_fp16 = const()[name = tensor("k_5_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10714816)))]; + tensor k_5_cast = linear(bias = k_5_bias_0_to_fp16, weight = var_209_to_fp16, x = var_194_cast); + tensor var_213_to_fp16 = const()[name = tensor("op_213_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10715904)))]; + tensor var_214_to_fp16 = const()[name = tensor("op_214_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11240256)))]; + tensor v_5_cast = linear(bias = var_214_to_fp16, weight = var_213_to_fp16, x = var_194_cast); + tensor var_222 = const()[name = tensor("op_222"), val = tensor([1, 1500, 8, -1])]; + tensor var_223_cast = reshape(shape = var_222, x = q_5_cast); + tensor const_44_to_fp16 = const()[name = tensor("const_44_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_7_cast = mul(x = var_223_cast, y = const_44_to_fp16); + tensor var_229 = const()[name = tensor("op_229"), val = tensor([1, 1500, 8, -1])]; + tensor var_230_cast = reshape(shape = var_229, x = k_5_cast); + tensor const_45_to_fp16 = const()[name = tensor("const_45_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_7_cast = mul(x = var_230_cast, y = const_45_to_fp16); + tensor var_236 = const()[name = tensor("op_236"), val = tensor([1, 1500, 8, -1])]; + tensor var_237_cast = reshape(shape = var_236, x = v_5_cast); + tensor var_238 = const()[name = tensor("op_238"), val = tensor([0, 2, 1, 3])]; + tensor qk_3_transpose_x_0 = const()[name = tensor("qk_3_transpose_x_0"), val = tensor(false)]; + tensor qk_3_transpose_y_0 = const()[name = tensor("qk_3_transpose_y_0"), val = tensor(false)]; + tensor transpose_14_perm_0 = const()[name = tensor("transpose_14_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_15_perm_0 = const()[name = tensor("transpose_15_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_41 = transpose(perm = transpose_15_perm_0, x = k_7_cast); + tensor transpose_42 = transpose(perm = transpose_14_perm_0, x = q_7_cast); + tensor qk_3_cast = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_42, y = transpose_41); + tensor var_242_cast = softmax(axis = var_177, x = qk_3_cast); + tensor var_244_transpose_x_0 = const()[name = tensor("op_244_transpose_x_0"), val = tensor(false)]; + tensor var_244_transpose_y_0 = const()[name = tensor("op_244_transpose_y_0"), val = tensor(false)]; + tensor transpose_43 = transpose(perm = var_238, x = var_237_cast); + tensor var_244_cast = matmul(transpose_x = var_244_transpose_x_0, transpose_y = var_244_transpose_y_0, x = var_242_cast, y = transpose_43); + tensor var_245 = const()[name = tensor("op_245"), val = tensor([0, 2, 1, 3])]; + tensor concat_1 = const()[name = tensor("concat_1"), val = tensor([1, 1500, 512])]; + tensor transpose_40 = transpose(perm = var_245, x = var_244_cast); + tensor x_23_cast = reshape(shape = concat_1, x = transpose_40); + tensor var_250_to_fp16 = const()[name = tensor("op_250_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11241344)))]; + tensor var_251_to_fp16 = const()[name = tensor("op_251_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11765696)))]; + tensor var_252_cast = linear(bias = var_251_to_fp16, weight = var_250_to_fp16, x = x_23_cast); + tensor x_25_cast = add(x = x_19_cast, y = var_252_cast); + tensor var_258_axes_0 = const()[name = tensor("op_258_axes_0"), val = tensor([-1])]; + tensor blocks_1_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11766784)))]; + tensor blocks_1_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11767872)))]; + tensor var_258_cast = layer_norm(axes = var_258_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_183_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast); + tensor var_267_to_fp16 = const()[name = tensor("op_267_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11768960)))]; + tensor var_268_to_fp16 = const()[name = tensor("op_268_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13866176)))]; + tensor input_17_cast = linear(bias = var_268_to_fp16, weight = var_267_to_fp16, x = var_258_cast); + tensor x_29_mode_0 = const()[name = tensor("x_29_mode_0"), val = tensor("EXACT")]; + tensor x_29_cast = gelu(mode = x_29_mode_0, x = input_17_cast); + tensor var_273_to_fp16 = const()[name = tensor("op_273_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13870336)))]; + tensor var_274_to_fp16 = const()[name = tensor("op_274_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15967552)))]; + tensor var_275_cast = linear(bias = var_274_to_fp16, weight = var_273_to_fp16, x = x_29_cast); + tensor x_31_cast = add(x = x_25_cast, y = var_275_cast); + tensor var_284 = const()[name = tensor("op_284"), val = tensor(-1)]; + tensor var_301_axes_0 = const()[name = tensor("op_301_axes_0"), val = tensor([-1])]; + tensor blocks_2_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_2_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15968640)))]; + tensor blocks_2_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_2_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15969728)))]; + tensor var_290_to_fp16 = const()[name = tensor("op_290_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_301_cast = layer_norm(axes = var_301_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_290_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast); + tensor var_312_to_fp16 = const()[name = tensor("op_312_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15970816)))]; + tensor var_313_to_fp16 = const()[name = tensor("op_313_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16495168)))]; + tensor q_9_cast = linear(bias = var_313_to_fp16, weight = var_312_to_fp16, x = var_301_cast); + tensor var_316_to_fp16 = const()[name = tensor("op_316_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16496256)))]; + tensor k_9_bias_0_to_fp16 = const()[name = tensor("k_9_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17020608)))]; + tensor k_9_cast = linear(bias = k_9_bias_0_to_fp16, weight = var_316_to_fp16, x = var_301_cast); + tensor var_320_to_fp16 = const()[name = tensor("op_320_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17021696)))]; + tensor var_321_to_fp16 = const()[name = tensor("op_321_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17546048)))]; + tensor v_9_cast = linear(bias = var_321_to_fp16, weight = var_320_to_fp16, x = var_301_cast); + tensor var_329 = const()[name = tensor("op_329"), val = tensor([1, 1500, 8, -1])]; + tensor var_330_cast = reshape(shape = var_329, x = q_9_cast); + tensor const_46_to_fp16 = const()[name = tensor("const_46_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_11_cast = mul(x = var_330_cast, y = const_46_to_fp16); + tensor var_336 = const()[name = tensor("op_336"), val = tensor([1, 1500, 8, -1])]; + tensor var_337_cast = reshape(shape = var_336, x = k_9_cast); + tensor const_47_to_fp16 = const()[name = tensor("const_47_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_11_cast = mul(x = var_337_cast, y = const_47_to_fp16); + tensor var_343 = const()[name = tensor("op_343"), val = tensor([1, 1500, 8, -1])]; + tensor var_344_cast = reshape(shape = var_343, x = v_9_cast); + tensor var_345 = const()[name = tensor("op_345"), val = tensor([0, 2, 1, 3])]; + tensor qk_5_transpose_x_0 = const()[name = tensor("qk_5_transpose_x_0"), val = tensor(false)]; + tensor qk_5_transpose_y_0 = const()[name = tensor("qk_5_transpose_y_0"), val = tensor(false)]; + tensor transpose_16_perm_0 = const()[name = tensor("transpose_16_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_17_perm_0 = const()[name = tensor("transpose_17_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_37 = transpose(perm = transpose_17_perm_0, x = k_11_cast); + tensor transpose_38 = transpose(perm = transpose_16_perm_0, x = q_11_cast); + tensor qk_5_cast = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_38, y = transpose_37); + tensor var_349_cast = softmax(axis = var_284, x = qk_5_cast); + tensor var_351_transpose_x_0 = const()[name = tensor("op_351_transpose_x_0"), val = tensor(false)]; + tensor var_351_transpose_y_0 = const()[name = tensor("op_351_transpose_y_0"), val = tensor(false)]; + tensor transpose_39 = transpose(perm = var_345, x = var_344_cast); + tensor var_351_cast = matmul(transpose_x = var_351_transpose_x_0, transpose_y = var_351_transpose_y_0, x = var_349_cast, y = transpose_39); + tensor var_352 = const()[name = tensor("op_352"), val = tensor([0, 2, 1, 3])]; + tensor concat_2 = const()[name = tensor("concat_2"), val = tensor([1, 1500, 512])]; + tensor transpose_36 = transpose(perm = var_352, x = var_351_cast); + tensor x_35_cast = reshape(shape = concat_2, x = transpose_36); + tensor var_357_to_fp16 = const()[name = tensor("op_357_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17547136)))]; + tensor var_358_to_fp16 = const()[name = tensor("op_358_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18071488)))]; + tensor var_359_cast = linear(bias = var_358_to_fp16, weight = var_357_to_fp16, x = x_35_cast); + tensor x_37_cast = add(x = x_31_cast, y = var_359_cast); + tensor var_365_axes_0 = const()[name = tensor("op_365_axes_0"), val = tensor([-1])]; + tensor blocks_2_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18072576)))]; + tensor blocks_2_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18073664)))]; + tensor var_365_cast = layer_norm(axes = var_365_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_290_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast); + tensor var_374_to_fp16 = const()[name = tensor("op_374_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18074752)))]; + tensor var_375_to_fp16 = const()[name = tensor("op_375_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20171968)))]; + tensor input_25_cast = linear(bias = var_375_to_fp16, weight = var_374_to_fp16, x = var_365_cast); + tensor x_41_mode_0 = const()[name = tensor("x_41_mode_0"), val = tensor("EXACT")]; + tensor x_41_cast = gelu(mode = x_41_mode_0, x = input_25_cast); + tensor var_380_to_fp16 = const()[name = tensor("op_380_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20176128)))]; + tensor var_381_to_fp16 = const()[name = tensor("op_381_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22273344)))]; + tensor var_382_cast = linear(bias = var_381_to_fp16, weight = var_380_to_fp16, x = x_41_cast); + tensor x_43_cast = add(x = x_37_cast, y = var_382_cast); + tensor var_391 = const()[name = tensor("op_391"), val = tensor(-1)]; + tensor var_408_axes_0 = const()[name = tensor("op_408_axes_0"), val = tensor([-1])]; + tensor blocks_3_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_3_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22274432)))]; + tensor blocks_3_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_3_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22275520)))]; + tensor var_397_to_fp16 = const()[name = tensor("op_397_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_408_cast = layer_norm(axes = var_408_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_397_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast); + tensor var_419_to_fp16 = const()[name = tensor("op_419_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22276608)))]; + tensor var_420_to_fp16 = const()[name = tensor("op_420_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22800960)))]; + tensor q_13_cast = linear(bias = var_420_to_fp16, weight = var_419_to_fp16, x = var_408_cast); + tensor var_423_to_fp16 = const()[name = tensor("op_423_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22802048)))]; + tensor k_13_bias_0_to_fp16 = const()[name = tensor("k_13_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23326400)))]; + tensor k_13_cast = linear(bias = k_13_bias_0_to_fp16, weight = var_423_to_fp16, x = var_408_cast); + tensor var_427_to_fp16 = const()[name = tensor("op_427_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23327488)))]; + tensor var_428_to_fp16 = const()[name = tensor("op_428_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23851840)))]; + tensor v_13_cast = linear(bias = var_428_to_fp16, weight = var_427_to_fp16, x = var_408_cast); + tensor var_436 = const()[name = tensor("op_436"), val = tensor([1, 1500, 8, -1])]; + tensor var_437_cast = reshape(shape = var_436, x = q_13_cast); + tensor const_48_to_fp16 = const()[name = tensor("const_48_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_15_cast = mul(x = var_437_cast, y = const_48_to_fp16); + tensor var_443 = const()[name = tensor("op_443"), val = tensor([1, 1500, 8, -1])]; + tensor var_444_cast = reshape(shape = var_443, x = k_13_cast); + tensor const_49_to_fp16 = const()[name = tensor("const_49_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_15_cast = mul(x = var_444_cast, y = const_49_to_fp16); + tensor var_450 = const()[name = tensor("op_450"), val = tensor([1, 1500, 8, -1])]; + tensor var_451_cast = reshape(shape = var_450, x = v_13_cast); + tensor var_452 = const()[name = tensor("op_452"), val = tensor([0, 2, 1, 3])]; + tensor qk_7_transpose_x_0 = const()[name = tensor("qk_7_transpose_x_0"), val = tensor(false)]; + tensor qk_7_transpose_y_0 = const()[name = tensor("qk_7_transpose_y_0"), val = tensor(false)]; + tensor transpose_18_perm_0 = const()[name = tensor("transpose_18_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_19_perm_0 = const()[name = tensor("transpose_19_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_33 = transpose(perm = transpose_19_perm_0, x = k_15_cast); + tensor transpose_34 = transpose(perm = transpose_18_perm_0, x = q_15_cast); + tensor qk_7_cast = matmul(transpose_x = qk_7_transpose_x_0, transpose_y = qk_7_transpose_y_0, x = transpose_34, y = transpose_33); + tensor var_456_cast = softmax(axis = var_391, x = qk_7_cast); + tensor var_458_transpose_x_0 = const()[name = tensor("op_458_transpose_x_0"), val = tensor(false)]; + tensor var_458_transpose_y_0 = const()[name = tensor("op_458_transpose_y_0"), val = tensor(false)]; + tensor transpose_35 = transpose(perm = var_452, x = var_451_cast); + tensor var_458_cast = matmul(transpose_x = var_458_transpose_x_0, transpose_y = var_458_transpose_y_0, x = var_456_cast, y = transpose_35); + tensor var_459 = const()[name = tensor("op_459"), val = tensor([0, 2, 1, 3])]; + tensor concat_3 = const()[name = tensor("concat_3"), val = tensor([1, 1500, 512])]; + tensor transpose_32 = transpose(perm = var_459, x = var_458_cast); + tensor x_47_cast = reshape(shape = concat_3, x = transpose_32); + tensor var_464_to_fp16 = const()[name = tensor("op_464_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23852928)))]; + tensor var_465_to_fp16 = const()[name = tensor("op_465_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24377280)))]; + tensor var_466_cast = linear(bias = var_465_to_fp16, weight = var_464_to_fp16, x = x_47_cast); + tensor x_49_cast = add(x = x_43_cast, y = var_466_cast); + tensor var_472_axes_0 = const()[name = tensor("op_472_axes_0"), val = tensor([-1])]; + tensor blocks_3_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24378368)))]; + tensor blocks_3_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24379456)))]; + tensor var_472_cast = layer_norm(axes = var_472_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_397_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast); + tensor var_481_to_fp16 = const()[name = tensor("op_481_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24380544)))]; + tensor var_482_to_fp16 = const()[name = tensor("op_482_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26477760)))]; + tensor input_33_cast = linear(bias = var_482_to_fp16, weight = var_481_to_fp16, x = var_472_cast); + tensor x_53_mode_0 = const()[name = tensor("x_53_mode_0"), val = tensor("EXACT")]; + tensor x_53_cast = gelu(mode = x_53_mode_0, x = input_33_cast); + tensor var_487_to_fp16 = const()[name = tensor("op_487_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26481920)))]; + tensor var_488_to_fp16 = const()[name = tensor("op_488_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28579136)))]; + tensor var_489_cast = linear(bias = var_488_to_fp16, weight = var_487_to_fp16, x = x_53_cast); + tensor x_55_cast = add(x = x_49_cast, y = var_489_cast); + tensor var_498 = const()[name = tensor("op_498"), val = tensor(-1)]; + tensor var_515_axes_0 = const()[name = tensor("op_515_axes_0"), val = tensor([-1])]; + tensor blocks_4_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_4_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28580224)))]; + tensor blocks_4_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_4_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28581312)))]; + tensor var_504_to_fp16 = const()[name = tensor("op_504_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_515_cast = layer_norm(axes = var_515_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_504_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast); + tensor var_526_to_fp16 = const()[name = tensor("op_526_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28582400)))]; + tensor var_527_to_fp16 = const()[name = tensor("op_527_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29106752)))]; + tensor q_17_cast = linear(bias = var_527_to_fp16, weight = var_526_to_fp16, x = var_515_cast); + tensor var_530_to_fp16 = const()[name = tensor("op_530_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29107840)))]; + tensor k_17_bias_0_to_fp16 = const()[name = tensor("k_17_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29632192)))]; + tensor k_17_cast = linear(bias = k_17_bias_0_to_fp16, weight = var_530_to_fp16, x = var_515_cast); + tensor var_534_to_fp16 = const()[name = tensor("op_534_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29633280)))]; + tensor var_535_to_fp16 = const()[name = tensor("op_535_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30157632)))]; + tensor v_17_cast = linear(bias = var_535_to_fp16, weight = var_534_to_fp16, x = var_515_cast); + tensor var_543 = const()[name = tensor("op_543"), val = tensor([1, 1500, 8, -1])]; + tensor var_544_cast = reshape(shape = var_543, x = q_17_cast); + tensor const_50_to_fp16 = const()[name = tensor("const_50_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_19_cast = mul(x = var_544_cast, y = const_50_to_fp16); + tensor var_550 = const()[name = tensor("op_550"), val = tensor([1, 1500, 8, -1])]; + tensor var_551_cast = reshape(shape = var_550, x = k_17_cast); + tensor const_51_to_fp16 = const()[name = tensor("const_51_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_19_cast = mul(x = var_551_cast, y = const_51_to_fp16); + tensor var_557 = const()[name = tensor("op_557"), val = tensor([1, 1500, 8, -1])]; + tensor var_558_cast = reshape(shape = var_557, x = v_17_cast); + tensor var_559 = const()[name = tensor("op_559"), val = tensor([0, 2, 1, 3])]; + tensor qk_9_transpose_x_0 = const()[name = tensor("qk_9_transpose_x_0"), val = tensor(false)]; + tensor qk_9_transpose_y_0 = const()[name = tensor("qk_9_transpose_y_0"), val = tensor(false)]; + tensor transpose_20_perm_0 = const()[name = tensor("transpose_20_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_21_perm_0 = const()[name = tensor("transpose_21_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_29 = transpose(perm = transpose_21_perm_0, x = k_19_cast); + tensor transpose_30 = transpose(perm = transpose_20_perm_0, x = q_19_cast); + tensor qk_9_cast = matmul(transpose_x = qk_9_transpose_x_0, transpose_y = qk_9_transpose_y_0, x = transpose_30, y = transpose_29); + tensor var_563_cast = softmax(axis = var_498, x = qk_9_cast); + tensor var_565_transpose_x_0 = const()[name = tensor("op_565_transpose_x_0"), val = tensor(false)]; + tensor var_565_transpose_y_0 = const()[name = tensor("op_565_transpose_y_0"), val = tensor(false)]; + tensor transpose_31 = transpose(perm = var_559, x = var_558_cast); + tensor var_565_cast = matmul(transpose_x = var_565_transpose_x_0, transpose_y = var_565_transpose_y_0, x = var_563_cast, y = transpose_31); + tensor var_566 = const()[name = tensor("op_566"), val = tensor([0, 2, 1, 3])]; + tensor concat_4 = const()[name = tensor("concat_4"), val = tensor([1, 1500, 512])]; + tensor transpose_28 = transpose(perm = var_566, x = var_565_cast); + tensor x_59_cast = reshape(shape = concat_4, x = transpose_28); + tensor var_571_to_fp16 = const()[name = tensor("op_571_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30158720)))]; + tensor var_572_to_fp16 = const()[name = tensor("op_572_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30683072)))]; + tensor var_573_cast = linear(bias = var_572_to_fp16, weight = var_571_to_fp16, x = x_59_cast); + tensor x_61_cast = add(x = x_55_cast, y = var_573_cast); + tensor var_579_axes_0 = const()[name = tensor("op_579_axes_0"), val = tensor([-1])]; + tensor blocks_4_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30684160)))]; + tensor blocks_4_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30685248)))]; + tensor var_579_cast = layer_norm(axes = var_579_axes_0, beta = blocks_4_mlp_ln_bias_to_fp16, epsilon = var_504_to_fp16, gamma = blocks_4_mlp_ln_weight_to_fp16, x = x_61_cast); + tensor var_588_to_fp16 = const()[name = tensor("op_588_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30686336)))]; + tensor var_589_to_fp16 = const()[name = tensor("op_589_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(32783552)))]; + tensor input_41_cast = linear(bias = var_589_to_fp16, weight = var_588_to_fp16, x = var_579_cast); + tensor x_65_mode_0 = const()[name = tensor("x_65_mode_0"), val = tensor("EXACT")]; + tensor x_65_cast = gelu(mode = x_65_mode_0, x = input_41_cast); + tensor var_594_to_fp16 = const()[name = tensor("op_594_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(32787712)))]; + tensor var_595_to_fp16 = const()[name = tensor("op_595_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34884928)))]; + tensor var_596_cast = linear(bias = var_595_to_fp16, weight = var_594_to_fp16, x = x_65_cast); + tensor x_67_cast = add(x = x_61_cast, y = var_596_cast); + tensor var_605 = const()[name = tensor("op_605"), val = tensor(-1)]; + tensor var_622_axes_0 = const()[name = tensor("op_622_axes_0"), val = tensor([-1])]; + tensor blocks_5_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_5_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34886016)))]; + tensor blocks_5_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_5_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34887104)))]; + tensor var_611_to_fp16 = const()[name = tensor("op_611_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_622_cast = layer_norm(axes = var_622_axes_0, beta = blocks_5_attn_ln_bias_to_fp16, epsilon = var_611_to_fp16, gamma = blocks_5_attn_ln_weight_to_fp16, x = x_67_cast); + tensor var_633_to_fp16 = const()[name = tensor("op_633_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34888192)))]; + tensor var_634_to_fp16 = const()[name = tensor("op_634_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35412544)))]; + tensor q_21_cast = linear(bias = var_634_to_fp16, weight = var_633_to_fp16, x = var_622_cast); + tensor var_637_to_fp16 = const()[name = tensor("op_637_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35413632)))]; + tensor k_21_bias_0_to_fp16 = const()[name = tensor("k_21_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35937984)))]; + tensor k_21_cast = linear(bias = k_21_bias_0_to_fp16, weight = var_637_to_fp16, x = var_622_cast); + tensor var_641_to_fp16 = const()[name = tensor("op_641_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35939072)))]; + tensor var_642_to_fp16 = const()[name = tensor("op_642_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36463424)))]; + tensor v_21_cast = linear(bias = var_642_to_fp16, weight = var_641_to_fp16, x = var_622_cast); + tensor var_650 = const()[name = tensor("op_650"), val = tensor([1, 1500, 8, -1])]; + tensor var_651_cast = reshape(shape = var_650, x = q_21_cast); + tensor const_52_to_fp16 = const()[name = tensor("const_52_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_cast = mul(x = var_651_cast, y = const_52_to_fp16); + tensor var_657 = const()[name = tensor("op_657"), val = tensor([1, 1500, 8, -1])]; + tensor var_658_cast = reshape(shape = var_657, x = k_21_cast); + tensor const_53_to_fp16 = const()[name = tensor("const_53_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_cast = mul(x = var_658_cast, y = const_53_to_fp16); + tensor var_664 = const()[name = tensor("op_664"), val = tensor([1, 1500, 8, -1])]; + tensor var_665_cast = reshape(shape = var_664, x = v_21_cast); + tensor var_666 = const()[name = tensor("op_666"), val = tensor([0, 2, 1, 3])]; + tensor qk_transpose_x_0 = const()[name = tensor("qk_transpose_x_0"), val = tensor(false)]; + tensor qk_transpose_y_0 = const()[name = tensor("qk_transpose_y_0"), val = tensor(false)]; + tensor transpose_22_perm_0 = const()[name = tensor("transpose_22_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_23_perm_0 = const()[name = tensor("transpose_23_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_25 = transpose(perm = transpose_23_perm_0, x = k_cast); + tensor transpose_26 = transpose(perm = transpose_22_perm_0, x = q_cast); + tensor qk_cast = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_26, y = transpose_25); + tensor var_670_cast = softmax(axis = var_605, x = qk_cast); + tensor var_672_transpose_x_0 = const()[name = tensor("op_672_transpose_x_0"), val = tensor(false)]; + tensor var_672_transpose_y_0 = const()[name = tensor("op_672_transpose_y_0"), val = tensor(false)]; + tensor transpose_27 = transpose(perm = var_666, x = var_665_cast); + tensor var_672_cast = matmul(transpose_x = var_672_transpose_x_0, transpose_y = var_672_transpose_y_0, x = var_670_cast, y = transpose_27); + tensor var_673 = const()[name = tensor("op_673"), val = tensor([0, 2, 1, 3])]; + tensor concat_5 = const()[name = tensor("concat_5"), val = tensor([1, 1500, 512])]; + tensor transpose_24 = transpose(perm = var_673, x = var_672_cast); + tensor x_71_cast = reshape(shape = concat_5, x = transpose_24); + tensor var_678_to_fp16 = const()[name = tensor("op_678_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36464512)))]; + tensor var_679_to_fp16 = const()[name = tensor("op_679_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36988864)))]; + tensor var_680_cast = linear(bias = var_679_to_fp16, weight = var_678_to_fp16, x = x_71_cast); + tensor x_73_cast = add(x = x_67_cast, y = var_680_cast); + tensor var_686_axes_0 = const()[name = tensor("op_686_axes_0"), val = tensor([-1])]; + tensor blocks_5_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36989952)))]; + tensor blocks_5_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36991040)))]; + tensor var_686_cast = layer_norm(axes = var_686_axes_0, beta = blocks_5_mlp_ln_bias_to_fp16, epsilon = var_611_to_fp16, gamma = blocks_5_mlp_ln_weight_to_fp16, x = x_73_cast); + tensor var_695_to_fp16 = const()[name = tensor("op_695_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36992128)))]; + tensor var_696_to_fp16 = const()[name = tensor("op_696_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39089344)))]; + tensor input_49_cast = linear(bias = var_696_to_fp16, weight = var_695_to_fp16, x = var_686_cast); + tensor x_77_mode_0 = const()[name = tensor("x_77_mode_0"), val = tensor("EXACT")]; + tensor x_77_cast = gelu(mode = x_77_mode_0, x = input_49_cast); + tensor var_701_to_fp16 = const()[name = tensor("op_701_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39093504)))]; + tensor var_702_to_fp16 = const()[name = tensor("op_702_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(41190720)))]; + tensor var_703_cast = linear(bias = var_702_to_fp16, weight = var_701_to_fp16, x = x_77_cast); + tensor x_cast = add(x = x_73_cast, y = var_703_cast); + tensor var_716_axes_0 = const()[name = tensor("op_716_axes_0"), val = tensor([-1])]; + tensor ln_post_weight_to_fp16 = const()[name = tensor("ln_post_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(41191808)))]; + tensor ln_post_bias_to_fp16 = const()[name = tensor("ln_post_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(41192896)))]; + tensor var_707_to_fp16 = const()[name = tensor("op_707_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_716_cast = layer_norm(axes = var_716_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_707_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast); + tensor var_716_cast_to_fp32_dtype_0 = const()[name = tensor("op_716_cast_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor output = cast(dtype = var_716_cast_to_fp32_dtype_0, x = var_716_cast); + } -> (output); +} \ No newline at end of file diff --git a/ggml-base-encoder.mlmodelc/weights/weight.bin b/ggml-base-encoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..34fb6f861394395ee81e8b21a6d9df8fa5b9a46d --- /dev/null +++ b/ggml-base-encoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:aa38c9d90f01effad1990d1f17e67cd8382c6486b8e5a4dbacf44439cd838a38 +size 41193984 diff --git a/ggml-base.en-encoder.mlmodelc.zip b/ggml-base.en-encoder.mlmodelc.zip index 2ef628f0c91930593908d9ac23c18660b01c275b..9e46d5de2fa95740c188c84cd0fd80f705bd8d1a 100644 --- a/ggml-base.en-encoder.mlmodelc.zip +++ b/ggml-base.en-encoder.mlmodelc.zip @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:8cf860309e2449e2bdc8be834cf838ab2565747ecc8c0ef914ef5975115e192b +oid sha256:0634e109c4ec900dc6ffc3305ac8367822ade13fcc10425ec0db0dbfcebb12aa size 37950917 diff --git a/ggml-base.en-encoder.mlmodelc/analytics/coremldata.bin b/ggml-base.en-encoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..3f5eb083dfd6b2e7283b81a31687c1b9ea8a3dc3 --- /dev/null +++ b/ggml-base.en-encoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2441ae34fc7d12946dba7b63379063e856ffc7c3e11ba5f7533efb1450562ca6 +size 207 diff --git a/ggml-base.en-encoder.mlmodelc/coremldata.bin b/ggml-base.en-encoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..c732ed7b8caeac3c15acda1c2836f3e1f525a5a8 --- /dev/null +++ b/ggml-base.en-encoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:05fe28591b40616fa0c34ad7b853133623f5300923ec812acb11459c411acf3b +size 149 diff --git a/ggml-base.en-encoder.mlmodelc/metadata.json b/ggml-base.en-encoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..d08d52976d7269d52e5f1283ff74c505ab57d50f --- /dev/null +++ b/ggml-base.en-encoder.mlmodelc/metadata.json @@ -0,0 +1,64 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float32", + "formattedType" : "MultiArray (Float32)", + "shortDescription" : "", + "shape" : "[]", + "name" : "output", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 6, + "mlProgramOperationTypeHistogram" : { + "Linear" : 36, + "Matmul" : 12, + "Cast" : 2, + "Conv" : 2, + "Softmax" : 6, + "Add" : 13, + "LayerNorm" : 13, + "Mul" : 12, + "Transpose" : 25, + "Gelu" : 8, + "Reshape" : 24 + }, + "computePrecision" : "Mixed (Float16, Float32, Int32)", + "isUpdatable" : "0", + "availability" : { + "macOS" : "12.0", + "tvOS" : "15.0", + "watchOS" : "8.0", + "iOS" : "15.0", + "macCatalyst" : "15.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float32", + "formattedType" : "MultiArray (Float32 1 × 80 × 3000)", + "shortDescription" : "", + "shape" : "[1, 80, 3000]", + "name" : "logmel_data", + "type" : "MultiArray" + } + ], + "generatedClassName" : "coreml_encoder_base_en", + "method" : "predict" + } +] \ No newline at end of file diff --git a/ggml-base.en-encoder.mlmodelc/model.mil b/ggml-base.en-encoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..1a556286ea11b368906c6d7a5613793051843109 --- /dev/null +++ b/ggml-base.en-encoder.mlmodelc/model.mil @@ -0,0 +1,393 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "4.28.4"}, {"coremlc-version", "1436.100.10"}})] +{ + func main(tensor logmel_data) { + tensor var_20 = const()[name = tensor("op_20"), val = tensor(1)]; + tensor var_28 = const()[name = tensor("op_28"), val = tensor([1])]; + tensor var_30 = const()[name = tensor("op_30"), val = tensor([1])]; + tensor var_32_pad_type_0 = const()[name = tensor("op_32_pad_type_0"), val = tensor("custom")]; + tensor var_32_pad_0 = const()[name = tensor("op_32_pad_0"), val = tensor([1, 1])]; + tensor logmel_data_to_fp16_dtype_0 = const()[name = tensor("logmel_data_to_fp16_dtype_0"), val = tensor("fp16")]; + tensor weight_3_to_fp16 = const()[name = tensor("weight_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor bias_3_to_fp16 = const()[name = tensor("bias_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(245888)))]; + tensor cast_187 = cast(dtype = logmel_data_to_fp16_dtype_0, x = logmel_data); + tensor var_32_cast = conv(bias = bias_3_to_fp16, dilations = var_30, groups = var_20, pad = var_32_pad_0, pad_type = var_32_pad_type_0, strides = var_28, weight = weight_3_to_fp16, x = cast_187); + tensor input_1_mode_0 = const()[name = tensor("input_1_mode_0"), val = tensor("EXACT")]; + tensor input_1_cast = gelu(mode = input_1_mode_0, x = var_32_cast); + tensor var_36 = const()[name = tensor("op_36"), val = tensor(1)]; + tensor var_45 = const()[name = tensor("op_45"), val = tensor([2])]; + tensor var_47 = const()[name = tensor("op_47"), val = tensor([1])]; + tensor var_49_pad_type_0 = const()[name = tensor("op_49_pad_type_0"), val = tensor("custom")]; + tensor var_49_pad_0 = const()[name = tensor("op_49_pad_0"), val = tensor([1, 1])]; + tensor weight_7_to_fp16 = const()[name = tensor("weight_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(246976)))]; + tensor bias_7_to_fp16 = const()[name = tensor("bias_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1819904)))]; + tensor var_49_cast = conv(bias = bias_7_to_fp16, dilations = var_47, groups = var_36, pad = var_49_pad_0, pad_type = var_49_pad_type_0, strides = var_45, weight = weight_7_to_fp16, x = input_1_cast); + tensor x_3_mode_0 = const()[name = tensor("x_3_mode_0"), val = tensor("EXACT")]; + tensor x_3_cast = gelu(mode = x_3_mode_0, x = var_49_cast); + tensor var_54 = const()[name = tensor("op_54"), val = tensor([0, 2, 1])]; + tensor positional_embedding_to_fp16 = const()[name = tensor("positional_embedding_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1820992)))]; + tensor transpose_48 = transpose(perm = var_54, x = x_3_cast); + tensor var_57_cast = add(x = transpose_48, y = positional_embedding_to_fp16); + tensor var_70 = const()[name = tensor("op_70"), val = tensor(-1)]; + tensor var_87_axes_0 = const()[name = tensor("op_87_axes_0"), val = tensor([-1])]; + tensor blocks_0_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_0_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3357056)))]; + tensor blocks_0_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_0_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3358144)))]; + tensor var_76_to_fp16 = const()[name = tensor("op_76_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_87_cast = layer_norm(axes = var_87_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_76_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_57_cast); + tensor var_98_to_fp16 = const()[name = tensor("op_98_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3359232)))]; + tensor var_99_to_fp16 = const()[name = tensor("op_99_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3883584)))]; + tensor q_1_cast = linear(bias = var_99_to_fp16, weight = var_98_to_fp16, x = var_87_cast); + tensor var_102_to_fp16 = const()[name = tensor("op_102_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3884672)))]; + tensor k_1_bias_0_to_fp16 = const()[name = tensor("k_1_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4409024)))]; + tensor k_1_cast = linear(bias = k_1_bias_0_to_fp16, weight = var_102_to_fp16, x = var_87_cast); + tensor var_106_to_fp16 = const()[name = tensor("op_106_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4410112)))]; + tensor var_107_to_fp16 = const()[name = tensor("op_107_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4934464)))]; + tensor v_1_cast = linear(bias = var_107_to_fp16, weight = var_106_to_fp16, x = var_87_cast); + tensor var_115 = const()[name = tensor("op_115"), val = tensor([1, 1500, 8, -1])]; + tensor var_116_cast = reshape(shape = var_115, x = q_1_cast); + tensor const_42_to_fp16 = const()[name = tensor("const_42_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_3_cast = mul(x = var_116_cast, y = const_42_to_fp16); + tensor var_122 = const()[name = tensor("op_122"), val = tensor([1, 1500, 8, -1])]; + tensor var_123_cast = reshape(shape = var_122, x = k_1_cast); + tensor const_43_to_fp16 = const()[name = tensor("const_43_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_3_cast = mul(x = var_123_cast, y = const_43_to_fp16); + tensor var_129 = const()[name = tensor("op_129"), val = tensor([1, 1500, 8, -1])]; + tensor var_130_cast = reshape(shape = var_129, x = v_1_cast); + tensor var_131 = const()[name = tensor("op_131"), val = tensor([0, 2, 1, 3])]; + tensor qk_1_transpose_x_0 = const()[name = tensor("qk_1_transpose_x_0"), val = tensor(false)]; + tensor qk_1_transpose_y_0 = const()[name = tensor("qk_1_transpose_y_0"), val = tensor(false)]; + tensor transpose_12_perm_0 = const()[name = tensor("transpose_12_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_13_perm_0 = const()[name = tensor("transpose_13_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_45 = transpose(perm = transpose_13_perm_0, x = k_3_cast); + tensor transpose_46 = transpose(perm = transpose_12_perm_0, x = q_3_cast); + tensor qk_1_cast = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_46, y = transpose_45); + tensor var_135_cast = softmax(axis = var_70, x = qk_1_cast); + tensor var_137_transpose_x_0 = const()[name = tensor("op_137_transpose_x_0"), val = tensor(false)]; + tensor var_137_transpose_y_0 = const()[name = tensor("op_137_transpose_y_0"), val = tensor(false)]; + tensor transpose_47 = transpose(perm = var_131, x = var_130_cast); + tensor var_137_cast = matmul(transpose_x = var_137_transpose_x_0, transpose_y = var_137_transpose_y_0, x = var_135_cast, y = transpose_47); + tensor var_138 = const()[name = tensor("op_138"), val = tensor([0, 2, 1, 3])]; + tensor concat_0 = const()[name = tensor("concat_0"), val = tensor([1, 1500, 512])]; + tensor transpose_44 = transpose(perm = var_138, x = var_137_cast); + tensor x_11_cast = reshape(shape = concat_0, x = transpose_44); + tensor var_143_to_fp16 = const()[name = tensor("op_143_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4935552)))]; + tensor var_144_to_fp16 = const()[name = tensor("op_144_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5459904)))]; + tensor var_145_cast = linear(bias = var_144_to_fp16, weight = var_143_to_fp16, x = x_11_cast); + tensor x_13_cast = add(x = var_57_cast, y = var_145_cast); + tensor var_151_axes_0 = const()[name = tensor("op_151_axes_0"), val = tensor([-1])]; + tensor blocks_0_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5460992)))]; + tensor blocks_0_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5462080)))]; + tensor var_151_cast = layer_norm(axes = var_151_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_76_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast); + tensor var_160_to_fp16 = const()[name = tensor("op_160_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5463168)))]; + tensor var_161_to_fp16 = const()[name = tensor("op_161_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7560384)))]; + tensor input_9_cast = linear(bias = var_161_to_fp16, weight = var_160_to_fp16, x = var_151_cast); + tensor x_17_mode_0 = const()[name = tensor("x_17_mode_0"), val = tensor("EXACT")]; + tensor x_17_cast = gelu(mode = x_17_mode_0, x = input_9_cast); + tensor var_166_to_fp16 = const()[name = tensor("op_166_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7564544)))]; + tensor var_167_to_fp16 = const()[name = tensor("op_167_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9661760)))]; + tensor var_168_cast = linear(bias = var_167_to_fp16, weight = var_166_to_fp16, x = x_17_cast); + tensor x_19_cast = add(x = x_13_cast, y = var_168_cast); + tensor var_177 = const()[name = tensor("op_177"), val = tensor(-1)]; + tensor var_194_axes_0 = const()[name = tensor("op_194_axes_0"), val = tensor([-1])]; + tensor blocks_1_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_1_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9662848)))]; + tensor blocks_1_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_1_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9663936)))]; + tensor var_183_to_fp16 = const()[name = tensor("op_183_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_194_cast = layer_norm(axes = var_194_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_183_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast); + tensor var_205_to_fp16 = const()[name = tensor("op_205_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9665024)))]; + tensor var_206_to_fp16 = const()[name = tensor("op_206_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10189376)))]; + tensor q_5_cast = linear(bias = var_206_to_fp16, weight = var_205_to_fp16, x = var_194_cast); + tensor var_209_to_fp16 = const()[name = tensor("op_209_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10190464)))]; + tensor k_5_bias_0_to_fp16 = const()[name = tensor("k_5_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10714816)))]; + tensor k_5_cast = linear(bias = k_5_bias_0_to_fp16, weight = var_209_to_fp16, x = var_194_cast); + tensor var_213_to_fp16 = const()[name = tensor("op_213_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10715904)))]; + tensor var_214_to_fp16 = const()[name = tensor("op_214_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11240256)))]; + tensor v_5_cast = linear(bias = var_214_to_fp16, weight = var_213_to_fp16, x = var_194_cast); + tensor var_222 = const()[name = tensor("op_222"), val = tensor([1, 1500, 8, -1])]; + tensor var_223_cast = reshape(shape = var_222, x = q_5_cast); + tensor const_44_to_fp16 = const()[name = tensor("const_44_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_7_cast = mul(x = var_223_cast, y = const_44_to_fp16); + tensor var_229 = const()[name = tensor("op_229"), val = tensor([1, 1500, 8, -1])]; + tensor var_230_cast = reshape(shape = var_229, x = k_5_cast); + tensor const_45_to_fp16 = const()[name = tensor("const_45_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_7_cast = mul(x = var_230_cast, y = const_45_to_fp16); + tensor var_236 = const()[name = tensor("op_236"), val = tensor([1, 1500, 8, -1])]; + tensor var_237_cast = reshape(shape = var_236, x = v_5_cast); + tensor var_238 = const()[name = tensor("op_238"), val = tensor([0, 2, 1, 3])]; + tensor qk_3_transpose_x_0 = const()[name = tensor("qk_3_transpose_x_0"), val = tensor(false)]; + tensor qk_3_transpose_y_0 = const()[name = tensor("qk_3_transpose_y_0"), val = tensor(false)]; + tensor transpose_14_perm_0 = const()[name = tensor("transpose_14_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_15_perm_0 = const()[name = tensor("transpose_15_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_41 = transpose(perm = transpose_15_perm_0, x = k_7_cast); + tensor transpose_42 = transpose(perm = transpose_14_perm_0, x = q_7_cast); + tensor qk_3_cast = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_42, y = transpose_41); + tensor var_242_cast = softmax(axis = var_177, x = qk_3_cast); + tensor var_244_transpose_x_0 = const()[name = tensor("op_244_transpose_x_0"), val = tensor(false)]; + tensor var_244_transpose_y_0 = const()[name = tensor("op_244_transpose_y_0"), val = tensor(false)]; + tensor transpose_43 = transpose(perm = var_238, x = var_237_cast); + tensor var_244_cast = matmul(transpose_x = var_244_transpose_x_0, transpose_y = var_244_transpose_y_0, x = var_242_cast, y = transpose_43); + tensor var_245 = const()[name = tensor("op_245"), val = tensor([0, 2, 1, 3])]; + tensor concat_1 = const()[name = tensor("concat_1"), val = tensor([1, 1500, 512])]; + tensor transpose_40 = transpose(perm = var_245, x = var_244_cast); + tensor x_23_cast = reshape(shape = concat_1, x = transpose_40); + tensor var_250_to_fp16 = const()[name = tensor("op_250_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11241344)))]; + tensor var_251_to_fp16 = const()[name = tensor("op_251_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11765696)))]; + tensor var_252_cast = linear(bias = var_251_to_fp16, weight = var_250_to_fp16, x = x_23_cast); + tensor x_25_cast = add(x = x_19_cast, y = var_252_cast); + tensor var_258_axes_0 = const()[name = tensor("op_258_axes_0"), val = tensor([-1])]; + tensor blocks_1_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11766784)))]; + tensor blocks_1_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11767872)))]; + tensor var_258_cast = layer_norm(axes = var_258_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_183_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast); + tensor var_267_to_fp16 = const()[name = tensor("op_267_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11768960)))]; + tensor var_268_to_fp16 = const()[name = tensor("op_268_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13866176)))]; + tensor input_17_cast = linear(bias = var_268_to_fp16, weight = var_267_to_fp16, x = var_258_cast); + tensor x_29_mode_0 = const()[name = tensor("x_29_mode_0"), val = tensor("EXACT")]; + tensor x_29_cast = gelu(mode = x_29_mode_0, x = input_17_cast); + tensor var_273_to_fp16 = const()[name = tensor("op_273_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13870336)))]; + tensor var_274_to_fp16 = const()[name = tensor("op_274_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15967552)))]; + tensor var_275_cast = linear(bias = var_274_to_fp16, weight = var_273_to_fp16, x = x_29_cast); + tensor x_31_cast = add(x = x_25_cast, y = var_275_cast); + tensor var_284 = const()[name = tensor("op_284"), val = tensor(-1)]; + tensor var_301_axes_0 = const()[name = tensor("op_301_axes_0"), val = tensor([-1])]; + tensor blocks_2_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_2_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15968640)))]; + tensor blocks_2_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_2_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15969728)))]; + tensor var_290_to_fp16 = const()[name = tensor("op_290_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_301_cast = layer_norm(axes = var_301_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_290_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast); + tensor var_312_to_fp16 = const()[name = tensor("op_312_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15970816)))]; + tensor var_313_to_fp16 = const()[name = tensor("op_313_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16495168)))]; + tensor q_9_cast = linear(bias = var_313_to_fp16, weight = var_312_to_fp16, x = var_301_cast); + tensor var_316_to_fp16 = const()[name = tensor("op_316_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16496256)))]; + tensor k_9_bias_0_to_fp16 = const()[name = tensor("k_9_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17020608)))]; + tensor k_9_cast = linear(bias = k_9_bias_0_to_fp16, weight = var_316_to_fp16, x = var_301_cast); + tensor var_320_to_fp16 = const()[name = tensor("op_320_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17021696)))]; + tensor var_321_to_fp16 = const()[name = tensor("op_321_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17546048)))]; + tensor v_9_cast = linear(bias = var_321_to_fp16, weight = var_320_to_fp16, x = var_301_cast); + tensor var_329 = const()[name = tensor("op_329"), val = tensor([1, 1500, 8, -1])]; + tensor var_330_cast = reshape(shape = var_329, x = q_9_cast); + tensor const_46_to_fp16 = const()[name = tensor("const_46_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_11_cast = mul(x = var_330_cast, y = const_46_to_fp16); + tensor var_336 = const()[name = tensor("op_336"), val = tensor([1, 1500, 8, -1])]; + tensor var_337_cast = reshape(shape = var_336, x = k_9_cast); + tensor const_47_to_fp16 = const()[name = tensor("const_47_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_11_cast = mul(x = var_337_cast, y = const_47_to_fp16); + tensor var_343 = const()[name = tensor("op_343"), val = tensor([1, 1500, 8, -1])]; + tensor var_344_cast = reshape(shape = var_343, x = v_9_cast); + tensor var_345 = const()[name = tensor("op_345"), val = tensor([0, 2, 1, 3])]; + tensor qk_5_transpose_x_0 = const()[name = tensor("qk_5_transpose_x_0"), val = tensor(false)]; + tensor qk_5_transpose_y_0 = const()[name = tensor("qk_5_transpose_y_0"), val = tensor(false)]; + tensor transpose_16_perm_0 = const()[name = tensor("transpose_16_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_17_perm_0 = const()[name = tensor("transpose_17_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_37 = transpose(perm = transpose_17_perm_0, x = k_11_cast); + tensor transpose_38 = transpose(perm = transpose_16_perm_0, x = q_11_cast); + tensor qk_5_cast = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_38, y = transpose_37); + tensor var_349_cast = softmax(axis = var_284, x = qk_5_cast); + tensor var_351_transpose_x_0 = const()[name = tensor("op_351_transpose_x_0"), val = tensor(false)]; + tensor var_351_transpose_y_0 = const()[name = tensor("op_351_transpose_y_0"), val = tensor(false)]; + tensor transpose_39 = transpose(perm = var_345, x = var_344_cast); + tensor var_351_cast = matmul(transpose_x = var_351_transpose_x_0, transpose_y = var_351_transpose_y_0, x = var_349_cast, y = transpose_39); + tensor var_352 = const()[name = tensor("op_352"), val = tensor([0, 2, 1, 3])]; + tensor concat_2 = const()[name = tensor("concat_2"), val = tensor([1, 1500, 512])]; + tensor transpose_36 = transpose(perm = var_352, x = var_351_cast); + tensor x_35_cast = reshape(shape = concat_2, x = transpose_36); + tensor var_357_to_fp16 = const()[name = tensor("op_357_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17547136)))]; + tensor var_358_to_fp16 = const()[name = tensor("op_358_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18071488)))]; + tensor var_359_cast = linear(bias = var_358_to_fp16, weight = var_357_to_fp16, x = x_35_cast); + tensor x_37_cast = add(x = x_31_cast, y = var_359_cast); + tensor var_365_axes_0 = const()[name = tensor("op_365_axes_0"), val = tensor([-1])]; + tensor blocks_2_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18072576)))]; + tensor blocks_2_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18073664)))]; + tensor var_365_cast = layer_norm(axes = var_365_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_290_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast); + tensor var_374_to_fp16 = const()[name = tensor("op_374_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18074752)))]; + tensor var_375_to_fp16 = const()[name = tensor("op_375_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20171968)))]; + tensor input_25_cast = linear(bias = var_375_to_fp16, weight = var_374_to_fp16, x = var_365_cast); + tensor x_41_mode_0 = const()[name = tensor("x_41_mode_0"), val = tensor("EXACT")]; + tensor x_41_cast = gelu(mode = x_41_mode_0, x = input_25_cast); + tensor var_380_to_fp16 = const()[name = tensor("op_380_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20176128)))]; + tensor var_381_to_fp16 = const()[name = tensor("op_381_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22273344)))]; + tensor var_382_cast = linear(bias = var_381_to_fp16, weight = var_380_to_fp16, x = x_41_cast); + tensor x_43_cast = add(x = x_37_cast, y = var_382_cast); + tensor var_391 = const()[name = tensor("op_391"), val = tensor(-1)]; + tensor var_408_axes_0 = const()[name = tensor("op_408_axes_0"), val = tensor([-1])]; + tensor blocks_3_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_3_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22274432)))]; + tensor blocks_3_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_3_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22275520)))]; + tensor var_397_to_fp16 = const()[name = tensor("op_397_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_408_cast = layer_norm(axes = var_408_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_397_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast); + tensor var_419_to_fp16 = const()[name = tensor("op_419_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22276608)))]; + tensor var_420_to_fp16 = const()[name = tensor("op_420_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22800960)))]; + tensor q_13_cast = linear(bias = var_420_to_fp16, weight = var_419_to_fp16, x = var_408_cast); + tensor var_423_to_fp16 = const()[name = tensor("op_423_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22802048)))]; + tensor k_13_bias_0_to_fp16 = const()[name = tensor("k_13_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23326400)))]; + tensor k_13_cast = linear(bias = k_13_bias_0_to_fp16, weight = var_423_to_fp16, x = var_408_cast); + tensor var_427_to_fp16 = const()[name = tensor("op_427_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23327488)))]; + tensor var_428_to_fp16 = const()[name = tensor("op_428_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23851840)))]; + tensor v_13_cast = linear(bias = var_428_to_fp16, weight = var_427_to_fp16, x = var_408_cast); + tensor var_436 = const()[name = tensor("op_436"), val = tensor([1, 1500, 8, -1])]; + tensor var_437_cast = reshape(shape = var_436, x = q_13_cast); + tensor const_48_to_fp16 = const()[name = tensor("const_48_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_15_cast = mul(x = var_437_cast, y = const_48_to_fp16); + tensor var_443 = const()[name = tensor("op_443"), val = tensor([1, 1500, 8, -1])]; + tensor var_444_cast = reshape(shape = var_443, x = k_13_cast); + tensor const_49_to_fp16 = const()[name = tensor("const_49_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_15_cast = mul(x = var_444_cast, y = const_49_to_fp16); + tensor var_450 = const()[name = tensor("op_450"), val = tensor([1, 1500, 8, -1])]; + tensor var_451_cast = reshape(shape = var_450, x = v_13_cast); + tensor var_452 = const()[name = tensor("op_452"), val = tensor([0, 2, 1, 3])]; + tensor qk_7_transpose_x_0 = const()[name = tensor("qk_7_transpose_x_0"), val = tensor(false)]; + tensor qk_7_transpose_y_0 = const()[name = tensor("qk_7_transpose_y_0"), val = tensor(false)]; + tensor transpose_18_perm_0 = const()[name = tensor("transpose_18_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_19_perm_0 = const()[name = tensor("transpose_19_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_33 = transpose(perm = transpose_19_perm_0, x = k_15_cast); + tensor transpose_34 = transpose(perm = transpose_18_perm_0, x = q_15_cast); + tensor qk_7_cast = matmul(transpose_x = qk_7_transpose_x_0, transpose_y = qk_7_transpose_y_0, x = transpose_34, y = transpose_33); + tensor var_456_cast = softmax(axis = var_391, x = qk_7_cast); + tensor var_458_transpose_x_0 = const()[name = tensor("op_458_transpose_x_0"), val = tensor(false)]; + tensor var_458_transpose_y_0 = const()[name = tensor("op_458_transpose_y_0"), val = tensor(false)]; + tensor transpose_35 = transpose(perm = var_452, x = var_451_cast); + tensor var_458_cast = matmul(transpose_x = var_458_transpose_x_0, transpose_y = var_458_transpose_y_0, x = var_456_cast, y = transpose_35); + tensor var_459 = const()[name = tensor("op_459"), val = tensor([0, 2, 1, 3])]; + tensor concat_3 = const()[name = tensor("concat_3"), val = tensor([1, 1500, 512])]; + tensor transpose_32 = transpose(perm = var_459, x = var_458_cast); + tensor x_47_cast = reshape(shape = concat_3, x = transpose_32); + tensor var_464_to_fp16 = const()[name = tensor("op_464_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23852928)))]; + tensor var_465_to_fp16 = const()[name = tensor("op_465_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24377280)))]; + tensor var_466_cast = linear(bias = var_465_to_fp16, weight = var_464_to_fp16, x = x_47_cast); + tensor x_49_cast = add(x = x_43_cast, y = var_466_cast); + tensor var_472_axes_0 = const()[name = tensor("op_472_axes_0"), val = tensor([-1])]; + tensor blocks_3_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24378368)))]; + tensor blocks_3_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24379456)))]; + tensor var_472_cast = layer_norm(axes = var_472_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_397_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast); + tensor var_481_to_fp16 = const()[name = tensor("op_481_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24380544)))]; + tensor var_482_to_fp16 = const()[name = tensor("op_482_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26477760)))]; + tensor input_33_cast = linear(bias = var_482_to_fp16, weight = var_481_to_fp16, x = var_472_cast); + tensor x_53_mode_0 = const()[name = tensor("x_53_mode_0"), val = tensor("EXACT")]; + tensor x_53_cast = gelu(mode = x_53_mode_0, x = input_33_cast); + tensor var_487_to_fp16 = const()[name = tensor("op_487_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26481920)))]; + tensor var_488_to_fp16 = const()[name = tensor("op_488_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28579136)))]; + tensor var_489_cast = linear(bias = var_488_to_fp16, weight = var_487_to_fp16, x = x_53_cast); + tensor x_55_cast = add(x = x_49_cast, y = var_489_cast); + tensor var_498 = const()[name = tensor("op_498"), val = tensor(-1)]; + tensor var_515_axes_0 = const()[name = tensor("op_515_axes_0"), val = tensor([-1])]; + tensor blocks_4_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_4_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28580224)))]; + tensor blocks_4_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_4_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28581312)))]; + tensor var_504_to_fp16 = const()[name = tensor("op_504_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_515_cast = layer_norm(axes = var_515_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_504_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast); + tensor var_526_to_fp16 = const()[name = tensor("op_526_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28582400)))]; + tensor var_527_to_fp16 = const()[name = tensor("op_527_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29106752)))]; + tensor q_17_cast = linear(bias = var_527_to_fp16, weight = var_526_to_fp16, x = var_515_cast); + tensor var_530_to_fp16 = const()[name = tensor("op_530_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29107840)))]; + tensor k_17_bias_0_to_fp16 = const()[name = tensor("k_17_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29632192)))]; + tensor k_17_cast = linear(bias = k_17_bias_0_to_fp16, weight = var_530_to_fp16, x = var_515_cast); + tensor var_534_to_fp16 = const()[name = tensor("op_534_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29633280)))]; + tensor var_535_to_fp16 = const()[name = tensor("op_535_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30157632)))]; + tensor v_17_cast = linear(bias = var_535_to_fp16, weight = var_534_to_fp16, x = var_515_cast); + tensor var_543 = const()[name = tensor("op_543"), val = tensor([1, 1500, 8, -1])]; + tensor var_544_cast = reshape(shape = var_543, x = q_17_cast); + tensor const_50_to_fp16 = const()[name = tensor("const_50_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_19_cast = mul(x = var_544_cast, y = const_50_to_fp16); + tensor var_550 = const()[name = tensor("op_550"), val = tensor([1, 1500, 8, -1])]; + tensor var_551_cast = reshape(shape = var_550, x = k_17_cast); + tensor const_51_to_fp16 = const()[name = tensor("const_51_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_19_cast = mul(x = var_551_cast, y = const_51_to_fp16); + tensor var_557 = const()[name = tensor("op_557"), val = tensor([1, 1500, 8, -1])]; + tensor var_558_cast = reshape(shape = var_557, x = v_17_cast); + tensor var_559 = const()[name = tensor("op_559"), val = tensor([0, 2, 1, 3])]; + tensor qk_9_transpose_x_0 = const()[name = tensor("qk_9_transpose_x_0"), val = tensor(false)]; + tensor qk_9_transpose_y_0 = const()[name = tensor("qk_9_transpose_y_0"), val = tensor(false)]; + tensor transpose_20_perm_0 = const()[name = tensor("transpose_20_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_21_perm_0 = const()[name = tensor("transpose_21_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_29 = transpose(perm = transpose_21_perm_0, x = k_19_cast); + tensor transpose_30 = transpose(perm = transpose_20_perm_0, x = q_19_cast); + tensor qk_9_cast = matmul(transpose_x = qk_9_transpose_x_0, transpose_y = qk_9_transpose_y_0, x = transpose_30, y = transpose_29); + tensor var_563_cast = softmax(axis = var_498, x = qk_9_cast); + tensor var_565_transpose_x_0 = const()[name = tensor("op_565_transpose_x_0"), val = tensor(false)]; + tensor var_565_transpose_y_0 = const()[name = tensor("op_565_transpose_y_0"), val = tensor(false)]; + tensor transpose_31 = transpose(perm = var_559, x = var_558_cast); + tensor var_565_cast = matmul(transpose_x = var_565_transpose_x_0, transpose_y = var_565_transpose_y_0, x = var_563_cast, y = transpose_31); + tensor var_566 = const()[name = tensor("op_566"), val = tensor([0, 2, 1, 3])]; + tensor concat_4 = const()[name = tensor("concat_4"), val = tensor([1, 1500, 512])]; + tensor transpose_28 = transpose(perm = var_566, x = var_565_cast); + tensor x_59_cast = reshape(shape = concat_4, x = transpose_28); + tensor var_571_to_fp16 = const()[name = tensor("op_571_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30158720)))]; + tensor var_572_to_fp16 = const()[name = tensor("op_572_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30683072)))]; + tensor var_573_cast = linear(bias = var_572_to_fp16, weight = var_571_to_fp16, x = x_59_cast); + tensor x_61_cast = add(x = x_55_cast, y = var_573_cast); + tensor var_579_axes_0 = const()[name = tensor("op_579_axes_0"), val = tensor([-1])]; + tensor blocks_4_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30684160)))]; + tensor blocks_4_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30685248)))]; + tensor var_579_cast = layer_norm(axes = var_579_axes_0, beta = blocks_4_mlp_ln_bias_to_fp16, epsilon = var_504_to_fp16, gamma = blocks_4_mlp_ln_weight_to_fp16, x = x_61_cast); + tensor var_588_to_fp16 = const()[name = tensor("op_588_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30686336)))]; + tensor var_589_to_fp16 = const()[name = tensor("op_589_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(32783552)))]; + tensor input_41_cast = linear(bias = var_589_to_fp16, weight = var_588_to_fp16, x = var_579_cast); + tensor x_65_mode_0 = const()[name = tensor("x_65_mode_0"), val = tensor("EXACT")]; + tensor x_65_cast = gelu(mode = x_65_mode_0, x = input_41_cast); + tensor var_594_to_fp16 = const()[name = tensor("op_594_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(32787712)))]; + tensor var_595_to_fp16 = const()[name = tensor("op_595_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34884928)))]; + tensor var_596_cast = linear(bias = var_595_to_fp16, weight = var_594_to_fp16, x = x_65_cast); + tensor x_67_cast = add(x = x_61_cast, y = var_596_cast); + tensor var_605 = const()[name = tensor("op_605"), val = tensor(-1)]; + tensor var_622_axes_0 = const()[name = tensor("op_622_axes_0"), val = tensor([-1])]; + tensor blocks_5_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_5_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34886016)))]; + tensor blocks_5_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_5_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34887104)))]; + tensor var_611_to_fp16 = const()[name = tensor("op_611_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_622_cast = layer_norm(axes = var_622_axes_0, beta = blocks_5_attn_ln_bias_to_fp16, epsilon = var_611_to_fp16, gamma = blocks_5_attn_ln_weight_to_fp16, x = x_67_cast); + tensor var_633_to_fp16 = const()[name = tensor("op_633_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34888192)))]; + tensor var_634_to_fp16 = const()[name = tensor("op_634_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35412544)))]; + tensor q_21_cast = linear(bias = var_634_to_fp16, weight = var_633_to_fp16, x = var_622_cast); + tensor var_637_to_fp16 = const()[name = tensor("op_637_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35413632)))]; + tensor k_21_bias_0_to_fp16 = const()[name = tensor("k_21_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35937984)))]; + tensor k_21_cast = linear(bias = k_21_bias_0_to_fp16, weight = var_637_to_fp16, x = var_622_cast); + tensor var_641_to_fp16 = const()[name = tensor("op_641_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35939072)))]; + tensor var_642_to_fp16 = const()[name = tensor("op_642_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36463424)))]; + tensor v_21_cast = linear(bias = var_642_to_fp16, weight = var_641_to_fp16, x = var_622_cast); + tensor var_650 = const()[name = tensor("op_650"), val = tensor([1, 1500, 8, -1])]; + tensor var_651_cast = reshape(shape = var_650, x = q_21_cast); + tensor const_52_to_fp16 = const()[name = tensor("const_52_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_cast = mul(x = var_651_cast, y = const_52_to_fp16); + tensor var_657 = const()[name = tensor("op_657"), val = tensor([1, 1500, 8, -1])]; + tensor var_658_cast = reshape(shape = var_657, x = k_21_cast); + tensor const_53_to_fp16 = const()[name = tensor("const_53_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_cast = mul(x = var_658_cast, y = const_53_to_fp16); + tensor var_664 = const()[name = tensor("op_664"), val = tensor([1, 1500, 8, -1])]; + tensor var_665_cast = reshape(shape = var_664, x = v_21_cast); + tensor var_666 = const()[name = tensor("op_666"), val = tensor([0, 2, 1, 3])]; + tensor qk_transpose_x_0 = const()[name = tensor("qk_transpose_x_0"), val = tensor(false)]; + tensor qk_transpose_y_0 = const()[name = tensor("qk_transpose_y_0"), val = tensor(false)]; + tensor transpose_22_perm_0 = const()[name = tensor("transpose_22_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_23_perm_0 = const()[name = tensor("transpose_23_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_25 = transpose(perm = transpose_23_perm_0, x = k_cast); + tensor transpose_26 = transpose(perm = transpose_22_perm_0, x = q_cast); + tensor qk_cast = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_26, y = transpose_25); + tensor var_670_cast = softmax(axis = var_605, x = qk_cast); + tensor var_672_transpose_x_0 = const()[name = tensor("op_672_transpose_x_0"), val = tensor(false)]; + tensor var_672_transpose_y_0 = const()[name = tensor("op_672_transpose_y_0"), val = tensor(false)]; + tensor transpose_27 = transpose(perm = var_666, x = var_665_cast); + tensor var_672_cast = matmul(transpose_x = var_672_transpose_x_0, transpose_y = var_672_transpose_y_0, x = var_670_cast, y = transpose_27); + tensor var_673 = const()[name = tensor("op_673"), val = tensor([0, 2, 1, 3])]; + tensor concat_5 = const()[name = tensor("concat_5"), val = tensor([1, 1500, 512])]; + tensor transpose_24 = transpose(perm = var_673, x = var_672_cast); + tensor x_71_cast = reshape(shape = concat_5, x = transpose_24); + tensor var_678_to_fp16 = const()[name = tensor("op_678_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36464512)))]; + tensor var_679_to_fp16 = const()[name = tensor("op_679_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36988864)))]; + tensor var_680_cast = linear(bias = var_679_to_fp16, weight = var_678_to_fp16, x = x_71_cast); + tensor x_73_cast = add(x = x_67_cast, y = var_680_cast); + tensor var_686_axes_0 = const()[name = tensor("op_686_axes_0"), val = tensor([-1])]; + tensor blocks_5_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36989952)))]; + tensor blocks_5_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36991040)))]; + tensor var_686_cast = layer_norm(axes = var_686_axes_0, beta = blocks_5_mlp_ln_bias_to_fp16, epsilon = var_611_to_fp16, gamma = blocks_5_mlp_ln_weight_to_fp16, x = x_73_cast); + tensor var_695_to_fp16 = const()[name = tensor("op_695_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36992128)))]; + tensor var_696_to_fp16 = const()[name = tensor("op_696_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39089344)))]; + tensor input_49_cast = linear(bias = var_696_to_fp16, weight = var_695_to_fp16, x = var_686_cast); + tensor x_77_mode_0 = const()[name = tensor("x_77_mode_0"), val = tensor("EXACT")]; + tensor x_77_cast = gelu(mode = x_77_mode_0, x = input_49_cast); + tensor var_701_to_fp16 = const()[name = tensor("op_701_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39093504)))]; + tensor var_702_to_fp16 = const()[name = tensor("op_702_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(41190720)))]; + tensor var_703_cast = linear(bias = var_702_to_fp16, weight = var_701_to_fp16, x = x_77_cast); + tensor x_cast = add(x = x_73_cast, y = var_703_cast); + tensor var_716_axes_0 = const()[name = tensor("op_716_axes_0"), val = tensor([-1])]; + tensor ln_post_weight_to_fp16 = const()[name = tensor("ln_post_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(41191808)))]; + tensor ln_post_bias_to_fp16 = const()[name = tensor("ln_post_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(41192896)))]; + tensor var_707_to_fp16 = const()[name = tensor("op_707_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_716_cast = layer_norm(axes = var_716_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_707_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast); + tensor var_716_cast_to_fp32_dtype_0 = const()[name = tensor("op_716_cast_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor output = cast(dtype = var_716_cast_to_fp32_dtype_0, x = var_716_cast); + } -> (output); +} \ No newline at end of file diff --git a/ggml-base.en-encoder.mlmodelc/weights/weight.bin b/ggml-base.en-encoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..5a4e6dc2f9e2a8593c315ff37e6c4c2c548dcdcc --- /dev/null +++ b/ggml-base.en-encoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1578c2b8925c0ff867a8bb96dd56c0eb75df3b909ca9507dc2aa1e437c296dc9 +size 41193984 diff --git a/ggml-large-encoder.mlmodelc.zip b/ggml-large-encoder.mlmodelc.zip deleted file mode 100644 index 90a40a16eb970b25bded31221b842037369014cc..0000000000000000000000000000000000000000 --- a/ggml-large-encoder.mlmodelc.zip +++ /dev/null @@ -1,3 +0,0 @@ -version https://git-lfs.github.com/spec/v1 -oid sha256:c65d28737f51d09dff3b9da9d26c2ab6f5d82857c9b049ce383d64c2502a5541 -size 1174643458 diff --git a/ggml-large-v1-encoder.mlmodelc.zip b/ggml-large-v1-encoder.mlmodelc.zip index fdd1cc51de663e3c24bb0b1e8fca78c69d98ce4c..30971c393b0fd7cdf0b767ee09e61fa761fe717b 100644 --- a/ggml-large-v1-encoder.mlmodelc.zip +++ b/ggml-large-v1-encoder.mlmodelc.zip @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:d10d24a4272572fee07843ffc2e4e4ceebd1d0b0831a142dd2fdb3d56d5bf56e -size 1177529527 +oid sha256:205f888f1f31bd06ff6650bec836e332e6f84f28dfbc2c00378559162b3584f4 +size 1177530049 diff --git a/ggml-large-v1-encoder.mlmodelc/analytics/coremldata.bin b/ggml-large-v1-encoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..75f9ab6ce4798c0794f6210393494b522c7901e4 --- /dev/null +++ b/ggml-large-v1-encoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a9b73481ce4a6af49154956c7f9c72dc02d96448fa69651f5c5a2124edfc8e31 +size 243 diff --git a/ggml-large-v1-encoder.mlmodelc/coremldata.bin b/ggml-large-v1-encoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..c732ed7b8caeac3c15acda1c2836f3e1f525a5a8 --- /dev/null +++ b/ggml-large-v1-encoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:05fe28591b40616fa0c34ad7b853133623f5300923ec812acb11459c411acf3b +size 149 diff --git a/ggml-large-v1-encoder.mlmodelc/metadata.json b/ggml-large-v1-encoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..7f8fd9e3c55f3fc13dea00474097f5d012f483b4 --- /dev/null +++ b/ggml-large-v1-encoder.mlmodelc/metadata.json @@ -0,0 +1,65 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float32", + "formattedType" : "MultiArray (Float32)", + "shortDescription" : "", + "shape" : "[]", + "name" : "output", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 6, + "mlProgramOperationTypeHistogram" : { + "Linear" : 192, + "Matmul" : 64, + "Cast" : 2, + "Conv" : 2, + "Softmax" : 32, + "Add" : 65, + "LayerNorm" : 65, + "Mul" : 64, + "Transpose" : 129, + "Gelu" : 34, + "Reshape" : 128 + }, + "computePrecision" : "Mixed (Float16, Float32, Int32)", + "isUpdatable" : "0", + "availability" : { + "macOS" : "12.0", + "tvOS" : "15.0", + "visionOS" : "1.0", + "watchOS" : "8.0", + "iOS" : "15.0", + "macCatalyst" : "15.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float32", + "formattedType" : "MultiArray (Float32 1 × 80 × 3000)", + "shortDescription" : "", + "shape" : "[1, 80, 3000]", + "name" : "logmel_data", + "type" : "MultiArray" + } + ], + "generatedClassName" : "coreml_encoder_large_v1", + "method" : "predict" + } +] \ No newline at end of file diff --git a/ggml-large-v1-encoder.mlmodelc/model.mil b/ggml-large-v1-encoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..488706ca043b4413a7addd38f7dc67c20e77e409 --- /dev/null +++ b/ggml-large-v1-encoder.mlmodelc/model.mil @@ -0,0 +1,1927 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "5.33.5"}, {"coremlc-version", "1877.40.3"}})] +{ + func main(tensor logmel_data) { + tensor var_72 = const()[name = tensor("op_72"), val = tensor(1)]; + tensor var_80 = const()[name = tensor("op_80"), val = tensor([1])]; + tensor var_82 = const()[name = tensor("op_82"), val = tensor([1])]; + tensor var_84_pad_type_0 = const()[name = tensor("op_84_pad_type_0"), val = tensor("custom")]; + tensor var_84_pad_0 = const()[name = tensor("op_84_pad_0"), val = tensor([1, 1])]; + tensor logmel_data_to_fp16_dtype_0 = const()[name = tensor("logmel_data_to_fp16_dtype_0"), val = tensor("fp16")]; + tensor weight_3_to_fp16 = const()[name = tensor("weight_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor bias_3_to_fp16 = const()[name = tensor("bias_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614528)))]; + tensor cast_967 = cast(dtype = logmel_data_to_fp16_dtype_0, x = logmel_data); + tensor var_84_cast = conv(bias = bias_3_to_fp16, dilations = var_82, groups = var_72, pad = var_84_pad_0, pad_type = var_84_pad_type_0, strides = var_80, weight = weight_3_to_fp16, x = cast_967); + tensor input_1_mode_0 = const()[name = tensor("input_1_mode_0"), val = tensor("EXACT")]; + tensor input_1_cast = gelu(mode = input_1_mode_0, x = var_84_cast); + tensor var_88 = const()[name = tensor("op_88"), val = tensor(1)]; + tensor var_97 = const()[name = tensor("op_97"), val = tensor([2])]; + tensor var_99 = const()[name = tensor("op_99"), val = tensor([1])]; + tensor var_101_pad_type_0 = const()[name = tensor("op_101_pad_type_0"), val = tensor("custom")]; + tensor var_101_pad_0 = const()[name = tensor("op_101_pad_0"), val = tensor([1, 1])]; + tensor weight_7_to_fp16 = const()[name = tensor("weight_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617152)))]; + tensor bias_7_to_fp16 = const()[name = tensor("bias_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10447616)))]; + tensor var_101_cast = conv(bias = bias_7_to_fp16, dilations = var_99, groups = var_88, pad = var_101_pad_0, pad_type = var_101_pad_type_0, strides = var_97, weight = weight_7_to_fp16, x = input_1_cast); + tensor x_3_mode_0 = const()[name = tensor("x_3_mode_0"), val = tensor("EXACT")]; + tensor x_3_cast = gelu(mode = x_3_mode_0, x = var_101_cast); + tensor var_106 = const()[name = tensor("op_106"), val = tensor([0, 2, 1])]; + tensor positional_embedding_to_fp16 = const()[name = tensor("positional_embedding_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10450240)))]; + tensor transpose_256 = transpose(perm = var_106, x = x_3_cast); + tensor var_109_cast = add(x = transpose_256, y = positional_embedding_to_fp16); + tensor var_122 = const()[name = tensor("op_122"), val = tensor(-1)]; + tensor var_139_axes_0 = const()[name = tensor("op_139_axes_0"), val = tensor([-1])]; + tensor blocks_0_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_0_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14290304)))]; + tensor blocks_0_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_0_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14292928)))]; + tensor var_128_to_fp16 = const()[name = tensor("op_128_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_139_cast = layer_norm(axes = var_139_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_128_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_109_cast); + tensor var_150_to_fp16 = const()[name = tensor("op_150_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14295552)))]; + tensor var_151_to_fp16 = const()[name = tensor("op_151_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17572416)))]; + tensor q_1_cast = linear(bias = var_151_to_fp16, weight = var_150_to_fp16, x = var_139_cast); + tensor var_154_to_fp16 = const()[name = tensor("op_154_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17575040)))]; + tensor k_1_bias_0_to_fp16 = const()[name = tensor("k_1_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20851904)))]; + tensor k_1_cast = linear(bias = k_1_bias_0_to_fp16, weight = var_154_to_fp16, x = var_139_cast); + tensor var_158_to_fp16 = const()[name = tensor("op_158_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20854528)))]; + tensor var_159_to_fp16 = const()[name = tensor("op_159_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24131392)))]; + tensor v_1_cast = linear(bias = var_159_to_fp16, weight = var_158_to_fp16, x = var_139_cast); + tensor var_167 = const()[name = tensor("op_167"), val = tensor([1, 1500, 20, -1])]; + tensor var_168_cast = reshape(shape = var_167, x = q_1_cast); + tensor const_224_to_fp16 = const()[name = tensor("const_224_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_3_cast = mul(x = var_168_cast, y = const_224_to_fp16); + tensor var_174 = const()[name = tensor("op_174"), val = tensor([1, 1500, 20, -1])]; + tensor var_175_cast = reshape(shape = var_174, x = k_1_cast); + tensor const_225_to_fp16 = const()[name = tensor("const_225_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_3_cast = mul(x = var_175_cast, y = const_225_to_fp16); + tensor var_181 = const()[name = tensor("op_181"), val = tensor([1, 1500, 20, -1])]; + tensor var_182_cast = reshape(shape = var_181, x = v_1_cast); + tensor var_183 = const()[name = tensor("op_183"), val = tensor([0, 2, 1, 3])]; + tensor qk_1_transpose_x_0 = const()[name = tensor("qk_1_transpose_x_0"), val = tensor(false)]; + tensor qk_1_transpose_y_0 = const()[name = tensor("qk_1_transpose_y_0"), val = tensor(false)]; + tensor transpose_64_perm_0 = const()[name = tensor("transpose_64_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_65_perm_0 = const()[name = tensor("transpose_65_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_253 = transpose(perm = transpose_65_perm_0, x = k_3_cast); + tensor transpose_254 = transpose(perm = transpose_64_perm_0, x = q_3_cast); + tensor qk_1_cast = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_254, y = transpose_253); + tensor var_187_cast = softmax(axis = var_122, x = qk_1_cast); + tensor var_189_transpose_x_0 = const()[name = tensor("op_189_transpose_x_0"), val = tensor(false)]; + tensor var_189_transpose_y_0 = const()[name = tensor("op_189_transpose_y_0"), val = tensor(false)]; + tensor transpose_255 = transpose(perm = var_183, x = var_182_cast); + tensor var_189_cast = matmul(transpose_x = var_189_transpose_x_0, transpose_y = var_189_transpose_y_0, x = var_187_cast, y = transpose_255); + tensor var_190 = const()[name = tensor("op_190"), val = tensor([0, 2, 1, 3])]; + tensor concat_0 = const()[name = tensor("concat_0"), val = tensor([1, 1500, 1280])]; + tensor transpose_252 = transpose(perm = var_190, x = var_189_cast); + tensor x_11_cast = reshape(shape = concat_0, x = transpose_252); + tensor var_195_to_fp16 = const()[name = tensor("op_195_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24134016)))]; + tensor var_196_to_fp16 = const()[name = tensor("op_196_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27410880)))]; + tensor var_197_cast = linear(bias = var_196_to_fp16, weight = var_195_to_fp16, x = x_11_cast); + tensor x_13_cast = add(x = var_109_cast, y = var_197_cast); + tensor var_203_axes_0 = const()[name = tensor("op_203_axes_0"), val = tensor([-1])]; + tensor blocks_0_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27413504)))]; + tensor blocks_0_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27416128)))]; + tensor var_203_cast = layer_norm(axes = var_203_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_128_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast); + tensor var_212_to_fp16 = const()[name = tensor("op_212_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27418752)))]; + tensor var_213_to_fp16 = const()[name = tensor("op_213_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40526016)))]; + tensor input_9_cast = linear(bias = var_213_to_fp16, weight = var_212_to_fp16, x = var_203_cast); + tensor x_17_mode_0 = const()[name = tensor("x_17_mode_0"), val = tensor("EXACT")]; + tensor x_17_cast = gelu(mode = x_17_mode_0, x = input_9_cast); + tensor var_218_to_fp16 = const()[name = tensor("op_218_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40536320)))]; + tensor var_219_to_fp16 = const()[name = tensor("op_219_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53643584)))]; + tensor var_220_cast = linear(bias = var_219_to_fp16, weight = var_218_to_fp16, x = x_17_cast); + tensor x_19_cast = add(x = x_13_cast, y = var_220_cast); + tensor var_229 = const()[name = tensor("op_229"), val = tensor(-1)]; + tensor var_246_axes_0 = const()[name = tensor("op_246_axes_0"), val = tensor([-1])]; + tensor blocks_1_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_1_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53646208)))]; + tensor blocks_1_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_1_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53648832)))]; + tensor var_235_to_fp16 = const()[name = tensor("op_235_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_246_cast = layer_norm(axes = var_246_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_235_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast); + tensor var_257_to_fp16 = const()[name = tensor("op_257_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53651456)))]; + tensor var_258_to_fp16 = const()[name = tensor("op_258_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56928320)))]; + tensor q_5_cast = linear(bias = var_258_to_fp16, weight = var_257_to_fp16, x = var_246_cast); + tensor var_261_to_fp16 = const()[name = tensor("op_261_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56930944)))]; + tensor k_5_bias_0_to_fp16 = const()[name = tensor("k_5_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60207808)))]; + tensor k_5_cast = linear(bias = k_5_bias_0_to_fp16, weight = var_261_to_fp16, x = var_246_cast); + tensor var_265_to_fp16 = const()[name = tensor("op_265_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60210432)))]; + tensor var_266_to_fp16 = const()[name = tensor("op_266_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63487296)))]; + tensor v_5_cast = linear(bias = var_266_to_fp16, weight = var_265_to_fp16, x = var_246_cast); + tensor var_274 = const()[name = tensor("op_274"), val = tensor([1, 1500, 20, -1])]; + tensor var_275_cast = reshape(shape = var_274, x = q_5_cast); + tensor const_226_to_fp16 = const()[name = tensor("const_226_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_7_cast = mul(x = var_275_cast, y = const_226_to_fp16); + tensor var_281 = const()[name = tensor("op_281"), val = tensor([1, 1500, 20, -1])]; + tensor var_282_cast = reshape(shape = var_281, x = k_5_cast); + tensor const_227_to_fp16 = const()[name = tensor("const_227_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_7_cast = mul(x = var_282_cast, y = const_227_to_fp16); + tensor var_288 = const()[name = tensor("op_288"), val = tensor([1, 1500, 20, -1])]; + tensor var_289_cast = reshape(shape = var_288, x = v_5_cast); + tensor var_290 = const()[name = tensor("op_290"), val = tensor([0, 2, 1, 3])]; + tensor qk_3_transpose_x_0 = const()[name = tensor("qk_3_transpose_x_0"), val = tensor(false)]; + tensor qk_3_transpose_y_0 = const()[name = tensor("qk_3_transpose_y_0"), val = tensor(false)]; + tensor transpose_66_perm_0 = const()[name = tensor("transpose_66_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_67_perm_0 = const()[name = tensor("transpose_67_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_249 = transpose(perm = transpose_67_perm_0, x = k_7_cast); + tensor transpose_250 = transpose(perm = transpose_66_perm_0, x = q_7_cast); + tensor qk_3_cast = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_250, y = transpose_249); + tensor var_294_cast = softmax(axis = var_229, x = qk_3_cast); + tensor var_296_transpose_x_0 = const()[name = tensor("op_296_transpose_x_0"), val = tensor(false)]; + tensor var_296_transpose_y_0 = const()[name = tensor("op_296_transpose_y_0"), val = tensor(false)]; + tensor transpose_251 = transpose(perm = var_290, x = var_289_cast); + tensor var_296_cast = matmul(transpose_x = var_296_transpose_x_0, transpose_y = var_296_transpose_y_0, x = var_294_cast, y = transpose_251); + tensor var_297 = const()[name = tensor("op_297"), val = tensor([0, 2, 1, 3])]; + tensor concat_1 = const()[name = tensor("concat_1"), val = tensor([1, 1500, 1280])]; + tensor transpose_248 = transpose(perm = var_297, x = var_296_cast); + tensor x_23_cast = reshape(shape = concat_1, x = transpose_248); + tensor var_302_to_fp16 = const()[name = tensor("op_302_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63489920)))]; + tensor var_303_to_fp16 = const()[name = tensor("op_303_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66766784)))]; + tensor var_304_cast = linear(bias = var_303_to_fp16, weight = var_302_to_fp16, x = x_23_cast); + tensor x_25_cast = add(x = x_19_cast, y = var_304_cast); + tensor var_310_axes_0 = const()[name = tensor("op_310_axes_0"), val = tensor([-1])]; + tensor blocks_1_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66769408)))]; + tensor blocks_1_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66772032)))]; + tensor var_310_cast = layer_norm(axes = var_310_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_235_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast); + tensor var_319_to_fp16 = const()[name = tensor("op_319_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66774656)))]; + tensor var_320_to_fp16 = const()[name = tensor("op_320_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79881920)))]; + tensor input_17_cast = linear(bias = var_320_to_fp16, weight = var_319_to_fp16, x = var_310_cast); + tensor x_29_mode_0 = const()[name = tensor("x_29_mode_0"), val = tensor("EXACT")]; + tensor x_29_cast = gelu(mode = x_29_mode_0, x = input_17_cast); + tensor var_325_to_fp16 = const()[name = tensor("op_325_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79892224)))]; + tensor var_326_to_fp16 = const()[name = tensor("op_326_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92999488)))]; + tensor var_327_cast = linear(bias = var_326_to_fp16, weight = var_325_to_fp16, x = x_29_cast); + tensor x_31_cast = add(x = x_25_cast, y = var_327_cast); + tensor var_336 = const()[name = tensor("op_336"), val = tensor(-1)]; + tensor var_353_axes_0 = const()[name = tensor("op_353_axes_0"), val = tensor([-1])]; + tensor blocks_2_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_2_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93002112)))]; + tensor blocks_2_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_2_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93004736)))]; + tensor var_342_to_fp16 = const()[name = tensor("op_342_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_353_cast = layer_norm(axes = var_353_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_342_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast); + tensor var_364_to_fp16 = const()[name = tensor("op_364_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93007360)))]; + tensor var_365_to_fp16 = const()[name = tensor("op_365_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96284224)))]; + tensor q_9_cast = linear(bias = var_365_to_fp16, weight = var_364_to_fp16, x = var_353_cast); + tensor var_368_to_fp16 = const()[name = tensor("op_368_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96286848)))]; + tensor k_9_bias_0_to_fp16 = const()[name = tensor("k_9_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99563712)))]; + tensor k_9_cast = linear(bias = k_9_bias_0_to_fp16, weight = var_368_to_fp16, x = var_353_cast); + tensor var_372_to_fp16 = const()[name = tensor("op_372_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99566336)))]; + tensor var_373_to_fp16 = const()[name = tensor("op_373_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102843200)))]; + tensor v_9_cast = linear(bias = var_373_to_fp16, weight = var_372_to_fp16, x = var_353_cast); + tensor var_381 = const()[name = tensor("op_381"), val = tensor([1, 1500, 20, -1])]; + tensor var_382_cast = reshape(shape = var_381, x = q_9_cast); + tensor const_228_to_fp16 = const()[name = tensor("const_228_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_11_cast = mul(x = var_382_cast, y = const_228_to_fp16); + tensor var_388 = const()[name = tensor("op_388"), val = tensor([1, 1500, 20, -1])]; + tensor var_389_cast = reshape(shape = var_388, x = k_9_cast); + tensor const_229_to_fp16 = const()[name = tensor("const_229_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_11_cast = mul(x = var_389_cast, y = const_229_to_fp16); + tensor var_395 = const()[name = tensor("op_395"), val = tensor([1, 1500, 20, -1])]; + tensor var_396_cast = reshape(shape = var_395, x = v_9_cast); + tensor var_397 = const()[name = tensor("op_397"), val = tensor([0, 2, 1, 3])]; + tensor qk_5_transpose_x_0 = const()[name = tensor("qk_5_transpose_x_0"), val = tensor(false)]; + tensor qk_5_transpose_y_0 = const()[name = tensor("qk_5_transpose_y_0"), val = tensor(false)]; + tensor transpose_68_perm_0 = const()[name = tensor("transpose_68_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_69_perm_0 = const()[name = tensor("transpose_69_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_245 = transpose(perm = transpose_69_perm_0, x = k_11_cast); + tensor transpose_246 = transpose(perm = transpose_68_perm_0, x = q_11_cast); + tensor qk_5_cast = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_246, y = transpose_245); + tensor var_401_cast = softmax(axis = var_336, x = qk_5_cast); + tensor var_403_transpose_x_0 = const()[name = tensor("op_403_transpose_x_0"), val = tensor(false)]; + tensor var_403_transpose_y_0 = const()[name = tensor("op_403_transpose_y_0"), val = tensor(false)]; + tensor transpose_247 = transpose(perm = var_397, x = var_396_cast); + tensor var_403_cast = matmul(transpose_x = var_403_transpose_x_0, transpose_y = var_403_transpose_y_0, x = var_401_cast, y = transpose_247); + tensor var_404 = const()[name = tensor("op_404"), val = tensor([0, 2, 1, 3])]; + tensor concat_2 = const()[name = tensor("concat_2"), val = tensor([1, 1500, 1280])]; + tensor transpose_244 = transpose(perm = var_404, x = var_403_cast); + tensor x_35_cast = reshape(shape = concat_2, x = transpose_244); + tensor var_409_to_fp16 = const()[name = tensor("op_409_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102845824)))]; + tensor var_410_to_fp16 = const()[name = tensor("op_410_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106122688)))]; + tensor var_411_cast = linear(bias = var_410_to_fp16, weight = var_409_to_fp16, x = x_35_cast); + tensor x_37_cast = add(x = x_31_cast, y = var_411_cast); + tensor var_417_axes_0 = const()[name = tensor("op_417_axes_0"), val = tensor([-1])]; + tensor blocks_2_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106125312)))]; + tensor blocks_2_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106127936)))]; + tensor var_417_cast = layer_norm(axes = var_417_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_342_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast); + tensor var_426_to_fp16 = const()[name = tensor("op_426_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106130560)))]; + tensor var_427_to_fp16 = const()[name = tensor("op_427_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119237824)))]; + tensor input_25_cast = linear(bias = var_427_to_fp16, weight = var_426_to_fp16, x = var_417_cast); + tensor x_41_mode_0 = const()[name = tensor("x_41_mode_0"), val = tensor("EXACT")]; + tensor x_41_cast = gelu(mode = x_41_mode_0, x = input_25_cast); + tensor var_432_to_fp16 = const()[name = tensor("op_432_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119248128)))]; + tensor var_433_to_fp16 = const()[name = tensor("op_433_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132355392)))]; + tensor var_434_cast = linear(bias = var_433_to_fp16, weight = var_432_to_fp16, x = x_41_cast); + tensor x_43_cast = add(x = x_37_cast, y = var_434_cast); + tensor var_443 = const()[name = tensor("op_443"), val = tensor(-1)]; + tensor var_460_axes_0 = const()[name = tensor("op_460_axes_0"), val = tensor([-1])]; + tensor blocks_3_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_3_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132358016)))]; + tensor blocks_3_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_3_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132360640)))]; + tensor var_449_to_fp16 = const()[name = tensor("op_449_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_460_cast = layer_norm(axes = var_460_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_449_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast); + tensor var_471_to_fp16 = const()[name = tensor("op_471_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132363264)))]; + tensor var_472_to_fp16 = const()[name = tensor("op_472_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135640128)))]; + tensor q_13_cast = linear(bias = var_472_to_fp16, weight = var_471_to_fp16, x = var_460_cast); + tensor var_475_to_fp16 = const()[name = tensor("op_475_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135642752)))]; + tensor k_13_bias_0_to_fp16 = const()[name = tensor("k_13_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138919616)))]; + tensor k_13_cast = linear(bias = k_13_bias_0_to_fp16, weight = var_475_to_fp16, x = var_460_cast); + tensor var_479_to_fp16 = const()[name = tensor("op_479_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138922240)))]; + tensor var_480_to_fp16 = const()[name = tensor("op_480_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142199104)))]; + tensor v_13_cast = linear(bias = var_480_to_fp16, weight = var_479_to_fp16, x = var_460_cast); + tensor var_488 = const()[name = tensor("op_488"), val = tensor([1, 1500, 20, -1])]; + tensor var_489_cast = reshape(shape = var_488, x = q_13_cast); + tensor const_230_to_fp16 = const()[name = tensor("const_230_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_15_cast = mul(x = var_489_cast, y = const_230_to_fp16); + tensor var_495 = const()[name = tensor("op_495"), val = tensor([1, 1500, 20, -1])]; + tensor var_496_cast = reshape(shape = var_495, x = k_13_cast); + tensor const_231_to_fp16 = const()[name = tensor("const_231_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_15_cast = mul(x = var_496_cast, y = const_231_to_fp16); + tensor var_502 = const()[name = tensor("op_502"), val = tensor([1, 1500, 20, -1])]; + tensor var_503_cast = reshape(shape = var_502, x = v_13_cast); + tensor var_504 = const()[name = tensor("op_504"), val = tensor([0, 2, 1, 3])]; + tensor qk_7_transpose_x_0 = const()[name = tensor("qk_7_transpose_x_0"), val = tensor(false)]; + tensor qk_7_transpose_y_0 = const()[name = tensor("qk_7_transpose_y_0"), val = tensor(false)]; + tensor transpose_70_perm_0 = const()[name = tensor("transpose_70_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_71_perm_0 = const()[name = tensor("transpose_71_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_241 = transpose(perm = transpose_71_perm_0, x = k_15_cast); + tensor transpose_242 = transpose(perm = transpose_70_perm_0, x = q_15_cast); + tensor qk_7_cast = matmul(transpose_x = qk_7_transpose_x_0, transpose_y = qk_7_transpose_y_0, x = transpose_242, y = transpose_241); + tensor var_508_cast = softmax(axis = var_443, x = qk_7_cast); + tensor var_510_transpose_x_0 = const()[name = tensor("op_510_transpose_x_0"), val = tensor(false)]; + tensor var_510_transpose_y_0 = const()[name = tensor("op_510_transpose_y_0"), val = tensor(false)]; + tensor transpose_243 = transpose(perm = var_504, x = var_503_cast); + tensor var_510_cast = matmul(transpose_x = var_510_transpose_x_0, transpose_y = var_510_transpose_y_0, x = var_508_cast, y = transpose_243); + tensor var_511 = const()[name = tensor("op_511"), val = tensor([0, 2, 1, 3])]; + tensor concat_3 = const()[name = tensor("concat_3"), val = tensor([1, 1500, 1280])]; + tensor transpose_240 = transpose(perm = var_511, x = var_510_cast); + tensor x_47_cast = reshape(shape = concat_3, x = transpose_240); + tensor var_516_to_fp16 = const()[name = tensor("op_516_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142201728)))]; + tensor var_517_to_fp16 = const()[name = tensor("op_517_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145478592)))]; + tensor var_518_cast = linear(bias = var_517_to_fp16, weight = var_516_to_fp16, x = x_47_cast); + tensor x_49_cast = add(x = x_43_cast, y = var_518_cast); + tensor var_524_axes_0 = const()[name = tensor("op_524_axes_0"), val = tensor([-1])]; + tensor blocks_3_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145481216)))]; + tensor blocks_3_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145483840)))]; + tensor var_524_cast = layer_norm(axes = var_524_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_449_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast); + tensor var_533_to_fp16 = const()[name = tensor("op_533_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145486464)))]; + tensor var_534_to_fp16 = const()[name = tensor("op_534_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158593728)))]; + tensor input_33_cast = linear(bias = var_534_to_fp16, weight = var_533_to_fp16, x = var_524_cast); + tensor x_53_mode_0 = const()[name = tensor("x_53_mode_0"), val = tensor("EXACT")]; + tensor x_53_cast = gelu(mode = x_53_mode_0, x = input_33_cast); + tensor var_539_to_fp16 = const()[name = tensor("op_539_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158604032)))]; + tensor var_540_to_fp16 = const()[name = tensor("op_540_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171711296)))]; + tensor var_541_cast = linear(bias = var_540_to_fp16, weight = var_539_to_fp16, x = x_53_cast); + tensor x_55_cast = add(x = x_49_cast, y = var_541_cast); + tensor var_550 = const()[name = tensor("op_550"), val = tensor(-1)]; + tensor var_567_axes_0 = const()[name = tensor("op_567_axes_0"), val = tensor([-1])]; + tensor blocks_4_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_4_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171713920)))]; + tensor blocks_4_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_4_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171716544)))]; + tensor var_556_to_fp16 = const()[name = tensor("op_556_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_567_cast = layer_norm(axes = var_567_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_556_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast); + tensor var_578_to_fp16 = const()[name = tensor("op_578_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171719168)))]; + tensor var_579_to_fp16 = const()[name = tensor("op_579_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174996032)))]; + tensor q_17_cast = linear(bias = var_579_to_fp16, weight = var_578_to_fp16, x = var_567_cast); + tensor var_582_to_fp16 = const()[name = tensor("op_582_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174998656)))]; + tensor k_17_bias_0_to_fp16 = const()[name = tensor("k_17_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178275520)))]; + tensor k_17_cast = linear(bias = k_17_bias_0_to_fp16, weight = var_582_to_fp16, x = var_567_cast); + tensor var_586_to_fp16 = const()[name = tensor("op_586_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178278144)))]; + tensor var_587_to_fp16 = const()[name = tensor("op_587_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(181555008)))]; + tensor v_17_cast = linear(bias = var_587_to_fp16, weight = var_586_to_fp16, x = var_567_cast); + tensor var_595 = const()[name = tensor("op_595"), val = tensor([1, 1500, 20, -1])]; + tensor var_596_cast = reshape(shape = var_595, x = q_17_cast); + tensor const_232_to_fp16 = const()[name = tensor("const_232_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_19_cast = mul(x = var_596_cast, y = const_232_to_fp16); + tensor var_602 = const()[name = tensor("op_602"), val = tensor([1, 1500, 20, -1])]; + tensor var_603_cast = reshape(shape = var_602, x = k_17_cast); + tensor const_233_to_fp16 = const()[name = tensor("const_233_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_19_cast = mul(x = var_603_cast, y = const_233_to_fp16); + tensor var_609 = const()[name = tensor("op_609"), val = tensor([1, 1500, 20, -1])]; + tensor var_610_cast = reshape(shape = var_609, x = v_17_cast); + tensor var_611 = const()[name = tensor("op_611"), val = tensor([0, 2, 1, 3])]; + tensor qk_9_transpose_x_0 = const()[name = tensor("qk_9_transpose_x_0"), val = tensor(false)]; + tensor qk_9_transpose_y_0 = const()[name = tensor("qk_9_transpose_y_0"), val = tensor(false)]; + tensor transpose_72_perm_0 = const()[name = tensor("transpose_72_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_73_perm_0 = const()[name = tensor("transpose_73_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_237 = transpose(perm = transpose_73_perm_0, x = k_19_cast); + tensor transpose_238 = transpose(perm = transpose_72_perm_0, x = q_19_cast); + tensor qk_9_cast = matmul(transpose_x = qk_9_transpose_x_0, transpose_y = qk_9_transpose_y_0, x = transpose_238, y = transpose_237); + tensor var_615_cast = softmax(axis = var_550, x = qk_9_cast); + tensor var_617_transpose_x_0 = const()[name = tensor("op_617_transpose_x_0"), val = tensor(false)]; + tensor var_617_transpose_y_0 = const()[name = tensor("op_617_transpose_y_0"), val = tensor(false)]; + tensor transpose_239 = transpose(perm = var_611, x = var_610_cast); + tensor var_617_cast = matmul(transpose_x = var_617_transpose_x_0, transpose_y = var_617_transpose_y_0, x = var_615_cast, y = transpose_239); + tensor var_618 = const()[name = tensor("op_618"), val = tensor([0, 2, 1, 3])]; + tensor concat_4 = const()[name = tensor("concat_4"), val = tensor([1, 1500, 1280])]; + tensor transpose_236 = transpose(perm = var_618, x = var_617_cast); + tensor x_59_cast = reshape(shape = concat_4, x = transpose_236); + tensor var_623_to_fp16 = const()[name = tensor("op_623_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(181557632)))]; + tensor var_624_to_fp16 = const()[name = tensor("op_624_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184834496)))]; + tensor var_625_cast = linear(bias = var_624_to_fp16, weight = var_623_to_fp16, x = x_59_cast); + tensor x_61_cast = add(x = x_55_cast, y = var_625_cast); + tensor var_631_axes_0 = const()[name = tensor("op_631_axes_0"), val = tensor([-1])]; + tensor blocks_4_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184837120)))]; + tensor blocks_4_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184839744)))]; + tensor var_631_cast = layer_norm(axes = var_631_axes_0, beta = blocks_4_mlp_ln_bias_to_fp16, epsilon = var_556_to_fp16, gamma = blocks_4_mlp_ln_weight_to_fp16, x = x_61_cast); + tensor var_640_to_fp16 = const()[name = tensor("op_640_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184842368)))]; + tensor var_641_to_fp16 = const()[name = tensor("op_641_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(197949632)))]; + tensor input_41_cast = linear(bias = var_641_to_fp16, weight = var_640_to_fp16, x = var_631_cast); + tensor x_65_mode_0 = const()[name = tensor("x_65_mode_0"), val = tensor("EXACT")]; + tensor x_65_cast = gelu(mode = x_65_mode_0, x = input_41_cast); + tensor var_646_to_fp16 = const()[name = tensor("op_646_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(197959936)))]; + tensor var_647_to_fp16 = const()[name = tensor("op_647_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211067200)))]; + tensor var_648_cast = linear(bias = var_647_to_fp16, weight = var_646_to_fp16, x = x_65_cast); + tensor x_67_cast = add(x = x_61_cast, y = var_648_cast); + tensor var_657 = const()[name = tensor("op_657"), val = tensor(-1)]; + tensor var_674_axes_0 = const()[name = tensor("op_674_axes_0"), val = tensor([-1])]; + tensor blocks_5_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_5_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211069824)))]; + tensor blocks_5_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_5_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211072448)))]; + tensor var_663_to_fp16 = const()[name = tensor("op_663_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_674_cast = layer_norm(axes = var_674_axes_0, beta = blocks_5_attn_ln_bias_to_fp16, epsilon = var_663_to_fp16, gamma = blocks_5_attn_ln_weight_to_fp16, x = x_67_cast); + tensor var_685_to_fp16 = const()[name = tensor("op_685_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211075072)))]; + tensor var_686_to_fp16 = const()[name = tensor("op_686_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214351936)))]; + tensor q_21_cast = linear(bias = var_686_to_fp16, weight = var_685_to_fp16, x = var_674_cast); + tensor var_689_to_fp16 = const()[name = tensor("op_689_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214354560)))]; + tensor k_21_bias_0_to_fp16 = const()[name = tensor("k_21_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217631424)))]; + tensor k_21_cast = linear(bias = k_21_bias_0_to_fp16, weight = var_689_to_fp16, x = var_674_cast); + tensor var_693_to_fp16 = const()[name = tensor("op_693_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217634048)))]; + tensor var_694_to_fp16 = const()[name = tensor("op_694_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(220910912)))]; + tensor v_21_cast = linear(bias = var_694_to_fp16, weight = var_693_to_fp16, x = var_674_cast); + tensor var_702 = const()[name = tensor("op_702"), val = tensor([1, 1500, 20, -1])]; + tensor var_703_cast = reshape(shape = var_702, x = q_21_cast); + tensor const_234_to_fp16 = const()[name = tensor("const_234_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_23_cast = mul(x = var_703_cast, y = const_234_to_fp16); + tensor var_709 = const()[name = tensor("op_709"), val = tensor([1, 1500, 20, -1])]; + tensor var_710_cast = reshape(shape = var_709, x = k_21_cast); + tensor const_235_to_fp16 = const()[name = tensor("const_235_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_23_cast = mul(x = var_710_cast, y = const_235_to_fp16); + tensor var_716 = const()[name = tensor("op_716"), val = tensor([1, 1500, 20, -1])]; + tensor var_717_cast = reshape(shape = var_716, x = v_21_cast); + tensor var_718 = const()[name = tensor("op_718"), val = tensor([0, 2, 1, 3])]; + tensor qk_11_transpose_x_0 = const()[name = tensor("qk_11_transpose_x_0"), val = tensor(false)]; + tensor qk_11_transpose_y_0 = const()[name = tensor("qk_11_transpose_y_0"), val = tensor(false)]; + tensor transpose_74_perm_0 = const()[name = tensor("transpose_74_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_75_perm_0 = const()[name = tensor("transpose_75_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_233 = transpose(perm = transpose_75_perm_0, x = k_23_cast); + tensor transpose_234 = transpose(perm = transpose_74_perm_0, x = q_23_cast); + tensor qk_11_cast = matmul(transpose_x = qk_11_transpose_x_0, transpose_y = qk_11_transpose_y_0, x = transpose_234, y = transpose_233); + tensor var_722_cast = softmax(axis = var_657, x = qk_11_cast); + tensor var_724_transpose_x_0 = const()[name = tensor("op_724_transpose_x_0"), val = tensor(false)]; + tensor var_724_transpose_y_0 = const()[name = tensor("op_724_transpose_y_0"), val = tensor(false)]; + tensor transpose_235 = transpose(perm = var_718, x = var_717_cast); + tensor var_724_cast = matmul(transpose_x = var_724_transpose_x_0, transpose_y = var_724_transpose_y_0, x = var_722_cast, y = transpose_235); + tensor var_725 = const()[name = tensor("op_725"), val = tensor([0, 2, 1, 3])]; + tensor concat_5 = const()[name = tensor("concat_5"), val = tensor([1, 1500, 1280])]; + tensor transpose_232 = transpose(perm = var_725, x = var_724_cast); + tensor x_71_cast = reshape(shape = concat_5, x = transpose_232); + tensor var_730_to_fp16 = const()[name = tensor("op_730_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(220913536)))]; + tensor var_731_to_fp16 = const()[name = tensor("op_731_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224190400)))]; + tensor var_732_cast = linear(bias = var_731_to_fp16, weight = var_730_to_fp16, x = x_71_cast); + tensor x_73_cast = add(x = x_67_cast, y = var_732_cast); + tensor var_738_axes_0 = const()[name = tensor("op_738_axes_0"), val = tensor([-1])]; + tensor blocks_5_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224193024)))]; + tensor blocks_5_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224195648)))]; + tensor var_738_cast = layer_norm(axes = var_738_axes_0, beta = blocks_5_mlp_ln_bias_to_fp16, epsilon = var_663_to_fp16, gamma = blocks_5_mlp_ln_weight_to_fp16, x = x_73_cast); + tensor var_747_to_fp16 = const()[name = tensor("op_747_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224198272)))]; + tensor var_748_to_fp16 = const()[name = tensor("op_748_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237305536)))]; + tensor input_49_cast = linear(bias = var_748_to_fp16, weight = var_747_to_fp16, x = var_738_cast); + tensor x_77_mode_0 = const()[name = tensor("x_77_mode_0"), val = tensor("EXACT")]; + tensor x_77_cast = gelu(mode = x_77_mode_0, x = input_49_cast); + tensor var_753_to_fp16 = const()[name = tensor("op_753_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237315840)))]; + tensor var_754_to_fp16 = const()[name = tensor("op_754_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250423104)))]; + tensor var_755_cast = linear(bias = var_754_to_fp16, weight = var_753_to_fp16, x = x_77_cast); + tensor x_79_cast = add(x = x_73_cast, y = var_755_cast); + tensor var_764 = const()[name = tensor("op_764"), val = tensor(-1)]; + tensor var_781_axes_0 = const()[name = tensor("op_781_axes_0"), val = tensor([-1])]; + tensor blocks_6_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_6_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250425728)))]; + tensor blocks_6_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_6_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250428352)))]; + tensor var_770_to_fp16 = const()[name = tensor("op_770_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_781_cast = layer_norm(axes = var_781_axes_0, beta = blocks_6_attn_ln_bias_to_fp16, epsilon = var_770_to_fp16, gamma = blocks_6_attn_ln_weight_to_fp16, x = x_79_cast); + tensor var_792_to_fp16 = const()[name = tensor("op_792_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250430976)))]; + tensor var_793_to_fp16 = const()[name = tensor("op_793_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(253707840)))]; + tensor q_25_cast = linear(bias = var_793_to_fp16, weight = var_792_to_fp16, x = var_781_cast); + tensor var_796_to_fp16 = const()[name = tensor("op_796_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(253710464)))]; + tensor k_25_bias_0_to_fp16 = const()[name = tensor("k_25_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(256987328)))]; + tensor k_25_cast = linear(bias = k_25_bias_0_to_fp16, weight = var_796_to_fp16, x = var_781_cast); + tensor var_800_to_fp16 = const()[name = tensor("op_800_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(256989952)))]; + tensor var_801_to_fp16 = const()[name = tensor("op_801_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260266816)))]; + tensor v_25_cast = linear(bias = var_801_to_fp16, weight = var_800_to_fp16, x = var_781_cast); + tensor var_809 = const()[name = tensor("op_809"), val = tensor([1, 1500, 20, -1])]; + tensor var_810_cast = reshape(shape = var_809, x = q_25_cast); + tensor const_236_to_fp16 = const()[name = tensor("const_236_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_27_cast = mul(x = var_810_cast, y = const_236_to_fp16); + tensor var_816 = const()[name = tensor("op_816"), val = tensor([1, 1500, 20, -1])]; + tensor var_817_cast = reshape(shape = var_816, x = k_25_cast); + tensor const_237_to_fp16 = const()[name = tensor("const_237_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_27_cast = mul(x = var_817_cast, y = const_237_to_fp16); + tensor var_823 = const()[name = tensor("op_823"), val = tensor([1, 1500, 20, -1])]; + tensor var_824_cast = reshape(shape = var_823, x = v_25_cast); + tensor var_825 = const()[name = tensor("op_825"), val = tensor([0, 2, 1, 3])]; + tensor qk_13_transpose_x_0 = const()[name = tensor("qk_13_transpose_x_0"), val = tensor(false)]; + tensor qk_13_transpose_y_0 = const()[name = tensor("qk_13_transpose_y_0"), val = tensor(false)]; + tensor transpose_76_perm_0 = const()[name = tensor("transpose_76_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_77_perm_0 = const()[name = tensor("transpose_77_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_229 = transpose(perm = transpose_77_perm_0, x = k_27_cast); + tensor transpose_230 = transpose(perm = transpose_76_perm_0, x = q_27_cast); + tensor qk_13_cast = matmul(transpose_x = qk_13_transpose_x_0, transpose_y = qk_13_transpose_y_0, x = transpose_230, y = transpose_229); + tensor var_829_cast = softmax(axis = var_764, x = qk_13_cast); + tensor var_831_transpose_x_0 = const()[name = tensor("op_831_transpose_x_0"), val = tensor(false)]; + tensor var_831_transpose_y_0 = const()[name = tensor("op_831_transpose_y_0"), val = tensor(false)]; + tensor transpose_231 = transpose(perm = var_825, x = var_824_cast); + tensor var_831_cast = matmul(transpose_x = var_831_transpose_x_0, transpose_y = var_831_transpose_y_0, x = var_829_cast, y = transpose_231); + tensor var_832 = const()[name = tensor("op_832"), val = tensor([0, 2, 1, 3])]; + tensor concat_6 = const()[name = tensor("concat_6"), val = tensor([1, 1500, 1280])]; + tensor transpose_228 = transpose(perm = var_832, x = var_831_cast); + tensor x_83_cast = reshape(shape = concat_6, x = transpose_228); + tensor var_837_to_fp16 = const()[name = tensor("op_837_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260269440)))]; + tensor var_838_to_fp16 = const()[name = tensor("op_838_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263546304)))]; + tensor var_839_cast = linear(bias = var_838_to_fp16, weight = var_837_to_fp16, x = x_83_cast); + tensor x_85_cast = add(x = x_79_cast, y = var_839_cast); + tensor var_845_axes_0 = const()[name = tensor("op_845_axes_0"), val = tensor([-1])]; + tensor blocks_6_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_6_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263548928)))]; + tensor blocks_6_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_6_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263551552)))]; + tensor var_845_cast = layer_norm(axes = var_845_axes_0, beta = blocks_6_mlp_ln_bias_to_fp16, epsilon = var_770_to_fp16, gamma = blocks_6_mlp_ln_weight_to_fp16, x = x_85_cast); + tensor var_854_to_fp16 = const()[name = tensor("op_854_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263554176)))]; + tensor var_855_to_fp16 = const()[name = tensor("op_855_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(276661440)))]; + tensor input_57_cast = linear(bias = var_855_to_fp16, weight = var_854_to_fp16, x = var_845_cast); + tensor x_89_mode_0 = const()[name = tensor("x_89_mode_0"), val = tensor("EXACT")]; + tensor x_89_cast = gelu(mode = x_89_mode_0, x = input_57_cast); + tensor var_860_to_fp16 = const()[name = tensor("op_860_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(276671744)))]; + tensor var_861_to_fp16 = const()[name = tensor("op_861_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289779008)))]; + tensor var_862_cast = linear(bias = var_861_to_fp16, weight = var_860_to_fp16, x = x_89_cast); + tensor x_91_cast = add(x = x_85_cast, y = var_862_cast); + tensor var_871 = const()[name = tensor("op_871"), val = tensor(-1)]; + tensor var_888_axes_0 = const()[name = tensor("op_888_axes_0"), val = tensor([-1])]; + tensor blocks_7_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_7_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289781632)))]; + tensor blocks_7_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_7_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289784256)))]; + tensor var_877_to_fp16 = const()[name = tensor("op_877_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_888_cast = layer_norm(axes = var_888_axes_0, beta = blocks_7_attn_ln_bias_to_fp16, epsilon = var_877_to_fp16, gamma = blocks_7_attn_ln_weight_to_fp16, x = x_91_cast); + tensor var_899_to_fp16 = const()[name = tensor("op_899_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289786880)))]; + tensor var_900_to_fp16 = const()[name = tensor("op_900_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293063744)))]; + tensor q_29_cast = linear(bias = var_900_to_fp16, weight = var_899_to_fp16, x = var_888_cast); + tensor var_903_to_fp16 = const()[name = tensor("op_903_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293066368)))]; + tensor k_29_bias_0_to_fp16 = const()[name = tensor("k_29_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(296343232)))]; + tensor k_29_cast = linear(bias = k_29_bias_0_to_fp16, weight = var_903_to_fp16, x = var_888_cast); + tensor var_907_to_fp16 = const()[name = tensor("op_907_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(296345856)))]; + tensor var_908_to_fp16 = const()[name = tensor("op_908_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(299622720)))]; + tensor v_29_cast = linear(bias = var_908_to_fp16, weight = var_907_to_fp16, x = var_888_cast); + tensor var_916 = const()[name = tensor("op_916"), val = tensor([1, 1500, 20, -1])]; + tensor var_917_cast = reshape(shape = var_916, x = q_29_cast); + tensor const_238_to_fp16 = const()[name = tensor("const_238_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_31_cast = mul(x = var_917_cast, y = const_238_to_fp16); + tensor var_923 = const()[name = tensor("op_923"), val = tensor([1, 1500, 20, -1])]; + tensor var_924_cast = reshape(shape = var_923, x = k_29_cast); + tensor const_239_to_fp16 = const()[name = tensor("const_239_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_31_cast = mul(x = var_924_cast, y = const_239_to_fp16); + tensor var_930 = const()[name = tensor("op_930"), val = tensor([1, 1500, 20, -1])]; + tensor var_931_cast = reshape(shape = var_930, x = v_29_cast); + tensor var_932 = const()[name = tensor("op_932"), val = tensor([0, 2, 1, 3])]; + tensor qk_15_transpose_x_0 = const()[name = tensor("qk_15_transpose_x_0"), val = tensor(false)]; + tensor qk_15_transpose_y_0 = const()[name = tensor("qk_15_transpose_y_0"), val = tensor(false)]; + tensor transpose_78_perm_0 = const()[name = tensor("transpose_78_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_79_perm_0 = const()[name = tensor("transpose_79_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_225 = transpose(perm = transpose_79_perm_0, x = k_31_cast); + tensor transpose_226 = transpose(perm = transpose_78_perm_0, x = q_31_cast); + tensor qk_15_cast = matmul(transpose_x = qk_15_transpose_x_0, transpose_y = qk_15_transpose_y_0, x = transpose_226, y = transpose_225); + tensor var_936_cast = softmax(axis = var_871, x = qk_15_cast); + tensor var_938_transpose_x_0 = const()[name = tensor("op_938_transpose_x_0"), val = tensor(false)]; + tensor var_938_transpose_y_0 = const()[name = tensor("op_938_transpose_y_0"), val = tensor(false)]; + tensor transpose_227 = transpose(perm = var_932, x = var_931_cast); + tensor var_938_cast = matmul(transpose_x = var_938_transpose_x_0, transpose_y = var_938_transpose_y_0, x = var_936_cast, y = transpose_227); + tensor var_939 = const()[name = tensor("op_939"), val = tensor([0, 2, 1, 3])]; + tensor concat_7 = const()[name = tensor("concat_7"), val = tensor([1, 1500, 1280])]; + tensor transpose_224 = transpose(perm = var_939, x = var_938_cast); + tensor x_95_cast = reshape(shape = concat_7, x = transpose_224); + tensor var_944_to_fp16 = const()[name = tensor("op_944_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(299625344)))]; + tensor var_945_to_fp16 = const()[name = tensor("op_945_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302902208)))]; + tensor var_946_cast = linear(bias = var_945_to_fp16, weight = var_944_to_fp16, x = x_95_cast); + tensor x_97_cast = add(x = x_91_cast, y = var_946_cast); + tensor var_952_axes_0 = const()[name = tensor("op_952_axes_0"), val = tensor([-1])]; + tensor blocks_7_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_7_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302904832)))]; + tensor blocks_7_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_7_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302907456)))]; + tensor var_952_cast = layer_norm(axes = var_952_axes_0, beta = blocks_7_mlp_ln_bias_to_fp16, epsilon = var_877_to_fp16, gamma = blocks_7_mlp_ln_weight_to_fp16, x = x_97_cast); + tensor var_961_to_fp16 = const()[name = tensor("op_961_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302910080)))]; + tensor var_962_to_fp16 = const()[name = tensor("op_962_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316017344)))]; + tensor input_65_cast = linear(bias = var_962_to_fp16, weight = var_961_to_fp16, x = var_952_cast); + tensor x_101_mode_0 = const()[name = tensor("x_101_mode_0"), val = tensor("EXACT")]; + tensor x_101_cast = gelu(mode = x_101_mode_0, x = input_65_cast); + tensor var_967_to_fp16 = const()[name = tensor("op_967_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316027648)))]; + tensor var_968_to_fp16 = const()[name = tensor("op_968_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329134912)))]; + tensor var_969_cast = linear(bias = var_968_to_fp16, weight = var_967_to_fp16, x = x_101_cast); + tensor x_103_cast = add(x = x_97_cast, y = var_969_cast); + tensor var_978 = const()[name = tensor("op_978"), val = tensor(-1)]; + tensor var_995_axes_0 = const()[name = tensor("op_995_axes_0"), val = tensor([-1])]; + tensor blocks_8_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_8_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329137536)))]; + tensor blocks_8_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_8_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329140160)))]; + tensor var_984_to_fp16 = const()[name = tensor("op_984_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_995_cast = layer_norm(axes = var_995_axes_0, beta = blocks_8_attn_ln_bias_to_fp16, epsilon = var_984_to_fp16, gamma = blocks_8_attn_ln_weight_to_fp16, x = x_103_cast); + tensor var_1006_to_fp16 = const()[name = tensor("op_1006_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329142784)))]; + tensor var_1007_to_fp16 = const()[name = tensor("op_1007_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(332419648)))]; + tensor q_33_cast = linear(bias = var_1007_to_fp16, weight = var_1006_to_fp16, x = var_995_cast); + tensor var_1010_to_fp16 = const()[name = tensor("op_1010_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(332422272)))]; + tensor k_33_bias_0_to_fp16 = const()[name = tensor("k_33_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335699136)))]; + tensor k_33_cast = linear(bias = k_33_bias_0_to_fp16, weight = var_1010_to_fp16, x = var_995_cast); + tensor var_1014_to_fp16 = const()[name = tensor("op_1014_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335701760)))]; + tensor var_1015_to_fp16 = const()[name = tensor("op_1015_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(338978624)))]; + tensor v_33_cast = linear(bias = var_1015_to_fp16, weight = var_1014_to_fp16, x = var_995_cast); + tensor var_1023 = const()[name = tensor("op_1023"), val = tensor([1, 1500, 20, -1])]; + tensor var_1024_cast = reshape(shape = var_1023, x = q_33_cast); + tensor const_240_to_fp16 = const()[name = tensor("const_240_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_35_cast = mul(x = var_1024_cast, y = const_240_to_fp16); + tensor var_1030 = const()[name = tensor("op_1030"), val = tensor([1, 1500, 20, -1])]; + tensor var_1031_cast = reshape(shape = var_1030, x = k_33_cast); + tensor const_241_to_fp16 = const()[name = tensor("const_241_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_35_cast = mul(x = var_1031_cast, y = const_241_to_fp16); + tensor var_1037 = const()[name = tensor("op_1037"), val = tensor([1, 1500, 20, -1])]; + tensor var_1038_cast = reshape(shape = var_1037, x = v_33_cast); + tensor var_1039 = const()[name = tensor("op_1039"), val = tensor([0, 2, 1, 3])]; + tensor qk_17_transpose_x_0 = const()[name = tensor("qk_17_transpose_x_0"), val = tensor(false)]; + tensor qk_17_transpose_y_0 = const()[name = tensor("qk_17_transpose_y_0"), val = tensor(false)]; + tensor transpose_80_perm_0 = const()[name = tensor("transpose_80_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_81_perm_0 = const()[name = tensor("transpose_81_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_221 = transpose(perm = transpose_81_perm_0, x = k_35_cast); + tensor transpose_222 = transpose(perm = transpose_80_perm_0, x = q_35_cast); + tensor qk_17_cast = matmul(transpose_x = qk_17_transpose_x_0, transpose_y = qk_17_transpose_y_0, x = transpose_222, y = transpose_221); + tensor var_1043_cast = softmax(axis = var_978, x = qk_17_cast); + tensor var_1045_transpose_x_0 = const()[name = tensor("op_1045_transpose_x_0"), val = tensor(false)]; + tensor var_1045_transpose_y_0 = const()[name = tensor("op_1045_transpose_y_0"), val = tensor(false)]; + tensor transpose_223 = transpose(perm = var_1039, x = var_1038_cast); + tensor var_1045_cast = matmul(transpose_x = var_1045_transpose_x_0, transpose_y = var_1045_transpose_y_0, x = var_1043_cast, y = transpose_223); + tensor var_1046 = const()[name = tensor("op_1046"), val = tensor([0, 2, 1, 3])]; + tensor concat_8 = const()[name = tensor("concat_8"), val = tensor([1, 1500, 1280])]; + tensor transpose_220 = transpose(perm = var_1046, x = var_1045_cast); + tensor x_107_cast = reshape(shape = concat_8, x = transpose_220); + tensor var_1051_to_fp16 = const()[name = tensor("op_1051_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(338981248)))]; + tensor var_1052_to_fp16 = const()[name = tensor("op_1052_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342258112)))]; + tensor var_1053_cast = linear(bias = var_1052_to_fp16, weight = var_1051_to_fp16, x = x_107_cast); + tensor x_109_cast = add(x = x_103_cast, y = var_1053_cast); + tensor var_1059_axes_0 = const()[name = tensor("op_1059_axes_0"), val = tensor([-1])]; + tensor blocks_8_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_8_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342260736)))]; + tensor blocks_8_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_8_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342263360)))]; + tensor var_1059_cast = layer_norm(axes = var_1059_axes_0, beta = blocks_8_mlp_ln_bias_to_fp16, epsilon = var_984_to_fp16, gamma = blocks_8_mlp_ln_weight_to_fp16, x = x_109_cast); + tensor var_1068_to_fp16 = const()[name = tensor("op_1068_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342265984)))]; + tensor var_1069_to_fp16 = const()[name = tensor("op_1069_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(355373248)))]; + tensor input_73_cast = linear(bias = var_1069_to_fp16, weight = var_1068_to_fp16, x = var_1059_cast); + tensor x_113_mode_0 = const()[name = tensor("x_113_mode_0"), val = tensor("EXACT")]; + tensor x_113_cast = gelu(mode = x_113_mode_0, x = input_73_cast); + tensor var_1074_to_fp16 = const()[name = tensor("op_1074_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(355383552)))]; + tensor var_1075_to_fp16 = const()[name = tensor("op_1075_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368490816)))]; + tensor var_1076_cast = linear(bias = var_1075_to_fp16, weight = var_1074_to_fp16, x = x_113_cast); + tensor x_115_cast = add(x = x_109_cast, y = var_1076_cast); + tensor var_1085 = const()[name = tensor("op_1085"), val = tensor(-1)]; + tensor var_1102_axes_0 = const()[name = tensor("op_1102_axes_0"), val = tensor([-1])]; + tensor blocks_9_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_9_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368493440)))]; + tensor blocks_9_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_9_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368496064)))]; + tensor var_1091_to_fp16 = const()[name = tensor("op_1091_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1102_cast = layer_norm(axes = var_1102_axes_0, beta = blocks_9_attn_ln_bias_to_fp16, epsilon = var_1091_to_fp16, gamma = blocks_9_attn_ln_weight_to_fp16, x = x_115_cast); + tensor var_1113_to_fp16 = const()[name = tensor("op_1113_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368498688)))]; + tensor var_1114_to_fp16 = const()[name = tensor("op_1114_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(371775552)))]; + tensor q_37_cast = linear(bias = var_1114_to_fp16, weight = var_1113_to_fp16, x = var_1102_cast); + tensor var_1117_to_fp16 = const()[name = tensor("op_1117_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(371778176)))]; + tensor k_37_bias_0_to_fp16 = const()[name = tensor("k_37_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(375055040)))]; + tensor k_37_cast = linear(bias = k_37_bias_0_to_fp16, weight = var_1117_to_fp16, x = var_1102_cast); + tensor var_1121_to_fp16 = const()[name = tensor("op_1121_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(375057664)))]; + tensor var_1122_to_fp16 = const()[name = tensor("op_1122_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(378334528)))]; + tensor v_37_cast = linear(bias = var_1122_to_fp16, weight = var_1121_to_fp16, x = var_1102_cast); + tensor var_1130 = const()[name = tensor("op_1130"), val = tensor([1, 1500, 20, -1])]; + tensor var_1131_cast = reshape(shape = var_1130, x = q_37_cast); + tensor const_242_to_fp16 = const()[name = tensor("const_242_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_39_cast = mul(x = var_1131_cast, y = const_242_to_fp16); + tensor var_1137 = const()[name = tensor("op_1137"), val = tensor([1, 1500, 20, -1])]; + tensor var_1138_cast = reshape(shape = var_1137, x = k_37_cast); + tensor const_243_to_fp16 = const()[name = tensor("const_243_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_39_cast = mul(x = var_1138_cast, y = const_243_to_fp16); + tensor var_1144 = const()[name = tensor("op_1144"), val = tensor([1, 1500, 20, -1])]; + tensor var_1145_cast = reshape(shape = var_1144, x = v_37_cast); + tensor var_1146 = const()[name = tensor("op_1146"), val = tensor([0, 2, 1, 3])]; + tensor qk_19_transpose_x_0 = const()[name = tensor("qk_19_transpose_x_0"), val = tensor(false)]; + tensor qk_19_transpose_y_0 = const()[name = tensor("qk_19_transpose_y_0"), val = tensor(false)]; + tensor transpose_82_perm_0 = const()[name = tensor("transpose_82_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_83_perm_0 = const()[name = tensor("transpose_83_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_217 = transpose(perm = transpose_83_perm_0, x = k_39_cast); + tensor transpose_218 = transpose(perm = transpose_82_perm_0, x = q_39_cast); + tensor qk_19_cast = matmul(transpose_x = qk_19_transpose_x_0, transpose_y = qk_19_transpose_y_0, x = transpose_218, y = transpose_217); + tensor var_1150_cast = softmax(axis = var_1085, x = qk_19_cast); + tensor var_1152_transpose_x_0 = const()[name = tensor("op_1152_transpose_x_0"), val = tensor(false)]; + tensor var_1152_transpose_y_0 = const()[name = tensor("op_1152_transpose_y_0"), val = tensor(false)]; + tensor transpose_219 = transpose(perm = var_1146, x = var_1145_cast); + tensor var_1152_cast = matmul(transpose_x = var_1152_transpose_x_0, transpose_y = var_1152_transpose_y_0, x = var_1150_cast, y = transpose_219); + tensor var_1153 = const()[name = tensor("op_1153"), val = tensor([0, 2, 1, 3])]; + tensor concat_9 = const()[name = tensor("concat_9"), val = tensor([1, 1500, 1280])]; + tensor transpose_216 = transpose(perm = var_1153, x = var_1152_cast); + tensor x_119_cast = reshape(shape = concat_9, x = transpose_216); + tensor var_1158_to_fp16 = const()[name = tensor("op_1158_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(378337152)))]; + tensor var_1159_to_fp16 = const()[name = tensor("op_1159_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381614016)))]; + tensor var_1160_cast = linear(bias = var_1159_to_fp16, weight = var_1158_to_fp16, x = x_119_cast); + tensor x_121_cast = add(x = x_115_cast, y = var_1160_cast); + tensor var_1166_axes_0 = const()[name = tensor("op_1166_axes_0"), val = tensor([-1])]; + tensor blocks_9_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_9_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381616640)))]; + tensor blocks_9_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_9_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381619264)))]; + tensor var_1166_cast = layer_norm(axes = var_1166_axes_0, beta = blocks_9_mlp_ln_bias_to_fp16, epsilon = var_1091_to_fp16, gamma = blocks_9_mlp_ln_weight_to_fp16, x = x_121_cast); + tensor var_1175_to_fp16 = const()[name = tensor("op_1175_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381621888)))]; + tensor var_1176_to_fp16 = const()[name = tensor("op_1176_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(394729152)))]; + tensor input_81_cast = linear(bias = var_1176_to_fp16, weight = var_1175_to_fp16, x = var_1166_cast); + tensor x_125_mode_0 = const()[name = tensor("x_125_mode_0"), val = tensor("EXACT")]; + tensor x_125_cast = gelu(mode = x_125_mode_0, x = input_81_cast); + tensor var_1181_to_fp16 = const()[name = tensor("op_1181_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(394739456)))]; + tensor var_1182_to_fp16 = const()[name = tensor("op_1182_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407846720)))]; + tensor var_1183_cast = linear(bias = var_1182_to_fp16, weight = var_1181_to_fp16, x = x_125_cast); + tensor x_127_cast = add(x = x_121_cast, y = var_1183_cast); + tensor var_1192 = const()[name = tensor("op_1192"), val = tensor(-1)]; + tensor var_1209_axes_0 = const()[name = tensor("op_1209_axes_0"), val = tensor([-1])]; + tensor blocks_10_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_10_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407849344)))]; + tensor blocks_10_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_10_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407851968)))]; + tensor var_1198_to_fp16 = const()[name = tensor("op_1198_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1209_cast = layer_norm(axes = var_1209_axes_0, beta = blocks_10_attn_ln_bias_to_fp16, epsilon = var_1198_to_fp16, gamma = blocks_10_attn_ln_weight_to_fp16, x = x_127_cast); + tensor var_1220_to_fp16 = const()[name = tensor("op_1220_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407854592)))]; + tensor var_1221_to_fp16 = const()[name = tensor("op_1221_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411131456)))]; + tensor q_41_cast = linear(bias = var_1221_to_fp16, weight = var_1220_to_fp16, x = var_1209_cast); + tensor var_1224_to_fp16 = const()[name = tensor("op_1224_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411134080)))]; + tensor k_41_bias_0_to_fp16 = const()[name = tensor("k_41_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(414410944)))]; + tensor k_41_cast = linear(bias = k_41_bias_0_to_fp16, weight = var_1224_to_fp16, x = var_1209_cast); + tensor var_1228_to_fp16 = const()[name = tensor("op_1228_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(414413568)))]; + tensor var_1229_to_fp16 = const()[name = tensor("op_1229_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(417690432)))]; + tensor v_41_cast = linear(bias = var_1229_to_fp16, weight = var_1228_to_fp16, x = var_1209_cast); + tensor var_1237 = const()[name = tensor("op_1237"), val = tensor([1, 1500, 20, -1])]; + tensor var_1238_cast = reshape(shape = var_1237, x = q_41_cast); + tensor const_244_to_fp16 = const()[name = tensor("const_244_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_43_cast = mul(x = var_1238_cast, y = const_244_to_fp16); + tensor var_1244 = const()[name = tensor("op_1244"), val = tensor([1, 1500, 20, -1])]; + tensor var_1245_cast = reshape(shape = var_1244, x = k_41_cast); + tensor const_245_to_fp16 = const()[name = tensor("const_245_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_43_cast = mul(x = var_1245_cast, y = const_245_to_fp16); + tensor var_1251 = const()[name = tensor("op_1251"), val = tensor([1, 1500, 20, -1])]; + tensor var_1252_cast = reshape(shape = var_1251, x = v_41_cast); + tensor var_1253 = const()[name = tensor("op_1253"), val = tensor([0, 2, 1, 3])]; + tensor qk_21_transpose_x_0 = const()[name = tensor("qk_21_transpose_x_0"), val = tensor(false)]; + tensor qk_21_transpose_y_0 = const()[name = tensor("qk_21_transpose_y_0"), val = tensor(false)]; + tensor transpose_84_perm_0 = const()[name = tensor("transpose_84_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_85_perm_0 = const()[name = tensor("transpose_85_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_213 = transpose(perm = transpose_85_perm_0, x = k_43_cast); + tensor transpose_214 = transpose(perm = transpose_84_perm_0, x = q_43_cast); + tensor qk_21_cast = matmul(transpose_x = qk_21_transpose_x_0, transpose_y = qk_21_transpose_y_0, x = transpose_214, y = transpose_213); + tensor var_1257_cast = softmax(axis = var_1192, x = qk_21_cast); + tensor var_1259_transpose_x_0 = const()[name = tensor("op_1259_transpose_x_0"), val = tensor(false)]; + tensor var_1259_transpose_y_0 = const()[name = tensor("op_1259_transpose_y_0"), val = tensor(false)]; + tensor transpose_215 = transpose(perm = var_1253, x = var_1252_cast); + tensor var_1259_cast = matmul(transpose_x = var_1259_transpose_x_0, transpose_y = var_1259_transpose_y_0, x = var_1257_cast, y = transpose_215); + tensor var_1260 = const()[name = tensor("op_1260"), val = tensor([0, 2, 1, 3])]; + tensor concat_10 = const()[name = tensor("concat_10"), val = tensor([1, 1500, 1280])]; + tensor transpose_212 = transpose(perm = var_1260, x = var_1259_cast); + tensor x_131_cast = reshape(shape = concat_10, x = transpose_212); + tensor var_1265_to_fp16 = const()[name = tensor("op_1265_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(417693056)))]; + tensor var_1266_to_fp16 = const()[name = tensor("op_1266_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(420969920)))]; + tensor var_1267_cast = linear(bias = var_1266_to_fp16, weight = var_1265_to_fp16, x = x_131_cast); + tensor x_133_cast = add(x = x_127_cast, y = var_1267_cast); + tensor var_1273_axes_0 = const()[name = tensor("op_1273_axes_0"), val = tensor([-1])]; + tensor blocks_10_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_10_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(420972544)))]; + tensor blocks_10_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_10_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(420975168)))]; + tensor var_1273_cast = layer_norm(axes = var_1273_axes_0, beta = blocks_10_mlp_ln_bias_to_fp16, epsilon = var_1198_to_fp16, gamma = blocks_10_mlp_ln_weight_to_fp16, x = x_133_cast); + tensor var_1282_to_fp16 = const()[name = tensor("op_1282_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(420977792)))]; + tensor var_1283_to_fp16 = const()[name = tensor("op_1283_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(434085056)))]; + tensor input_89_cast = linear(bias = var_1283_to_fp16, weight = var_1282_to_fp16, x = var_1273_cast); + tensor x_137_mode_0 = const()[name = tensor("x_137_mode_0"), val = tensor("EXACT")]; + tensor x_137_cast = gelu(mode = x_137_mode_0, x = input_89_cast); + tensor var_1288_to_fp16 = const()[name = tensor("op_1288_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(434095360)))]; + tensor var_1289_to_fp16 = const()[name = tensor("op_1289_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447202624)))]; + tensor var_1290_cast = linear(bias = var_1289_to_fp16, weight = var_1288_to_fp16, x = x_137_cast); + tensor x_139_cast = add(x = x_133_cast, y = var_1290_cast); + tensor var_1299 = const()[name = tensor("op_1299"), val = tensor(-1)]; + tensor var_1316_axes_0 = const()[name = tensor("op_1316_axes_0"), val = tensor([-1])]; + tensor blocks_11_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_11_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447205248)))]; + tensor blocks_11_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_11_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447207872)))]; + tensor var_1305_to_fp16 = const()[name = tensor("op_1305_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1316_cast = layer_norm(axes = var_1316_axes_0, beta = blocks_11_attn_ln_bias_to_fp16, epsilon = var_1305_to_fp16, gamma = blocks_11_attn_ln_weight_to_fp16, x = x_139_cast); + tensor var_1327_to_fp16 = const()[name = tensor("op_1327_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447210496)))]; + tensor var_1328_to_fp16 = const()[name = tensor("op_1328_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(450487360)))]; + tensor q_45_cast = linear(bias = var_1328_to_fp16, weight = var_1327_to_fp16, x = var_1316_cast); + tensor var_1331_to_fp16 = const()[name = tensor("op_1331_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(450489984)))]; + tensor k_45_bias_0_to_fp16 = const()[name = tensor("k_45_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(453766848)))]; + tensor k_45_cast = linear(bias = k_45_bias_0_to_fp16, weight = var_1331_to_fp16, x = var_1316_cast); + tensor var_1335_to_fp16 = const()[name = tensor("op_1335_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(453769472)))]; + tensor var_1336_to_fp16 = const()[name = tensor("op_1336_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(457046336)))]; + tensor v_45_cast = linear(bias = var_1336_to_fp16, weight = var_1335_to_fp16, x = var_1316_cast); + tensor var_1344 = const()[name = tensor("op_1344"), val = tensor([1, 1500, 20, -1])]; + tensor var_1345_cast = reshape(shape = var_1344, x = q_45_cast); + tensor const_246_to_fp16 = const()[name = tensor("const_246_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_47_cast = mul(x = var_1345_cast, y = const_246_to_fp16); + tensor var_1351 = const()[name = tensor("op_1351"), val = tensor([1, 1500, 20, -1])]; + tensor var_1352_cast = reshape(shape = var_1351, x = k_45_cast); + tensor const_247_to_fp16 = const()[name = tensor("const_247_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_47_cast = mul(x = var_1352_cast, y = const_247_to_fp16); + tensor var_1358 = const()[name = tensor("op_1358"), val = tensor([1, 1500, 20, -1])]; + tensor var_1359_cast = reshape(shape = var_1358, x = v_45_cast); + tensor var_1360 = const()[name = tensor("op_1360"), val = tensor([0, 2, 1, 3])]; + tensor qk_23_transpose_x_0 = const()[name = tensor("qk_23_transpose_x_0"), val = tensor(false)]; + tensor qk_23_transpose_y_0 = const()[name = tensor("qk_23_transpose_y_0"), val = tensor(false)]; + tensor transpose_86_perm_0 = const()[name = tensor("transpose_86_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_87_perm_0 = const()[name = tensor("transpose_87_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_209 = transpose(perm = transpose_87_perm_0, x = k_47_cast); + tensor transpose_210 = transpose(perm = transpose_86_perm_0, x = q_47_cast); + tensor qk_23_cast = matmul(transpose_x = qk_23_transpose_x_0, transpose_y = qk_23_transpose_y_0, x = transpose_210, y = transpose_209); + tensor var_1364_cast = softmax(axis = var_1299, x = qk_23_cast); + tensor var_1366_transpose_x_0 = const()[name = tensor("op_1366_transpose_x_0"), val = tensor(false)]; + tensor var_1366_transpose_y_0 = const()[name = tensor("op_1366_transpose_y_0"), val = tensor(false)]; + tensor transpose_211 = transpose(perm = var_1360, x = var_1359_cast); + tensor var_1366_cast = matmul(transpose_x = var_1366_transpose_x_0, transpose_y = var_1366_transpose_y_0, x = var_1364_cast, y = transpose_211); + tensor var_1367 = const()[name = tensor("op_1367"), val = tensor([0, 2, 1, 3])]; + tensor concat_11 = const()[name = tensor("concat_11"), val = tensor([1, 1500, 1280])]; + tensor transpose_208 = transpose(perm = var_1367, x = var_1366_cast); + tensor x_143_cast = reshape(shape = concat_11, x = transpose_208); + tensor var_1372_to_fp16 = const()[name = tensor("op_1372_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(457048960)))]; + tensor var_1373_to_fp16 = const()[name = tensor("op_1373_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460325824)))]; + tensor var_1374_cast = linear(bias = var_1373_to_fp16, weight = var_1372_to_fp16, x = x_143_cast); + tensor x_145_cast = add(x = x_139_cast, y = var_1374_cast); + tensor var_1380_axes_0 = const()[name = tensor("op_1380_axes_0"), val = tensor([-1])]; + tensor blocks_11_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_11_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460328448)))]; + tensor blocks_11_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_11_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460331072)))]; + tensor var_1380_cast = layer_norm(axes = var_1380_axes_0, beta = blocks_11_mlp_ln_bias_to_fp16, epsilon = var_1305_to_fp16, gamma = blocks_11_mlp_ln_weight_to_fp16, x = x_145_cast); + tensor var_1389_to_fp16 = const()[name = tensor("op_1389_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460333696)))]; + tensor var_1390_to_fp16 = const()[name = tensor("op_1390_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473440960)))]; + tensor input_97_cast = linear(bias = var_1390_to_fp16, weight = var_1389_to_fp16, x = var_1380_cast); + tensor x_149_mode_0 = const()[name = tensor("x_149_mode_0"), val = tensor("EXACT")]; + tensor x_149_cast = gelu(mode = x_149_mode_0, x = input_97_cast); + tensor var_1395_to_fp16 = const()[name = tensor("op_1395_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473451264)))]; + tensor var_1396_to_fp16 = const()[name = tensor("op_1396_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486558528)))]; + tensor var_1397_cast = linear(bias = var_1396_to_fp16, weight = var_1395_to_fp16, x = x_149_cast); + tensor x_151_cast = add(x = x_145_cast, y = var_1397_cast); + tensor var_1406 = const()[name = tensor("op_1406"), val = tensor(-1)]; + tensor var_1423_axes_0 = const()[name = tensor("op_1423_axes_0"), val = tensor([-1])]; + tensor blocks_12_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_12_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486561152)))]; + tensor blocks_12_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_12_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486563776)))]; + tensor var_1412_to_fp16 = const()[name = tensor("op_1412_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1423_cast = layer_norm(axes = var_1423_axes_0, beta = blocks_12_attn_ln_bias_to_fp16, epsilon = var_1412_to_fp16, gamma = blocks_12_attn_ln_weight_to_fp16, x = x_151_cast); + tensor var_1434_to_fp16 = const()[name = tensor("op_1434_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486566400)))]; + tensor var_1435_to_fp16 = const()[name = tensor("op_1435_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(489843264)))]; + tensor q_49_cast = linear(bias = var_1435_to_fp16, weight = var_1434_to_fp16, x = var_1423_cast); + tensor var_1438_to_fp16 = const()[name = tensor("op_1438_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(489845888)))]; + tensor k_49_bias_0_to_fp16 = const()[name = tensor("k_49_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(493122752)))]; + tensor k_49_cast = linear(bias = k_49_bias_0_to_fp16, weight = var_1438_to_fp16, x = var_1423_cast); + tensor var_1442_to_fp16 = const()[name = tensor("op_1442_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(493125376)))]; + tensor var_1443_to_fp16 = const()[name = tensor("op_1443_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496402240)))]; + tensor v_49_cast = linear(bias = var_1443_to_fp16, weight = var_1442_to_fp16, x = var_1423_cast); + tensor var_1451 = const()[name = tensor("op_1451"), val = tensor([1, 1500, 20, -1])]; + tensor var_1452_cast = reshape(shape = var_1451, x = q_49_cast); + tensor const_248_to_fp16 = const()[name = tensor("const_248_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_51_cast = mul(x = var_1452_cast, y = const_248_to_fp16); + tensor var_1458 = const()[name = tensor("op_1458"), val = tensor([1, 1500, 20, -1])]; + tensor var_1459_cast = reshape(shape = var_1458, x = k_49_cast); + tensor const_249_to_fp16 = const()[name = tensor("const_249_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_51_cast = mul(x = var_1459_cast, y = const_249_to_fp16); + tensor var_1465 = const()[name = tensor("op_1465"), val = tensor([1, 1500, 20, -1])]; + tensor var_1466_cast = reshape(shape = var_1465, x = v_49_cast); + tensor var_1467 = const()[name = tensor("op_1467"), val = tensor([0, 2, 1, 3])]; + tensor qk_25_transpose_x_0 = const()[name = tensor("qk_25_transpose_x_0"), val = tensor(false)]; + tensor qk_25_transpose_y_0 = const()[name = tensor("qk_25_transpose_y_0"), val = tensor(false)]; + tensor transpose_88_perm_0 = const()[name = tensor("transpose_88_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_89_perm_0 = const()[name = tensor("transpose_89_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_205 = transpose(perm = transpose_89_perm_0, x = k_51_cast); + tensor transpose_206 = transpose(perm = transpose_88_perm_0, x = q_51_cast); + tensor qk_25_cast = matmul(transpose_x = qk_25_transpose_x_0, transpose_y = qk_25_transpose_y_0, x = transpose_206, y = transpose_205); + tensor var_1471_cast = softmax(axis = var_1406, x = qk_25_cast); + tensor var_1473_transpose_x_0 = const()[name = tensor("op_1473_transpose_x_0"), val = tensor(false)]; + tensor var_1473_transpose_y_0 = const()[name = tensor("op_1473_transpose_y_0"), val = tensor(false)]; + tensor transpose_207 = transpose(perm = var_1467, x = var_1466_cast); + tensor var_1473_cast = matmul(transpose_x = var_1473_transpose_x_0, transpose_y = var_1473_transpose_y_0, x = var_1471_cast, y = transpose_207); + tensor var_1474 = const()[name = tensor("op_1474"), val = tensor([0, 2, 1, 3])]; + tensor concat_12 = const()[name = tensor("concat_12"), val = tensor([1, 1500, 1280])]; + tensor transpose_204 = transpose(perm = var_1474, x = var_1473_cast); + tensor x_155_cast = reshape(shape = concat_12, x = transpose_204); + tensor var_1479_to_fp16 = const()[name = tensor("op_1479_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496404864)))]; + tensor var_1480_to_fp16 = const()[name = tensor("op_1480_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(499681728)))]; + tensor var_1481_cast = linear(bias = var_1480_to_fp16, weight = var_1479_to_fp16, x = x_155_cast); + tensor x_157_cast = add(x = x_151_cast, y = var_1481_cast); + tensor var_1487_axes_0 = const()[name = tensor("op_1487_axes_0"), val = tensor([-1])]; + tensor blocks_12_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_12_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(499684352)))]; + tensor blocks_12_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_12_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(499686976)))]; + tensor var_1487_cast = layer_norm(axes = var_1487_axes_0, beta = blocks_12_mlp_ln_bias_to_fp16, epsilon = var_1412_to_fp16, gamma = blocks_12_mlp_ln_weight_to_fp16, x = x_157_cast); + tensor var_1496_to_fp16 = const()[name = tensor("op_1496_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(499689600)))]; + tensor var_1497_to_fp16 = const()[name = tensor("op_1497_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(512796864)))]; + tensor input_105_cast = linear(bias = var_1497_to_fp16, weight = var_1496_to_fp16, x = var_1487_cast); + tensor x_161_mode_0 = const()[name = tensor("x_161_mode_0"), val = tensor("EXACT")]; + tensor x_161_cast = gelu(mode = x_161_mode_0, x = input_105_cast); + tensor var_1502_to_fp16 = const()[name = tensor("op_1502_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(512807168)))]; + tensor var_1503_to_fp16 = const()[name = tensor("op_1503_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(525914432)))]; + tensor var_1504_cast = linear(bias = var_1503_to_fp16, weight = var_1502_to_fp16, x = x_161_cast); + tensor x_163_cast = add(x = x_157_cast, y = var_1504_cast); + tensor var_1513 = const()[name = tensor("op_1513"), val = tensor(-1)]; + tensor var_1530_axes_0 = const()[name = tensor("op_1530_axes_0"), val = tensor([-1])]; + tensor blocks_13_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_13_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(525917056)))]; + tensor blocks_13_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_13_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(525919680)))]; + tensor var_1519_to_fp16 = const()[name = tensor("op_1519_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1530_cast = layer_norm(axes = var_1530_axes_0, beta = blocks_13_attn_ln_bias_to_fp16, epsilon = var_1519_to_fp16, gamma = blocks_13_attn_ln_weight_to_fp16, x = x_163_cast); + tensor var_1541_to_fp16 = const()[name = tensor("op_1541_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(525922304)))]; + tensor var_1542_to_fp16 = const()[name = tensor("op_1542_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529199168)))]; + tensor q_53_cast = linear(bias = var_1542_to_fp16, weight = var_1541_to_fp16, x = var_1530_cast); + tensor var_1545_to_fp16 = const()[name = tensor("op_1545_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529201792)))]; + tensor k_53_bias_0_to_fp16 = const()[name = tensor("k_53_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(532478656)))]; + tensor k_53_cast = linear(bias = k_53_bias_0_to_fp16, weight = var_1545_to_fp16, x = var_1530_cast); + tensor var_1549_to_fp16 = const()[name = tensor("op_1549_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(532481280)))]; + tensor var_1550_to_fp16 = const()[name = tensor("op_1550_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(535758144)))]; + tensor v_53_cast = linear(bias = var_1550_to_fp16, weight = var_1549_to_fp16, x = var_1530_cast); + tensor var_1558 = const()[name = tensor("op_1558"), val = tensor([1, 1500, 20, -1])]; + tensor var_1559_cast = reshape(shape = var_1558, x = q_53_cast); + tensor const_250_to_fp16 = const()[name = tensor("const_250_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_55_cast = mul(x = var_1559_cast, y = const_250_to_fp16); + tensor var_1565 = const()[name = tensor("op_1565"), val = tensor([1, 1500, 20, -1])]; + tensor var_1566_cast = reshape(shape = var_1565, x = k_53_cast); + tensor const_251_to_fp16 = const()[name = tensor("const_251_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_55_cast = mul(x = var_1566_cast, y = const_251_to_fp16); + tensor var_1572 = const()[name = tensor("op_1572"), val = tensor([1, 1500, 20, -1])]; + tensor var_1573_cast = reshape(shape = var_1572, x = v_53_cast); + tensor var_1574 = const()[name = tensor("op_1574"), val = tensor([0, 2, 1, 3])]; + tensor qk_27_transpose_x_0 = const()[name = tensor("qk_27_transpose_x_0"), val = tensor(false)]; + tensor qk_27_transpose_y_0 = const()[name = tensor("qk_27_transpose_y_0"), val = tensor(false)]; + tensor transpose_90_perm_0 = const()[name = tensor("transpose_90_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_91_perm_0 = const()[name = tensor("transpose_91_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_201 = transpose(perm = transpose_91_perm_0, x = k_55_cast); + tensor transpose_202 = transpose(perm = transpose_90_perm_0, x = q_55_cast); + tensor qk_27_cast = matmul(transpose_x = qk_27_transpose_x_0, transpose_y = qk_27_transpose_y_0, x = transpose_202, y = transpose_201); + tensor var_1578_cast = softmax(axis = var_1513, x = qk_27_cast); + tensor var_1580_transpose_x_0 = const()[name = tensor("op_1580_transpose_x_0"), val = tensor(false)]; + tensor var_1580_transpose_y_0 = const()[name = tensor("op_1580_transpose_y_0"), val = tensor(false)]; + tensor transpose_203 = transpose(perm = var_1574, x = var_1573_cast); + tensor var_1580_cast = matmul(transpose_x = var_1580_transpose_x_0, transpose_y = var_1580_transpose_y_0, x = var_1578_cast, y = transpose_203); + tensor var_1581 = const()[name = tensor("op_1581"), val = tensor([0, 2, 1, 3])]; + tensor concat_13 = const()[name = tensor("concat_13"), val = tensor([1, 1500, 1280])]; + tensor transpose_200 = transpose(perm = var_1581, x = var_1580_cast); + tensor x_167_cast = reshape(shape = concat_13, x = transpose_200); + tensor var_1586_to_fp16 = const()[name = tensor("op_1586_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(535760768)))]; + tensor var_1587_to_fp16 = const()[name = tensor("op_1587_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(539037632)))]; + tensor var_1588_cast = linear(bias = var_1587_to_fp16, weight = var_1586_to_fp16, x = x_167_cast); + tensor x_169_cast = add(x = x_163_cast, y = var_1588_cast); + tensor var_1594_axes_0 = const()[name = tensor("op_1594_axes_0"), val = tensor([-1])]; + tensor blocks_13_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_13_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(539040256)))]; + tensor blocks_13_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_13_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(539042880)))]; + tensor var_1594_cast = layer_norm(axes = var_1594_axes_0, beta = blocks_13_mlp_ln_bias_to_fp16, epsilon = var_1519_to_fp16, gamma = blocks_13_mlp_ln_weight_to_fp16, x = x_169_cast); + tensor var_1603_to_fp16 = const()[name = tensor("op_1603_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(539045504)))]; + tensor var_1604_to_fp16 = const()[name = tensor("op_1604_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552152768)))]; + tensor input_113_cast = linear(bias = var_1604_to_fp16, weight = var_1603_to_fp16, x = var_1594_cast); + tensor x_173_mode_0 = const()[name = tensor("x_173_mode_0"), val = tensor("EXACT")]; + tensor x_173_cast = gelu(mode = x_173_mode_0, x = input_113_cast); + tensor var_1609_to_fp16 = const()[name = tensor("op_1609_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552163072)))]; + tensor var_1610_to_fp16 = const()[name = tensor("op_1610_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565270336)))]; + tensor var_1611_cast = linear(bias = var_1610_to_fp16, weight = var_1609_to_fp16, x = x_173_cast); + tensor x_175_cast = add(x = x_169_cast, y = var_1611_cast); + tensor var_1620 = const()[name = tensor("op_1620"), val = tensor(-1)]; + tensor var_1637_axes_0 = const()[name = tensor("op_1637_axes_0"), val = tensor([-1])]; + tensor blocks_14_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_14_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565272960)))]; + tensor blocks_14_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_14_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565275584)))]; + tensor var_1626_to_fp16 = const()[name = tensor("op_1626_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1637_cast = layer_norm(axes = var_1637_axes_0, beta = blocks_14_attn_ln_bias_to_fp16, epsilon = var_1626_to_fp16, gamma = blocks_14_attn_ln_weight_to_fp16, x = x_175_cast); + tensor var_1648_to_fp16 = const()[name = tensor("op_1648_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565278208)))]; + tensor var_1649_to_fp16 = const()[name = tensor("op_1649_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568555072)))]; + tensor q_57_cast = linear(bias = var_1649_to_fp16, weight = var_1648_to_fp16, x = var_1637_cast); + tensor var_1652_to_fp16 = const()[name = tensor("op_1652_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568557696)))]; + tensor k_57_bias_0_to_fp16 = const()[name = tensor("k_57_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(571834560)))]; + tensor k_57_cast = linear(bias = k_57_bias_0_to_fp16, weight = var_1652_to_fp16, x = var_1637_cast); + tensor var_1656_to_fp16 = const()[name = tensor("op_1656_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(571837184)))]; + tensor var_1657_to_fp16 = const()[name = tensor("op_1657_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(575114048)))]; + tensor v_57_cast = linear(bias = var_1657_to_fp16, weight = var_1656_to_fp16, x = var_1637_cast); + tensor var_1665 = const()[name = tensor("op_1665"), val = tensor([1, 1500, 20, -1])]; + tensor var_1666_cast = reshape(shape = var_1665, x = q_57_cast); + tensor const_252_to_fp16 = const()[name = tensor("const_252_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_59_cast = mul(x = var_1666_cast, y = const_252_to_fp16); + tensor var_1672 = const()[name = tensor("op_1672"), val = tensor([1, 1500, 20, -1])]; + tensor var_1673_cast = reshape(shape = var_1672, x = k_57_cast); + tensor const_253_to_fp16 = const()[name = tensor("const_253_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_59_cast = mul(x = var_1673_cast, y = const_253_to_fp16); + tensor var_1679 = const()[name = tensor("op_1679"), val = tensor([1, 1500, 20, -1])]; + tensor var_1680_cast = reshape(shape = var_1679, x = v_57_cast); + tensor var_1681 = const()[name = tensor("op_1681"), val = tensor([0, 2, 1, 3])]; + tensor qk_29_transpose_x_0 = const()[name = tensor("qk_29_transpose_x_0"), val = tensor(false)]; + tensor qk_29_transpose_y_0 = const()[name = tensor("qk_29_transpose_y_0"), val = tensor(false)]; + tensor transpose_92_perm_0 = const()[name = tensor("transpose_92_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_93_perm_0 = const()[name = tensor("transpose_93_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_197 = transpose(perm = transpose_93_perm_0, x = k_59_cast); + tensor transpose_198 = transpose(perm = transpose_92_perm_0, x = q_59_cast); + tensor qk_29_cast = matmul(transpose_x = qk_29_transpose_x_0, transpose_y = qk_29_transpose_y_0, x = transpose_198, y = transpose_197); + tensor var_1685_cast = softmax(axis = var_1620, x = qk_29_cast); + tensor var_1687_transpose_x_0 = const()[name = tensor("op_1687_transpose_x_0"), val = tensor(false)]; + tensor var_1687_transpose_y_0 = const()[name = tensor("op_1687_transpose_y_0"), val = tensor(false)]; + tensor transpose_199 = transpose(perm = var_1681, x = var_1680_cast); + tensor var_1687_cast = matmul(transpose_x = var_1687_transpose_x_0, transpose_y = var_1687_transpose_y_0, x = var_1685_cast, y = transpose_199); + tensor var_1688 = const()[name = tensor("op_1688"), val = tensor([0, 2, 1, 3])]; + tensor concat_14 = const()[name = tensor("concat_14"), val = tensor([1, 1500, 1280])]; + tensor transpose_196 = transpose(perm = var_1688, x = var_1687_cast); + tensor x_179_cast = reshape(shape = concat_14, x = transpose_196); + tensor var_1693_to_fp16 = const()[name = tensor("op_1693_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(575116672)))]; + tensor var_1694_to_fp16 = const()[name = tensor("op_1694_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578393536)))]; + tensor var_1695_cast = linear(bias = var_1694_to_fp16, weight = var_1693_to_fp16, x = x_179_cast); + tensor x_181_cast = add(x = x_175_cast, y = var_1695_cast); + tensor var_1701_axes_0 = const()[name = tensor("op_1701_axes_0"), val = tensor([-1])]; + tensor blocks_14_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_14_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578396160)))]; + tensor blocks_14_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_14_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578398784)))]; + tensor var_1701_cast = layer_norm(axes = var_1701_axes_0, beta = blocks_14_mlp_ln_bias_to_fp16, epsilon = var_1626_to_fp16, gamma = blocks_14_mlp_ln_weight_to_fp16, x = x_181_cast); + tensor var_1710_to_fp16 = const()[name = tensor("op_1710_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578401408)))]; + tensor var_1711_to_fp16 = const()[name = tensor("op_1711_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(591508672)))]; + tensor input_121_cast = linear(bias = var_1711_to_fp16, weight = var_1710_to_fp16, x = var_1701_cast); + tensor x_185_mode_0 = const()[name = tensor("x_185_mode_0"), val = tensor("EXACT")]; + tensor x_185_cast = gelu(mode = x_185_mode_0, x = input_121_cast); + tensor var_1716_to_fp16 = const()[name = tensor("op_1716_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(591518976)))]; + tensor var_1717_to_fp16 = const()[name = tensor("op_1717_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604626240)))]; + tensor var_1718_cast = linear(bias = var_1717_to_fp16, weight = var_1716_to_fp16, x = x_185_cast); + tensor x_187_cast = add(x = x_181_cast, y = var_1718_cast); + tensor var_1727 = const()[name = tensor("op_1727"), val = tensor(-1)]; + tensor var_1744_axes_0 = const()[name = tensor("op_1744_axes_0"), val = tensor([-1])]; + tensor blocks_15_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_15_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604628864)))]; + tensor blocks_15_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_15_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604631488)))]; + tensor var_1733_to_fp16 = const()[name = tensor("op_1733_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1744_cast = layer_norm(axes = var_1744_axes_0, beta = blocks_15_attn_ln_bias_to_fp16, epsilon = var_1733_to_fp16, gamma = blocks_15_attn_ln_weight_to_fp16, x = x_187_cast); + tensor var_1755_to_fp16 = const()[name = tensor("op_1755_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604634112)))]; + tensor var_1756_to_fp16 = const()[name = tensor("op_1756_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(607910976)))]; + tensor q_61_cast = linear(bias = var_1756_to_fp16, weight = var_1755_to_fp16, x = var_1744_cast); + tensor var_1759_to_fp16 = const()[name = tensor("op_1759_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(607913600)))]; + tensor k_61_bias_0_to_fp16 = const()[name = tensor("k_61_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(611190464)))]; + tensor k_61_cast = linear(bias = k_61_bias_0_to_fp16, weight = var_1759_to_fp16, x = var_1744_cast); + tensor var_1763_to_fp16 = const()[name = tensor("op_1763_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(611193088)))]; + tensor var_1764_to_fp16 = const()[name = tensor("op_1764_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614469952)))]; + tensor v_61_cast = linear(bias = var_1764_to_fp16, weight = var_1763_to_fp16, x = var_1744_cast); + tensor var_1772 = const()[name = tensor("op_1772"), val = tensor([1, 1500, 20, -1])]; + tensor var_1773_cast = reshape(shape = var_1772, x = q_61_cast); + tensor const_254_to_fp16 = const()[name = tensor("const_254_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_63_cast = mul(x = var_1773_cast, y = const_254_to_fp16); + tensor var_1779 = const()[name = tensor("op_1779"), val = tensor([1, 1500, 20, -1])]; + tensor var_1780_cast = reshape(shape = var_1779, x = k_61_cast); + tensor const_255_to_fp16 = const()[name = tensor("const_255_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_63_cast = mul(x = var_1780_cast, y = const_255_to_fp16); + tensor var_1786 = const()[name = tensor("op_1786"), val = tensor([1, 1500, 20, -1])]; + tensor var_1787_cast = reshape(shape = var_1786, x = v_61_cast); + tensor var_1788 = const()[name = tensor("op_1788"), val = tensor([0, 2, 1, 3])]; + tensor qk_31_transpose_x_0 = const()[name = tensor("qk_31_transpose_x_0"), val = tensor(false)]; + tensor qk_31_transpose_y_0 = const()[name = tensor("qk_31_transpose_y_0"), val = tensor(false)]; + tensor transpose_94_perm_0 = const()[name = tensor("transpose_94_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_95_perm_0 = const()[name = tensor("transpose_95_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_193 = transpose(perm = transpose_95_perm_0, x = k_63_cast); + tensor transpose_194 = transpose(perm = transpose_94_perm_0, x = q_63_cast); + tensor qk_31_cast = matmul(transpose_x = qk_31_transpose_x_0, transpose_y = qk_31_transpose_y_0, x = transpose_194, y = transpose_193); + tensor var_1792_cast = softmax(axis = var_1727, x = qk_31_cast); + tensor var_1794_transpose_x_0 = const()[name = tensor("op_1794_transpose_x_0"), val = tensor(false)]; + tensor var_1794_transpose_y_0 = const()[name = tensor("op_1794_transpose_y_0"), val = tensor(false)]; + tensor transpose_195 = transpose(perm = var_1788, x = var_1787_cast); + tensor var_1794_cast = matmul(transpose_x = var_1794_transpose_x_0, transpose_y = var_1794_transpose_y_0, x = var_1792_cast, y = transpose_195); + tensor var_1795 = const()[name = tensor("op_1795"), val = tensor([0, 2, 1, 3])]; + tensor concat_15 = const()[name = tensor("concat_15"), val = tensor([1, 1500, 1280])]; + tensor transpose_192 = transpose(perm = var_1795, x = var_1794_cast); + tensor x_191_cast = reshape(shape = concat_15, x = transpose_192); + tensor var_1800_to_fp16 = const()[name = tensor("op_1800_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614472576)))]; + tensor var_1801_to_fp16 = const()[name = tensor("op_1801_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617749440)))]; + tensor var_1802_cast = linear(bias = var_1801_to_fp16, weight = var_1800_to_fp16, x = x_191_cast); + tensor x_193_cast = add(x = x_187_cast, y = var_1802_cast); + tensor var_1808_axes_0 = const()[name = tensor("op_1808_axes_0"), val = tensor([-1])]; + tensor blocks_15_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_15_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617752064)))]; + tensor blocks_15_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_15_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617754688)))]; + tensor var_1808_cast = layer_norm(axes = var_1808_axes_0, beta = blocks_15_mlp_ln_bias_to_fp16, epsilon = var_1733_to_fp16, gamma = blocks_15_mlp_ln_weight_to_fp16, x = x_193_cast); + tensor var_1817_to_fp16 = const()[name = tensor("op_1817_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617757312)))]; + tensor var_1818_to_fp16 = const()[name = tensor("op_1818_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(630864576)))]; + tensor input_129_cast = linear(bias = var_1818_to_fp16, weight = var_1817_to_fp16, x = var_1808_cast); + tensor x_197_mode_0 = const()[name = tensor("x_197_mode_0"), val = tensor("EXACT")]; + tensor x_197_cast = gelu(mode = x_197_mode_0, x = input_129_cast); + tensor var_1823_to_fp16 = const()[name = tensor("op_1823_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(630874880)))]; + tensor var_1824_to_fp16 = const()[name = tensor("op_1824_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643982144)))]; + tensor var_1825_cast = linear(bias = var_1824_to_fp16, weight = var_1823_to_fp16, x = x_197_cast); + tensor x_199_cast = add(x = x_193_cast, y = var_1825_cast); + tensor var_1834 = const()[name = tensor("op_1834"), val = tensor(-1)]; + tensor var_1851_axes_0 = const()[name = tensor("op_1851_axes_0"), val = tensor([-1])]; + tensor blocks_16_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_16_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643984768)))]; + tensor blocks_16_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_16_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643987392)))]; + tensor var_1840_to_fp16 = const()[name = tensor("op_1840_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1851_cast = layer_norm(axes = var_1851_axes_0, beta = blocks_16_attn_ln_bias_to_fp16, epsilon = var_1840_to_fp16, gamma = blocks_16_attn_ln_weight_to_fp16, x = x_199_cast); + tensor var_1862_to_fp16 = const()[name = tensor("op_1862_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643990016)))]; + tensor var_1863_to_fp16 = const()[name = tensor("op_1863_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647266880)))]; + tensor q_65_cast = linear(bias = var_1863_to_fp16, weight = var_1862_to_fp16, x = var_1851_cast); + tensor var_1866_to_fp16 = const()[name = tensor("op_1866_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647269504)))]; + tensor k_65_bias_0_to_fp16 = const()[name = tensor("k_65_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(650546368)))]; + tensor k_65_cast = linear(bias = k_65_bias_0_to_fp16, weight = var_1866_to_fp16, x = var_1851_cast); + tensor var_1870_to_fp16 = const()[name = tensor("op_1870_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(650548992)))]; + tensor var_1871_to_fp16 = const()[name = tensor("op_1871_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(653825856)))]; + tensor v_65_cast = linear(bias = var_1871_to_fp16, weight = var_1870_to_fp16, x = var_1851_cast); + tensor var_1879 = const()[name = tensor("op_1879"), val = tensor([1, 1500, 20, -1])]; + tensor var_1880_cast = reshape(shape = var_1879, x = q_65_cast); + tensor const_256_to_fp16 = const()[name = tensor("const_256_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_67_cast = mul(x = var_1880_cast, y = const_256_to_fp16); + tensor var_1886 = const()[name = tensor("op_1886"), val = tensor([1, 1500, 20, -1])]; + tensor var_1887_cast = reshape(shape = var_1886, x = k_65_cast); + tensor const_257_to_fp16 = const()[name = tensor("const_257_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_67_cast = mul(x = var_1887_cast, y = const_257_to_fp16); + tensor var_1893 = const()[name = tensor("op_1893"), val = tensor([1, 1500, 20, -1])]; + tensor var_1894_cast = reshape(shape = var_1893, x = v_65_cast); + tensor var_1895 = const()[name = tensor("op_1895"), val = tensor([0, 2, 1, 3])]; + tensor qk_33_transpose_x_0 = const()[name = tensor("qk_33_transpose_x_0"), val = tensor(false)]; + tensor qk_33_transpose_y_0 = const()[name = tensor("qk_33_transpose_y_0"), val = tensor(false)]; + tensor transpose_96_perm_0 = const()[name = tensor("transpose_96_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_97_perm_0 = const()[name = tensor("transpose_97_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_189 = transpose(perm = transpose_97_perm_0, x = k_67_cast); + tensor transpose_190 = transpose(perm = transpose_96_perm_0, x = q_67_cast); + tensor qk_33_cast = matmul(transpose_x = qk_33_transpose_x_0, transpose_y = qk_33_transpose_y_0, x = transpose_190, y = transpose_189); + tensor var_1899_cast = softmax(axis = var_1834, x = qk_33_cast); + tensor var_1901_transpose_x_0 = const()[name = tensor("op_1901_transpose_x_0"), val = tensor(false)]; + tensor var_1901_transpose_y_0 = const()[name = tensor("op_1901_transpose_y_0"), val = tensor(false)]; + tensor transpose_191 = transpose(perm = var_1895, x = var_1894_cast); + tensor var_1901_cast = matmul(transpose_x = var_1901_transpose_x_0, transpose_y = var_1901_transpose_y_0, x = var_1899_cast, y = transpose_191); + tensor var_1902 = const()[name = tensor("op_1902"), val = tensor([0, 2, 1, 3])]; + tensor concat_16 = const()[name = tensor("concat_16"), val = tensor([1, 1500, 1280])]; + tensor transpose_188 = transpose(perm = var_1902, x = var_1901_cast); + tensor x_203_cast = reshape(shape = concat_16, x = transpose_188); + tensor var_1907_to_fp16 = const()[name = tensor("op_1907_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(653828480)))]; + tensor var_1908_to_fp16 = const()[name = tensor("op_1908_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657105344)))]; + tensor var_1909_cast = linear(bias = var_1908_to_fp16, weight = var_1907_to_fp16, x = x_203_cast); + tensor x_205_cast = add(x = x_199_cast, y = var_1909_cast); + tensor var_1915_axes_0 = const()[name = tensor("op_1915_axes_0"), val = tensor([-1])]; + tensor blocks_16_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_16_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657107968)))]; + tensor blocks_16_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_16_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657110592)))]; + tensor var_1915_cast = layer_norm(axes = var_1915_axes_0, beta = blocks_16_mlp_ln_bias_to_fp16, epsilon = var_1840_to_fp16, gamma = blocks_16_mlp_ln_weight_to_fp16, x = x_205_cast); + tensor var_1924_to_fp16 = const()[name = tensor("op_1924_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657113216)))]; + tensor var_1925_to_fp16 = const()[name = tensor("op_1925_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(670220480)))]; + tensor input_137_cast = linear(bias = var_1925_to_fp16, weight = var_1924_to_fp16, x = var_1915_cast); + tensor x_209_mode_0 = const()[name = tensor("x_209_mode_0"), val = tensor("EXACT")]; + tensor x_209_cast = gelu(mode = x_209_mode_0, x = input_137_cast); + tensor var_1930_to_fp16 = const()[name = tensor("op_1930_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(670230784)))]; + tensor var_1931_to_fp16 = const()[name = tensor("op_1931_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(683338048)))]; + tensor var_1932_cast = linear(bias = var_1931_to_fp16, weight = var_1930_to_fp16, x = x_209_cast); + tensor x_211_cast = add(x = x_205_cast, y = var_1932_cast); + tensor var_1941 = const()[name = tensor("op_1941"), val = tensor(-1)]; + tensor var_1958_axes_0 = const()[name = tensor("op_1958_axes_0"), val = tensor([-1])]; + tensor blocks_17_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_17_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(683340672)))]; + tensor blocks_17_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_17_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(683343296)))]; + tensor var_1947_to_fp16 = const()[name = tensor("op_1947_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1958_cast = layer_norm(axes = var_1958_axes_0, beta = blocks_17_attn_ln_bias_to_fp16, epsilon = var_1947_to_fp16, gamma = blocks_17_attn_ln_weight_to_fp16, x = x_211_cast); + tensor var_1969_to_fp16 = const()[name = tensor("op_1969_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(683345920)))]; + tensor var_1970_to_fp16 = const()[name = tensor("op_1970_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(686622784)))]; + tensor q_69_cast = linear(bias = var_1970_to_fp16, weight = var_1969_to_fp16, x = var_1958_cast); + tensor var_1973_to_fp16 = const()[name = tensor("op_1973_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(686625408)))]; + tensor k_69_bias_0_to_fp16 = const()[name = tensor("k_69_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(689902272)))]; + tensor k_69_cast = linear(bias = k_69_bias_0_to_fp16, weight = var_1973_to_fp16, x = var_1958_cast); + tensor var_1977_to_fp16 = const()[name = tensor("op_1977_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(689904896)))]; + tensor var_1978_to_fp16 = const()[name = tensor("op_1978_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(693181760)))]; + tensor v_69_cast = linear(bias = var_1978_to_fp16, weight = var_1977_to_fp16, x = var_1958_cast); + tensor var_1986 = const()[name = tensor("op_1986"), val = tensor([1, 1500, 20, -1])]; + tensor var_1987_cast = reshape(shape = var_1986, x = q_69_cast); + tensor const_258_to_fp16 = const()[name = tensor("const_258_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_71_cast = mul(x = var_1987_cast, y = const_258_to_fp16); + tensor var_1993 = const()[name = tensor("op_1993"), val = tensor([1, 1500, 20, -1])]; + tensor var_1994_cast = reshape(shape = var_1993, x = k_69_cast); + tensor const_259_to_fp16 = const()[name = tensor("const_259_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_71_cast = mul(x = var_1994_cast, y = const_259_to_fp16); + tensor var_2000 = const()[name = tensor("op_2000"), val = tensor([1, 1500, 20, -1])]; + tensor var_2001_cast = reshape(shape = var_2000, x = v_69_cast); + tensor var_2002 = const()[name = tensor("op_2002"), val = tensor([0, 2, 1, 3])]; + tensor qk_35_transpose_x_0 = const()[name = tensor("qk_35_transpose_x_0"), val = tensor(false)]; + tensor qk_35_transpose_y_0 = const()[name = tensor("qk_35_transpose_y_0"), val = tensor(false)]; + tensor transpose_98_perm_0 = const()[name = tensor("transpose_98_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_99_perm_0 = const()[name = tensor("transpose_99_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_185 = transpose(perm = transpose_99_perm_0, x = k_71_cast); + tensor transpose_186 = transpose(perm = transpose_98_perm_0, x = q_71_cast); + tensor qk_35_cast = matmul(transpose_x = qk_35_transpose_x_0, transpose_y = qk_35_transpose_y_0, x = transpose_186, y = transpose_185); + tensor var_2006_cast = softmax(axis = var_1941, x = qk_35_cast); + tensor var_2008_transpose_x_0 = const()[name = tensor("op_2008_transpose_x_0"), val = tensor(false)]; + tensor var_2008_transpose_y_0 = const()[name = tensor("op_2008_transpose_y_0"), val = tensor(false)]; + tensor transpose_187 = transpose(perm = var_2002, x = var_2001_cast); + tensor var_2008_cast = matmul(transpose_x = var_2008_transpose_x_0, transpose_y = var_2008_transpose_y_0, x = var_2006_cast, y = transpose_187); + tensor var_2009 = const()[name = tensor("op_2009"), val = tensor([0, 2, 1, 3])]; + tensor concat_17 = const()[name = tensor("concat_17"), val = tensor([1, 1500, 1280])]; + tensor transpose_184 = transpose(perm = var_2009, x = var_2008_cast); + tensor x_215_cast = reshape(shape = concat_17, x = transpose_184); + tensor var_2014_to_fp16 = const()[name = tensor("op_2014_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(693184384)))]; + tensor var_2015_to_fp16 = const()[name = tensor("op_2015_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696461248)))]; + tensor var_2016_cast = linear(bias = var_2015_to_fp16, weight = var_2014_to_fp16, x = x_215_cast); + tensor x_217_cast = add(x = x_211_cast, y = var_2016_cast); + tensor var_2022_axes_0 = const()[name = tensor("op_2022_axes_0"), val = tensor([-1])]; + tensor blocks_17_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_17_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696463872)))]; + tensor blocks_17_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_17_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696466496)))]; + tensor var_2022_cast = layer_norm(axes = var_2022_axes_0, beta = blocks_17_mlp_ln_bias_to_fp16, epsilon = var_1947_to_fp16, gamma = blocks_17_mlp_ln_weight_to_fp16, x = x_217_cast); + tensor var_2031_to_fp16 = const()[name = tensor("op_2031_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696469120)))]; + tensor var_2032_to_fp16 = const()[name = tensor("op_2032_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(709576384)))]; + tensor input_145_cast = linear(bias = var_2032_to_fp16, weight = var_2031_to_fp16, x = var_2022_cast); + tensor x_221_mode_0 = const()[name = tensor("x_221_mode_0"), val = tensor("EXACT")]; + tensor x_221_cast = gelu(mode = x_221_mode_0, x = input_145_cast); + tensor var_2037_to_fp16 = const()[name = tensor("op_2037_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(709586688)))]; + tensor var_2038_to_fp16 = const()[name = tensor("op_2038_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(722693952)))]; + tensor var_2039_cast = linear(bias = var_2038_to_fp16, weight = var_2037_to_fp16, x = x_221_cast); + tensor x_223_cast = add(x = x_217_cast, y = var_2039_cast); + tensor var_2048 = const()[name = tensor("op_2048"), val = tensor(-1)]; + tensor var_2065_axes_0 = const()[name = tensor("op_2065_axes_0"), val = tensor([-1])]; + tensor blocks_18_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_18_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(722696576)))]; + tensor blocks_18_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_18_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(722699200)))]; + tensor var_2054_to_fp16 = const()[name = tensor("op_2054_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2065_cast = layer_norm(axes = var_2065_axes_0, beta = blocks_18_attn_ln_bias_to_fp16, epsilon = var_2054_to_fp16, gamma = blocks_18_attn_ln_weight_to_fp16, x = x_223_cast); + tensor var_2076_to_fp16 = const()[name = tensor("op_2076_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(722701824)))]; + tensor var_2077_to_fp16 = const()[name = tensor("op_2077_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(725978688)))]; + tensor q_73_cast = linear(bias = var_2077_to_fp16, weight = var_2076_to_fp16, x = var_2065_cast); + tensor var_2080_to_fp16 = const()[name = tensor("op_2080_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(725981312)))]; + tensor k_73_bias_0_to_fp16 = const()[name = tensor("k_73_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(729258176)))]; + tensor k_73_cast = linear(bias = k_73_bias_0_to_fp16, weight = var_2080_to_fp16, x = var_2065_cast); + tensor var_2084_to_fp16 = const()[name = tensor("op_2084_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(729260800)))]; + tensor var_2085_to_fp16 = const()[name = tensor("op_2085_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(732537664)))]; + tensor v_73_cast = linear(bias = var_2085_to_fp16, weight = var_2084_to_fp16, x = var_2065_cast); + tensor var_2093 = const()[name = tensor("op_2093"), val = tensor([1, 1500, 20, -1])]; + tensor var_2094_cast = reshape(shape = var_2093, x = q_73_cast); + tensor const_260_to_fp16 = const()[name = tensor("const_260_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_75_cast = mul(x = var_2094_cast, y = const_260_to_fp16); + tensor var_2100 = const()[name = tensor("op_2100"), val = tensor([1, 1500, 20, -1])]; + tensor var_2101_cast = reshape(shape = var_2100, x = k_73_cast); + tensor const_261_to_fp16 = const()[name = tensor("const_261_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_75_cast = mul(x = var_2101_cast, y = const_261_to_fp16); + tensor var_2107 = const()[name = tensor("op_2107"), val = tensor([1, 1500, 20, -1])]; + tensor var_2108_cast = reshape(shape = var_2107, x = v_73_cast); + tensor var_2109 = const()[name = tensor("op_2109"), val = tensor([0, 2, 1, 3])]; + tensor qk_37_transpose_x_0 = const()[name = tensor("qk_37_transpose_x_0"), val = tensor(false)]; + tensor qk_37_transpose_y_0 = const()[name = tensor("qk_37_transpose_y_0"), val = tensor(false)]; + tensor transpose_100_perm_0 = const()[name = tensor("transpose_100_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_101_perm_0 = const()[name = tensor("transpose_101_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_181 = transpose(perm = transpose_101_perm_0, x = k_75_cast); + tensor transpose_182 = transpose(perm = transpose_100_perm_0, x = q_75_cast); + tensor qk_37_cast = matmul(transpose_x = qk_37_transpose_x_0, transpose_y = qk_37_transpose_y_0, x = transpose_182, y = transpose_181); + tensor var_2113_cast = softmax(axis = var_2048, x = qk_37_cast); + tensor var_2115_transpose_x_0 = const()[name = tensor("op_2115_transpose_x_0"), val = tensor(false)]; + tensor var_2115_transpose_y_0 = const()[name = tensor("op_2115_transpose_y_0"), val = tensor(false)]; + tensor transpose_183 = transpose(perm = var_2109, x = var_2108_cast); + tensor var_2115_cast = matmul(transpose_x = var_2115_transpose_x_0, transpose_y = var_2115_transpose_y_0, x = var_2113_cast, y = transpose_183); + tensor var_2116 = const()[name = tensor("op_2116"), val = tensor([0, 2, 1, 3])]; + tensor concat_18 = const()[name = tensor("concat_18"), val = tensor([1, 1500, 1280])]; + tensor transpose_180 = transpose(perm = var_2116, x = var_2115_cast); + tensor x_227_cast = reshape(shape = concat_18, x = transpose_180); + tensor var_2121_to_fp16 = const()[name = tensor("op_2121_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(732540288)))]; + tensor var_2122_to_fp16 = const()[name = tensor("op_2122_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(735817152)))]; + tensor var_2123_cast = linear(bias = var_2122_to_fp16, weight = var_2121_to_fp16, x = x_227_cast); + tensor x_229_cast = add(x = x_223_cast, y = var_2123_cast); + tensor var_2129_axes_0 = const()[name = tensor("op_2129_axes_0"), val = tensor([-1])]; + tensor blocks_18_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_18_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(735819776)))]; + tensor blocks_18_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_18_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(735822400)))]; + tensor var_2129_cast = layer_norm(axes = var_2129_axes_0, beta = blocks_18_mlp_ln_bias_to_fp16, epsilon = var_2054_to_fp16, gamma = blocks_18_mlp_ln_weight_to_fp16, x = x_229_cast); + tensor var_2138_to_fp16 = const()[name = tensor("op_2138_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(735825024)))]; + tensor var_2139_to_fp16 = const()[name = tensor("op_2139_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(748932288)))]; + tensor input_153_cast = linear(bias = var_2139_to_fp16, weight = var_2138_to_fp16, x = var_2129_cast); + tensor x_233_mode_0 = const()[name = tensor("x_233_mode_0"), val = tensor("EXACT")]; + tensor x_233_cast = gelu(mode = x_233_mode_0, x = input_153_cast); + tensor var_2144_to_fp16 = const()[name = tensor("op_2144_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(748942592)))]; + tensor var_2145_to_fp16 = const()[name = tensor("op_2145_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762049856)))]; + tensor var_2146_cast = linear(bias = var_2145_to_fp16, weight = var_2144_to_fp16, x = x_233_cast); + tensor x_235_cast = add(x = x_229_cast, y = var_2146_cast); + tensor var_2155 = const()[name = tensor("op_2155"), val = tensor(-1)]; + tensor var_2172_axes_0 = const()[name = tensor("op_2172_axes_0"), val = tensor([-1])]; + tensor blocks_19_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_19_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762052480)))]; + tensor blocks_19_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_19_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762055104)))]; + tensor var_2161_to_fp16 = const()[name = tensor("op_2161_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2172_cast = layer_norm(axes = var_2172_axes_0, beta = blocks_19_attn_ln_bias_to_fp16, epsilon = var_2161_to_fp16, gamma = blocks_19_attn_ln_weight_to_fp16, x = x_235_cast); + tensor var_2183_to_fp16 = const()[name = tensor("op_2183_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762057728)))]; + tensor var_2184_to_fp16 = const()[name = tensor("op_2184_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(765334592)))]; + tensor q_77_cast = linear(bias = var_2184_to_fp16, weight = var_2183_to_fp16, x = var_2172_cast); + tensor var_2187_to_fp16 = const()[name = tensor("op_2187_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(765337216)))]; + tensor k_77_bias_0_to_fp16 = const()[name = tensor("k_77_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(768614080)))]; + tensor k_77_cast = linear(bias = k_77_bias_0_to_fp16, weight = var_2187_to_fp16, x = var_2172_cast); + tensor var_2191_to_fp16 = const()[name = tensor("op_2191_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(768616704)))]; + tensor var_2192_to_fp16 = const()[name = tensor("op_2192_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(771893568)))]; + tensor v_77_cast = linear(bias = var_2192_to_fp16, weight = var_2191_to_fp16, x = var_2172_cast); + tensor var_2200 = const()[name = tensor("op_2200"), val = tensor([1, 1500, 20, -1])]; + tensor var_2201_cast = reshape(shape = var_2200, x = q_77_cast); + tensor const_262_to_fp16 = const()[name = tensor("const_262_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_79_cast = mul(x = var_2201_cast, y = const_262_to_fp16); + tensor var_2207 = const()[name = tensor("op_2207"), val = tensor([1, 1500, 20, -1])]; + tensor var_2208_cast = reshape(shape = var_2207, x = k_77_cast); + tensor const_263_to_fp16 = const()[name = tensor("const_263_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_79_cast = mul(x = var_2208_cast, y = const_263_to_fp16); + tensor var_2214 = const()[name = tensor("op_2214"), val = tensor([1, 1500, 20, -1])]; + tensor var_2215_cast = reshape(shape = var_2214, x = v_77_cast); + tensor var_2216 = const()[name = tensor("op_2216"), val = tensor([0, 2, 1, 3])]; + tensor qk_39_transpose_x_0 = const()[name = tensor("qk_39_transpose_x_0"), val = tensor(false)]; + tensor qk_39_transpose_y_0 = const()[name = tensor("qk_39_transpose_y_0"), val = tensor(false)]; + tensor transpose_102_perm_0 = const()[name = tensor("transpose_102_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_103_perm_0 = const()[name = tensor("transpose_103_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_177 = transpose(perm = transpose_103_perm_0, x = k_79_cast); + tensor transpose_178 = transpose(perm = transpose_102_perm_0, x = q_79_cast); + tensor qk_39_cast = matmul(transpose_x = qk_39_transpose_x_0, transpose_y = qk_39_transpose_y_0, x = transpose_178, y = transpose_177); + tensor var_2220_cast = softmax(axis = var_2155, x = qk_39_cast); + tensor var_2222_transpose_x_0 = const()[name = tensor("op_2222_transpose_x_0"), val = tensor(false)]; + tensor var_2222_transpose_y_0 = const()[name = tensor("op_2222_transpose_y_0"), val = tensor(false)]; + tensor transpose_179 = transpose(perm = var_2216, x = var_2215_cast); + tensor var_2222_cast = matmul(transpose_x = var_2222_transpose_x_0, transpose_y = var_2222_transpose_y_0, x = var_2220_cast, y = transpose_179); + tensor var_2223 = const()[name = tensor("op_2223"), val = tensor([0, 2, 1, 3])]; + tensor concat_19 = const()[name = tensor("concat_19"), val = tensor([1, 1500, 1280])]; + tensor transpose_176 = transpose(perm = var_2223, x = var_2222_cast); + tensor x_239_cast = reshape(shape = concat_19, x = transpose_176); + tensor var_2228_to_fp16 = const()[name = tensor("op_2228_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(771896192)))]; + tensor var_2229_to_fp16 = const()[name = tensor("op_2229_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775173056)))]; + tensor var_2230_cast = linear(bias = var_2229_to_fp16, weight = var_2228_to_fp16, x = x_239_cast); + tensor x_241_cast = add(x = x_235_cast, y = var_2230_cast); + tensor var_2236_axes_0 = const()[name = tensor("op_2236_axes_0"), val = tensor([-1])]; + tensor blocks_19_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_19_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775175680)))]; + tensor blocks_19_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_19_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775178304)))]; + tensor var_2236_cast = layer_norm(axes = var_2236_axes_0, beta = blocks_19_mlp_ln_bias_to_fp16, epsilon = var_2161_to_fp16, gamma = blocks_19_mlp_ln_weight_to_fp16, x = x_241_cast); + tensor var_2245_to_fp16 = const()[name = tensor("op_2245_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775180928)))]; + tensor var_2246_to_fp16 = const()[name = tensor("op_2246_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(788288192)))]; + tensor input_161_cast = linear(bias = var_2246_to_fp16, weight = var_2245_to_fp16, x = var_2236_cast); + tensor x_245_mode_0 = const()[name = tensor("x_245_mode_0"), val = tensor("EXACT")]; + tensor x_245_cast = gelu(mode = x_245_mode_0, x = input_161_cast); + tensor var_2251_to_fp16 = const()[name = tensor("op_2251_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(788298496)))]; + tensor var_2252_to_fp16 = const()[name = tensor("op_2252_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(801405760)))]; + tensor var_2253_cast = linear(bias = var_2252_to_fp16, weight = var_2251_to_fp16, x = x_245_cast); + tensor x_247_cast = add(x = x_241_cast, y = var_2253_cast); + tensor var_2262 = const()[name = tensor("op_2262"), val = tensor(-1)]; + tensor var_2279_axes_0 = const()[name = tensor("op_2279_axes_0"), val = tensor([-1])]; + tensor blocks_20_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_20_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(801408384)))]; + tensor blocks_20_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_20_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(801411008)))]; + tensor var_2268_to_fp16 = const()[name = tensor("op_2268_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2279_cast = layer_norm(axes = var_2279_axes_0, beta = blocks_20_attn_ln_bias_to_fp16, epsilon = var_2268_to_fp16, gamma = blocks_20_attn_ln_weight_to_fp16, x = x_247_cast); + tensor var_2290_to_fp16 = const()[name = tensor("op_2290_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(801413632)))]; + tensor var_2291_to_fp16 = const()[name = tensor("op_2291_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(804690496)))]; + tensor q_81_cast = linear(bias = var_2291_to_fp16, weight = var_2290_to_fp16, x = var_2279_cast); + tensor var_2294_to_fp16 = const()[name = tensor("op_2294_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(804693120)))]; + tensor k_81_bias_0_to_fp16 = const()[name = tensor("k_81_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(807969984)))]; + tensor k_81_cast = linear(bias = k_81_bias_0_to_fp16, weight = var_2294_to_fp16, x = var_2279_cast); + tensor var_2298_to_fp16 = const()[name = tensor("op_2298_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(807972608)))]; + tensor var_2299_to_fp16 = const()[name = tensor("op_2299_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(811249472)))]; + tensor v_81_cast = linear(bias = var_2299_to_fp16, weight = var_2298_to_fp16, x = var_2279_cast); + tensor var_2307 = const()[name = tensor("op_2307"), val = tensor([1, 1500, 20, -1])]; + tensor var_2308_cast = reshape(shape = var_2307, x = q_81_cast); + tensor const_264_to_fp16 = const()[name = tensor("const_264_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_83_cast = mul(x = var_2308_cast, y = const_264_to_fp16); + tensor var_2314 = const()[name = tensor("op_2314"), val = tensor([1, 1500, 20, -1])]; + tensor var_2315_cast = reshape(shape = var_2314, x = k_81_cast); + tensor const_265_to_fp16 = const()[name = tensor("const_265_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_83_cast = mul(x = var_2315_cast, y = const_265_to_fp16); + tensor var_2321 = const()[name = tensor("op_2321"), val = tensor([1, 1500, 20, -1])]; + tensor var_2322_cast = reshape(shape = var_2321, x = v_81_cast); + tensor var_2323 = const()[name = tensor("op_2323"), val = tensor([0, 2, 1, 3])]; + tensor qk_41_transpose_x_0 = const()[name = tensor("qk_41_transpose_x_0"), val = tensor(false)]; + tensor qk_41_transpose_y_0 = const()[name = tensor("qk_41_transpose_y_0"), val = tensor(false)]; + tensor transpose_104_perm_0 = const()[name = tensor("transpose_104_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_105_perm_0 = const()[name = tensor("transpose_105_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_173 = transpose(perm = transpose_105_perm_0, x = k_83_cast); + tensor transpose_174 = transpose(perm = transpose_104_perm_0, x = q_83_cast); + tensor qk_41_cast = matmul(transpose_x = qk_41_transpose_x_0, transpose_y = qk_41_transpose_y_0, x = transpose_174, y = transpose_173); + tensor var_2327_cast = softmax(axis = var_2262, x = qk_41_cast); + tensor var_2329_transpose_x_0 = const()[name = tensor("op_2329_transpose_x_0"), val = tensor(false)]; + tensor var_2329_transpose_y_0 = const()[name = tensor("op_2329_transpose_y_0"), val = tensor(false)]; + tensor transpose_175 = transpose(perm = var_2323, x = var_2322_cast); + tensor var_2329_cast = matmul(transpose_x = var_2329_transpose_x_0, transpose_y = var_2329_transpose_y_0, x = var_2327_cast, y = transpose_175); + tensor var_2330 = const()[name = tensor("op_2330"), val = tensor([0, 2, 1, 3])]; + tensor concat_20 = const()[name = tensor("concat_20"), val = tensor([1, 1500, 1280])]; + tensor transpose_172 = transpose(perm = var_2330, x = var_2329_cast); + tensor x_251_cast = reshape(shape = concat_20, x = transpose_172); + tensor var_2335_to_fp16 = const()[name = tensor("op_2335_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(811252096)))]; + tensor var_2336_to_fp16 = const()[name = tensor("op_2336_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814528960)))]; + tensor var_2337_cast = linear(bias = var_2336_to_fp16, weight = var_2335_to_fp16, x = x_251_cast); + tensor x_253_cast = add(x = x_247_cast, y = var_2337_cast); + tensor var_2343_axes_0 = const()[name = tensor("op_2343_axes_0"), val = tensor([-1])]; + tensor blocks_20_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_20_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814531584)))]; + tensor blocks_20_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_20_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814534208)))]; + tensor var_2343_cast = layer_norm(axes = var_2343_axes_0, beta = blocks_20_mlp_ln_bias_to_fp16, epsilon = var_2268_to_fp16, gamma = blocks_20_mlp_ln_weight_to_fp16, x = x_253_cast); + tensor var_2352_to_fp16 = const()[name = tensor("op_2352_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814536832)))]; + tensor var_2353_to_fp16 = const()[name = tensor("op_2353_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(827644096)))]; + tensor input_169_cast = linear(bias = var_2353_to_fp16, weight = var_2352_to_fp16, x = var_2343_cast); + tensor x_257_mode_0 = const()[name = tensor("x_257_mode_0"), val = tensor("EXACT")]; + tensor x_257_cast = gelu(mode = x_257_mode_0, x = input_169_cast); + tensor var_2358_to_fp16 = const()[name = tensor("op_2358_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(827654400)))]; + tensor var_2359_to_fp16 = const()[name = tensor("op_2359_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(840761664)))]; + tensor var_2360_cast = linear(bias = var_2359_to_fp16, weight = var_2358_to_fp16, x = x_257_cast); + tensor x_259_cast = add(x = x_253_cast, y = var_2360_cast); + tensor var_2369 = const()[name = tensor("op_2369"), val = tensor(-1)]; + tensor var_2386_axes_0 = const()[name = tensor("op_2386_axes_0"), val = tensor([-1])]; + tensor blocks_21_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_21_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(840764288)))]; + tensor blocks_21_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_21_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(840766912)))]; + tensor var_2375_to_fp16 = const()[name = tensor("op_2375_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2386_cast = layer_norm(axes = var_2386_axes_0, beta = blocks_21_attn_ln_bias_to_fp16, epsilon = var_2375_to_fp16, gamma = blocks_21_attn_ln_weight_to_fp16, x = x_259_cast); + tensor var_2397_to_fp16 = const()[name = tensor("op_2397_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(840769536)))]; + tensor var_2398_to_fp16 = const()[name = tensor("op_2398_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(844046400)))]; + tensor q_85_cast = linear(bias = var_2398_to_fp16, weight = var_2397_to_fp16, x = var_2386_cast); + tensor var_2401_to_fp16 = const()[name = tensor("op_2401_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(844049024)))]; + tensor k_85_bias_0_to_fp16 = const()[name = tensor("k_85_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(847325888)))]; + tensor k_85_cast = linear(bias = k_85_bias_0_to_fp16, weight = var_2401_to_fp16, x = var_2386_cast); + tensor var_2405_to_fp16 = const()[name = tensor("op_2405_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(847328512)))]; + tensor var_2406_to_fp16 = const()[name = tensor("op_2406_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(850605376)))]; + tensor v_85_cast = linear(bias = var_2406_to_fp16, weight = var_2405_to_fp16, x = var_2386_cast); + tensor var_2414 = const()[name = tensor("op_2414"), val = tensor([1, 1500, 20, -1])]; + tensor var_2415_cast = reshape(shape = var_2414, x = q_85_cast); + tensor const_266_to_fp16 = const()[name = tensor("const_266_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_87_cast = mul(x = var_2415_cast, y = const_266_to_fp16); + tensor var_2421 = const()[name = tensor("op_2421"), val = tensor([1, 1500, 20, -1])]; + tensor var_2422_cast = reshape(shape = var_2421, x = k_85_cast); + tensor const_267_to_fp16 = const()[name = tensor("const_267_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_87_cast = mul(x = var_2422_cast, y = const_267_to_fp16); + tensor var_2428 = const()[name = tensor("op_2428"), val = tensor([1, 1500, 20, -1])]; + tensor var_2429_cast = reshape(shape = var_2428, x = v_85_cast); + tensor var_2430 = const()[name = tensor("op_2430"), val = tensor([0, 2, 1, 3])]; + tensor qk_43_transpose_x_0 = const()[name = tensor("qk_43_transpose_x_0"), val = tensor(false)]; + tensor qk_43_transpose_y_0 = const()[name = tensor("qk_43_transpose_y_0"), val = tensor(false)]; + tensor transpose_106_perm_0 = const()[name = tensor("transpose_106_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_107_perm_0 = const()[name = tensor("transpose_107_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_169 = transpose(perm = transpose_107_perm_0, x = k_87_cast); + tensor transpose_170 = transpose(perm = transpose_106_perm_0, x = q_87_cast); + tensor qk_43_cast = matmul(transpose_x = qk_43_transpose_x_0, transpose_y = qk_43_transpose_y_0, x = transpose_170, y = transpose_169); + tensor var_2434_cast = softmax(axis = var_2369, x = qk_43_cast); + tensor var_2436_transpose_x_0 = const()[name = tensor("op_2436_transpose_x_0"), val = tensor(false)]; + tensor var_2436_transpose_y_0 = const()[name = tensor("op_2436_transpose_y_0"), val = tensor(false)]; + tensor transpose_171 = transpose(perm = var_2430, x = var_2429_cast); + tensor var_2436_cast = matmul(transpose_x = var_2436_transpose_x_0, transpose_y = var_2436_transpose_y_0, x = var_2434_cast, y = transpose_171); + tensor var_2437 = const()[name = tensor("op_2437"), val = tensor([0, 2, 1, 3])]; + tensor concat_21 = const()[name = tensor("concat_21"), val = tensor([1, 1500, 1280])]; + tensor transpose_168 = transpose(perm = var_2437, x = var_2436_cast); + tensor x_263_cast = reshape(shape = concat_21, x = transpose_168); + tensor var_2442_to_fp16 = const()[name = tensor("op_2442_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(850608000)))]; + tensor var_2443_to_fp16 = const()[name = tensor("op_2443_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(853884864)))]; + tensor var_2444_cast = linear(bias = var_2443_to_fp16, weight = var_2442_to_fp16, x = x_263_cast); + tensor x_265_cast = add(x = x_259_cast, y = var_2444_cast); + tensor var_2450_axes_0 = const()[name = tensor("op_2450_axes_0"), val = tensor([-1])]; + tensor blocks_21_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_21_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(853887488)))]; + tensor blocks_21_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_21_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(853890112)))]; + tensor var_2450_cast = layer_norm(axes = var_2450_axes_0, beta = blocks_21_mlp_ln_bias_to_fp16, epsilon = var_2375_to_fp16, gamma = blocks_21_mlp_ln_weight_to_fp16, x = x_265_cast); + tensor var_2459_to_fp16 = const()[name = tensor("op_2459_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(853892736)))]; + tensor var_2460_to_fp16 = const()[name = tensor("op_2460_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(867000000)))]; + tensor input_177_cast = linear(bias = var_2460_to_fp16, weight = var_2459_to_fp16, x = var_2450_cast); + tensor x_269_mode_0 = const()[name = tensor("x_269_mode_0"), val = tensor("EXACT")]; + tensor x_269_cast = gelu(mode = x_269_mode_0, x = input_177_cast); + tensor var_2465_to_fp16 = const()[name = tensor("op_2465_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(867010304)))]; + tensor var_2466_to_fp16 = const()[name = tensor("op_2466_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880117568)))]; + tensor var_2467_cast = linear(bias = var_2466_to_fp16, weight = var_2465_to_fp16, x = x_269_cast); + tensor x_271_cast = add(x = x_265_cast, y = var_2467_cast); + tensor var_2476 = const()[name = tensor("op_2476"), val = tensor(-1)]; + tensor var_2493_axes_0 = const()[name = tensor("op_2493_axes_0"), val = tensor([-1])]; + tensor blocks_22_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_22_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880120192)))]; + tensor blocks_22_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_22_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880122816)))]; + tensor var_2482_to_fp16 = const()[name = tensor("op_2482_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2493_cast = layer_norm(axes = var_2493_axes_0, beta = blocks_22_attn_ln_bias_to_fp16, epsilon = var_2482_to_fp16, gamma = blocks_22_attn_ln_weight_to_fp16, x = x_271_cast); + tensor var_2504_to_fp16 = const()[name = tensor("op_2504_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880125440)))]; + tensor var_2505_to_fp16 = const()[name = tensor("op_2505_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(883402304)))]; + tensor q_89_cast = linear(bias = var_2505_to_fp16, weight = var_2504_to_fp16, x = var_2493_cast); + tensor var_2508_to_fp16 = const()[name = tensor("op_2508_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(883404928)))]; + tensor k_89_bias_0_to_fp16 = const()[name = tensor("k_89_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(886681792)))]; + tensor k_89_cast = linear(bias = k_89_bias_0_to_fp16, weight = var_2508_to_fp16, x = var_2493_cast); + tensor var_2512_to_fp16 = const()[name = tensor("op_2512_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(886684416)))]; + tensor var_2513_to_fp16 = const()[name = tensor("op_2513_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(889961280)))]; + tensor v_89_cast = linear(bias = var_2513_to_fp16, weight = var_2512_to_fp16, x = var_2493_cast); + tensor var_2521 = const()[name = tensor("op_2521"), val = tensor([1, 1500, 20, -1])]; + tensor var_2522_cast = reshape(shape = var_2521, x = q_89_cast); + tensor const_268_to_fp16 = const()[name = tensor("const_268_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_91_cast = mul(x = var_2522_cast, y = const_268_to_fp16); + tensor var_2528 = const()[name = tensor("op_2528"), val = tensor([1, 1500, 20, -1])]; + tensor var_2529_cast = reshape(shape = var_2528, x = k_89_cast); + tensor const_269_to_fp16 = const()[name = tensor("const_269_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_91_cast = mul(x = var_2529_cast, y = const_269_to_fp16); + tensor var_2535 = const()[name = tensor("op_2535"), val = tensor([1, 1500, 20, -1])]; + tensor var_2536_cast = reshape(shape = var_2535, x = v_89_cast); + tensor var_2537 = const()[name = tensor("op_2537"), val = tensor([0, 2, 1, 3])]; + tensor qk_45_transpose_x_0 = const()[name = tensor("qk_45_transpose_x_0"), val = tensor(false)]; + tensor qk_45_transpose_y_0 = const()[name = tensor("qk_45_transpose_y_0"), val = tensor(false)]; + tensor transpose_108_perm_0 = const()[name = tensor("transpose_108_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_109_perm_0 = const()[name = tensor("transpose_109_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_165 = transpose(perm = transpose_109_perm_0, x = k_91_cast); + tensor transpose_166 = transpose(perm = transpose_108_perm_0, x = q_91_cast); + tensor qk_45_cast = matmul(transpose_x = qk_45_transpose_x_0, transpose_y = qk_45_transpose_y_0, x = transpose_166, y = transpose_165); + tensor var_2541_cast = softmax(axis = var_2476, x = qk_45_cast); + tensor var_2543_transpose_x_0 = const()[name = tensor("op_2543_transpose_x_0"), val = tensor(false)]; + tensor var_2543_transpose_y_0 = const()[name = tensor("op_2543_transpose_y_0"), val = tensor(false)]; + tensor transpose_167 = transpose(perm = var_2537, x = var_2536_cast); + tensor var_2543_cast = matmul(transpose_x = var_2543_transpose_x_0, transpose_y = var_2543_transpose_y_0, x = var_2541_cast, y = transpose_167); + tensor var_2544 = const()[name = tensor("op_2544"), val = tensor([0, 2, 1, 3])]; + tensor concat_22 = const()[name = tensor("concat_22"), val = tensor([1, 1500, 1280])]; + tensor transpose_164 = transpose(perm = var_2544, x = var_2543_cast); + tensor x_275_cast = reshape(shape = concat_22, x = transpose_164); + tensor var_2549_to_fp16 = const()[name = tensor("op_2549_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(889963904)))]; + tensor var_2550_to_fp16 = const()[name = tensor("op_2550_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893240768)))]; + tensor var_2551_cast = linear(bias = var_2550_to_fp16, weight = var_2549_to_fp16, x = x_275_cast); + tensor x_277_cast = add(x = x_271_cast, y = var_2551_cast); + tensor var_2557_axes_0 = const()[name = tensor("op_2557_axes_0"), val = tensor([-1])]; + tensor blocks_22_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_22_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893243392)))]; + tensor blocks_22_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_22_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893246016)))]; + tensor var_2557_cast = layer_norm(axes = var_2557_axes_0, beta = blocks_22_mlp_ln_bias_to_fp16, epsilon = var_2482_to_fp16, gamma = blocks_22_mlp_ln_weight_to_fp16, x = x_277_cast); + tensor var_2566_to_fp16 = const()[name = tensor("op_2566_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893248640)))]; + tensor var_2567_to_fp16 = const()[name = tensor("op_2567_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(906355904)))]; + tensor input_185_cast = linear(bias = var_2567_to_fp16, weight = var_2566_to_fp16, x = var_2557_cast); + tensor x_281_mode_0 = const()[name = tensor("x_281_mode_0"), val = tensor("EXACT")]; + tensor x_281_cast = gelu(mode = x_281_mode_0, x = input_185_cast); + tensor var_2572_to_fp16 = const()[name = tensor("op_2572_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(906366208)))]; + tensor var_2573_to_fp16 = const()[name = tensor("op_2573_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(919473472)))]; + tensor var_2574_cast = linear(bias = var_2573_to_fp16, weight = var_2572_to_fp16, x = x_281_cast); + tensor x_283_cast = add(x = x_277_cast, y = var_2574_cast); + tensor var_2583 = const()[name = tensor("op_2583"), val = tensor(-1)]; + tensor var_2600_axes_0 = const()[name = tensor("op_2600_axes_0"), val = tensor([-1])]; + tensor blocks_23_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_23_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(919476096)))]; + tensor blocks_23_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_23_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(919478720)))]; + tensor var_2589_to_fp16 = const()[name = tensor("op_2589_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2600_cast = layer_norm(axes = var_2600_axes_0, beta = blocks_23_attn_ln_bias_to_fp16, epsilon = var_2589_to_fp16, gamma = blocks_23_attn_ln_weight_to_fp16, x = x_283_cast); + tensor var_2611_to_fp16 = const()[name = tensor("op_2611_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(919481344)))]; + tensor var_2612_to_fp16 = const()[name = tensor("op_2612_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(922758208)))]; + tensor q_93_cast = linear(bias = var_2612_to_fp16, weight = var_2611_to_fp16, x = var_2600_cast); + tensor var_2615_to_fp16 = const()[name = tensor("op_2615_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(922760832)))]; + tensor k_93_bias_0_to_fp16 = const()[name = tensor("k_93_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(926037696)))]; + tensor k_93_cast = linear(bias = k_93_bias_0_to_fp16, weight = var_2615_to_fp16, x = var_2600_cast); + tensor var_2619_to_fp16 = const()[name = tensor("op_2619_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(926040320)))]; + tensor var_2620_to_fp16 = const()[name = tensor("op_2620_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(929317184)))]; + tensor v_93_cast = linear(bias = var_2620_to_fp16, weight = var_2619_to_fp16, x = var_2600_cast); + tensor var_2628 = const()[name = tensor("op_2628"), val = tensor([1, 1500, 20, -1])]; + tensor var_2629_cast = reshape(shape = var_2628, x = q_93_cast); + tensor const_270_to_fp16 = const()[name = tensor("const_270_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_95_cast = mul(x = var_2629_cast, y = const_270_to_fp16); + tensor var_2635 = const()[name = tensor("op_2635"), val = tensor([1, 1500, 20, -1])]; + tensor var_2636_cast = reshape(shape = var_2635, x = k_93_cast); + tensor const_271_to_fp16 = const()[name = tensor("const_271_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_95_cast = mul(x = var_2636_cast, y = const_271_to_fp16); + tensor var_2642 = const()[name = tensor("op_2642"), val = tensor([1, 1500, 20, -1])]; + tensor var_2643_cast = reshape(shape = var_2642, x = v_93_cast); + tensor var_2644 = const()[name = tensor("op_2644"), val = tensor([0, 2, 1, 3])]; + tensor qk_47_transpose_x_0 = const()[name = tensor("qk_47_transpose_x_0"), val = tensor(false)]; + tensor qk_47_transpose_y_0 = const()[name = tensor("qk_47_transpose_y_0"), val = tensor(false)]; + tensor transpose_110_perm_0 = const()[name = tensor("transpose_110_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_111_perm_0 = const()[name = tensor("transpose_111_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_161 = transpose(perm = transpose_111_perm_0, x = k_95_cast); + tensor transpose_162 = transpose(perm = transpose_110_perm_0, x = q_95_cast); + tensor qk_47_cast = matmul(transpose_x = qk_47_transpose_x_0, transpose_y = qk_47_transpose_y_0, x = transpose_162, y = transpose_161); + tensor var_2648_cast = softmax(axis = var_2583, x = qk_47_cast); + tensor var_2650_transpose_x_0 = const()[name = tensor("op_2650_transpose_x_0"), val = tensor(false)]; + tensor var_2650_transpose_y_0 = const()[name = tensor("op_2650_transpose_y_0"), val = tensor(false)]; + tensor transpose_163 = transpose(perm = var_2644, x = var_2643_cast); + tensor var_2650_cast = matmul(transpose_x = var_2650_transpose_x_0, transpose_y = var_2650_transpose_y_0, x = var_2648_cast, y = transpose_163); + tensor var_2651 = const()[name = tensor("op_2651"), val = tensor([0, 2, 1, 3])]; + tensor concat_23 = const()[name = tensor("concat_23"), val = tensor([1, 1500, 1280])]; + tensor transpose_160 = transpose(perm = var_2651, x = var_2650_cast); + tensor x_287_cast = reshape(shape = concat_23, x = transpose_160); + tensor var_2656_to_fp16 = const()[name = tensor("op_2656_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(929319808)))]; + tensor var_2657_to_fp16 = const()[name = tensor("op_2657_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(932596672)))]; + tensor var_2658_cast = linear(bias = var_2657_to_fp16, weight = var_2656_to_fp16, x = x_287_cast); + tensor x_289_cast = add(x = x_283_cast, y = var_2658_cast); + tensor var_2664_axes_0 = const()[name = tensor("op_2664_axes_0"), val = tensor([-1])]; + tensor blocks_23_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_23_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(932599296)))]; + tensor blocks_23_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_23_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(932601920)))]; + tensor var_2664_cast = layer_norm(axes = var_2664_axes_0, beta = blocks_23_mlp_ln_bias_to_fp16, epsilon = var_2589_to_fp16, gamma = blocks_23_mlp_ln_weight_to_fp16, x = x_289_cast); + tensor var_2673_to_fp16 = const()[name = tensor("op_2673_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(932604544)))]; + tensor var_2674_to_fp16 = const()[name = tensor("op_2674_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(945711808)))]; + tensor input_193_cast = linear(bias = var_2674_to_fp16, weight = var_2673_to_fp16, x = var_2664_cast); + tensor x_293_mode_0 = const()[name = tensor("x_293_mode_0"), val = tensor("EXACT")]; + tensor x_293_cast = gelu(mode = x_293_mode_0, x = input_193_cast); + tensor var_2679_to_fp16 = const()[name = tensor("op_2679_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(945722112)))]; + tensor var_2680_to_fp16 = const()[name = tensor("op_2680_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(958829376)))]; + tensor var_2681_cast = linear(bias = var_2680_to_fp16, weight = var_2679_to_fp16, x = x_293_cast); + tensor x_295_cast = add(x = x_289_cast, y = var_2681_cast); + tensor var_2690 = const()[name = tensor("op_2690"), val = tensor(-1)]; + tensor var_2707_axes_0 = const()[name = tensor("op_2707_axes_0"), val = tensor([-1])]; + tensor blocks_24_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_24_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(958832000)))]; + tensor blocks_24_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_24_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(958834624)))]; + tensor var_2696_to_fp16 = const()[name = tensor("op_2696_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2707_cast = layer_norm(axes = var_2707_axes_0, beta = blocks_24_attn_ln_bias_to_fp16, epsilon = var_2696_to_fp16, gamma = blocks_24_attn_ln_weight_to_fp16, x = x_295_cast); + tensor var_2718_to_fp16 = const()[name = tensor("op_2718_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(958837248)))]; + tensor var_2719_to_fp16 = const()[name = tensor("op_2719_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(962114112)))]; + tensor q_97_cast = linear(bias = var_2719_to_fp16, weight = var_2718_to_fp16, x = var_2707_cast); + tensor var_2722_to_fp16 = const()[name = tensor("op_2722_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(962116736)))]; + tensor k_97_bias_0_to_fp16 = const()[name = tensor("k_97_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(965393600)))]; + tensor k_97_cast = linear(bias = k_97_bias_0_to_fp16, weight = var_2722_to_fp16, x = var_2707_cast); + tensor var_2726_to_fp16 = const()[name = tensor("op_2726_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(965396224)))]; + tensor var_2727_to_fp16 = const()[name = tensor("op_2727_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(968673088)))]; + tensor v_97_cast = linear(bias = var_2727_to_fp16, weight = var_2726_to_fp16, x = var_2707_cast); + tensor var_2735 = const()[name = tensor("op_2735"), val = tensor([1, 1500, 20, -1])]; + tensor var_2736_cast = reshape(shape = var_2735, x = q_97_cast); + tensor const_272_to_fp16 = const()[name = tensor("const_272_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_99_cast = mul(x = var_2736_cast, y = const_272_to_fp16); + tensor var_2742 = const()[name = tensor("op_2742"), val = tensor([1, 1500, 20, -1])]; + tensor var_2743_cast = reshape(shape = var_2742, x = k_97_cast); + tensor const_273_to_fp16 = const()[name = tensor("const_273_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_99_cast = mul(x = var_2743_cast, y = const_273_to_fp16); + tensor var_2749 = const()[name = tensor("op_2749"), val = tensor([1, 1500, 20, -1])]; + tensor var_2750_cast = reshape(shape = var_2749, x = v_97_cast); + tensor var_2751 = const()[name = tensor("op_2751"), val = tensor([0, 2, 1, 3])]; + tensor qk_49_transpose_x_0 = const()[name = tensor("qk_49_transpose_x_0"), val = tensor(false)]; + tensor qk_49_transpose_y_0 = const()[name = tensor("qk_49_transpose_y_0"), val = tensor(false)]; + tensor transpose_112_perm_0 = const()[name = tensor("transpose_112_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_113_perm_0 = const()[name = tensor("transpose_113_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_157 = transpose(perm = transpose_113_perm_0, x = k_99_cast); + tensor transpose_158 = transpose(perm = transpose_112_perm_0, x = q_99_cast); + tensor qk_49_cast = matmul(transpose_x = qk_49_transpose_x_0, transpose_y = qk_49_transpose_y_0, x = transpose_158, y = transpose_157); + tensor var_2755_cast = softmax(axis = var_2690, x = qk_49_cast); + tensor var_2757_transpose_x_0 = const()[name = tensor("op_2757_transpose_x_0"), val = tensor(false)]; + tensor var_2757_transpose_y_0 = const()[name = tensor("op_2757_transpose_y_0"), val = tensor(false)]; + tensor transpose_159 = transpose(perm = var_2751, x = var_2750_cast); + tensor var_2757_cast = matmul(transpose_x = var_2757_transpose_x_0, transpose_y = var_2757_transpose_y_0, x = var_2755_cast, y = transpose_159); + tensor var_2758 = const()[name = tensor("op_2758"), val = tensor([0, 2, 1, 3])]; + tensor concat_24 = const()[name = tensor("concat_24"), val = tensor([1, 1500, 1280])]; + tensor transpose_156 = transpose(perm = var_2758, x = var_2757_cast); + tensor x_299_cast = reshape(shape = concat_24, x = transpose_156); + tensor var_2763_to_fp16 = const()[name = tensor("op_2763_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(968675712)))]; + tensor var_2764_to_fp16 = const()[name = tensor("op_2764_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(971952576)))]; + tensor var_2765_cast = linear(bias = var_2764_to_fp16, weight = var_2763_to_fp16, x = x_299_cast); + tensor x_301_cast = add(x = x_295_cast, y = var_2765_cast); + tensor var_2771_axes_0 = const()[name = tensor("op_2771_axes_0"), val = tensor([-1])]; + tensor blocks_24_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_24_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(971955200)))]; + tensor blocks_24_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_24_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(971957824)))]; + tensor var_2771_cast = layer_norm(axes = var_2771_axes_0, beta = blocks_24_mlp_ln_bias_to_fp16, epsilon = var_2696_to_fp16, gamma = blocks_24_mlp_ln_weight_to_fp16, x = x_301_cast); + tensor var_2780_to_fp16 = const()[name = tensor("op_2780_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(971960448)))]; + tensor var_2781_to_fp16 = const()[name = tensor("op_2781_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(985067712)))]; + tensor input_201_cast = linear(bias = var_2781_to_fp16, weight = var_2780_to_fp16, x = var_2771_cast); + tensor x_305_mode_0 = const()[name = tensor("x_305_mode_0"), val = tensor("EXACT")]; + tensor x_305_cast = gelu(mode = x_305_mode_0, x = input_201_cast); + tensor var_2786_to_fp16 = const()[name = tensor("op_2786_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(985078016)))]; + tensor var_2787_to_fp16 = const()[name = tensor("op_2787_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(998185280)))]; + tensor var_2788_cast = linear(bias = var_2787_to_fp16, weight = var_2786_to_fp16, x = x_305_cast); + tensor x_307_cast = add(x = x_301_cast, y = var_2788_cast); + tensor var_2797 = const()[name = tensor("op_2797"), val = tensor(-1)]; + tensor var_2814_axes_0 = const()[name = tensor("op_2814_axes_0"), val = tensor([-1])]; + tensor blocks_25_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_25_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(998187904)))]; + tensor blocks_25_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_25_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(998190528)))]; + tensor var_2803_to_fp16 = const()[name = tensor("op_2803_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2814_cast = layer_norm(axes = var_2814_axes_0, beta = blocks_25_attn_ln_bias_to_fp16, epsilon = var_2803_to_fp16, gamma = blocks_25_attn_ln_weight_to_fp16, x = x_307_cast); + tensor var_2825_to_fp16 = const()[name = tensor("op_2825_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(998193152)))]; + tensor var_2826_to_fp16 = const()[name = tensor("op_2826_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1001470016)))]; + tensor q_101_cast = linear(bias = var_2826_to_fp16, weight = var_2825_to_fp16, x = var_2814_cast); + tensor var_2829_to_fp16 = const()[name = tensor("op_2829_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1001472640)))]; + tensor k_101_bias_0_to_fp16 = const()[name = tensor("k_101_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1004749504)))]; + tensor k_101_cast = linear(bias = k_101_bias_0_to_fp16, weight = var_2829_to_fp16, x = var_2814_cast); + tensor var_2833_to_fp16 = const()[name = tensor("op_2833_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1004752128)))]; + tensor var_2834_to_fp16 = const()[name = tensor("op_2834_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1008028992)))]; + tensor v_101_cast = linear(bias = var_2834_to_fp16, weight = var_2833_to_fp16, x = var_2814_cast); + tensor var_2842 = const()[name = tensor("op_2842"), val = tensor([1, 1500, 20, -1])]; + tensor var_2843_cast = reshape(shape = var_2842, x = q_101_cast); + tensor const_274_to_fp16 = const()[name = tensor("const_274_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_103_cast = mul(x = var_2843_cast, y = const_274_to_fp16); + tensor var_2849 = const()[name = tensor("op_2849"), val = tensor([1, 1500, 20, -1])]; + tensor var_2850_cast = reshape(shape = var_2849, x = k_101_cast); + tensor const_275_to_fp16 = const()[name = tensor("const_275_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_103_cast = mul(x = var_2850_cast, y = const_275_to_fp16); + tensor var_2856 = const()[name = tensor("op_2856"), val = tensor([1, 1500, 20, -1])]; + tensor var_2857_cast = reshape(shape = var_2856, x = v_101_cast); + tensor var_2858 = const()[name = tensor("op_2858"), val = tensor([0, 2, 1, 3])]; + tensor qk_51_transpose_x_0 = const()[name = tensor("qk_51_transpose_x_0"), val = tensor(false)]; + tensor qk_51_transpose_y_0 = const()[name = tensor("qk_51_transpose_y_0"), val = tensor(false)]; + tensor transpose_114_perm_0 = const()[name = tensor("transpose_114_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_115_perm_0 = const()[name = tensor("transpose_115_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_153 = transpose(perm = transpose_115_perm_0, x = k_103_cast); + tensor transpose_154 = transpose(perm = transpose_114_perm_0, x = q_103_cast); + tensor qk_51_cast = matmul(transpose_x = qk_51_transpose_x_0, transpose_y = qk_51_transpose_y_0, x = transpose_154, y = transpose_153); + tensor var_2862_cast = softmax(axis = var_2797, x = qk_51_cast); + tensor var_2864_transpose_x_0 = const()[name = tensor("op_2864_transpose_x_0"), val = tensor(false)]; + tensor var_2864_transpose_y_0 = const()[name = tensor("op_2864_transpose_y_0"), val = tensor(false)]; + tensor transpose_155 = transpose(perm = var_2858, x = var_2857_cast); + tensor var_2864_cast = matmul(transpose_x = var_2864_transpose_x_0, transpose_y = var_2864_transpose_y_0, x = var_2862_cast, y = transpose_155); + tensor var_2865 = const()[name = tensor("op_2865"), val = tensor([0, 2, 1, 3])]; + tensor concat_25 = const()[name = tensor("concat_25"), val = tensor([1, 1500, 1280])]; + tensor transpose_152 = transpose(perm = var_2865, x = var_2864_cast); + tensor x_311_cast = reshape(shape = concat_25, x = transpose_152); + tensor var_2870_to_fp16 = const()[name = tensor("op_2870_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1008031616)))]; + tensor var_2871_to_fp16 = const()[name = tensor("op_2871_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1011308480)))]; + tensor var_2872_cast = linear(bias = var_2871_to_fp16, weight = var_2870_to_fp16, x = x_311_cast); + tensor x_313_cast = add(x = x_307_cast, y = var_2872_cast); + tensor var_2878_axes_0 = const()[name = tensor("op_2878_axes_0"), val = tensor([-1])]; + tensor blocks_25_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_25_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1011311104)))]; + tensor blocks_25_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_25_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1011313728)))]; + tensor var_2878_cast = layer_norm(axes = var_2878_axes_0, beta = blocks_25_mlp_ln_bias_to_fp16, epsilon = var_2803_to_fp16, gamma = blocks_25_mlp_ln_weight_to_fp16, x = x_313_cast); + tensor var_2887_to_fp16 = const()[name = tensor("op_2887_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1011316352)))]; + tensor var_2888_to_fp16 = const()[name = tensor("op_2888_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1024423616)))]; + tensor input_209_cast = linear(bias = var_2888_to_fp16, weight = var_2887_to_fp16, x = var_2878_cast); + tensor x_317_mode_0 = const()[name = tensor("x_317_mode_0"), val = tensor("EXACT")]; + tensor x_317_cast = gelu(mode = x_317_mode_0, x = input_209_cast); + tensor var_2893_to_fp16 = const()[name = tensor("op_2893_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1024433920)))]; + tensor var_2894_to_fp16 = const()[name = tensor("op_2894_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1037541184)))]; + tensor var_2895_cast = linear(bias = var_2894_to_fp16, weight = var_2893_to_fp16, x = x_317_cast); + tensor x_319_cast = add(x = x_313_cast, y = var_2895_cast); + tensor var_2904 = const()[name = tensor("op_2904"), val = tensor(-1)]; + tensor var_2921_axes_0 = const()[name = tensor("op_2921_axes_0"), val = tensor([-1])]; + tensor blocks_26_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_26_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1037543808)))]; + tensor blocks_26_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_26_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1037546432)))]; + tensor var_2910_to_fp16 = const()[name = tensor("op_2910_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2921_cast = layer_norm(axes = var_2921_axes_0, beta = blocks_26_attn_ln_bias_to_fp16, epsilon = var_2910_to_fp16, gamma = blocks_26_attn_ln_weight_to_fp16, x = x_319_cast); + tensor var_2932_to_fp16 = const()[name = tensor("op_2932_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1037549056)))]; + tensor var_2933_to_fp16 = const()[name = tensor("op_2933_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1040825920)))]; + tensor q_105_cast = linear(bias = var_2933_to_fp16, weight = var_2932_to_fp16, x = var_2921_cast); + tensor var_2936_to_fp16 = const()[name = tensor("op_2936_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1040828544)))]; + tensor k_105_bias_0_to_fp16 = const()[name = tensor("k_105_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1044105408)))]; + tensor k_105_cast = linear(bias = k_105_bias_0_to_fp16, weight = var_2936_to_fp16, x = var_2921_cast); + tensor var_2940_to_fp16 = const()[name = tensor("op_2940_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1044108032)))]; + tensor var_2941_to_fp16 = const()[name = tensor("op_2941_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1047384896)))]; + tensor v_105_cast = linear(bias = var_2941_to_fp16, weight = var_2940_to_fp16, x = var_2921_cast); + tensor var_2949 = const()[name = tensor("op_2949"), val = tensor([1, 1500, 20, -1])]; + tensor var_2950_cast = reshape(shape = var_2949, x = q_105_cast); + tensor const_276_to_fp16 = const()[name = tensor("const_276_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_107_cast = mul(x = var_2950_cast, y = const_276_to_fp16); + tensor var_2956 = const()[name = tensor("op_2956"), val = tensor([1, 1500, 20, -1])]; + tensor var_2957_cast = reshape(shape = var_2956, x = k_105_cast); + tensor const_277_to_fp16 = const()[name = tensor("const_277_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_107_cast = mul(x = var_2957_cast, y = const_277_to_fp16); + tensor var_2963 = const()[name = tensor("op_2963"), val = tensor([1, 1500, 20, -1])]; + tensor var_2964_cast = reshape(shape = var_2963, x = v_105_cast); + tensor var_2965 = const()[name = tensor("op_2965"), val = tensor([0, 2, 1, 3])]; + tensor qk_53_transpose_x_0 = const()[name = tensor("qk_53_transpose_x_0"), val = tensor(false)]; + tensor qk_53_transpose_y_0 = const()[name = tensor("qk_53_transpose_y_0"), val = tensor(false)]; + tensor transpose_116_perm_0 = const()[name = tensor("transpose_116_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_117_perm_0 = const()[name = tensor("transpose_117_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_149 = transpose(perm = transpose_117_perm_0, x = k_107_cast); + tensor transpose_150 = transpose(perm = transpose_116_perm_0, x = q_107_cast); + tensor qk_53_cast = matmul(transpose_x = qk_53_transpose_x_0, transpose_y = qk_53_transpose_y_0, x = transpose_150, y = transpose_149); + tensor var_2969_cast = softmax(axis = var_2904, x = qk_53_cast); + tensor var_2971_transpose_x_0 = const()[name = tensor("op_2971_transpose_x_0"), val = tensor(false)]; + tensor var_2971_transpose_y_0 = const()[name = tensor("op_2971_transpose_y_0"), val = tensor(false)]; + tensor transpose_151 = transpose(perm = var_2965, x = var_2964_cast); + tensor var_2971_cast = matmul(transpose_x = var_2971_transpose_x_0, transpose_y = var_2971_transpose_y_0, x = var_2969_cast, y = transpose_151); + tensor var_2972 = const()[name = tensor("op_2972"), val = tensor([0, 2, 1, 3])]; + tensor concat_26 = const()[name = tensor("concat_26"), val = tensor([1, 1500, 1280])]; + tensor transpose_148 = transpose(perm = var_2972, x = var_2971_cast); + tensor x_323_cast = reshape(shape = concat_26, x = transpose_148); + tensor var_2977_to_fp16 = const()[name = tensor("op_2977_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1047387520)))]; + tensor var_2978_to_fp16 = const()[name = tensor("op_2978_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1050664384)))]; + tensor var_2979_cast = linear(bias = var_2978_to_fp16, weight = var_2977_to_fp16, x = x_323_cast); + tensor x_325_cast = add(x = x_319_cast, y = var_2979_cast); + tensor var_2985_axes_0 = const()[name = tensor("op_2985_axes_0"), val = tensor([-1])]; + tensor blocks_26_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_26_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1050667008)))]; + tensor blocks_26_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_26_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1050669632)))]; + tensor var_2985_cast = layer_norm(axes = var_2985_axes_0, beta = blocks_26_mlp_ln_bias_to_fp16, epsilon = var_2910_to_fp16, gamma = blocks_26_mlp_ln_weight_to_fp16, x = x_325_cast); + tensor var_2994_to_fp16 = const()[name = tensor("op_2994_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1050672256)))]; + tensor var_2995_to_fp16 = const()[name = tensor("op_2995_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1063779520)))]; + tensor input_217_cast = linear(bias = var_2995_to_fp16, weight = var_2994_to_fp16, x = var_2985_cast); + tensor x_329_mode_0 = const()[name = tensor("x_329_mode_0"), val = tensor("EXACT")]; + tensor x_329_cast = gelu(mode = x_329_mode_0, x = input_217_cast); + tensor var_3000_to_fp16 = const()[name = tensor("op_3000_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1063789824)))]; + tensor var_3001_to_fp16 = const()[name = tensor("op_3001_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1076897088)))]; + tensor var_3002_cast = linear(bias = var_3001_to_fp16, weight = var_3000_to_fp16, x = x_329_cast); + tensor x_331_cast = add(x = x_325_cast, y = var_3002_cast); + tensor var_3011 = const()[name = tensor("op_3011"), val = tensor(-1)]; + tensor var_3028_axes_0 = const()[name = tensor("op_3028_axes_0"), val = tensor([-1])]; + tensor blocks_27_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_27_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1076899712)))]; + tensor blocks_27_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_27_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1076902336)))]; + tensor var_3017_to_fp16 = const()[name = tensor("op_3017_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3028_cast = layer_norm(axes = var_3028_axes_0, beta = blocks_27_attn_ln_bias_to_fp16, epsilon = var_3017_to_fp16, gamma = blocks_27_attn_ln_weight_to_fp16, x = x_331_cast); + tensor var_3039_to_fp16 = const()[name = tensor("op_3039_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1076904960)))]; + tensor var_3040_to_fp16 = const()[name = tensor("op_3040_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1080181824)))]; + tensor q_109_cast = linear(bias = var_3040_to_fp16, weight = var_3039_to_fp16, x = var_3028_cast); + tensor var_3043_to_fp16 = const()[name = tensor("op_3043_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1080184448)))]; + tensor k_109_bias_0_to_fp16 = const()[name = tensor("k_109_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1083461312)))]; + tensor k_109_cast = linear(bias = k_109_bias_0_to_fp16, weight = var_3043_to_fp16, x = var_3028_cast); + tensor var_3047_to_fp16 = const()[name = tensor("op_3047_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1083463936)))]; + tensor var_3048_to_fp16 = const()[name = tensor("op_3048_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1086740800)))]; + tensor v_109_cast = linear(bias = var_3048_to_fp16, weight = var_3047_to_fp16, x = var_3028_cast); + tensor var_3056 = const()[name = tensor("op_3056"), val = tensor([1, 1500, 20, -1])]; + tensor var_3057_cast = reshape(shape = var_3056, x = q_109_cast); + tensor const_278_to_fp16 = const()[name = tensor("const_278_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_111_cast = mul(x = var_3057_cast, y = const_278_to_fp16); + tensor var_3063 = const()[name = tensor("op_3063"), val = tensor([1, 1500, 20, -1])]; + tensor var_3064_cast = reshape(shape = var_3063, x = k_109_cast); + tensor const_279_to_fp16 = const()[name = tensor("const_279_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_111_cast = mul(x = var_3064_cast, y = const_279_to_fp16); + tensor var_3070 = const()[name = tensor("op_3070"), val = tensor([1, 1500, 20, -1])]; + tensor var_3071_cast = reshape(shape = var_3070, x = v_109_cast); + tensor var_3072 = const()[name = tensor("op_3072"), val = tensor([0, 2, 1, 3])]; + tensor qk_55_transpose_x_0 = const()[name = tensor("qk_55_transpose_x_0"), val = tensor(false)]; + tensor qk_55_transpose_y_0 = const()[name = tensor("qk_55_transpose_y_0"), val = tensor(false)]; + tensor transpose_118_perm_0 = const()[name = tensor("transpose_118_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_119_perm_0 = const()[name = tensor("transpose_119_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_145 = transpose(perm = transpose_119_perm_0, x = k_111_cast); + tensor transpose_146 = transpose(perm = transpose_118_perm_0, x = q_111_cast); + tensor qk_55_cast = matmul(transpose_x = qk_55_transpose_x_0, transpose_y = qk_55_transpose_y_0, x = transpose_146, y = transpose_145); + tensor var_3076_cast = softmax(axis = var_3011, x = qk_55_cast); + tensor var_3078_transpose_x_0 = const()[name = tensor("op_3078_transpose_x_0"), val = tensor(false)]; + tensor var_3078_transpose_y_0 = const()[name = tensor("op_3078_transpose_y_0"), val = tensor(false)]; + tensor transpose_147 = transpose(perm = var_3072, x = var_3071_cast); + tensor var_3078_cast = matmul(transpose_x = var_3078_transpose_x_0, transpose_y = var_3078_transpose_y_0, x = var_3076_cast, y = transpose_147); + tensor var_3079 = const()[name = tensor("op_3079"), val = tensor([0, 2, 1, 3])]; + tensor concat_27 = const()[name = tensor("concat_27"), val = tensor([1, 1500, 1280])]; + tensor transpose_144 = transpose(perm = var_3079, x = var_3078_cast); + tensor x_335_cast = reshape(shape = concat_27, x = transpose_144); + tensor var_3084_to_fp16 = const()[name = tensor("op_3084_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1086743424)))]; + tensor var_3085_to_fp16 = const()[name = tensor("op_3085_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1090020288)))]; + tensor var_3086_cast = linear(bias = var_3085_to_fp16, weight = var_3084_to_fp16, x = x_335_cast); + tensor x_337_cast = add(x = x_331_cast, y = var_3086_cast); + tensor var_3092_axes_0 = const()[name = tensor("op_3092_axes_0"), val = tensor([-1])]; + tensor blocks_27_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_27_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1090022912)))]; + tensor blocks_27_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_27_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1090025536)))]; + tensor var_3092_cast = layer_norm(axes = var_3092_axes_0, beta = blocks_27_mlp_ln_bias_to_fp16, epsilon = var_3017_to_fp16, gamma = blocks_27_mlp_ln_weight_to_fp16, x = x_337_cast); + tensor var_3101_to_fp16 = const()[name = tensor("op_3101_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1090028160)))]; + tensor var_3102_to_fp16 = const()[name = tensor("op_3102_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1103135424)))]; + tensor input_225_cast = linear(bias = var_3102_to_fp16, weight = var_3101_to_fp16, x = var_3092_cast); + tensor x_341_mode_0 = const()[name = tensor("x_341_mode_0"), val = tensor("EXACT")]; + tensor x_341_cast = gelu(mode = x_341_mode_0, x = input_225_cast); + tensor var_3107_to_fp16 = const()[name = tensor("op_3107_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1103145728)))]; + tensor var_3108_to_fp16 = const()[name = tensor("op_3108_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116252992)))]; + tensor var_3109_cast = linear(bias = var_3108_to_fp16, weight = var_3107_to_fp16, x = x_341_cast); + tensor x_343_cast = add(x = x_337_cast, y = var_3109_cast); + tensor var_3118 = const()[name = tensor("op_3118"), val = tensor(-1)]; + tensor var_3135_axes_0 = const()[name = tensor("op_3135_axes_0"), val = tensor([-1])]; + tensor blocks_28_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_28_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116255616)))]; + tensor blocks_28_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_28_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116258240)))]; + tensor var_3124_to_fp16 = const()[name = tensor("op_3124_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3135_cast = layer_norm(axes = var_3135_axes_0, beta = blocks_28_attn_ln_bias_to_fp16, epsilon = var_3124_to_fp16, gamma = blocks_28_attn_ln_weight_to_fp16, x = x_343_cast); + tensor var_3146_to_fp16 = const()[name = tensor("op_3146_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116260864)))]; + tensor var_3147_to_fp16 = const()[name = tensor("op_3147_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1119537728)))]; + tensor q_113_cast = linear(bias = var_3147_to_fp16, weight = var_3146_to_fp16, x = var_3135_cast); + tensor var_3150_to_fp16 = const()[name = tensor("op_3150_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1119540352)))]; + tensor k_113_bias_0_to_fp16 = const()[name = tensor("k_113_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1122817216)))]; + tensor k_113_cast = linear(bias = k_113_bias_0_to_fp16, weight = var_3150_to_fp16, x = var_3135_cast); + tensor var_3154_to_fp16 = const()[name = tensor("op_3154_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1122819840)))]; + tensor var_3155_to_fp16 = const()[name = tensor("op_3155_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1126096704)))]; + tensor v_113_cast = linear(bias = var_3155_to_fp16, weight = var_3154_to_fp16, x = var_3135_cast); + tensor var_3163 = const()[name = tensor("op_3163"), val = tensor([1, 1500, 20, -1])]; + tensor var_3164_cast = reshape(shape = var_3163, x = q_113_cast); + tensor const_280_to_fp16 = const()[name = tensor("const_280_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_115_cast = mul(x = var_3164_cast, y = const_280_to_fp16); + tensor var_3170 = const()[name = tensor("op_3170"), val = tensor([1, 1500, 20, -1])]; + tensor var_3171_cast = reshape(shape = var_3170, x = k_113_cast); + tensor const_281_to_fp16 = const()[name = tensor("const_281_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_115_cast = mul(x = var_3171_cast, y = const_281_to_fp16); + tensor var_3177 = const()[name = tensor("op_3177"), val = tensor([1, 1500, 20, -1])]; + tensor var_3178_cast = reshape(shape = var_3177, x = v_113_cast); + tensor var_3179 = const()[name = tensor("op_3179"), val = tensor([0, 2, 1, 3])]; + tensor qk_57_transpose_x_0 = const()[name = tensor("qk_57_transpose_x_0"), val = tensor(false)]; + tensor qk_57_transpose_y_0 = const()[name = tensor("qk_57_transpose_y_0"), val = tensor(false)]; + tensor transpose_120_perm_0 = const()[name = tensor("transpose_120_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_121_perm_0 = const()[name = tensor("transpose_121_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_141 = transpose(perm = transpose_121_perm_0, x = k_115_cast); + tensor transpose_142 = transpose(perm = transpose_120_perm_0, x = q_115_cast); + tensor qk_57_cast = matmul(transpose_x = qk_57_transpose_x_0, transpose_y = qk_57_transpose_y_0, x = transpose_142, y = transpose_141); + tensor var_3183_cast = softmax(axis = var_3118, x = qk_57_cast); + tensor var_3185_transpose_x_0 = const()[name = tensor("op_3185_transpose_x_0"), val = tensor(false)]; + tensor var_3185_transpose_y_0 = const()[name = tensor("op_3185_transpose_y_0"), val = tensor(false)]; + tensor transpose_143 = transpose(perm = var_3179, x = var_3178_cast); + tensor var_3185_cast = matmul(transpose_x = var_3185_transpose_x_0, transpose_y = var_3185_transpose_y_0, x = var_3183_cast, y = transpose_143); + tensor var_3186 = const()[name = tensor("op_3186"), val = tensor([0, 2, 1, 3])]; + tensor concat_28 = const()[name = tensor("concat_28"), val = tensor([1, 1500, 1280])]; + tensor transpose_140 = transpose(perm = var_3186, x = var_3185_cast); + tensor x_347_cast = reshape(shape = concat_28, x = transpose_140); + tensor var_3191_to_fp16 = const()[name = tensor("op_3191_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1126099328)))]; + tensor var_3192_to_fp16 = const()[name = tensor("op_3192_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1129376192)))]; + tensor var_3193_cast = linear(bias = var_3192_to_fp16, weight = var_3191_to_fp16, x = x_347_cast); + tensor x_349_cast = add(x = x_343_cast, y = var_3193_cast); + tensor var_3199_axes_0 = const()[name = tensor("op_3199_axes_0"), val = tensor([-1])]; + tensor blocks_28_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_28_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1129378816)))]; + tensor blocks_28_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_28_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1129381440)))]; + tensor var_3199_cast = layer_norm(axes = var_3199_axes_0, beta = blocks_28_mlp_ln_bias_to_fp16, epsilon = var_3124_to_fp16, gamma = blocks_28_mlp_ln_weight_to_fp16, x = x_349_cast); + tensor var_3208_to_fp16 = const()[name = tensor("op_3208_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1129384064)))]; + tensor var_3209_to_fp16 = const()[name = tensor("op_3209_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1142491328)))]; + tensor input_233_cast = linear(bias = var_3209_to_fp16, weight = var_3208_to_fp16, x = var_3199_cast); + tensor x_353_mode_0 = const()[name = tensor("x_353_mode_0"), val = tensor("EXACT")]; + tensor x_353_cast = gelu(mode = x_353_mode_0, x = input_233_cast); + tensor var_3214_to_fp16 = const()[name = tensor("op_3214_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1142501632)))]; + tensor var_3215_to_fp16 = const()[name = tensor("op_3215_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1155608896)))]; + tensor var_3216_cast = linear(bias = var_3215_to_fp16, weight = var_3214_to_fp16, x = x_353_cast); + tensor x_355_cast = add(x = x_349_cast, y = var_3216_cast); + tensor var_3225 = const()[name = tensor("op_3225"), val = tensor(-1)]; + tensor var_3242_axes_0 = const()[name = tensor("op_3242_axes_0"), val = tensor([-1])]; + tensor blocks_29_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_29_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1155611520)))]; + tensor blocks_29_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_29_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1155614144)))]; + tensor var_3231_to_fp16 = const()[name = tensor("op_3231_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3242_cast = layer_norm(axes = var_3242_axes_0, beta = blocks_29_attn_ln_bias_to_fp16, epsilon = var_3231_to_fp16, gamma = blocks_29_attn_ln_weight_to_fp16, x = x_355_cast); + tensor var_3253_to_fp16 = const()[name = tensor("op_3253_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1155616768)))]; + tensor var_3254_to_fp16 = const()[name = tensor("op_3254_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1158893632)))]; + tensor q_117_cast = linear(bias = var_3254_to_fp16, weight = var_3253_to_fp16, x = var_3242_cast); + tensor var_3257_to_fp16 = const()[name = tensor("op_3257_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1158896256)))]; + tensor k_117_bias_0_to_fp16 = const()[name = tensor("k_117_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1162173120)))]; + tensor k_117_cast = linear(bias = k_117_bias_0_to_fp16, weight = var_3257_to_fp16, x = var_3242_cast); + tensor var_3261_to_fp16 = const()[name = tensor("op_3261_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1162175744)))]; + tensor var_3262_to_fp16 = const()[name = tensor("op_3262_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1165452608)))]; + tensor v_117_cast = linear(bias = var_3262_to_fp16, weight = var_3261_to_fp16, x = var_3242_cast); + tensor var_3270 = const()[name = tensor("op_3270"), val = tensor([1, 1500, 20, -1])]; + tensor var_3271_cast = reshape(shape = var_3270, x = q_117_cast); + tensor const_282_to_fp16 = const()[name = tensor("const_282_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_119_cast = mul(x = var_3271_cast, y = const_282_to_fp16); + tensor var_3277 = const()[name = tensor("op_3277"), val = tensor([1, 1500, 20, -1])]; + tensor var_3278_cast = reshape(shape = var_3277, x = k_117_cast); + tensor const_283_to_fp16 = const()[name = tensor("const_283_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_119_cast = mul(x = var_3278_cast, y = const_283_to_fp16); + tensor var_3284 = const()[name = tensor("op_3284"), val = tensor([1, 1500, 20, -1])]; + tensor var_3285_cast = reshape(shape = var_3284, x = v_117_cast); + tensor var_3286 = const()[name = tensor("op_3286"), val = tensor([0, 2, 1, 3])]; + tensor qk_59_transpose_x_0 = const()[name = tensor("qk_59_transpose_x_0"), val = tensor(false)]; + tensor qk_59_transpose_y_0 = const()[name = tensor("qk_59_transpose_y_0"), val = tensor(false)]; + tensor transpose_122_perm_0 = const()[name = tensor("transpose_122_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_123_perm_0 = const()[name = tensor("transpose_123_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_137 = transpose(perm = transpose_123_perm_0, x = k_119_cast); + tensor transpose_138 = transpose(perm = transpose_122_perm_0, x = q_119_cast); + tensor qk_59_cast = matmul(transpose_x = qk_59_transpose_x_0, transpose_y = qk_59_transpose_y_0, x = transpose_138, y = transpose_137); + tensor var_3290_cast = softmax(axis = var_3225, x = qk_59_cast); + tensor var_3292_transpose_x_0 = const()[name = tensor("op_3292_transpose_x_0"), val = tensor(false)]; + tensor var_3292_transpose_y_0 = const()[name = tensor("op_3292_transpose_y_0"), val = tensor(false)]; + tensor transpose_139 = transpose(perm = var_3286, x = var_3285_cast); + tensor var_3292_cast = matmul(transpose_x = var_3292_transpose_x_0, transpose_y = var_3292_transpose_y_0, x = var_3290_cast, y = transpose_139); + tensor var_3293 = const()[name = tensor("op_3293"), val = tensor([0, 2, 1, 3])]; + tensor concat_29 = const()[name = tensor("concat_29"), val = tensor([1, 1500, 1280])]; + tensor transpose_136 = transpose(perm = var_3293, x = var_3292_cast); + tensor x_359_cast = reshape(shape = concat_29, x = transpose_136); + tensor var_3298_to_fp16 = const()[name = tensor("op_3298_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1165455232)))]; + tensor var_3299_to_fp16 = const()[name = tensor("op_3299_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1168732096)))]; + tensor var_3300_cast = linear(bias = var_3299_to_fp16, weight = var_3298_to_fp16, x = x_359_cast); + tensor x_361_cast = add(x = x_355_cast, y = var_3300_cast); + tensor var_3306_axes_0 = const()[name = tensor("op_3306_axes_0"), val = tensor([-1])]; + tensor blocks_29_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_29_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1168734720)))]; + tensor blocks_29_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_29_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1168737344)))]; + tensor var_3306_cast = layer_norm(axes = var_3306_axes_0, beta = blocks_29_mlp_ln_bias_to_fp16, epsilon = var_3231_to_fp16, gamma = blocks_29_mlp_ln_weight_to_fp16, x = x_361_cast); + tensor var_3315_to_fp16 = const()[name = tensor("op_3315_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1168739968)))]; + tensor var_3316_to_fp16 = const()[name = tensor("op_3316_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1181847232)))]; + tensor input_241_cast = linear(bias = var_3316_to_fp16, weight = var_3315_to_fp16, x = var_3306_cast); + tensor x_365_mode_0 = const()[name = tensor("x_365_mode_0"), val = tensor("EXACT")]; + tensor x_365_cast = gelu(mode = x_365_mode_0, x = input_241_cast); + tensor var_3321_to_fp16 = const()[name = tensor("op_3321_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1181857536)))]; + tensor var_3322_to_fp16 = const()[name = tensor("op_3322_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1194964800)))]; + tensor var_3323_cast = linear(bias = var_3322_to_fp16, weight = var_3321_to_fp16, x = x_365_cast); + tensor x_367_cast = add(x = x_361_cast, y = var_3323_cast); + tensor var_3332 = const()[name = tensor("op_3332"), val = tensor(-1)]; + tensor var_3349_axes_0 = const()[name = tensor("op_3349_axes_0"), val = tensor([-1])]; + tensor blocks_30_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_30_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1194967424)))]; + tensor blocks_30_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_30_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1194970048)))]; + tensor var_3338_to_fp16 = const()[name = tensor("op_3338_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3349_cast = layer_norm(axes = var_3349_axes_0, beta = blocks_30_attn_ln_bias_to_fp16, epsilon = var_3338_to_fp16, gamma = blocks_30_attn_ln_weight_to_fp16, x = x_367_cast); + tensor var_3360_to_fp16 = const()[name = tensor("op_3360_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1194972672)))]; + tensor var_3361_to_fp16 = const()[name = tensor("op_3361_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1198249536)))]; + tensor q_121_cast = linear(bias = var_3361_to_fp16, weight = var_3360_to_fp16, x = var_3349_cast); + tensor var_3364_to_fp16 = const()[name = tensor("op_3364_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1198252160)))]; + tensor k_121_bias_0_to_fp16 = const()[name = tensor("k_121_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1201529024)))]; + tensor k_121_cast = linear(bias = k_121_bias_0_to_fp16, weight = var_3364_to_fp16, x = var_3349_cast); + tensor var_3368_to_fp16 = const()[name = tensor("op_3368_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1201531648)))]; + tensor var_3369_to_fp16 = const()[name = tensor("op_3369_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1204808512)))]; + tensor v_121_cast = linear(bias = var_3369_to_fp16, weight = var_3368_to_fp16, x = var_3349_cast); + tensor var_3377 = const()[name = tensor("op_3377"), val = tensor([1, 1500, 20, -1])]; + tensor var_3378_cast = reshape(shape = var_3377, x = q_121_cast); + tensor const_284_to_fp16 = const()[name = tensor("const_284_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_123_cast = mul(x = var_3378_cast, y = const_284_to_fp16); + tensor var_3384 = const()[name = tensor("op_3384"), val = tensor([1, 1500, 20, -1])]; + tensor var_3385_cast = reshape(shape = var_3384, x = k_121_cast); + tensor const_285_to_fp16 = const()[name = tensor("const_285_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_123_cast = mul(x = var_3385_cast, y = const_285_to_fp16); + tensor var_3391 = const()[name = tensor("op_3391"), val = tensor([1, 1500, 20, -1])]; + tensor var_3392_cast = reshape(shape = var_3391, x = v_121_cast); + tensor var_3393 = const()[name = tensor("op_3393"), val = tensor([0, 2, 1, 3])]; + tensor qk_61_transpose_x_0 = const()[name = tensor("qk_61_transpose_x_0"), val = tensor(false)]; + tensor qk_61_transpose_y_0 = const()[name = tensor("qk_61_transpose_y_0"), val = tensor(false)]; + tensor transpose_124_perm_0 = const()[name = tensor("transpose_124_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_125_perm_0 = const()[name = tensor("transpose_125_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_133 = transpose(perm = transpose_125_perm_0, x = k_123_cast); + tensor transpose_134 = transpose(perm = transpose_124_perm_0, x = q_123_cast); + tensor qk_61_cast = matmul(transpose_x = qk_61_transpose_x_0, transpose_y = qk_61_transpose_y_0, x = transpose_134, y = transpose_133); + tensor var_3397_cast = softmax(axis = var_3332, x = qk_61_cast); + tensor var_3399_transpose_x_0 = const()[name = tensor("op_3399_transpose_x_0"), val = tensor(false)]; + tensor var_3399_transpose_y_0 = const()[name = tensor("op_3399_transpose_y_0"), val = tensor(false)]; + tensor transpose_135 = transpose(perm = var_3393, x = var_3392_cast); + tensor var_3399_cast = matmul(transpose_x = var_3399_transpose_x_0, transpose_y = var_3399_transpose_y_0, x = var_3397_cast, y = transpose_135); + tensor var_3400 = const()[name = tensor("op_3400"), val = tensor([0, 2, 1, 3])]; + tensor concat_30 = const()[name = tensor("concat_30"), val = tensor([1, 1500, 1280])]; + tensor transpose_132 = transpose(perm = var_3400, x = var_3399_cast); + tensor x_371_cast = reshape(shape = concat_30, x = transpose_132); + tensor var_3405_to_fp16 = const()[name = tensor("op_3405_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1204811136)))]; + tensor var_3406_to_fp16 = const()[name = tensor("op_3406_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208088000)))]; + tensor var_3407_cast = linear(bias = var_3406_to_fp16, weight = var_3405_to_fp16, x = x_371_cast); + tensor x_373_cast = add(x = x_367_cast, y = var_3407_cast); + tensor var_3413_axes_0 = const()[name = tensor("op_3413_axes_0"), val = tensor([-1])]; + tensor blocks_30_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_30_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208090624)))]; + tensor blocks_30_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_30_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208093248)))]; + tensor var_3413_cast = layer_norm(axes = var_3413_axes_0, beta = blocks_30_mlp_ln_bias_to_fp16, epsilon = var_3338_to_fp16, gamma = blocks_30_mlp_ln_weight_to_fp16, x = x_373_cast); + tensor var_3422_to_fp16 = const()[name = tensor("op_3422_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208095872)))]; + tensor var_3423_to_fp16 = const()[name = tensor("op_3423_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1221203136)))]; + tensor input_249_cast = linear(bias = var_3423_to_fp16, weight = var_3422_to_fp16, x = var_3413_cast); + tensor x_377_mode_0 = const()[name = tensor("x_377_mode_0"), val = tensor("EXACT")]; + tensor x_377_cast = gelu(mode = x_377_mode_0, x = input_249_cast); + tensor var_3428_to_fp16 = const()[name = tensor("op_3428_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1221213440)))]; + tensor var_3429_to_fp16 = const()[name = tensor("op_3429_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1234320704)))]; + tensor var_3430_cast = linear(bias = var_3429_to_fp16, weight = var_3428_to_fp16, x = x_377_cast); + tensor x_379_cast = add(x = x_373_cast, y = var_3430_cast); + tensor var_3439 = const()[name = tensor("op_3439"), val = tensor(-1)]; + tensor var_3456_axes_0 = const()[name = tensor("op_3456_axes_0"), val = tensor([-1])]; + tensor blocks_31_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_31_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1234323328)))]; + tensor blocks_31_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_31_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1234325952)))]; + tensor var_3445_to_fp16 = const()[name = tensor("op_3445_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3456_cast = layer_norm(axes = var_3456_axes_0, beta = blocks_31_attn_ln_bias_to_fp16, epsilon = var_3445_to_fp16, gamma = blocks_31_attn_ln_weight_to_fp16, x = x_379_cast); + tensor var_3467_to_fp16 = const()[name = tensor("op_3467_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1234328576)))]; + tensor var_3468_to_fp16 = const()[name = tensor("op_3468_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1237605440)))]; + tensor q_125_cast = linear(bias = var_3468_to_fp16, weight = var_3467_to_fp16, x = var_3456_cast); + tensor var_3471_to_fp16 = const()[name = tensor("op_3471_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1237608064)))]; + tensor k_125_bias_0_to_fp16 = const()[name = tensor("k_125_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1240884928)))]; + tensor k_125_cast = linear(bias = k_125_bias_0_to_fp16, weight = var_3471_to_fp16, x = var_3456_cast); + tensor var_3475_to_fp16 = const()[name = tensor("op_3475_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1240887552)))]; + tensor var_3476_to_fp16 = const()[name = tensor("op_3476_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1244164416)))]; + tensor v_125_cast = linear(bias = var_3476_to_fp16, weight = var_3475_to_fp16, x = var_3456_cast); + tensor var_3484 = const()[name = tensor("op_3484"), val = tensor([1, 1500, 20, -1])]; + tensor var_3485_cast = reshape(shape = var_3484, x = q_125_cast); + tensor const_286_to_fp16 = const()[name = tensor("const_286_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_cast = mul(x = var_3485_cast, y = const_286_to_fp16); + tensor var_3491 = const()[name = tensor("op_3491"), val = tensor([1, 1500, 20, -1])]; + tensor var_3492_cast = reshape(shape = var_3491, x = k_125_cast); + tensor const_287_to_fp16 = const()[name = tensor("const_287_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_cast = mul(x = var_3492_cast, y = const_287_to_fp16); + tensor var_3498 = const()[name = tensor("op_3498"), val = tensor([1, 1500, 20, -1])]; + tensor var_3499_cast = reshape(shape = var_3498, x = v_125_cast); + tensor var_3500 = const()[name = tensor("op_3500"), val = tensor([0, 2, 1, 3])]; + tensor qk_transpose_x_0 = const()[name = tensor("qk_transpose_x_0"), val = tensor(false)]; + tensor qk_transpose_y_0 = const()[name = tensor("qk_transpose_y_0"), val = tensor(false)]; + tensor transpose_126_perm_0 = const()[name = tensor("transpose_126_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_127_perm_0 = const()[name = tensor("transpose_127_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_129 = transpose(perm = transpose_127_perm_0, x = k_cast); + tensor transpose_130 = transpose(perm = transpose_126_perm_0, x = q_cast); + tensor qk_cast = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_130, y = transpose_129); + tensor var_3504_cast = softmax(axis = var_3439, x = qk_cast); + tensor var_3506_transpose_x_0 = const()[name = tensor("op_3506_transpose_x_0"), val = tensor(false)]; + tensor var_3506_transpose_y_0 = const()[name = tensor("op_3506_transpose_y_0"), val = tensor(false)]; + tensor transpose_131 = transpose(perm = var_3500, x = var_3499_cast); + tensor var_3506_cast = matmul(transpose_x = var_3506_transpose_x_0, transpose_y = var_3506_transpose_y_0, x = var_3504_cast, y = transpose_131); + tensor var_3507 = const()[name = tensor("op_3507"), val = tensor([0, 2, 1, 3])]; + tensor concat_31 = const()[name = tensor("concat_31"), val = tensor([1, 1500, 1280])]; + tensor transpose_128 = transpose(perm = var_3507, x = var_3506_cast); + tensor x_383_cast = reshape(shape = concat_31, x = transpose_128); + tensor var_3512_to_fp16 = const()[name = tensor("op_3512_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1244167040)))]; + tensor var_3513_to_fp16 = const()[name = tensor("op_3513_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1247443904)))]; + tensor var_3514_cast = linear(bias = var_3513_to_fp16, weight = var_3512_to_fp16, x = x_383_cast); + tensor x_385_cast = add(x = x_379_cast, y = var_3514_cast); + tensor var_3520_axes_0 = const()[name = tensor("op_3520_axes_0"), val = tensor([-1])]; + tensor blocks_31_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_31_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1247446528)))]; + tensor blocks_31_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_31_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1247449152)))]; + tensor var_3520_cast = layer_norm(axes = var_3520_axes_0, beta = blocks_31_mlp_ln_bias_to_fp16, epsilon = var_3445_to_fp16, gamma = blocks_31_mlp_ln_weight_to_fp16, x = x_385_cast); + tensor var_3529_to_fp16 = const()[name = tensor("op_3529_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1247451776)))]; + tensor var_3530_to_fp16 = const()[name = tensor("op_3530_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1260559040)))]; + tensor input_257_cast = linear(bias = var_3530_to_fp16, weight = var_3529_to_fp16, x = var_3520_cast); + tensor x_389_mode_0 = const()[name = tensor("x_389_mode_0"), val = tensor("EXACT")]; + tensor x_389_cast = gelu(mode = x_389_mode_0, x = input_257_cast); + tensor var_3535_to_fp16 = const()[name = tensor("op_3535_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1260569344)))]; + tensor var_3536_to_fp16 = const()[name = tensor("op_3536_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1273676608)))]; + tensor var_3537_cast = linear(bias = var_3536_to_fp16, weight = var_3535_to_fp16, x = x_389_cast); + tensor x_cast = add(x = x_385_cast, y = var_3537_cast); + tensor var_3550_axes_0 = const()[name = tensor("op_3550_axes_0"), val = tensor([-1])]; + tensor ln_post_weight_to_fp16 = const()[name = tensor("ln_post_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1273679232)))]; + tensor ln_post_bias_to_fp16 = const()[name = tensor("ln_post_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1273681856)))]; + tensor var_3541_to_fp16 = const()[name = tensor("op_3541_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3550_cast = layer_norm(axes = var_3550_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_3541_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast); + tensor var_3550_cast_to_fp32_dtype_0 = const()[name = tensor("op_3550_cast_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor output = cast(dtype = var_3550_cast_to_fp32_dtype_0, x = var_3550_cast); + } -> (output); +} \ No newline at end of file diff --git a/ggml-large-v1-encoder.mlmodelc/weights/weight.bin b/ggml-large-v1-encoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..cd67dbab4d1a900d4b808596e1af2998f9e498ca --- /dev/null +++ b/ggml-large-v1-encoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fe6163af221e89e96ebdd4163bddbf79670be7ee6aac34f1d662ce42f6fb014d +size 1273684480 diff --git a/ggml-large-v2-encoder.mlmodelc.zip b/ggml-large-v2-encoder.mlmodelc.zip index 90a40a16eb970b25bded31221b842037369014cc..872e90bdb5e4d68a73e118e99260de248e4adc3f 100644 --- a/ggml-large-v2-encoder.mlmodelc.zip +++ b/ggml-large-v2-encoder.mlmodelc.zip @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:c65d28737f51d09dff3b9da9d26c2ab6f5d82857c9b049ce383d64c2502a5541 -size 1174643458 +oid sha256:08d4306667aa7378a8ed9ecbdc94337fe01a2d71c9afe40c4e8dfb967f7bfd41 +size 1174643634 diff --git a/ggml-large-v2-encoder.mlmodelc/analytics/coremldata.bin b/ggml-large-v2-encoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..75f9ab6ce4798c0794f6210393494b522c7901e4 --- /dev/null +++ b/ggml-large-v2-encoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a9b73481ce4a6af49154956c7f9c72dc02d96448fa69651f5c5a2124edfc8e31 +size 243 diff --git a/ggml-large-v2-encoder.mlmodelc/coremldata.bin b/ggml-large-v2-encoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..c732ed7b8caeac3c15acda1c2836f3e1f525a5a8 --- /dev/null +++ b/ggml-large-v2-encoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:05fe28591b40616fa0c34ad7b853133623f5300923ec812acb11459c411acf3b +size 149 diff --git a/ggml-large-v2-encoder.mlmodelc/metadata.json b/ggml-large-v2-encoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..a8d90d9c1d15da639581e329405d75bac5cb0338 --- /dev/null +++ b/ggml-large-v2-encoder.mlmodelc/metadata.json @@ -0,0 +1,65 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float32", + "formattedType" : "MultiArray (Float32)", + "shortDescription" : "", + "shape" : "[]", + "name" : "output", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 6, + "mlProgramOperationTypeHistogram" : { + "Linear" : 192, + "Matmul" : 64, + "Cast" : 2, + "Conv" : 2, + "Softmax" : 32, + "Add" : 65, + "LayerNorm" : 65, + "Mul" : 64, + "Transpose" : 129, + "Gelu" : 34, + "Reshape" : 128 + }, + "computePrecision" : "Mixed (Float16, Float32, Int32)", + "isUpdatable" : "0", + "availability" : { + "macOS" : "12.0", + "tvOS" : "15.0", + "visionOS" : "1.0", + "watchOS" : "8.0", + "iOS" : "15.0", + "macCatalyst" : "15.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float32", + "formattedType" : "MultiArray (Float32 1 × 80 × 3000)", + "shortDescription" : "", + "shape" : "[1, 80, 3000]", + "name" : "logmel_data", + "type" : "MultiArray" + } + ], + "generatedClassName" : "coreml_encoder_large_v2", + "method" : "predict" + } +] \ No newline at end of file diff --git a/ggml-large-v2-encoder.mlmodelc/model.mil b/ggml-large-v2-encoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..488706ca043b4413a7addd38f7dc67c20e77e409 --- /dev/null +++ b/ggml-large-v2-encoder.mlmodelc/model.mil @@ -0,0 +1,1927 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "5.33.5"}, {"coremlc-version", "1877.40.3"}})] +{ + func main(tensor logmel_data) { + tensor var_72 = const()[name = tensor("op_72"), val = tensor(1)]; + tensor var_80 = const()[name = tensor("op_80"), val = tensor([1])]; + tensor var_82 = const()[name = tensor("op_82"), val = tensor([1])]; + tensor var_84_pad_type_0 = const()[name = tensor("op_84_pad_type_0"), val = tensor("custom")]; + tensor var_84_pad_0 = const()[name = tensor("op_84_pad_0"), val = tensor([1, 1])]; + tensor logmel_data_to_fp16_dtype_0 = const()[name = tensor("logmel_data_to_fp16_dtype_0"), val = tensor("fp16")]; + tensor weight_3_to_fp16 = const()[name = tensor("weight_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor bias_3_to_fp16 = const()[name = tensor("bias_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614528)))]; + tensor cast_967 = cast(dtype = logmel_data_to_fp16_dtype_0, x = logmel_data); + tensor var_84_cast = conv(bias = bias_3_to_fp16, dilations = var_82, groups = var_72, pad = var_84_pad_0, pad_type = var_84_pad_type_0, strides = var_80, weight = weight_3_to_fp16, x = cast_967); + tensor input_1_mode_0 = const()[name = tensor("input_1_mode_0"), val = tensor("EXACT")]; + tensor input_1_cast = gelu(mode = input_1_mode_0, x = var_84_cast); + tensor var_88 = const()[name = tensor("op_88"), val = tensor(1)]; + tensor var_97 = const()[name = tensor("op_97"), val = tensor([2])]; + tensor var_99 = const()[name = tensor("op_99"), val = tensor([1])]; + tensor var_101_pad_type_0 = const()[name = tensor("op_101_pad_type_0"), val = tensor("custom")]; + tensor var_101_pad_0 = const()[name = tensor("op_101_pad_0"), val = tensor([1, 1])]; + tensor weight_7_to_fp16 = const()[name = tensor("weight_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617152)))]; + tensor bias_7_to_fp16 = const()[name = tensor("bias_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10447616)))]; + tensor var_101_cast = conv(bias = bias_7_to_fp16, dilations = var_99, groups = var_88, pad = var_101_pad_0, pad_type = var_101_pad_type_0, strides = var_97, weight = weight_7_to_fp16, x = input_1_cast); + tensor x_3_mode_0 = const()[name = tensor("x_3_mode_0"), val = tensor("EXACT")]; + tensor x_3_cast = gelu(mode = x_3_mode_0, x = var_101_cast); + tensor var_106 = const()[name = tensor("op_106"), val = tensor([0, 2, 1])]; + tensor positional_embedding_to_fp16 = const()[name = tensor("positional_embedding_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10450240)))]; + tensor transpose_256 = transpose(perm = var_106, x = x_3_cast); + tensor var_109_cast = add(x = transpose_256, y = positional_embedding_to_fp16); + tensor var_122 = const()[name = tensor("op_122"), val = tensor(-1)]; + tensor var_139_axes_0 = const()[name = tensor("op_139_axes_0"), val = tensor([-1])]; + tensor blocks_0_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_0_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14290304)))]; + tensor blocks_0_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_0_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14292928)))]; + tensor var_128_to_fp16 = const()[name = tensor("op_128_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_139_cast = layer_norm(axes = var_139_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_128_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_109_cast); + tensor var_150_to_fp16 = const()[name = tensor("op_150_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14295552)))]; + tensor var_151_to_fp16 = const()[name = tensor("op_151_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17572416)))]; + tensor q_1_cast = linear(bias = var_151_to_fp16, weight = var_150_to_fp16, x = var_139_cast); + tensor var_154_to_fp16 = const()[name = tensor("op_154_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17575040)))]; + tensor k_1_bias_0_to_fp16 = const()[name = tensor("k_1_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20851904)))]; + tensor k_1_cast = linear(bias = k_1_bias_0_to_fp16, weight = var_154_to_fp16, x = var_139_cast); + tensor var_158_to_fp16 = const()[name = tensor("op_158_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20854528)))]; + tensor var_159_to_fp16 = const()[name = tensor("op_159_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24131392)))]; + tensor v_1_cast = linear(bias = var_159_to_fp16, weight = var_158_to_fp16, x = var_139_cast); + tensor var_167 = const()[name = tensor("op_167"), val = tensor([1, 1500, 20, -1])]; + tensor var_168_cast = reshape(shape = var_167, x = q_1_cast); + tensor const_224_to_fp16 = const()[name = tensor("const_224_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_3_cast = mul(x = var_168_cast, y = const_224_to_fp16); + tensor var_174 = const()[name = tensor("op_174"), val = tensor([1, 1500, 20, -1])]; + tensor var_175_cast = reshape(shape = var_174, x = k_1_cast); + tensor const_225_to_fp16 = const()[name = tensor("const_225_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_3_cast = mul(x = var_175_cast, y = const_225_to_fp16); + tensor var_181 = const()[name = tensor("op_181"), val = tensor([1, 1500, 20, -1])]; + tensor var_182_cast = reshape(shape = var_181, x = v_1_cast); + tensor var_183 = const()[name = tensor("op_183"), val = tensor([0, 2, 1, 3])]; + tensor qk_1_transpose_x_0 = const()[name = tensor("qk_1_transpose_x_0"), val = tensor(false)]; + tensor qk_1_transpose_y_0 = const()[name = tensor("qk_1_transpose_y_0"), val = tensor(false)]; + tensor transpose_64_perm_0 = const()[name = tensor("transpose_64_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_65_perm_0 = const()[name = tensor("transpose_65_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_253 = transpose(perm = transpose_65_perm_0, x = k_3_cast); + tensor transpose_254 = transpose(perm = transpose_64_perm_0, x = q_3_cast); + tensor qk_1_cast = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_254, y = transpose_253); + tensor var_187_cast = softmax(axis = var_122, x = qk_1_cast); + tensor var_189_transpose_x_0 = const()[name = tensor("op_189_transpose_x_0"), val = tensor(false)]; + tensor var_189_transpose_y_0 = const()[name = tensor("op_189_transpose_y_0"), val = tensor(false)]; + tensor transpose_255 = transpose(perm = var_183, x = var_182_cast); + tensor var_189_cast = matmul(transpose_x = var_189_transpose_x_0, transpose_y = var_189_transpose_y_0, x = var_187_cast, y = transpose_255); + tensor var_190 = const()[name = tensor("op_190"), val = tensor([0, 2, 1, 3])]; + tensor concat_0 = const()[name = tensor("concat_0"), val = tensor([1, 1500, 1280])]; + tensor transpose_252 = transpose(perm = var_190, x = var_189_cast); + tensor x_11_cast = reshape(shape = concat_0, x = transpose_252); + tensor var_195_to_fp16 = const()[name = tensor("op_195_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24134016)))]; + tensor var_196_to_fp16 = const()[name = tensor("op_196_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27410880)))]; + tensor var_197_cast = linear(bias = var_196_to_fp16, weight = var_195_to_fp16, x = x_11_cast); + tensor x_13_cast = add(x = var_109_cast, y = var_197_cast); + tensor var_203_axes_0 = const()[name = tensor("op_203_axes_0"), val = tensor([-1])]; + tensor blocks_0_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27413504)))]; + tensor blocks_0_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27416128)))]; + tensor var_203_cast = layer_norm(axes = var_203_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_128_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast); + tensor var_212_to_fp16 = const()[name = tensor("op_212_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27418752)))]; + tensor var_213_to_fp16 = const()[name = tensor("op_213_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40526016)))]; + tensor input_9_cast = linear(bias = var_213_to_fp16, weight = var_212_to_fp16, x = var_203_cast); + tensor x_17_mode_0 = const()[name = tensor("x_17_mode_0"), val = tensor("EXACT")]; + tensor x_17_cast = gelu(mode = x_17_mode_0, x = input_9_cast); + tensor var_218_to_fp16 = const()[name = tensor("op_218_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40536320)))]; + tensor var_219_to_fp16 = const()[name = tensor("op_219_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53643584)))]; + tensor var_220_cast = linear(bias = var_219_to_fp16, weight = var_218_to_fp16, x = x_17_cast); + tensor x_19_cast = add(x = x_13_cast, y = var_220_cast); + tensor var_229 = const()[name = tensor("op_229"), val = tensor(-1)]; + tensor var_246_axes_0 = const()[name = tensor("op_246_axes_0"), val = tensor([-1])]; + tensor blocks_1_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_1_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53646208)))]; + tensor blocks_1_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_1_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53648832)))]; + tensor var_235_to_fp16 = const()[name = tensor("op_235_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_246_cast = layer_norm(axes = var_246_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_235_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast); + tensor var_257_to_fp16 = const()[name = tensor("op_257_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53651456)))]; + tensor var_258_to_fp16 = const()[name = tensor("op_258_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56928320)))]; + tensor q_5_cast = linear(bias = var_258_to_fp16, weight = var_257_to_fp16, x = var_246_cast); + tensor var_261_to_fp16 = const()[name = tensor("op_261_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56930944)))]; + tensor k_5_bias_0_to_fp16 = const()[name = tensor("k_5_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60207808)))]; + tensor k_5_cast = linear(bias = k_5_bias_0_to_fp16, weight = var_261_to_fp16, x = var_246_cast); + tensor var_265_to_fp16 = const()[name = tensor("op_265_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60210432)))]; + tensor var_266_to_fp16 = const()[name = tensor("op_266_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63487296)))]; + tensor v_5_cast = linear(bias = var_266_to_fp16, weight = var_265_to_fp16, x = var_246_cast); + tensor var_274 = const()[name = tensor("op_274"), val = tensor([1, 1500, 20, -1])]; + tensor var_275_cast = reshape(shape = var_274, x = q_5_cast); + tensor const_226_to_fp16 = const()[name = tensor("const_226_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_7_cast = mul(x = var_275_cast, y = const_226_to_fp16); + tensor var_281 = const()[name = tensor("op_281"), val = tensor([1, 1500, 20, -1])]; + tensor var_282_cast = reshape(shape = var_281, x = k_5_cast); + tensor const_227_to_fp16 = const()[name = tensor("const_227_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_7_cast = mul(x = var_282_cast, y = const_227_to_fp16); + tensor var_288 = const()[name = tensor("op_288"), val = tensor([1, 1500, 20, -1])]; + tensor var_289_cast = reshape(shape = var_288, x = v_5_cast); + tensor var_290 = const()[name = tensor("op_290"), val = tensor([0, 2, 1, 3])]; + tensor qk_3_transpose_x_0 = const()[name = tensor("qk_3_transpose_x_0"), val = tensor(false)]; + tensor qk_3_transpose_y_0 = const()[name = tensor("qk_3_transpose_y_0"), val = tensor(false)]; + tensor transpose_66_perm_0 = const()[name = tensor("transpose_66_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_67_perm_0 = const()[name = tensor("transpose_67_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_249 = transpose(perm = transpose_67_perm_0, x = k_7_cast); + tensor transpose_250 = transpose(perm = transpose_66_perm_0, x = q_7_cast); + tensor qk_3_cast = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_250, y = transpose_249); + tensor var_294_cast = softmax(axis = var_229, x = qk_3_cast); + tensor var_296_transpose_x_0 = const()[name = tensor("op_296_transpose_x_0"), val = tensor(false)]; + tensor var_296_transpose_y_0 = const()[name = tensor("op_296_transpose_y_0"), val = tensor(false)]; + tensor transpose_251 = transpose(perm = var_290, x = var_289_cast); + tensor var_296_cast = matmul(transpose_x = var_296_transpose_x_0, transpose_y = var_296_transpose_y_0, x = var_294_cast, y = transpose_251); + tensor var_297 = const()[name = tensor("op_297"), val = tensor([0, 2, 1, 3])]; + tensor concat_1 = const()[name = tensor("concat_1"), val = tensor([1, 1500, 1280])]; + tensor transpose_248 = transpose(perm = var_297, x = var_296_cast); + tensor x_23_cast = reshape(shape = concat_1, x = transpose_248); + tensor var_302_to_fp16 = const()[name = tensor("op_302_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63489920)))]; + tensor var_303_to_fp16 = const()[name = tensor("op_303_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66766784)))]; + tensor var_304_cast = linear(bias = var_303_to_fp16, weight = var_302_to_fp16, x = x_23_cast); + tensor x_25_cast = add(x = x_19_cast, y = var_304_cast); + tensor var_310_axes_0 = const()[name = tensor("op_310_axes_0"), val = tensor([-1])]; + tensor blocks_1_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66769408)))]; + tensor blocks_1_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66772032)))]; + tensor var_310_cast = layer_norm(axes = var_310_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_235_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast); + tensor var_319_to_fp16 = const()[name = tensor("op_319_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66774656)))]; + tensor var_320_to_fp16 = const()[name = tensor("op_320_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79881920)))]; + tensor input_17_cast = linear(bias = var_320_to_fp16, weight = var_319_to_fp16, x = var_310_cast); + tensor x_29_mode_0 = const()[name = tensor("x_29_mode_0"), val = tensor("EXACT")]; + tensor x_29_cast = gelu(mode = x_29_mode_0, x = input_17_cast); + tensor var_325_to_fp16 = const()[name = tensor("op_325_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79892224)))]; + tensor var_326_to_fp16 = const()[name = tensor("op_326_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92999488)))]; + tensor var_327_cast = linear(bias = var_326_to_fp16, weight = var_325_to_fp16, x = x_29_cast); + tensor x_31_cast = add(x = x_25_cast, y = var_327_cast); + tensor var_336 = const()[name = tensor("op_336"), val = tensor(-1)]; + tensor var_353_axes_0 = const()[name = tensor("op_353_axes_0"), val = tensor([-1])]; + tensor blocks_2_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_2_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93002112)))]; + tensor blocks_2_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_2_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93004736)))]; + tensor var_342_to_fp16 = const()[name = tensor("op_342_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_353_cast = layer_norm(axes = var_353_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_342_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast); + tensor var_364_to_fp16 = const()[name = tensor("op_364_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93007360)))]; + tensor var_365_to_fp16 = const()[name = tensor("op_365_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96284224)))]; + tensor q_9_cast = linear(bias = var_365_to_fp16, weight = var_364_to_fp16, x = var_353_cast); + tensor var_368_to_fp16 = const()[name = tensor("op_368_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96286848)))]; + tensor k_9_bias_0_to_fp16 = const()[name = tensor("k_9_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99563712)))]; + tensor k_9_cast = linear(bias = k_9_bias_0_to_fp16, weight = var_368_to_fp16, x = var_353_cast); + tensor var_372_to_fp16 = const()[name = tensor("op_372_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99566336)))]; + tensor var_373_to_fp16 = const()[name = tensor("op_373_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102843200)))]; + tensor v_9_cast = linear(bias = var_373_to_fp16, weight = var_372_to_fp16, x = var_353_cast); + tensor var_381 = const()[name = tensor("op_381"), val = tensor([1, 1500, 20, -1])]; + tensor var_382_cast = reshape(shape = var_381, x = q_9_cast); + tensor const_228_to_fp16 = const()[name = tensor("const_228_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_11_cast = mul(x = var_382_cast, y = const_228_to_fp16); + tensor var_388 = const()[name = tensor("op_388"), val = tensor([1, 1500, 20, -1])]; + tensor var_389_cast = reshape(shape = var_388, x = k_9_cast); + tensor const_229_to_fp16 = const()[name = tensor("const_229_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_11_cast = mul(x = var_389_cast, y = const_229_to_fp16); + tensor var_395 = const()[name = tensor("op_395"), val = tensor([1, 1500, 20, -1])]; + tensor var_396_cast = reshape(shape = var_395, x = v_9_cast); + tensor var_397 = const()[name = tensor("op_397"), val = tensor([0, 2, 1, 3])]; + tensor qk_5_transpose_x_0 = const()[name = tensor("qk_5_transpose_x_0"), val = tensor(false)]; + tensor qk_5_transpose_y_0 = const()[name = tensor("qk_5_transpose_y_0"), val = tensor(false)]; + tensor transpose_68_perm_0 = const()[name = tensor("transpose_68_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_69_perm_0 = const()[name = tensor("transpose_69_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_245 = transpose(perm = transpose_69_perm_0, x = k_11_cast); + tensor transpose_246 = transpose(perm = transpose_68_perm_0, x = q_11_cast); + tensor qk_5_cast = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_246, y = transpose_245); + tensor var_401_cast = softmax(axis = var_336, x = qk_5_cast); + tensor var_403_transpose_x_0 = const()[name = tensor("op_403_transpose_x_0"), val = tensor(false)]; + tensor var_403_transpose_y_0 = const()[name = tensor("op_403_transpose_y_0"), val = tensor(false)]; + tensor transpose_247 = transpose(perm = var_397, x = var_396_cast); + tensor var_403_cast = matmul(transpose_x = var_403_transpose_x_0, transpose_y = var_403_transpose_y_0, x = var_401_cast, y = transpose_247); + tensor var_404 = const()[name = tensor("op_404"), val = tensor([0, 2, 1, 3])]; + tensor concat_2 = const()[name = tensor("concat_2"), val = tensor([1, 1500, 1280])]; + tensor transpose_244 = transpose(perm = var_404, x = var_403_cast); + tensor x_35_cast = reshape(shape = concat_2, x = transpose_244); + tensor var_409_to_fp16 = const()[name = tensor("op_409_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102845824)))]; + tensor var_410_to_fp16 = const()[name = tensor("op_410_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106122688)))]; + tensor var_411_cast = linear(bias = var_410_to_fp16, weight = var_409_to_fp16, x = x_35_cast); + tensor x_37_cast = add(x = x_31_cast, y = var_411_cast); + tensor var_417_axes_0 = const()[name = tensor("op_417_axes_0"), val = tensor([-1])]; + tensor blocks_2_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106125312)))]; + tensor blocks_2_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106127936)))]; + tensor var_417_cast = layer_norm(axes = var_417_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_342_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast); + tensor var_426_to_fp16 = const()[name = tensor("op_426_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106130560)))]; + tensor var_427_to_fp16 = const()[name = tensor("op_427_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119237824)))]; + tensor input_25_cast = linear(bias = var_427_to_fp16, weight = var_426_to_fp16, x = var_417_cast); + tensor x_41_mode_0 = const()[name = tensor("x_41_mode_0"), val = tensor("EXACT")]; + tensor x_41_cast = gelu(mode = x_41_mode_0, x = input_25_cast); + tensor var_432_to_fp16 = const()[name = tensor("op_432_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119248128)))]; + tensor var_433_to_fp16 = const()[name = tensor("op_433_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132355392)))]; + tensor var_434_cast = linear(bias = var_433_to_fp16, weight = var_432_to_fp16, x = x_41_cast); + tensor x_43_cast = add(x = x_37_cast, y = var_434_cast); + tensor var_443 = const()[name = tensor("op_443"), val = tensor(-1)]; + tensor var_460_axes_0 = const()[name = tensor("op_460_axes_0"), val = tensor([-1])]; + tensor blocks_3_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_3_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132358016)))]; + tensor blocks_3_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_3_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132360640)))]; + tensor var_449_to_fp16 = const()[name = tensor("op_449_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_460_cast = layer_norm(axes = var_460_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_449_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast); + tensor var_471_to_fp16 = const()[name = tensor("op_471_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132363264)))]; + tensor var_472_to_fp16 = const()[name = tensor("op_472_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135640128)))]; + tensor q_13_cast = linear(bias = var_472_to_fp16, weight = var_471_to_fp16, x = var_460_cast); + tensor var_475_to_fp16 = const()[name = tensor("op_475_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135642752)))]; + tensor k_13_bias_0_to_fp16 = const()[name = tensor("k_13_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138919616)))]; + tensor k_13_cast = linear(bias = k_13_bias_0_to_fp16, weight = var_475_to_fp16, x = var_460_cast); + tensor var_479_to_fp16 = const()[name = tensor("op_479_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138922240)))]; + tensor var_480_to_fp16 = const()[name = tensor("op_480_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142199104)))]; + tensor v_13_cast = linear(bias = var_480_to_fp16, weight = var_479_to_fp16, x = var_460_cast); + tensor var_488 = const()[name = tensor("op_488"), val = tensor([1, 1500, 20, -1])]; + tensor var_489_cast = reshape(shape = var_488, x = q_13_cast); + tensor const_230_to_fp16 = const()[name = tensor("const_230_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_15_cast = mul(x = var_489_cast, y = const_230_to_fp16); + tensor var_495 = const()[name = tensor("op_495"), val = tensor([1, 1500, 20, -1])]; + tensor var_496_cast = reshape(shape = var_495, x = k_13_cast); + tensor const_231_to_fp16 = const()[name = tensor("const_231_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_15_cast = mul(x = var_496_cast, y = const_231_to_fp16); + tensor var_502 = const()[name = tensor("op_502"), val = tensor([1, 1500, 20, -1])]; + tensor var_503_cast = reshape(shape = var_502, x = v_13_cast); + tensor var_504 = const()[name = tensor("op_504"), val = tensor([0, 2, 1, 3])]; + tensor qk_7_transpose_x_0 = const()[name = tensor("qk_7_transpose_x_0"), val = tensor(false)]; + tensor qk_7_transpose_y_0 = const()[name = tensor("qk_7_transpose_y_0"), val = tensor(false)]; + tensor transpose_70_perm_0 = const()[name = tensor("transpose_70_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_71_perm_0 = const()[name = tensor("transpose_71_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_241 = transpose(perm = transpose_71_perm_0, x = k_15_cast); + tensor transpose_242 = transpose(perm = transpose_70_perm_0, x = q_15_cast); + tensor qk_7_cast = matmul(transpose_x = qk_7_transpose_x_0, transpose_y = qk_7_transpose_y_0, x = transpose_242, y = transpose_241); + tensor var_508_cast = softmax(axis = var_443, x = qk_7_cast); + tensor var_510_transpose_x_0 = const()[name = tensor("op_510_transpose_x_0"), val = tensor(false)]; + tensor var_510_transpose_y_0 = const()[name = tensor("op_510_transpose_y_0"), val = tensor(false)]; + tensor transpose_243 = transpose(perm = var_504, x = var_503_cast); + tensor var_510_cast = matmul(transpose_x = var_510_transpose_x_0, transpose_y = var_510_transpose_y_0, x = var_508_cast, y = transpose_243); + tensor var_511 = const()[name = tensor("op_511"), val = tensor([0, 2, 1, 3])]; + tensor concat_3 = const()[name = tensor("concat_3"), val = tensor([1, 1500, 1280])]; + tensor transpose_240 = transpose(perm = var_511, x = var_510_cast); + tensor x_47_cast = reshape(shape = concat_3, x = transpose_240); + tensor var_516_to_fp16 = const()[name = tensor("op_516_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142201728)))]; + tensor var_517_to_fp16 = const()[name = tensor("op_517_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145478592)))]; + tensor var_518_cast = linear(bias = var_517_to_fp16, weight = var_516_to_fp16, x = x_47_cast); + tensor x_49_cast = add(x = x_43_cast, y = var_518_cast); + tensor var_524_axes_0 = const()[name = tensor("op_524_axes_0"), val = tensor([-1])]; + tensor blocks_3_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145481216)))]; + tensor blocks_3_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145483840)))]; + tensor var_524_cast = layer_norm(axes = var_524_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_449_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast); + tensor var_533_to_fp16 = const()[name = tensor("op_533_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145486464)))]; + tensor var_534_to_fp16 = const()[name = tensor("op_534_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158593728)))]; + tensor input_33_cast = linear(bias = var_534_to_fp16, weight = var_533_to_fp16, x = var_524_cast); + tensor x_53_mode_0 = const()[name = tensor("x_53_mode_0"), val = tensor("EXACT")]; + tensor x_53_cast = gelu(mode = x_53_mode_0, x = input_33_cast); + tensor var_539_to_fp16 = const()[name = tensor("op_539_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158604032)))]; + tensor var_540_to_fp16 = const()[name = tensor("op_540_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171711296)))]; + tensor var_541_cast = linear(bias = var_540_to_fp16, weight = var_539_to_fp16, x = x_53_cast); + tensor x_55_cast = add(x = x_49_cast, y = var_541_cast); + tensor var_550 = const()[name = tensor("op_550"), val = tensor(-1)]; + tensor var_567_axes_0 = const()[name = tensor("op_567_axes_0"), val = tensor([-1])]; + tensor blocks_4_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_4_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171713920)))]; + tensor blocks_4_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_4_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171716544)))]; + tensor var_556_to_fp16 = const()[name = tensor("op_556_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_567_cast = layer_norm(axes = var_567_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_556_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast); + tensor var_578_to_fp16 = const()[name = tensor("op_578_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171719168)))]; + tensor var_579_to_fp16 = const()[name = tensor("op_579_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174996032)))]; + tensor q_17_cast = linear(bias = var_579_to_fp16, weight = var_578_to_fp16, x = var_567_cast); + tensor var_582_to_fp16 = const()[name = tensor("op_582_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174998656)))]; + tensor k_17_bias_0_to_fp16 = const()[name = tensor("k_17_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178275520)))]; + tensor k_17_cast = linear(bias = k_17_bias_0_to_fp16, weight = var_582_to_fp16, x = var_567_cast); + tensor var_586_to_fp16 = const()[name = tensor("op_586_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178278144)))]; + tensor var_587_to_fp16 = const()[name = tensor("op_587_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(181555008)))]; + tensor v_17_cast = linear(bias = var_587_to_fp16, weight = var_586_to_fp16, x = var_567_cast); + tensor var_595 = const()[name = tensor("op_595"), val = tensor([1, 1500, 20, -1])]; + tensor var_596_cast = reshape(shape = var_595, x = q_17_cast); + tensor const_232_to_fp16 = const()[name = tensor("const_232_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_19_cast = mul(x = var_596_cast, y = const_232_to_fp16); + tensor var_602 = const()[name = tensor("op_602"), val = tensor([1, 1500, 20, -1])]; + tensor var_603_cast = reshape(shape = var_602, x = k_17_cast); + tensor const_233_to_fp16 = const()[name = tensor("const_233_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_19_cast = mul(x = var_603_cast, y = const_233_to_fp16); + tensor var_609 = const()[name = tensor("op_609"), val = tensor([1, 1500, 20, -1])]; + tensor var_610_cast = reshape(shape = var_609, x = v_17_cast); + tensor var_611 = const()[name = tensor("op_611"), val = tensor([0, 2, 1, 3])]; + tensor qk_9_transpose_x_0 = const()[name = tensor("qk_9_transpose_x_0"), val = tensor(false)]; + tensor qk_9_transpose_y_0 = const()[name = tensor("qk_9_transpose_y_0"), val = tensor(false)]; + tensor transpose_72_perm_0 = const()[name = tensor("transpose_72_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_73_perm_0 = const()[name = tensor("transpose_73_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_237 = transpose(perm = transpose_73_perm_0, x = k_19_cast); + tensor transpose_238 = transpose(perm = transpose_72_perm_0, x = q_19_cast); + tensor qk_9_cast = matmul(transpose_x = qk_9_transpose_x_0, transpose_y = qk_9_transpose_y_0, x = transpose_238, y = transpose_237); + tensor var_615_cast = softmax(axis = var_550, x = qk_9_cast); + tensor var_617_transpose_x_0 = const()[name = tensor("op_617_transpose_x_0"), val = tensor(false)]; + tensor var_617_transpose_y_0 = const()[name = tensor("op_617_transpose_y_0"), val = tensor(false)]; + tensor transpose_239 = transpose(perm = var_611, x = var_610_cast); + tensor var_617_cast = matmul(transpose_x = var_617_transpose_x_0, transpose_y = var_617_transpose_y_0, x = var_615_cast, y = transpose_239); + tensor var_618 = const()[name = tensor("op_618"), val = tensor([0, 2, 1, 3])]; + tensor concat_4 = const()[name = tensor("concat_4"), val = tensor([1, 1500, 1280])]; + tensor transpose_236 = transpose(perm = var_618, x = var_617_cast); + tensor x_59_cast = reshape(shape = concat_4, x = transpose_236); + tensor var_623_to_fp16 = const()[name = tensor("op_623_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(181557632)))]; + tensor var_624_to_fp16 = const()[name = tensor("op_624_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184834496)))]; + tensor var_625_cast = linear(bias = var_624_to_fp16, weight = var_623_to_fp16, x = x_59_cast); + tensor x_61_cast = add(x = x_55_cast, y = var_625_cast); + tensor var_631_axes_0 = const()[name = tensor("op_631_axes_0"), val = tensor([-1])]; + tensor blocks_4_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184837120)))]; + tensor blocks_4_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184839744)))]; + tensor var_631_cast = layer_norm(axes = var_631_axes_0, beta = blocks_4_mlp_ln_bias_to_fp16, epsilon = var_556_to_fp16, gamma = blocks_4_mlp_ln_weight_to_fp16, x = x_61_cast); + tensor var_640_to_fp16 = const()[name = tensor("op_640_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184842368)))]; + tensor var_641_to_fp16 = const()[name = tensor("op_641_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(197949632)))]; + tensor input_41_cast = linear(bias = var_641_to_fp16, weight = var_640_to_fp16, x = var_631_cast); + tensor x_65_mode_0 = const()[name = tensor("x_65_mode_0"), val = tensor("EXACT")]; + tensor x_65_cast = gelu(mode = x_65_mode_0, x = input_41_cast); + tensor var_646_to_fp16 = const()[name = tensor("op_646_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(197959936)))]; + tensor var_647_to_fp16 = const()[name = tensor("op_647_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211067200)))]; + tensor var_648_cast = linear(bias = var_647_to_fp16, weight = var_646_to_fp16, x = x_65_cast); + tensor x_67_cast = add(x = x_61_cast, y = var_648_cast); + tensor var_657 = const()[name = tensor("op_657"), val = tensor(-1)]; + tensor var_674_axes_0 = const()[name = tensor("op_674_axes_0"), val = tensor([-1])]; + tensor blocks_5_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_5_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211069824)))]; + tensor blocks_5_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_5_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211072448)))]; + tensor var_663_to_fp16 = const()[name = tensor("op_663_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_674_cast = layer_norm(axes = var_674_axes_0, beta = blocks_5_attn_ln_bias_to_fp16, epsilon = var_663_to_fp16, gamma = blocks_5_attn_ln_weight_to_fp16, x = x_67_cast); + tensor var_685_to_fp16 = const()[name = tensor("op_685_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211075072)))]; + tensor var_686_to_fp16 = const()[name = tensor("op_686_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214351936)))]; + tensor q_21_cast = linear(bias = var_686_to_fp16, weight = var_685_to_fp16, x = var_674_cast); + tensor var_689_to_fp16 = const()[name = tensor("op_689_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214354560)))]; + tensor k_21_bias_0_to_fp16 = const()[name = tensor("k_21_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217631424)))]; + tensor k_21_cast = linear(bias = k_21_bias_0_to_fp16, weight = var_689_to_fp16, x = var_674_cast); + tensor var_693_to_fp16 = const()[name = tensor("op_693_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217634048)))]; + tensor var_694_to_fp16 = const()[name = tensor("op_694_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(220910912)))]; + tensor v_21_cast = linear(bias = var_694_to_fp16, weight = var_693_to_fp16, x = var_674_cast); + tensor var_702 = const()[name = tensor("op_702"), val = tensor([1, 1500, 20, -1])]; + tensor var_703_cast = reshape(shape = var_702, x = q_21_cast); + tensor const_234_to_fp16 = const()[name = tensor("const_234_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_23_cast = mul(x = var_703_cast, y = const_234_to_fp16); + tensor var_709 = const()[name = tensor("op_709"), val = tensor([1, 1500, 20, -1])]; + tensor var_710_cast = reshape(shape = var_709, x = k_21_cast); + tensor const_235_to_fp16 = const()[name = tensor("const_235_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_23_cast = mul(x = var_710_cast, y = const_235_to_fp16); + tensor var_716 = const()[name = tensor("op_716"), val = tensor([1, 1500, 20, -1])]; + tensor var_717_cast = reshape(shape = var_716, x = v_21_cast); + tensor var_718 = const()[name = tensor("op_718"), val = tensor([0, 2, 1, 3])]; + tensor qk_11_transpose_x_0 = const()[name = tensor("qk_11_transpose_x_0"), val = tensor(false)]; + tensor qk_11_transpose_y_0 = const()[name = tensor("qk_11_transpose_y_0"), val = tensor(false)]; + tensor transpose_74_perm_0 = const()[name = tensor("transpose_74_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_75_perm_0 = const()[name = tensor("transpose_75_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_233 = transpose(perm = transpose_75_perm_0, x = k_23_cast); + tensor transpose_234 = transpose(perm = transpose_74_perm_0, x = q_23_cast); + tensor qk_11_cast = matmul(transpose_x = qk_11_transpose_x_0, transpose_y = qk_11_transpose_y_0, x = transpose_234, y = transpose_233); + tensor var_722_cast = softmax(axis = var_657, x = qk_11_cast); + tensor var_724_transpose_x_0 = const()[name = tensor("op_724_transpose_x_0"), val = tensor(false)]; + tensor var_724_transpose_y_0 = const()[name = tensor("op_724_transpose_y_0"), val = tensor(false)]; + tensor transpose_235 = transpose(perm = var_718, x = var_717_cast); + tensor var_724_cast = matmul(transpose_x = var_724_transpose_x_0, transpose_y = var_724_transpose_y_0, x = var_722_cast, y = transpose_235); + tensor var_725 = const()[name = tensor("op_725"), val = tensor([0, 2, 1, 3])]; + tensor concat_5 = const()[name = tensor("concat_5"), val = tensor([1, 1500, 1280])]; + tensor transpose_232 = transpose(perm = var_725, x = var_724_cast); + tensor x_71_cast = reshape(shape = concat_5, x = transpose_232); + tensor var_730_to_fp16 = const()[name = tensor("op_730_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(220913536)))]; + tensor var_731_to_fp16 = const()[name = tensor("op_731_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224190400)))]; + tensor var_732_cast = linear(bias = var_731_to_fp16, weight = var_730_to_fp16, x = x_71_cast); + tensor x_73_cast = add(x = x_67_cast, y = var_732_cast); + tensor var_738_axes_0 = const()[name = tensor("op_738_axes_0"), val = tensor([-1])]; + tensor blocks_5_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224193024)))]; + tensor blocks_5_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224195648)))]; + tensor var_738_cast = layer_norm(axes = var_738_axes_0, beta = blocks_5_mlp_ln_bias_to_fp16, epsilon = var_663_to_fp16, gamma = blocks_5_mlp_ln_weight_to_fp16, x = x_73_cast); + tensor var_747_to_fp16 = const()[name = tensor("op_747_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224198272)))]; + tensor var_748_to_fp16 = const()[name = tensor("op_748_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237305536)))]; + tensor input_49_cast = linear(bias = var_748_to_fp16, weight = var_747_to_fp16, x = var_738_cast); + tensor x_77_mode_0 = const()[name = tensor("x_77_mode_0"), val = tensor("EXACT")]; + tensor x_77_cast = gelu(mode = x_77_mode_0, x = input_49_cast); + tensor var_753_to_fp16 = const()[name = tensor("op_753_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237315840)))]; + tensor var_754_to_fp16 = const()[name = tensor("op_754_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250423104)))]; + tensor var_755_cast = linear(bias = var_754_to_fp16, weight = var_753_to_fp16, x = x_77_cast); + tensor x_79_cast = add(x = x_73_cast, y = var_755_cast); + tensor var_764 = const()[name = tensor("op_764"), val = tensor(-1)]; + tensor var_781_axes_0 = const()[name = tensor("op_781_axes_0"), val = tensor([-1])]; + tensor blocks_6_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_6_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250425728)))]; + tensor blocks_6_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_6_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250428352)))]; + tensor var_770_to_fp16 = const()[name = tensor("op_770_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_781_cast = layer_norm(axes = var_781_axes_0, beta = blocks_6_attn_ln_bias_to_fp16, epsilon = var_770_to_fp16, gamma = blocks_6_attn_ln_weight_to_fp16, x = x_79_cast); + tensor var_792_to_fp16 = const()[name = tensor("op_792_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250430976)))]; + tensor var_793_to_fp16 = const()[name = tensor("op_793_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(253707840)))]; + tensor q_25_cast = linear(bias = var_793_to_fp16, weight = var_792_to_fp16, x = var_781_cast); + tensor var_796_to_fp16 = const()[name = tensor("op_796_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(253710464)))]; + tensor k_25_bias_0_to_fp16 = const()[name = tensor("k_25_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(256987328)))]; + tensor k_25_cast = linear(bias = k_25_bias_0_to_fp16, weight = var_796_to_fp16, x = var_781_cast); + tensor var_800_to_fp16 = const()[name = tensor("op_800_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(256989952)))]; + tensor var_801_to_fp16 = const()[name = tensor("op_801_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260266816)))]; + tensor v_25_cast = linear(bias = var_801_to_fp16, weight = var_800_to_fp16, x = var_781_cast); + tensor var_809 = const()[name = tensor("op_809"), val = tensor([1, 1500, 20, -1])]; + tensor var_810_cast = reshape(shape = var_809, x = q_25_cast); + tensor const_236_to_fp16 = const()[name = tensor("const_236_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_27_cast = mul(x = var_810_cast, y = const_236_to_fp16); + tensor var_816 = const()[name = tensor("op_816"), val = tensor([1, 1500, 20, -1])]; + tensor var_817_cast = reshape(shape = var_816, x = k_25_cast); + tensor const_237_to_fp16 = const()[name = tensor("const_237_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_27_cast = mul(x = var_817_cast, y = const_237_to_fp16); + tensor var_823 = const()[name = tensor("op_823"), val = tensor([1, 1500, 20, -1])]; + tensor var_824_cast = reshape(shape = var_823, x = v_25_cast); + tensor var_825 = const()[name = tensor("op_825"), val = tensor([0, 2, 1, 3])]; + tensor qk_13_transpose_x_0 = const()[name = tensor("qk_13_transpose_x_0"), val = tensor(false)]; + tensor qk_13_transpose_y_0 = const()[name = tensor("qk_13_transpose_y_0"), val = tensor(false)]; + tensor transpose_76_perm_0 = const()[name = tensor("transpose_76_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_77_perm_0 = const()[name = tensor("transpose_77_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_229 = transpose(perm = transpose_77_perm_0, x = k_27_cast); + tensor transpose_230 = transpose(perm = transpose_76_perm_0, x = q_27_cast); + tensor qk_13_cast = matmul(transpose_x = qk_13_transpose_x_0, transpose_y = qk_13_transpose_y_0, x = transpose_230, y = transpose_229); + tensor var_829_cast = softmax(axis = var_764, x = qk_13_cast); + tensor var_831_transpose_x_0 = const()[name = tensor("op_831_transpose_x_0"), val = tensor(false)]; + tensor var_831_transpose_y_0 = const()[name = tensor("op_831_transpose_y_0"), val = tensor(false)]; + tensor transpose_231 = transpose(perm = var_825, x = var_824_cast); + tensor var_831_cast = matmul(transpose_x = var_831_transpose_x_0, transpose_y = var_831_transpose_y_0, x = var_829_cast, y = transpose_231); + tensor var_832 = const()[name = tensor("op_832"), val = tensor([0, 2, 1, 3])]; + tensor concat_6 = const()[name = tensor("concat_6"), val = tensor([1, 1500, 1280])]; + tensor transpose_228 = transpose(perm = var_832, x = var_831_cast); + tensor x_83_cast = reshape(shape = concat_6, x = transpose_228); + tensor var_837_to_fp16 = const()[name = tensor("op_837_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260269440)))]; + tensor var_838_to_fp16 = const()[name = tensor("op_838_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263546304)))]; + tensor var_839_cast = linear(bias = var_838_to_fp16, weight = var_837_to_fp16, x = x_83_cast); + tensor x_85_cast = add(x = x_79_cast, y = var_839_cast); + tensor var_845_axes_0 = const()[name = tensor("op_845_axes_0"), val = tensor([-1])]; + tensor blocks_6_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_6_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263548928)))]; + tensor blocks_6_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_6_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263551552)))]; + tensor var_845_cast = layer_norm(axes = var_845_axes_0, beta = blocks_6_mlp_ln_bias_to_fp16, epsilon = var_770_to_fp16, gamma = blocks_6_mlp_ln_weight_to_fp16, x = x_85_cast); + tensor var_854_to_fp16 = const()[name = tensor("op_854_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263554176)))]; + tensor var_855_to_fp16 = const()[name = tensor("op_855_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(276661440)))]; + tensor input_57_cast = linear(bias = var_855_to_fp16, weight = var_854_to_fp16, x = var_845_cast); + tensor x_89_mode_0 = const()[name = tensor("x_89_mode_0"), val = tensor("EXACT")]; + tensor x_89_cast = gelu(mode = x_89_mode_0, x = input_57_cast); + tensor var_860_to_fp16 = const()[name = tensor("op_860_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(276671744)))]; + tensor var_861_to_fp16 = const()[name = tensor("op_861_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289779008)))]; + tensor var_862_cast = linear(bias = var_861_to_fp16, weight = var_860_to_fp16, x = x_89_cast); + tensor x_91_cast = add(x = x_85_cast, y = var_862_cast); + tensor var_871 = const()[name = tensor("op_871"), val = tensor(-1)]; + tensor var_888_axes_0 = const()[name = tensor("op_888_axes_0"), val = tensor([-1])]; + tensor blocks_7_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_7_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289781632)))]; + tensor blocks_7_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_7_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289784256)))]; + tensor var_877_to_fp16 = const()[name = tensor("op_877_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_888_cast = layer_norm(axes = var_888_axes_0, beta = blocks_7_attn_ln_bias_to_fp16, epsilon = var_877_to_fp16, gamma = blocks_7_attn_ln_weight_to_fp16, x = x_91_cast); + tensor var_899_to_fp16 = const()[name = tensor("op_899_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289786880)))]; + tensor var_900_to_fp16 = const()[name = tensor("op_900_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293063744)))]; + tensor q_29_cast = linear(bias = var_900_to_fp16, weight = var_899_to_fp16, x = var_888_cast); + tensor var_903_to_fp16 = const()[name = tensor("op_903_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293066368)))]; + tensor k_29_bias_0_to_fp16 = const()[name = tensor("k_29_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(296343232)))]; + tensor k_29_cast = linear(bias = k_29_bias_0_to_fp16, weight = var_903_to_fp16, x = var_888_cast); + tensor var_907_to_fp16 = const()[name = tensor("op_907_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(296345856)))]; + tensor var_908_to_fp16 = const()[name = tensor("op_908_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(299622720)))]; + tensor v_29_cast = linear(bias = var_908_to_fp16, weight = var_907_to_fp16, x = var_888_cast); + tensor var_916 = const()[name = tensor("op_916"), val = tensor([1, 1500, 20, -1])]; + tensor var_917_cast = reshape(shape = var_916, x = q_29_cast); + tensor const_238_to_fp16 = const()[name = tensor("const_238_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_31_cast = mul(x = var_917_cast, y = const_238_to_fp16); + tensor var_923 = const()[name = tensor("op_923"), val = tensor([1, 1500, 20, -1])]; + tensor var_924_cast = reshape(shape = var_923, x = k_29_cast); + tensor const_239_to_fp16 = const()[name = tensor("const_239_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_31_cast = mul(x = var_924_cast, y = const_239_to_fp16); + tensor var_930 = const()[name = tensor("op_930"), val = tensor([1, 1500, 20, -1])]; + tensor var_931_cast = reshape(shape = var_930, x = v_29_cast); + tensor var_932 = const()[name = tensor("op_932"), val = tensor([0, 2, 1, 3])]; + tensor qk_15_transpose_x_0 = const()[name = tensor("qk_15_transpose_x_0"), val = tensor(false)]; + tensor qk_15_transpose_y_0 = const()[name = tensor("qk_15_transpose_y_0"), val = tensor(false)]; + tensor transpose_78_perm_0 = const()[name = tensor("transpose_78_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_79_perm_0 = const()[name = tensor("transpose_79_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_225 = transpose(perm = transpose_79_perm_0, x = k_31_cast); + tensor transpose_226 = transpose(perm = transpose_78_perm_0, x = q_31_cast); + tensor qk_15_cast = matmul(transpose_x = qk_15_transpose_x_0, transpose_y = qk_15_transpose_y_0, x = transpose_226, y = transpose_225); + tensor var_936_cast = softmax(axis = var_871, x = qk_15_cast); + tensor var_938_transpose_x_0 = const()[name = tensor("op_938_transpose_x_0"), val = tensor(false)]; + tensor var_938_transpose_y_0 = const()[name = tensor("op_938_transpose_y_0"), val = tensor(false)]; + tensor transpose_227 = transpose(perm = var_932, x = var_931_cast); + tensor var_938_cast = matmul(transpose_x = var_938_transpose_x_0, transpose_y = var_938_transpose_y_0, x = var_936_cast, y = transpose_227); + tensor var_939 = const()[name = tensor("op_939"), val = tensor([0, 2, 1, 3])]; + tensor concat_7 = const()[name = tensor("concat_7"), val = tensor([1, 1500, 1280])]; + tensor transpose_224 = transpose(perm = var_939, x = var_938_cast); + tensor x_95_cast = reshape(shape = concat_7, x = transpose_224); + tensor var_944_to_fp16 = const()[name = tensor("op_944_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(299625344)))]; + tensor var_945_to_fp16 = const()[name = tensor("op_945_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302902208)))]; + tensor var_946_cast = linear(bias = var_945_to_fp16, weight = var_944_to_fp16, x = x_95_cast); + tensor x_97_cast = add(x = x_91_cast, y = var_946_cast); + tensor var_952_axes_0 = const()[name = tensor("op_952_axes_0"), val = tensor([-1])]; + tensor blocks_7_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_7_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302904832)))]; + tensor blocks_7_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_7_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302907456)))]; + tensor var_952_cast = layer_norm(axes = var_952_axes_0, beta = blocks_7_mlp_ln_bias_to_fp16, epsilon = var_877_to_fp16, gamma = blocks_7_mlp_ln_weight_to_fp16, x = x_97_cast); + tensor var_961_to_fp16 = const()[name = tensor("op_961_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302910080)))]; + tensor var_962_to_fp16 = const()[name = tensor("op_962_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316017344)))]; + tensor input_65_cast = linear(bias = var_962_to_fp16, weight = var_961_to_fp16, x = var_952_cast); + tensor x_101_mode_0 = const()[name = tensor("x_101_mode_0"), val = tensor("EXACT")]; + tensor x_101_cast = gelu(mode = x_101_mode_0, x = input_65_cast); + tensor var_967_to_fp16 = const()[name = tensor("op_967_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316027648)))]; + tensor var_968_to_fp16 = const()[name = tensor("op_968_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329134912)))]; + tensor var_969_cast = linear(bias = var_968_to_fp16, weight = var_967_to_fp16, x = x_101_cast); + tensor x_103_cast = add(x = x_97_cast, y = var_969_cast); + tensor var_978 = const()[name = tensor("op_978"), val = tensor(-1)]; + tensor var_995_axes_0 = const()[name = tensor("op_995_axes_0"), val = tensor([-1])]; + tensor blocks_8_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_8_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329137536)))]; + tensor blocks_8_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_8_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329140160)))]; + tensor var_984_to_fp16 = const()[name = tensor("op_984_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_995_cast = layer_norm(axes = var_995_axes_0, beta = blocks_8_attn_ln_bias_to_fp16, epsilon = var_984_to_fp16, gamma = blocks_8_attn_ln_weight_to_fp16, x = x_103_cast); + tensor var_1006_to_fp16 = const()[name = tensor("op_1006_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329142784)))]; + tensor var_1007_to_fp16 = const()[name = tensor("op_1007_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(332419648)))]; + tensor q_33_cast = linear(bias = var_1007_to_fp16, weight = var_1006_to_fp16, x = var_995_cast); + tensor var_1010_to_fp16 = const()[name = tensor("op_1010_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(332422272)))]; + tensor k_33_bias_0_to_fp16 = const()[name = tensor("k_33_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335699136)))]; + tensor k_33_cast = linear(bias = k_33_bias_0_to_fp16, weight = var_1010_to_fp16, x = var_995_cast); + tensor var_1014_to_fp16 = const()[name = tensor("op_1014_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335701760)))]; + tensor var_1015_to_fp16 = const()[name = tensor("op_1015_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(338978624)))]; + tensor v_33_cast = linear(bias = var_1015_to_fp16, weight = var_1014_to_fp16, x = var_995_cast); + tensor var_1023 = const()[name = tensor("op_1023"), val = tensor([1, 1500, 20, -1])]; + tensor var_1024_cast = reshape(shape = var_1023, x = q_33_cast); + tensor const_240_to_fp16 = const()[name = tensor("const_240_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_35_cast = mul(x = var_1024_cast, y = const_240_to_fp16); + tensor var_1030 = const()[name = tensor("op_1030"), val = tensor([1, 1500, 20, -1])]; + tensor var_1031_cast = reshape(shape = var_1030, x = k_33_cast); + tensor const_241_to_fp16 = const()[name = tensor("const_241_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_35_cast = mul(x = var_1031_cast, y = const_241_to_fp16); + tensor var_1037 = const()[name = tensor("op_1037"), val = tensor([1, 1500, 20, -1])]; + tensor var_1038_cast = reshape(shape = var_1037, x = v_33_cast); + tensor var_1039 = const()[name = tensor("op_1039"), val = tensor([0, 2, 1, 3])]; + tensor qk_17_transpose_x_0 = const()[name = tensor("qk_17_transpose_x_0"), val = tensor(false)]; + tensor qk_17_transpose_y_0 = const()[name = tensor("qk_17_transpose_y_0"), val = tensor(false)]; + tensor transpose_80_perm_0 = const()[name = tensor("transpose_80_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_81_perm_0 = const()[name = tensor("transpose_81_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_221 = transpose(perm = transpose_81_perm_0, x = k_35_cast); + tensor transpose_222 = transpose(perm = transpose_80_perm_0, x = q_35_cast); + tensor qk_17_cast = matmul(transpose_x = qk_17_transpose_x_0, transpose_y = qk_17_transpose_y_0, x = transpose_222, y = transpose_221); + tensor var_1043_cast = softmax(axis = var_978, x = qk_17_cast); + tensor var_1045_transpose_x_0 = const()[name = tensor("op_1045_transpose_x_0"), val = tensor(false)]; + tensor var_1045_transpose_y_0 = const()[name = tensor("op_1045_transpose_y_0"), val = tensor(false)]; + tensor transpose_223 = transpose(perm = var_1039, x = var_1038_cast); + tensor var_1045_cast = matmul(transpose_x = var_1045_transpose_x_0, transpose_y = var_1045_transpose_y_0, x = var_1043_cast, y = transpose_223); + tensor var_1046 = const()[name = tensor("op_1046"), val = tensor([0, 2, 1, 3])]; + tensor concat_8 = const()[name = tensor("concat_8"), val = tensor([1, 1500, 1280])]; + tensor transpose_220 = transpose(perm = var_1046, x = var_1045_cast); + tensor x_107_cast = reshape(shape = concat_8, x = transpose_220); + tensor var_1051_to_fp16 = const()[name = tensor("op_1051_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(338981248)))]; + tensor var_1052_to_fp16 = const()[name = tensor("op_1052_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342258112)))]; + tensor var_1053_cast = linear(bias = var_1052_to_fp16, weight = var_1051_to_fp16, x = x_107_cast); + tensor x_109_cast = add(x = x_103_cast, y = var_1053_cast); + tensor var_1059_axes_0 = const()[name = tensor("op_1059_axes_0"), val = tensor([-1])]; + tensor blocks_8_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_8_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342260736)))]; + tensor blocks_8_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_8_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342263360)))]; + tensor var_1059_cast = layer_norm(axes = var_1059_axes_0, beta = blocks_8_mlp_ln_bias_to_fp16, epsilon = var_984_to_fp16, gamma = blocks_8_mlp_ln_weight_to_fp16, x = x_109_cast); + tensor var_1068_to_fp16 = const()[name = tensor("op_1068_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342265984)))]; + tensor var_1069_to_fp16 = const()[name = tensor("op_1069_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(355373248)))]; + tensor input_73_cast = linear(bias = var_1069_to_fp16, weight = var_1068_to_fp16, x = var_1059_cast); + tensor x_113_mode_0 = const()[name = tensor("x_113_mode_0"), val = tensor("EXACT")]; + tensor x_113_cast = gelu(mode = x_113_mode_0, x = input_73_cast); + tensor var_1074_to_fp16 = const()[name = tensor("op_1074_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(355383552)))]; + tensor var_1075_to_fp16 = const()[name = tensor("op_1075_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368490816)))]; + tensor var_1076_cast = linear(bias = var_1075_to_fp16, weight = var_1074_to_fp16, x = x_113_cast); + tensor x_115_cast = add(x = x_109_cast, y = var_1076_cast); + tensor var_1085 = const()[name = tensor("op_1085"), val = tensor(-1)]; + tensor var_1102_axes_0 = const()[name = tensor("op_1102_axes_0"), val = tensor([-1])]; + tensor blocks_9_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_9_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368493440)))]; + tensor blocks_9_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_9_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368496064)))]; + tensor var_1091_to_fp16 = const()[name = tensor("op_1091_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1102_cast = layer_norm(axes = var_1102_axes_0, beta = blocks_9_attn_ln_bias_to_fp16, epsilon = var_1091_to_fp16, gamma = blocks_9_attn_ln_weight_to_fp16, x = x_115_cast); + tensor var_1113_to_fp16 = const()[name = tensor("op_1113_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368498688)))]; + tensor var_1114_to_fp16 = const()[name = tensor("op_1114_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(371775552)))]; + tensor q_37_cast = linear(bias = var_1114_to_fp16, weight = var_1113_to_fp16, x = var_1102_cast); + tensor var_1117_to_fp16 = const()[name = tensor("op_1117_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(371778176)))]; + tensor k_37_bias_0_to_fp16 = const()[name = tensor("k_37_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(375055040)))]; + tensor k_37_cast = linear(bias = k_37_bias_0_to_fp16, weight = var_1117_to_fp16, x = var_1102_cast); + tensor var_1121_to_fp16 = const()[name = tensor("op_1121_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(375057664)))]; + tensor var_1122_to_fp16 = const()[name = tensor("op_1122_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(378334528)))]; + tensor v_37_cast = linear(bias = var_1122_to_fp16, weight = var_1121_to_fp16, x = var_1102_cast); + tensor var_1130 = const()[name = tensor("op_1130"), val = tensor([1, 1500, 20, -1])]; + tensor var_1131_cast = reshape(shape = var_1130, x = q_37_cast); + tensor const_242_to_fp16 = const()[name = tensor("const_242_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_39_cast = mul(x = var_1131_cast, y = const_242_to_fp16); + tensor var_1137 = const()[name = tensor("op_1137"), val = tensor([1, 1500, 20, -1])]; + tensor var_1138_cast = reshape(shape = var_1137, x = k_37_cast); + tensor const_243_to_fp16 = const()[name = tensor("const_243_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_39_cast = mul(x = var_1138_cast, y = const_243_to_fp16); + tensor var_1144 = const()[name = tensor("op_1144"), val = tensor([1, 1500, 20, -1])]; + tensor var_1145_cast = reshape(shape = var_1144, x = v_37_cast); + tensor var_1146 = const()[name = tensor("op_1146"), val = tensor([0, 2, 1, 3])]; + tensor qk_19_transpose_x_0 = const()[name = tensor("qk_19_transpose_x_0"), val = tensor(false)]; + tensor qk_19_transpose_y_0 = const()[name = tensor("qk_19_transpose_y_0"), val = tensor(false)]; + tensor transpose_82_perm_0 = const()[name = tensor("transpose_82_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_83_perm_0 = const()[name = tensor("transpose_83_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_217 = transpose(perm = transpose_83_perm_0, x = k_39_cast); + tensor transpose_218 = transpose(perm = transpose_82_perm_0, x = q_39_cast); + tensor qk_19_cast = matmul(transpose_x = qk_19_transpose_x_0, transpose_y = qk_19_transpose_y_0, x = transpose_218, y = transpose_217); + tensor var_1150_cast = softmax(axis = var_1085, x = qk_19_cast); + tensor var_1152_transpose_x_0 = const()[name = tensor("op_1152_transpose_x_0"), val = tensor(false)]; + tensor var_1152_transpose_y_0 = const()[name = tensor("op_1152_transpose_y_0"), val = tensor(false)]; + tensor transpose_219 = transpose(perm = var_1146, x = var_1145_cast); + tensor var_1152_cast = matmul(transpose_x = var_1152_transpose_x_0, transpose_y = var_1152_transpose_y_0, x = var_1150_cast, y = transpose_219); + tensor var_1153 = const()[name = tensor("op_1153"), val = tensor([0, 2, 1, 3])]; + tensor concat_9 = const()[name = tensor("concat_9"), val = tensor([1, 1500, 1280])]; + tensor transpose_216 = transpose(perm = var_1153, x = var_1152_cast); + tensor x_119_cast = reshape(shape = concat_9, x = transpose_216); + tensor var_1158_to_fp16 = const()[name = tensor("op_1158_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(378337152)))]; + tensor var_1159_to_fp16 = const()[name = tensor("op_1159_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381614016)))]; + tensor var_1160_cast = linear(bias = var_1159_to_fp16, weight = var_1158_to_fp16, x = x_119_cast); + tensor x_121_cast = add(x = x_115_cast, y = var_1160_cast); + tensor var_1166_axes_0 = const()[name = tensor("op_1166_axes_0"), val = tensor([-1])]; + tensor blocks_9_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_9_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381616640)))]; + tensor blocks_9_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_9_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381619264)))]; + tensor var_1166_cast = layer_norm(axes = var_1166_axes_0, beta = blocks_9_mlp_ln_bias_to_fp16, epsilon = var_1091_to_fp16, gamma = blocks_9_mlp_ln_weight_to_fp16, x = x_121_cast); + tensor var_1175_to_fp16 = const()[name = tensor("op_1175_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381621888)))]; + tensor var_1176_to_fp16 = const()[name = tensor("op_1176_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(394729152)))]; + tensor input_81_cast = linear(bias = var_1176_to_fp16, weight = var_1175_to_fp16, x = var_1166_cast); + tensor x_125_mode_0 = const()[name = tensor("x_125_mode_0"), val = tensor("EXACT")]; + tensor x_125_cast = gelu(mode = x_125_mode_0, x = input_81_cast); + tensor var_1181_to_fp16 = const()[name = tensor("op_1181_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(394739456)))]; + tensor var_1182_to_fp16 = const()[name = tensor("op_1182_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407846720)))]; + tensor var_1183_cast = linear(bias = var_1182_to_fp16, weight = var_1181_to_fp16, x = x_125_cast); + tensor x_127_cast = add(x = x_121_cast, y = var_1183_cast); + tensor var_1192 = const()[name = tensor("op_1192"), val = tensor(-1)]; + tensor var_1209_axes_0 = const()[name = tensor("op_1209_axes_0"), val = tensor([-1])]; + tensor blocks_10_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_10_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407849344)))]; + tensor blocks_10_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_10_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407851968)))]; + tensor var_1198_to_fp16 = const()[name = tensor("op_1198_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1209_cast = layer_norm(axes = var_1209_axes_0, beta = blocks_10_attn_ln_bias_to_fp16, epsilon = var_1198_to_fp16, gamma = blocks_10_attn_ln_weight_to_fp16, x = x_127_cast); + tensor var_1220_to_fp16 = const()[name = tensor("op_1220_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407854592)))]; + tensor var_1221_to_fp16 = const()[name = tensor("op_1221_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411131456)))]; + tensor q_41_cast = linear(bias = var_1221_to_fp16, weight = var_1220_to_fp16, x = var_1209_cast); + tensor var_1224_to_fp16 = const()[name = tensor("op_1224_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411134080)))]; + tensor k_41_bias_0_to_fp16 = const()[name = tensor("k_41_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(414410944)))]; + tensor k_41_cast = linear(bias = k_41_bias_0_to_fp16, weight = var_1224_to_fp16, x = var_1209_cast); + tensor var_1228_to_fp16 = const()[name = tensor("op_1228_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(414413568)))]; + tensor var_1229_to_fp16 = const()[name = tensor("op_1229_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(417690432)))]; + tensor v_41_cast = linear(bias = var_1229_to_fp16, weight = var_1228_to_fp16, x = var_1209_cast); + tensor var_1237 = const()[name = tensor("op_1237"), val = tensor([1, 1500, 20, -1])]; + tensor var_1238_cast = reshape(shape = var_1237, x = q_41_cast); + tensor const_244_to_fp16 = const()[name = tensor("const_244_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_43_cast = mul(x = var_1238_cast, y = const_244_to_fp16); + tensor var_1244 = const()[name = tensor("op_1244"), val = tensor([1, 1500, 20, -1])]; + tensor var_1245_cast = reshape(shape = var_1244, x = k_41_cast); + tensor const_245_to_fp16 = const()[name = tensor("const_245_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_43_cast = mul(x = var_1245_cast, y = const_245_to_fp16); + tensor var_1251 = const()[name = tensor("op_1251"), val = tensor([1, 1500, 20, -1])]; + tensor var_1252_cast = reshape(shape = var_1251, x = v_41_cast); + tensor var_1253 = const()[name = tensor("op_1253"), val = tensor([0, 2, 1, 3])]; + tensor qk_21_transpose_x_0 = const()[name = tensor("qk_21_transpose_x_0"), val = tensor(false)]; + tensor qk_21_transpose_y_0 = const()[name = tensor("qk_21_transpose_y_0"), val = tensor(false)]; + tensor transpose_84_perm_0 = const()[name = tensor("transpose_84_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_85_perm_0 = const()[name = tensor("transpose_85_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_213 = transpose(perm = transpose_85_perm_0, x = k_43_cast); + tensor transpose_214 = transpose(perm = transpose_84_perm_0, x = q_43_cast); + tensor qk_21_cast = matmul(transpose_x = qk_21_transpose_x_0, transpose_y = qk_21_transpose_y_0, x = transpose_214, y = transpose_213); + tensor var_1257_cast = softmax(axis = var_1192, x = qk_21_cast); + tensor var_1259_transpose_x_0 = const()[name = tensor("op_1259_transpose_x_0"), val = tensor(false)]; + tensor var_1259_transpose_y_0 = const()[name = tensor("op_1259_transpose_y_0"), val = tensor(false)]; + tensor transpose_215 = transpose(perm = var_1253, x = var_1252_cast); + tensor var_1259_cast = matmul(transpose_x = var_1259_transpose_x_0, transpose_y = var_1259_transpose_y_0, x = var_1257_cast, y = transpose_215); + tensor var_1260 = const()[name = tensor("op_1260"), val = tensor([0, 2, 1, 3])]; + tensor concat_10 = const()[name = tensor("concat_10"), val = tensor([1, 1500, 1280])]; + tensor transpose_212 = transpose(perm = var_1260, x = var_1259_cast); + tensor x_131_cast = reshape(shape = concat_10, x = transpose_212); + tensor var_1265_to_fp16 = const()[name = tensor("op_1265_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(417693056)))]; + tensor var_1266_to_fp16 = const()[name = tensor("op_1266_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(420969920)))]; + tensor var_1267_cast = linear(bias = var_1266_to_fp16, weight = var_1265_to_fp16, x = x_131_cast); + tensor x_133_cast = add(x = x_127_cast, y = var_1267_cast); + tensor var_1273_axes_0 = const()[name = tensor("op_1273_axes_0"), val = tensor([-1])]; + tensor blocks_10_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_10_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(420972544)))]; + tensor blocks_10_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_10_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(420975168)))]; + tensor var_1273_cast = layer_norm(axes = var_1273_axes_0, beta = blocks_10_mlp_ln_bias_to_fp16, epsilon = var_1198_to_fp16, gamma = blocks_10_mlp_ln_weight_to_fp16, x = x_133_cast); + tensor var_1282_to_fp16 = const()[name = tensor("op_1282_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(420977792)))]; + tensor var_1283_to_fp16 = const()[name = tensor("op_1283_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(434085056)))]; + tensor input_89_cast = linear(bias = var_1283_to_fp16, weight = var_1282_to_fp16, x = var_1273_cast); + tensor x_137_mode_0 = const()[name = tensor("x_137_mode_0"), val = tensor("EXACT")]; + tensor x_137_cast = gelu(mode = x_137_mode_0, x = input_89_cast); + tensor var_1288_to_fp16 = const()[name = tensor("op_1288_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(434095360)))]; + tensor var_1289_to_fp16 = const()[name = tensor("op_1289_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447202624)))]; + tensor var_1290_cast = linear(bias = var_1289_to_fp16, weight = var_1288_to_fp16, x = x_137_cast); + tensor x_139_cast = add(x = x_133_cast, y = var_1290_cast); + tensor var_1299 = const()[name = tensor("op_1299"), val = tensor(-1)]; + tensor var_1316_axes_0 = const()[name = tensor("op_1316_axes_0"), val = tensor([-1])]; + tensor blocks_11_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_11_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447205248)))]; + tensor blocks_11_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_11_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447207872)))]; + tensor var_1305_to_fp16 = const()[name = tensor("op_1305_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1316_cast = layer_norm(axes = var_1316_axes_0, beta = blocks_11_attn_ln_bias_to_fp16, epsilon = var_1305_to_fp16, gamma = blocks_11_attn_ln_weight_to_fp16, x = x_139_cast); + tensor var_1327_to_fp16 = const()[name = tensor("op_1327_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447210496)))]; + tensor var_1328_to_fp16 = const()[name = tensor("op_1328_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(450487360)))]; + tensor q_45_cast = linear(bias = var_1328_to_fp16, weight = var_1327_to_fp16, x = var_1316_cast); + tensor var_1331_to_fp16 = const()[name = tensor("op_1331_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(450489984)))]; + tensor k_45_bias_0_to_fp16 = const()[name = tensor("k_45_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(453766848)))]; + tensor k_45_cast = linear(bias = k_45_bias_0_to_fp16, weight = var_1331_to_fp16, x = var_1316_cast); + tensor var_1335_to_fp16 = const()[name = tensor("op_1335_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(453769472)))]; + tensor var_1336_to_fp16 = const()[name = tensor("op_1336_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(457046336)))]; + tensor v_45_cast = linear(bias = var_1336_to_fp16, weight = var_1335_to_fp16, x = var_1316_cast); + tensor var_1344 = const()[name = tensor("op_1344"), val = tensor([1, 1500, 20, -1])]; + tensor var_1345_cast = reshape(shape = var_1344, x = q_45_cast); + tensor const_246_to_fp16 = const()[name = tensor("const_246_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_47_cast = mul(x = var_1345_cast, y = const_246_to_fp16); + tensor var_1351 = const()[name = tensor("op_1351"), val = tensor([1, 1500, 20, -1])]; + tensor var_1352_cast = reshape(shape = var_1351, x = k_45_cast); + tensor const_247_to_fp16 = const()[name = tensor("const_247_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_47_cast = mul(x = var_1352_cast, y = const_247_to_fp16); + tensor var_1358 = const()[name = tensor("op_1358"), val = tensor([1, 1500, 20, -1])]; + tensor var_1359_cast = reshape(shape = var_1358, x = v_45_cast); + tensor var_1360 = const()[name = tensor("op_1360"), val = tensor([0, 2, 1, 3])]; + tensor qk_23_transpose_x_0 = const()[name = tensor("qk_23_transpose_x_0"), val = tensor(false)]; + tensor qk_23_transpose_y_0 = const()[name = tensor("qk_23_transpose_y_0"), val = tensor(false)]; + tensor transpose_86_perm_0 = const()[name = tensor("transpose_86_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_87_perm_0 = const()[name = tensor("transpose_87_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_209 = transpose(perm = transpose_87_perm_0, x = k_47_cast); + tensor transpose_210 = transpose(perm = transpose_86_perm_0, x = q_47_cast); + tensor qk_23_cast = matmul(transpose_x = qk_23_transpose_x_0, transpose_y = qk_23_transpose_y_0, x = transpose_210, y = transpose_209); + tensor var_1364_cast = softmax(axis = var_1299, x = qk_23_cast); + tensor var_1366_transpose_x_0 = const()[name = tensor("op_1366_transpose_x_0"), val = tensor(false)]; + tensor var_1366_transpose_y_0 = const()[name = tensor("op_1366_transpose_y_0"), val = tensor(false)]; + tensor transpose_211 = transpose(perm = var_1360, x = var_1359_cast); + tensor var_1366_cast = matmul(transpose_x = var_1366_transpose_x_0, transpose_y = var_1366_transpose_y_0, x = var_1364_cast, y = transpose_211); + tensor var_1367 = const()[name = tensor("op_1367"), val = tensor([0, 2, 1, 3])]; + tensor concat_11 = const()[name = tensor("concat_11"), val = tensor([1, 1500, 1280])]; + tensor transpose_208 = transpose(perm = var_1367, x = var_1366_cast); + tensor x_143_cast = reshape(shape = concat_11, x = transpose_208); + tensor var_1372_to_fp16 = const()[name = tensor("op_1372_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(457048960)))]; + tensor var_1373_to_fp16 = const()[name = tensor("op_1373_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460325824)))]; + tensor var_1374_cast = linear(bias = var_1373_to_fp16, weight = var_1372_to_fp16, x = x_143_cast); + tensor x_145_cast = add(x = x_139_cast, y = var_1374_cast); + tensor var_1380_axes_0 = const()[name = tensor("op_1380_axes_0"), val = tensor([-1])]; + tensor blocks_11_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_11_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460328448)))]; + tensor blocks_11_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_11_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460331072)))]; + tensor var_1380_cast = layer_norm(axes = var_1380_axes_0, beta = blocks_11_mlp_ln_bias_to_fp16, epsilon = var_1305_to_fp16, gamma = blocks_11_mlp_ln_weight_to_fp16, x = x_145_cast); + tensor var_1389_to_fp16 = const()[name = tensor("op_1389_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460333696)))]; + tensor var_1390_to_fp16 = const()[name = tensor("op_1390_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473440960)))]; + tensor input_97_cast = linear(bias = var_1390_to_fp16, weight = var_1389_to_fp16, x = var_1380_cast); + tensor x_149_mode_0 = const()[name = tensor("x_149_mode_0"), val = tensor("EXACT")]; + tensor x_149_cast = gelu(mode = x_149_mode_0, x = input_97_cast); + tensor var_1395_to_fp16 = const()[name = tensor("op_1395_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473451264)))]; + tensor var_1396_to_fp16 = const()[name = tensor("op_1396_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486558528)))]; + tensor var_1397_cast = linear(bias = var_1396_to_fp16, weight = var_1395_to_fp16, x = x_149_cast); + tensor x_151_cast = add(x = x_145_cast, y = var_1397_cast); + tensor var_1406 = const()[name = tensor("op_1406"), val = tensor(-1)]; + tensor var_1423_axes_0 = const()[name = tensor("op_1423_axes_0"), val = tensor([-1])]; + tensor blocks_12_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_12_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486561152)))]; + tensor blocks_12_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_12_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486563776)))]; + tensor var_1412_to_fp16 = const()[name = tensor("op_1412_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1423_cast = layer_norm(axes = var_1423_axes_0, beta = blocks_12_attn_ln_bias_to_fp16, epsilon = var_1412_to_fp16, gamma = blocks_12_attn_ln_weight_to_fp16, x = x_151_cast); + tensor var_1434_to_fp16 = const()[name = tensor("op_1434_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486566400)))]; + tensor var_1435_to_fp16 = const()[name = tensor("op_1435_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(489843264)))]; + tensor q_49_cast = linear(bias = var_1435_to_fp16, weight = var_1434_to_fp16, x = var_1423_cast); + tensor var_1438_to_fp16 = const()[name = tensor("op_1438_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(489845888)))]; + tensor k_49_bias_0_to_fp16 = const()[name = tensor("k_49_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(493122752)))]; + tensor k_49_cast = linear(bias = k_49_bias_0_to_fp16, weight = var_1438_to_fp16, x = var_1423_cast); + tensor var_1442_to_fp16 = const()[name = tensor("op_1442_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(493125376)))]; + tensor var_1443_to_fp16 = const()[name = tensor("op_1443_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496402240)))]; + tensor v_49_cast = linear(bias = var_1443_to_fp16, weight = var_1442_to_fp16, x = var_1423_cast); + tensor var_1451 = const()[name = tensor("op_1451"), val = tensor([1, 1500, 20, -1])]; + tensor var_1452_cast = reshape(shape = var_1451, x = q_49_cast); + tensor const_248_to_fp16 = const()[name = tensor("const_248_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_51_cast = mul(x = var_1452_cast, y = const_248_to_fp16); + tensor var_1458 = const()[name = tensor("op_1458"), val = tensor([1, 1500, 20, -1])]; + tensor var_1459_cast = reshape(shape = var_1458, x = k_49_cast); + tensor const_249_to_fp16 = const()[name = tensor("const_249_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_51_cast = mul(x = var_1459_cast, y = const_249_to_fp16); + tensor var_1465 = const()[name = tensor("op_1465"), val = tensor([1, 1500, 20, -1])]; + tensor var_1466_cast = reshape(shape = var_1465, x = v_49_cast); + tensor var_1467 = const()[name = tensor("op_1467"), val = tensor([0, 2, 1, 3])]; + tensor qk_25_transpose_x_0 = const()[name = tensor("qk_25_transpose_x_0"), val = tensor(false)]; + tensor qk_25_transpose_y_0 = const()[name = tensor("qk_25_transpose_y_0"), val = tensor(false)]; + tensor transpose_88_perm_0 = const()[name = tensor("transpose_88_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_89_perm_0 = const()[name = tensor("transpose_89_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_205 = transpose(perm = transpose_89_perm_0, x = k_51_cast); + tensor transpose_206 = transpose(perm = transpose_88_perm_0, x = q_51_cast); + tensor qk_25_cast = matmul(transpose_x = qk_25_transpose_x_0, transpose_y = qk_25_transpose_y_0, x = transpose_206, y = transpose_205); + tensor var_1471_cast = softmax(axis = var_1406, x = qk_25_cast); + tensor var_1473_transpose_x_0 = const()[name = tensor("op_1473_transpose_x_0"), val = tensor(false)]; + tensor var_1473_transpose_y_0 = const()[name = tensor("op_1473_transpose_y_0"), val = tensor(false)]; + tensor transpose_207 = transpose(perm = var_1467, x = var_1466_cast); + tensor var_1473_cast = matmul(transpose_x = var_1473_transpose_x_0, transpose_y = var_1473_transpose_y_0, x = var_1471_cast, y = transpose_207); + tensor var_1474 = const()[name = tensor("op_1474"), val = tensor([0, 2, 1, 3])]; + tensor concat_12 = const()[name = tensor("concat_12"), val = tensor([1, 1500, 1280])]; + tensor transpose_204 = transpose(perm = var_1474, x = var_1473_cast); + tensor x_155_cast = reshape(shape = concat_12, x = transpose_204); + tensor var_1479_to_fp16 = const()[name = tensor("op_1479_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496404864)))]; + tensor var_1480_to_fp16 = const()[name = tensor("op_1480_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(499681728)))]; + tensor var_1481_cast = linear(bias = var_1480_to_fp16, weight = var_1479_to_fp16, x = x_155_cast); + tensor x_157_cast = add(x = x_151_cast, y = var_1481_cast); + tensor var_1487_axes_0 = const()[name = tensor("op_1487_axes_0"), val = tensor([-1])]; + tensor blocks_12_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_12_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(499684352)))]; + tensor blocks_12_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_12_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(499686976)))]; + tensor var_1487_cast = layer_norm(axes = var_1487_axes_0, beta = blocks_12_mlp_ln_bias_to_fp16, epsilon = var_1412_to_fp16, gamma = blocks_12_mlp_ln_weight_to_fp16, x = x_157_cast); + tensor var_1496_to_fp16 = const()[name = tensor("op_1496_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(499689600)))]; + tensor var_1497_to_fp16 = const()[name = tensor("op_1497_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(512796864)))]; + tensor input_105_cast = linear(bias = var_1497_to_fp16, weight = var_1496_to_fp16, x = var_1487_cast); + tensor x_161_mode_0 = const()[name = tensor("x_161_mode_0"), val = tensor("EXACT")]; + tensor x_161_cast = gelu(mode = x_161_mode_0, x = input_105_cast); + tensor var_1502_to_fp16 = const()[name = tensor("op_1502_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(512807168)))]; + tensor var_1503_to_fp16 = const()[name = tensor("op_1503_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(525914432)))]; + tensor var_1504_cast = linear(bias = var_1503_to_fp16, weight = var_1502_to_fp16, x = x_161_cast); + tensor x_163_cast = add(x = x_157_cast, y = var_1504_cast); + tensor var_1513 = const()[name = tensor("op_1513"), val = tensor(-1)]; + tensor var_1530_axes_0 = const()[name = tensor("op_1530_axes_0"), val = tensor([-1])]; + tensor blocks_13_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_13_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(525917056)))]; + tensor blocks_13_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_13_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(525919680)))]; + tensor var_1519_to_fp16 = const()[name = tensor("op_1519_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1530_cast = layer_norm(axes = var_1530_axes_0, beta = blocks_13_attn_ln_bias_to_fp16, epsilon = var_1519_to_fp16, gamma = blocks_13_attn_ln_weight_to_fp16, x = x_163_cast); + tensor var_1541_to_fp16 = const()[name = tensor("op_1541_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(525922304)))]; + tensor var_1542_to_fp16 = const()[name = tensor("op_1542_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529199168)))]; + tensor q_53_cast = linear(bias = var_1542_to_fp16, weight = var_1541_to_fp16, x = var_1530_cast); + tensor var_1545_to_fp16 = const()[name = tensor("op_1545_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529201792)))]; + tensor k_53_bias_0_to_fp16 = const()[name = tensor("k_53_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(532478656)))]; + tensor k_53_cast = linear(bias = k_53_bias_0_to_fp16, weight = var_1545_to_fp16, x = var_1530_cast); + tensor var_1549_to_fp16 = const()[name = tensor("op_1549_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(532481280)))]; + tensor var_1550_to_fp16 = const()[name = tensor("op_1550_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(535758144)))]; + tensor v_53_cast = linear(bias = var_1550_to_fp16, weight = var_1549_to_fp16, x = var_1530_cast); + tensor var_1558 = const()[name = tensor("op_1558"), val = tensor([1, 1500, 20, -1])]; + tensor var_1559_cast = reshape(shape = var_1558, x = q_53_cast); + tensor const_250_to_fp16 = const()[name = tensor("const_250_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_55_cast = mul(x = var_1559_cast, y = const_250_to_fp16); + tensor var_1565 = const()[name = tensor("op_1565"), val = tensor([1, 1500, 20, -1])]; + tensor var_1566_cast = reshape(shape = var_1565, x = k_53_cast); + tensor const_251_to_fp16 = const()[name = tensor("const_251_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_55_cast = mul(x = var_1566_cast, y = const_251_to_fp16); + tensor var_1572 = const()[name = tensor("op_1572"), val = tensor([1, 1500, 20, -1])]; + tensor var_1573_cast = reshape(shape = var_1572, x = v_53_cast); + tensor var_1574 = const()[name = tensor("op_1574"), val = tensor([0, 2, 1, 3])]; + tensor qk_27_transpose_x_0 = const()[name = tensor("qk_27_transpose_x_0"), val = tensor(false)]; + tensor qk_27_transpose_y_0 = const()[name = tensor("qk_27_transpose_y_0"), val = tensor(false)]; + tensor transpose_90_perm_0 = const()[name = tensor("transpose_90_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_91_perm_0 = const()[name = tensor("transpose_91_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_201 = transpose(perm = transpose_91_perm_0, x = k_55_cast); + tensor transpose_202 = transpose(perm = transpose_90_perm_0, x = q_55_cast); + tensor qk_27_cast = matmul(transpose_x = qk_27_transpose_x_0, transpose_y = qk_27_transpose_y_0, x = transpose_202, y = transpose_201); + tensor var_1578_cast = softmax(axis = var_1513, x = qk_27_cast); + tensor var_1580_transpose_x_0 = const()[name = tensor("op_1580_transpose_x_0"), val = tensor(false)]; + tensor var_1580_transpose_y_0 = const()[name = tensor("op_1580_transpose_y_0"), val = tensor(false)]; + tensor transpose_203 = transpose(perm = var_1574, x = var_1573_cast); + tensor var_1580_cast = matmul(transpose_x = var_1580_transpose_x_0, transpose_y = var_1580_transpose_y_0, x = var_1578_cast, y = transpose_203); + tensor var_1581 = const()[name = tensor("op_1581"), val = tensor([0, 2, 1, 3])]; + tensor concat_13 = const()[name = tensor("concat_13"), val = tensor([1, 1500, 1280])]; + tensor transpose_200 = transpose(perm = var_1581, x = var_1580_cast); + tensor x_167_cast = reshape(shape = concat_13, x = transpose_200); + tensor var_1586_to_fp16 = const()[name = tensor("op_1586_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(535760768)))]; + tensor var_1587_to_fp16 = const()[name = tensor("op_1587_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(539037632)))]; + tensor var_1588_cast = linear(bias = var_1587_to_fp16, weight = var_1586_to_fp16, x = x_167_cast); + tensor x_169_cast = add(x = x_163_cast, y = var_1588_cast); + tensor var_1594_axes_0 = const()[name = tensor("op_1594_axes_0"), val = tensor([-1])]; + tensor blocks_13_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_13_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(539040256)))]; + tensor blocks_13_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_13_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(539042880)))]; + tensor var_1594_cast = layer_norm(axes = var_1594_axes_0, beta = blocks_13_mlp_ln_bias_to_fp16, epsilon = var_1519_to_fp16, gamma = blocks_13_mlp_ln_weight_to_fp16, x = x_169_cast); + tensor var_1603_to_fp16 = const()[name = tensor("op_1603_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(539045504)))]; + tensor var_1604_to_fp16 = const()[name = tensor("op_1604_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552152768)))]; + tensor input_113_cast = linear(bias = var_1604_to_fp16, weight = var_1603_to_fp16, x = var_1594_cast); + tensor x_173_mode_0 = const()[name = tensor("x_173_mode_0"), val = tensor("EXACT")]; + tensor x_173_cast = gelu(mode = x_173_mode_0, x = input_113_cast); + tensor var_1609_to_fp16 = const()[name = tensor("op_1609_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552163072)))]; + tensor var_1610_to_fp16 = const()[name = tensor("op_1610_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565270336)))]; + tensor var_1611_cast = linear(bias = var_1610_to_fp16, weight = var_1609_to_fp16, x = x_173_cast); + tensor x_175_cast = add(x = x_169_cast, y = var_1611_cast); + tensor var_1620 = const()[name = tensor("op_1620"), val = tensor(-1)]; + tensor var_1637_axes_0 = const()[name = tensor("op_1637_axes_0"), val = tensor([-1])]; + tensor blocks_14_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_14_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565272960)))]; + tensor blocks_14_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_14_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565275584)))]; + tensor var_1626_to_fp16 = const()[name = tensor("op_1626_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1637_cast = layer_norm(axes = var_1637_axes_0, beta = blocks_14_attn_ln_bias_to_fp16, epsilon = var_1626_to_fp16, gamma = blocks_14_attn_ln_weight_to_fp16, x = x_175_cast); + tensor var_1648_to_fp16 = const()[name = tensor("op_1648_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565278208)))]; + tensor var_1649_to_fp16 = const()[name = tensor("op_1649_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568555072)))]; + tensor q_57_cast = linear(bias = var_1649_to_fp16, weight = var_1648_to_fp16, x = var_1637_cast); + tensor var_1652_to_fp16 = const()[name = tensor("op_1652_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568557696)))]; + tensor k_57_bias_0_to_fp16 = const()[name = tensor("k_57_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(571834560)))]; + tensor k_57_cast = linear(bias = k_57_bias_0_to_fp16, weight = var_1652_to_fp16, x = var_1637_cast); + tensor var_1656_to_fp16 = const()[name = tensor("op_1656_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(571837184)))]; + tensor var_1657_to_fp16 = const()[name = tensor("op_1657_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(575114048)))]; + tensor v_57_cast = linear(bias = var_1657_to_fp16, weight = var_1656_to_fp16, x = var_1637_cast); + tensor var_1665 = const()[name = tensor("op_1665"), val = tensor([1, 1500, 20, -1])]; + tensor var_1666_cast = reshape(shape = var_1665, x = q_57_cast); + tensor const_252_to_fp16 = const()[name = tensor("const_252_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_59_cast = mul(x = var_1666_cast, y = const_252_to_fp16); + tensor var_1672 = const()[name = tensor("op_1672"), val = tensor([1, 1500, 20, -1])]; + tensor var_1673_cast = reshape(shape = var_1672, x = k_57_cast); + tensor const_253_to_fp16 = const()[name = tensor("const_253_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_59_cast = mul(x = var_1673_cast, y = const_253_to_fp16); + tensor var_1679 = const()[name = tensor("op_1679"), val = tensor([1, 1500, 20, -1])]; + tensor var_1680_cast = reshape(shape = var_1679, x = v_57_cast); + tensor var_1681 = const()[name = tensor("op_1681"), val = tensor([0, 2, 1, 3])]; + tensor qk_29_transpose_x_0 = const()[name = tensor("qk_29_transpose_x_0"), val = tensor(false)]; + tensor qk_29_transpose_y_0 = const()[name = tensor("qk_29_transpose_y_0"), val = tensor(false)]; + tensor transpose_92_perm_0 = const()[name = tensor("transpose_92_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_93_perm_0 = const()[name = tensor("transpose_93_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_197 = transpose(perm = transpose_93_perm_0, x = k_59_cast); + tensor transpose_198 = transpose(perm = transpose_92_perm_0, x = q_59_cast); + tensor qk_29_cast = matmul(transpose_x = qk_29_transpose_x_0, transpose_y = qk_29_transpose_y_0, x = transpose_198, y = transpose_197); + tensor var_1685_cast = softmax(axis = var_1620, x = qk_29_cast); + tensor var_1687_transpose_x_0 = const()[name = tensor("op_1687_transpose_x_0"), val = tensor(false)]; + tensor var_1687_transpose_y_0 = const()[name = tensor("op_1687_transpose_y_0"), val = tensor(false)]; + tensor transpose_199 = transpose(perm = var_1681, x = var_1680_cast); + tensor var_1687_cast = matmul(transpose_x = var_1687_transpose_x_0, transpose_y = var_1687_transpose_y_0, x = var_1685_cast, y = transpose_199); + tensor var_1688 = const()[name = tensor("op_1688"), val = tensor([0, 2, 1, 3])]; + tensor concat_14 = const()[name = tensor("concat_14"), val = tensor([1, 1500, 1280])]; + tensor transpose_196 = transpose(perm = var_1688, x = var_1687_cast); + tensor x_179_cast = reshape(shape = concat_14, x = transpose_196); + tensor var_1693_to_fp16 = const()[name = tensor("op_1693_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(575116672)))]; + tensor var_1694_to_fp16 = const()[name = tensor("op_1694_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578393536)))]; + tensor var_1695_cast = linear(bias = var_1694_to_fp16, weight = var_1693_to_fp16, x = x_179_cast); + tensor x_181_cast = add(x = x_175_cast, y = var_1695_cast); + tensor var_1701_axes_0 = const()[name = tensor("op_1701_axes_0"), val = tensor([-1])]; + tensor blocks_14_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_14_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578396160)))]; + tensor blocks_14_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_14_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578398784)))]; + tensor var_1701_cast = layer_norm(axes = var_1701_axes_0, beta = blocks_14_mlp_ln_bias_to_fp16, epsilon = var_1626_to_fp16, gamma = blocks_14_mlp_ln_weight_to_fp16, x = x_181_cast); + tensor var_1710_to_fp16 = const()[name = tensor("op_1710_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578401408)))]; + tensor var_1711_to_fp16 = const()[name = tensor("op_1711_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(591508672)))]; + tensor input_121_cast = linear(bias = var_1711_to_fp16, weight = var_1710_to_fp16, x = var_1701_cast); + tensor x_185_mode_0 = const()[name = tensor("x_185_mode_0"), val = tensor("EXACT")]; + tensor x_185_cast = gelu(mode = x_185_mode_0, x = input_121_cast); + tensor var_1716_to_fp16 = const()[name = tensor("op_1716_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(591518976)))]; + tensor var_1717_to_fp16 = const()[name = tensor("op_1717_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604626240)))]; + tensor var_1718_cast = linear(bias = var_1717_to_fp16, weight = var_1716_to_fp16, x = x_185_cast); + tensor x_187_cast = add(x = x_181_cast, y = var_1718_cast); + tensor var_1727 = const()[name = tensor("op_1727"), val = tensor(-1)]; + tensor var_1744_axes_0 = const()[name = tensor("op_1744_axes_0"), val = tensor([-1])]; + tensor blocks_15_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_15_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604628864)))]; + tensor blocks_15_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_15_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604631488)))]; + tensor var_1733_to_fp16 = const()[name = tensor("op_1733_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1744_cast = layer_norm(axes = var_1744_axes_0, beta = blocks_15_attn_ln_bias_to_fp16, epsilon = var_1733_to_fp16, gamma = blocks_15_attn_ln_weight_to_fp16, x = x_187_cast); + tensor var_1755_to_fp16 = const()[name = tensor("op_1755_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604634112)))]; + tensor var_1756_to_fp16 = const()[name = tensor("op_1756_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(607910976)))]; + tensor q_61_cast = linear(bias = var_1756_to_fp16, weight = var_1755_to_fp16, x = var_1744_cast); + tensor var_1759_to_fp16 = const()[name = tensor("op_1759_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(607913600)))]; + tensor k_61_bias_0_to_fp16 = const()[name = tensor("k_61_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(611190464)))]; + tensor k_61_cast = linear(bias = k_61_bias_0_to_fp16, weight = var_1759_to_fp16, x = var_1744_cast); + tensor var_1763_to_fp16 = const()[name = tensor("op_1763_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(611193088)))]; + tensor var_1764_to_fp16 = const()[name = tensor("op_1764_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614469952)))]; + tensor v_61_cast = linear(bias = var_1764_to_fp16, weight = var_1763_to_fp16, x = var_1744_cast); + tensor var_1772 = const()[name = tensor("op_1772"), val = tensor([1, 1500, 20, -1])]; + tensor var_1773_cast = reshape(shape = var_1772, x = q_61_cast); + tensor const_254_to_fp16 = const()[name = tensor("const_254_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_63_cast = mul(x = var_1773_cast, y = const_254_to_fp16); + tensor var_1779 = const()[name = tensor("op_1779"), val = tensor([1, 1500, 20, -1])]; + tensor var_1780_cast = reshape(shape = var_1779, x = k_61_cast); + tensor const_255_to_fp16 = const()[name = tensor("const_255_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_63_cast = mul(x = var_1780_cast, y = const_255_to_fp16); + tensor var_1786 = const()[name = tensor("op_1786"), val = tensor([1, 1500, 20, -1])]; + tensor var_1787_cast = reshape(shape = var_1786, x = v_61_cast); + tensor var_1788 = const()[name = tensor("op_1788"), val = tensor([0, 2, 1, 3])]; + tensor qk_31_transpose_x_0 = const()[name = tensor("qk_31_transpose_x_0"), val = tensor(false)]; + tensor qk_31_transpose_y_0 = const()[name = tensor("qk_31_transpose_y_0"), val = tensor(false)]; + tensor transpose_94_perm_0 = const()[name = tensor("transpose_94_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_95_perm_0 = const()[name = tensor("transpose_95_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_193 = transpose(perm = transpose_95_perm_0, x = k_63_cast); + tensor transpose_194 = transpose(perm = transpose_94_perm_0, x = q_63_cast); + tensor qk_31_cast = matmul(transpose_x = qk_31_transpose_x_0, transpose_y = qk_31_transpose_y_0, x = transpose_194, y = transpose_193); + tensor var_1792_cast = softmax(axis = var_1727, x = qk_31_cast); + tensor var_1794_transpose_x_0 = const()[name = tensor("op_1794_transpose_x_0"), val = tensor(false)]; + tensor var_1794_transpose_y_0 = const()[name = tensor("op_1794_transpose_y_0"), val = tensor(false)]; + tensor transpose_195 = transpose(perm = var_1788, x = var_1787_cast); + tensor var_1794_cast = matmul(transpose_x = var_1794_transpose_x_0, transpose_y = var_1794_transpose_y_0, x = var_1792_cast, y = transpose_195); + tensor var_1795 = const()[name = tensor("op_1795"), val = tensor([0, 2, 1, 3])]; + tensor concat_15 = const()[name = tensor("concat_15"), val = tensor([1, 1500, 1280])]; + tensor transpose_192 = transpose(perm = var_1795, x = var_1794_cast); + tensor x_191_cast = reshape(shape = concat_15, x = transpose_192); + tensor var_1800_to_fp16 = const()[name = tensor("op_1800_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614472576)))]; + tensor var_1801_to_fp16 = const()[name = tensor("op_1801_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617749440)))]; + tensor var_1802_cast = linear(bias = var_1801_to_fp16, weight = var_1800_to_fp16, x = x_191_cast); + tensor x_193_cast = add(x = x_187_cast, y = var_1802_cast); + tensor var_1808_axes_0 = const()[name = tensor("op_1808_axes_0"), val = tensor([-1])]; + tensor blocks_15_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_15_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617752064)))]; + tensor blocks_15_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_15_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617754688)))]; + tensor var_1808_cast = layer_norm(axes = var_1808_axes_0, beta = blocks_15_mlp_ln_bias_to_fp16, epsilon = var_1733_to_fp16, gamma = blocks_15_mlp_ln_weight_to_fp16, x = x_193_cast); + tensor var_1817_to_fp16 = const()[name = tensor("op_1817_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617757312)))]; + tensor var_1818_to_fp16 = const()[name = tensor("op_1818_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(630864576)))]; + tensor input_129_cast = linear(bias = var_1818_to_fp16, weight = var_1817_to_fp16, x = var_1808_cast); + tensor x_197_mode_0 = const()[name = tensor("x_197_mode_0"), val = tensor("EXACT")]; + tensor x_197_cast = gelu(mode = x_197_mode_0, x = input_129_cast); + tensor var_1823_to_fp16 = const()[name = tensor("op_1823_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(630874880)))]; + tensor var_1824_to_fp16 = const()[name = tensor("op_1824_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643982144)))]; + tensor var_1825_cast = linear(bias = var_1824_to_fp16, weight = var_1823_to_fp16, x = x_197_cast); + tensor x_199_cast = add(x = x_193_cast, y = var_1825_cast); + tensor var_1834 = const()[name = tensor("op_1834"), val = tensor(-1)]; + tensor var_1851_axes_0 = const()[name = tensor("op_1851_axes_0"), val = tensor([-1])]; + tensor blocks_16_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_16_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643984768)))]; + tensor blocks_16_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_16_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643987392)))]; + tensor var_1840_to_fp16 = const()[name = tensor("op_1840_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1851_cast = layer_norm(axes = var_1851_axes_0, beta = blocks_16_attn_ln_bias_to_fp16, epsilon = var_1840_to_fp16, gamma = blocks_16_attn_ln_weight_to_fp16, x = x_199_cast); + tensor var_1862_to_fp16 = const()[name = tensor("op_1862_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643990016)))]; + tensor var_1863_to_fp16 = const()[name = tensor("op_1863_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647266880)))]; + tensor q_65_cast = linear(bias = var_1863_to_fp16, weight = var_1862_to_fp16, x = var_1851_cast); + tensor var_1866_to_fp16 = const()[name = tensor("op_1866_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647269504)))]; + tensor k_65_bias_0_to_fp16 = const()[name = tensor("k_65_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(650546368)))]; + tensor k_65_cast = linear(bias = k_65_bias_0_to_fp16, weight = var_1866_to_fp16, x = var_1851_cast); + tensor var_1870_to_fp16 = const()[name = tensor("op_1870_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(650548992)))]; + tensor var_1871_to_fp16 = const()[name = tensor("op_1871_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(653825856)))]; + tensor v_65_cast = linear(bias = var_1871_to_fp16, weight = var_1870_to_fp16, x = var_1851_cast); + tensor var_1879 = const()[name = tensor("op_1879"), val = tensor([1, 1500, 20, -1])]; + tensor var_1880_cast = reshape(shape = var_1879, x = q_65_cast); + tensor const_256_to_fp16 = const()[name = tensor("const_256_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_67_cast = mul(x = var_1880_cast, y = const_256_to_fp16); + tensor var_1886 = const()[name = tensor("op_1886"), val = tensor([1, 1500, 20, -1])]; + tensor var_1887_cast = reshape(shape = var_1886, x = k_65_cast); + tensor const_257_to_fp16 = const()[name = tensor("const_257_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_67_cast = mul(x = var_1887_cast, y = const_257_to_fp16); + tensor var_1893 = const()[name = tensor("op_1893"), val = tensor([1, 1500, 20, -1])]; + tensor var_1894_cast = reshape(shape = var_1893, x = v_65_cast); + tensor var_1895 = const()[name = tensor("op_1895"), val = tensor([0, 2, 1, 3])]; + tensor qk_33_transpose_x_0 = const()[name = tensor("qk_33_transpose_x_0"), val = tensor(false)]; + tensor qk_33_transpose_y_0 = const()[name = tensor("qk_33_transpose_y_0"), val = tensor(false)]; + tensor transpose_96_perm_0 = const()[name = tensor("transpose_96_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_97_perm_0 = const()[name = tensor("transpose_97_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_189 = transpose(perm = transpose_97_perm_0, x = k_67_cast); + tensor transpose_190 = transpose(perm = transpose_96_perm_0, x = q_67_cast); + tensor qk_33_cast = matmul(transpose_x = qk_33_transpose_x_0, transpose_y = qk_33_transpose_y_0, x = transpose_190, y = transpose_189); + tensor var_1899_cast = softmax(axis = var_1834, x = qk_33_cast); + tensor var_1901_transpose_x_0 = const()[name = tensor("op_1901_transpose_x_0"), val = tensor(false)]; + tensor var_1901_transpose_y_0 = const()[name = tensor("op_1901_transpose_y_0"), val = tensor(false)]; + tensor transpose_191 = transpose(perm = var_1895, x = var_1894_cast); + tensor var_1901_cast = matmul(transpose_x = var_1901_transpose_x_0, transpose_y = var_1901_transpose_y_0, x = var_1899_cast, y = transpose_191); + tensor var_1902 = const()[name = tensor("op_1902"), val = tensor([0, 2, 1, 3])]; + tensor concat_16 = const()[name = tensor("concat_16"), val = tensor([1, 1500, 1280])]; + tensor transpose_188 = transpose(perm = var_1902, x = var_1901_cast); + tensor x_203_cast = reshape(shape = concat_16, x = transpose_188); + tensor var_1907_to_fp16 = const()[name = tensor("op_1907_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(653828480)))]; + tensor var_1908_to_fp16 = const()[name = tensor("op_1908_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657105344)))]; + tensor var_1909_cast = linear(bias = var_1908_to_fp16, weight = var_1907_to_fp16, x = x_203_cast); + tensor x_205_cast = add(x = x_199_cast, y = var_1909_cast); + tensor var_1915_axes_0 = const()[name = tensor("op_1915_axes_0"), val = tensor([-1])]; + tensor blocks_16_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_16_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657107968)))]; + tensor blocks_16_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_16_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657110592)))]; + tensor var_1915_cast = layer_norm(axes = var_1915_axes_0, beta = blocks_16_mlp_ln_bias_to_fp16, epsilon = var_1840_to_fp16, gamma = blocks_16_mlp_ln_weight_to_fp16, x = x_205_cast); + tensor var_1924_to_fp16 = const()[name = tensor("op_1924_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657113216)))]; + tensor var_1925_to_fp16 = const()[name = tensor("op_1925_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(670220480)))]; + tensor input_137_cast = linear(bias = var_1925_to_fp16, weight = var_1924_to_fp16, x = var_1915_cast); + tensor x_209_mode_0 = const()[name = tensor("x_209_mode_0"), val = tensor("EXACT")]; + tensor x_209_cast = gelu(mode = x_209_mode_0, x = input_137_cast); + tensor var_1930_to_fp16 = const()[name = tensor("op_1930_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(670230784)))]; + tensor var_1931_to_fp16 = const()[name = tensor("op_1931_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(683338048)))]; + tensor var_1932_cast = linear(bias = var_1931_to_fp16, weight = var_1930_to_fp16, x = x_209_cast); + tensor x_211_cast = add(x = x_205_cast, y = var_1932_cast); + tensor var_1941 = const()[name = tensor("op_1941"), val = tensor(-1)]; + tensor var_1958_axes_0 = const()[name = tensor("op_1958_axes_0"), val = tensor([-1])]; + tensor blocks_17_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_17_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(683340672)))]; + tensor blocks_17_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_17_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(683343296)))]; + tensor var_1947_to_fp16 = const()[name = tensor("op_1947_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1958_cast = layer_norm(axes = var_1958_axes_0, beta = blocks_17_attn_ln_bias_to_fp16, epsilon = var_1947_to_fp16, gamma = blocks_17_attn_ln_weight_to_fp16, x = x_211_cast); + tensor var_1969_to_fp16 = const()[name = tensor("op_1969_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(683345920)))]; + tensor var_1970_to_fp16 = const()[name = tensor("op_1970_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(686622784)))]; + tensor q_69_cast = linear(bias = var_1970_to_fp16, weight = var_1969_to_fp16, x = var_1958_cast); + tensor var_1973_to_fp16 = const()[name = tensor("op_1973_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(686625408)))]; + tensor k_69_bias_0_to_fp16 = const()[name = tensor("k_69_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(689902272)))]; + tensor k_69_cast = linear(bias = k_69_bias_0_to_fp16, weight = var_1973_to_fp16, x = var_1958_cast); + tensor var_1977_to_fp16 = const()[name = tensor("op_1977_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(689904896)))]; + tensor var_1978_to_fp16 = const()[name = tensor("op_1978_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(693181760)))]; + tensor v_69_cast = linear(bias = var_1978_to_fp16, weight = var_1977_to_fp16, x = var_1958_cast); + tensor var_1986 = const()[name = tensor("op_1986"), val = tensor([1, 1500, 20, -1])]; + tensor var_1987_cast = reshape(shape = var_1986, x = q_69_cast); + tensor const_258_to_fp16 = const()[name = tensor("const_258_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_71_cast = mul(x = var_1987_cast, y = const_258_to_fp16); + tensor var_1993 = const()[name = tensor("op_1993"), val = tensor([1, 1500, 20, -1])]; + tensor var_1994_cast = reshape(shape = var_1993, x = k_69_cast); + tensor const_259_to_fp16 = const()[name = tensor("const_259_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_71_cast = mul(x = var_1994_cast, y = const_259_to_fp16); + tensor var_2000 = const()[name = tensor("op_2000"), val = tensor([1, 1500, 20, -1])]; + tensor var_2001_cast = reshape(shape = var_2000, x = v_69_cast); + tensor var_2002 = const()[name = tensor("op_2002"), val = tensor([0, 2, 1, 3])]; + tensor qk_35_transpose_x_0 = const()[name = tensor("qk_35_transpose_x_0"), val = tensor(false)]; + tensor qk_35_transpose_y_0 = const()[name = tensor("qk_35_transpose_y_0"), val = tensor(false)]; + tensor transpose_98_perm_0 = const()[name = tensor("transpose_98_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_99_perm_0 = const()[name = tensor("transpose_99_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_185 = transpose(perm = transpose_99_perm_0, x = k_71_cast); + tensor transpose_186 = transpose(perm = transpose_98_perm_0, x = q_71_cast); + tensor qk_35_cast = matmul(transpose_x = qk_35_transpose_x_0, transpose_y = qk_35_transpose_y_0, x = transpose_186, y = transpose_185); + tensor var_2006_cast = softmax(axis = var_1941, x = qk_35_cast); + tensor var_2008_transpose_x_0 = const()[name = tensor("op_2008_transpose_x_0"), val = tensor(false)]; + tensor var_2008_transpose_y_0 = const()[name = tensor("op_2008_transpose_y_0"), val = tensor(false)]; + tensor transpose_187 = transpose(perm = var_2002, x = var_2001_cast); + tensor var_2008_cast = matmul(transpose_x = var_2008_transpose_x_0, transpose_y = var_2008_transpose_y_0, x = var_2006_cast, y = transpose_187); + tensor var_2009 = const()[name = tensor("op_2009"), val = tensor([0, 2, 1, 3])]; + tensor concat_17 = const()[name = tensor("concat_17"), val = tensor([1, 1500, 1280])]; + tensor transpose_184 = transpose(perm = var_2009, x = var_2008_cast); + tensor x_215_cast = reshape(shape = concat_17, x = transpose_184); + tensor var_2014_to_fp16 = const()[name = tensor("op_2014_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(693184384)))]; + tensor var_2015_to_fp16 = const()[name = tensor("op_2015_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696461248)))]; + tensor var_2016_cast = linear(bias = var_2015_to_fp16, weight = var_2014_to_fp16, x = x_215_cast); + tensor x_217_cast = add(x = x_211_cast, y = var_2016_cast); + tensor var_2022_axes_0 = const()[name = tensor("op_2022_axes_0"), val = tensor([-1])]; + tensor blocks_17_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_17_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696463872)))]; + tensor blocks_17_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_17_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696466496)))]; + tensor var_2022_cast = layer_norm(axes = var_2022_axes_0, beta = blocks_17_mlp_ln_bias_to_fp16, epsilon = var_1947_to_fp16, gamma = blocks_17_mlp_ln_weight_to_fp16, x = x_217_cast); + tensor var_2031_to_fp16 = const()[name = tensor("op_2031_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696469120)))]; + tensor var_2032_to_fp16 = const()[name = tensor("op_2032_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(709576384)))]; + tensor input_145_cast = linear(bias = var_2032_to_fp16, weight = var_2031_to_fp16, x = var_2022_cast); + tensor x_221_mode_0 = const()[name = tensor("x_221_mode_0"), val = tensor("EXACT")]; + tensor x_221_cast = gelu(mode = x_221_mode_0, x = input_145_cast); + tensor var_2037_to_fp16 = const()[name = tensor("op_2037_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(709586688)))]; + tensor var_2038_to_fp16 = const()[name = tensor("op_2038_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(722693952)))]; + tensor var_2039_cast = linear(bias = var_2038_to_fp16, weight = var_2037_to_fp16, x = x_221_cast); + tensor x_223_cast = add(x = x_217_cast, y = var_2039_cast); + tensor var_2048 = const()[name = tensor("op_2048"), val = tensor(-1)]; + tensor var_2065_axes_0 = const()[name = tensor("op_2065_axes_0"), val = tensor([-1])]; + tensor blocks_18_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_18_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(722696576)))]; + tensor blocks_18_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_18_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(722699200)))]; + tensor var_2054_to_fp16 = const()[name = tensor("op_2054_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2065_cast = layer_norm(axes = var_2065_axes_0, beta = blocks_18_attn_ln_bias_to_fp16, epsilon = var_2054_to_fp16, gamma = blocks_18_attn_ln_weight_to_fp16, x = x_223_cast); + tensor var_2076_to_fp16 = const()[name = tensor("op_2076_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(722701824)))]; + tensor var_2077_to_fp16 = const()[name = tensor("op_2077_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(725978688)))]; + tensor q_73_cast = linear(bias = var_2077_to_fp16, weight = var_2076_to_fp16, x = var_2065_cast); + tensor var_2080_to_fp16 = const()[name = tensor("op_2080_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(725981312)))]; + tensor k_73_bias_0_to_fp16 = const()[name = tensor("k_73_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(729258176)))]; + tensor k_73_cast = linear(bias = k_73_bias_0_to_fp16, weight = var_2080_to_fp16, x = var_2065_cast); + tensor var_2084_to_fp16 = const()[name = tensor("op_2084_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(729260800)))]; + tensor var_2085_to_fp16 = const()[name = tensor("op_2085_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(732537664)))]; + tensor v_73_cast = linear(bias = var_2085_to_fp16, weight = var_2084_to_fp16, x = var_2065_cast); + tensor var_2093 = const()[name = tensor("op_2093"), val = tensor([1, 1500, 20, -1])]; + tensor var_2094_cast = reshape(shape = var_2093, x = q_73_cast); + tensor const_260_to_fp16 = const()[name = tensor("const_260_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_75_cast = mul(x = var_2094_cast, y = const_260_to_fp16); + tensor var_2100 = const()[name = tensor("op_2100"), val = tensor([1, 1500, 20, -1])]; + tensor var_2101_cast = reshape(shape = var_2100, x = k_73_cast); + tensor const_261_to_fp16 = const()[name = tensor("const_261_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_75_cast = mul(x = var_2101_cast, y = const_261_to_fp16); + tensor var_2107 = const()[name = tensor("op_2107"), val = tensor([1, 1500, 20, -1])]; + tensor var_2108_cast = reshape(shape = var_2107, x = v_73_cast); + tensor var_2109 = const()[name = tensor("op_2109"), val = tensor([0, 2, 1, 3])]; + tensor qk_37_transpose_x_0 = const()[name = tensor("qk_37_transpose_x_0"), val = tensor(false)]; + tensor qk_37_transpose_y_0 = const()[name = tensor("qk_37_transpose_y_0"), val = tensor(false)]; + tensor transpose_100_perm_0 = const()[name = tensor("transpose_100_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_101_perm_0 = const()[name = tensor("transpose_101_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_181 = transpose(perm = transpose_101_perm_0, x = k_75_cast); + tensor transpose_182 = transpose(perm = transpose_100_perm_0, x = q_75_cast); + tensor qk_37_cast = matmul(transpose_x = qk_37_transpose_x_0, transpose_y = qk_37_transpose_y_0, x = transpose_182, y = transpose_181); + tensor var_2113_cast = softmax(axis = var_2048, x = qk_37_cast); + tensor var_2115_transpose_x_0 = const()[name = tensor("op_2115_transpose_x_0"), val = tensor(false)]; + tensor var_2115_transpose_y_0 = const()[name = tensor("op_2115_transpose_y_0"), val = tensor(false)]; + tensor transpose_183 = transpose(perm = var_2109, x = var_2108_cast); + tensor var_2115_cast = matmul(transpose_x = var_2115_transpose_x_0, transpose_y = var_2115_transpose_y_0, x = var_2113_cast, y = transpose_183); + tensor var_2116 = const()[name = tensor("op_2116"), val = tensor([0, 2, 1, 3])]; + tensor concat_18 = const()[name = tensor("concat_18"), val = tensor([1, 1500, 1280])]; + tensor transpose_180 = transpose(perm = var_2116, x = var_2115_cast); + tensor x_227_cast = reshape(shape = concat_18, x = transpose_180); + tensor var_2121_to_fp16 = const()[name = tensor("op_2121_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(732540288)))]; + tensor var_2122_to_fp16 = const()[name = tensor("op_2122_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(735817152)))]; + tensor var_2123_cast = linear(bias = var_2122_to_fp16, weight = var_2121_to_fp16, x = x_227_cast); + tensor x_229_cast = add(x = x_223_cast, y = var_2123_cast); + tensor var_2129_axes_0 = const()[name = tensor("op_2129_axes_0"), val = tensor([-1])]; + tensor blocks_18_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_18_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(735819776)))]; + tensor blocks_18_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_18_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(735822400)))]; + tensor var_2129_cast = layer_norm(axes = var_2129_axes_0, beta = blocks_18_mlp_ln_bias_to_fp16, epsilon = var_2054_to_fp16, gamma = blocks_18_mlp_ln_weight_to_fp16, x = x_229_cast); + tensor var_2138_to_fp16 = const()[name = tensor("op_2138_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(735825024)))]; + tensor var_2139_to_fp16 = const()[name = tensor("op_2139_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(748932288)))]; + tensor input_153_cast = linear(bias = var_2139_to_fp16, weight = var_2138_to_fp16, x = var_2129_cast); + tensor x_233_mode_0 = const()[name = tensor("x_233_mode_0"), val = tensor("EXACT")]; + tensor x_233_cast = gelu(mode = x_233_mode_0, x = input_153_cast); + tensor var_2144_to_fp16 = const()[name = tensor("op_2144_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(748942592)))]; + tensor var_2145_to_fp16 = const()[name = tensor("op_2145_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762049856)))]; + tensor var_2146_cast = linear(bias = var_2145_to_fp16, weight = var_2144_to_fp16, x = x_233_cast); + tensor x_235_cast = add(x = x_229_cast, y = var_2146_cast); + tensor var_2155 = const()[name = tensor("op_2155"), val = tensor(-1)]; + tensor var_2172_axes_0 = const()[name = tensor("op_2172_axes_0"), val = tensor([-1])]; + tensor blocks_19_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_19_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762052480)))]; + tensor blocks_19_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_19_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762055104)))]; + tensor var_2161_to_fp16 = const()[name = tensor("op_2161_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2172_cast = layer_norm(axes = var_2172_axes_0, beta = blocks_19_attn_ln_bias_to_fp16, epsilon = var_2161_to_fp16, gamma = blocks_19_attn_ln_weight_to_fp16, x = x_235_cast); + tensor var_2183_to_fp16 = const()[name = tensor("op_2183_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762057728)))]; + tensor var_2184_to_fp16 = const()[name = tensor("op_2184_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(765334592)))]; + tensor q_77_cast = linear(bias = var_2184_to_fp16, weight = var_2183_to_fp16, x = var_2172_cast); + tensor var_2187_to_fp16 = const()[name = tensor("op_2187_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(765337216)))]; + tensor k_77_bias_0_to_fp16 = const()[name = tensor("k_77_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(768614080)))]; + tensor k_77_cast = linear(bias = k_77_bias_0_to_fp16, weight = var_2187_to_fp16, x = var_2172_cast); + tensor var_2191_to_fp16 = const()[name = tensor("op_2191_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(768616704)))]; + tensor var_2192_to_fp16 = const()[name = tensor("op_2192_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(771893568)))]; + tensor v_77_cast = linear(bias = var_2192_to_fp16, weight = var_2191_to_fp16, x = var_2172_cast); + tensor var_2200 = const()[name = tensor("op_2200"), val = tensor([1, 1500, 20, -1])]; + tensor var_2201_cast = reshape(shape = var_2200, x = q_77_cast); + tensor const_262_to_fp16 = const()[name = tensor("const_262_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_79_cast = mul(x = var_2201_cast, y = const_262_to_fp16); + tensor var_2207 = const()[name = tensor("op_2207"), val = tensor([1, 1500, 20, -1])]; + tensor var_2208_cast = reshape(shape = var_2207, x = k_77_cast); + tensor const_263_to_fp16 = const()[name = tensor("const_263_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_79_cast = mul(x = var_2208_cast, y = const_263_to_fp16); + tensor var_2214 = const()[name = tensor("op_2214"), val = tensor([1, 1500, 20, -1])]; + tensor var_2215_cast = reshape(shape = var_2214, x = v_77_cast); + tensor var_2216 = const()[name = tensor("op_2216"), val = tensor([0, 2, 1, 3])]; + tensor qk_39_transpose_x_0 = const()[name = tensor("qk_39_transpose_x_0"), val = tensor(false)]; + tensor qk_39_transpose_y_0 = const()[name = tensor("qk_39_transpose_y_0"), val = tensor(false)]; + tensor transpose_102_perm_0 = const()[name = tensor("transpose_102_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_103_perm_0 = const()[name = tensor("transpose_103_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_177 = transpose(perm = transpose_103_perm_0, x = k_79_cast); + tensor transpose_178 = transpose(perm = transpose_102_perm_0, x = q_79_cast); + tensor qk_39_cast = matmul(transpose_x = qk_39_transpose_x_0, transpose_y = qk_39_transpose_y_0, x = transpose_178, y = transpose_177); + tensor var_2220_cast = softmax(axis = var_2155, x = qk_39_cast); + tensor var_2222_transpose_x_0 = const()[name = tensor("op_2222_transpose_x_0"), val = tensor(false)]; + tensor var_2222_transpose_y_0 = const()[name = tensor("op_2222_transpose_y_0"), val = tensor(false)]; + tensor transpose_179 = transpose(perm = var_2216, x = var_2215_cast); + tensor var_2222_cast = matmul(transpose_x = var_2222_transpose_x_0, transpose_y = var_2222_transpose_y_0, x = var_2220_cast, y = transpose_179); + tensor var_2223 = const()[name = tensor("op_2223"), val = tensor([0, 2, 1, 3])]; + tensor concat_19 = const()[name = tensor("concat_19"), val = tensor([1, 1500, 1280])]; + tensor transpose_176 = transpose(perm = var_2223, x = var_2222_cast); + tensor x_239_cast = reshape(shape = concat_19, x = transpose_176); + tensor var_2228_to_fp16 = const()[name = tensor("op_2228_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(771896192)))]; + tensor var_2229_to_fp16 = const()[name = tensor("op_2229_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775173056)))]; + tensor var_2230_cast = linear(bias = var_2229_to_fp16, weight = var_2228_to_fp16, x = x_239_cast); + tensor x_241_cast = add(x = x_235_cast, y = var_2230_cast); + tensor var_2236_axes_0 = const()[name = tensor("op_2236_axes_0"), val = tensor([-1])]; + tensor blocks_19_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_19_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775175680)))]; + tensor blocks_19_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_19_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775178304)))]; + tensor var_2236_cast = layer_norm(axes = var_2236_axes_0, beta = blocks_19_mlp_ln_bias_to_fp16, epsilon = var_2161_to_fp16, gamma = blocks_19_mlp_ln_weight_to_fp16, x = x_241_cast); + tensor var_2245_to_fp16 = const()[name = tensor("op_2245_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775180928)))]; + tensor var_2246_to_fp16 = const()[name = tensor("op_2246_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(788288192)))]; + tensor input_161_cast = linear(bias = var_2246_to_fp16, weight = var_2245_to_fp16, x = var_2236_cast); + tensor x_245_mode_0 = const()[name = tensor("x_245_mode_0"), val = tensor("EXACT")]; + tensor x_245_cast = gelu(mode = x_245_mode_0, x = input_161_cast); + tensor var_2251_to_fp16 = const()[name = tensor("op_2251_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(788298496)))]; + tensor var_2252_to_fp16 = const()[name = tensor("op_2252_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(801405760)))]; + tensor var_2253_cast = linear(bias = var_2252_to_fp16, weight = var_2251_to_fp16, x = x_245_cast); + tensor x_247_cast = add(x = x_241_cast, y = var_2253_cast); + tensor var_2262 = const()[name = tensor("op_2262"), val = tensor(-1)]; + tensor var_2279_axes_0 = const()[name = tensor("op_2279_axes_0"), val = tensor([-1])]; + tensor blocks_20_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_20_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(801408384)))]; + tensor blocks_20_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_20_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(801411008)))]; + tensor var_2268_to_fp16 = const()[name = tensor("op_2268_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2279_cast = layer_norm(axes = var_2279_axes_0, beta = blocks_20_attn_ln_bias_to_fp16, epsilon = var_2268_to_fp16, gamma = blocks_20_attn_ln_weight_to_fp16, x = x_247_cast); + tensor var_2290_to_fp16 = const()[name = tensor("op_2290_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(801413632)))]; + tensor var_2291_to_fp16 = const()[name = tensor("op_2291_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(804690496)))]; + tensor q_81_cast = linear(bias = var_2291_to_fp16, weight = var_2290_to_fp16, x = var_2279_cast); + tensor var_2294_to_fp16 = const()[name = tensor("op_2294_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(804693120)))]; + tensor k_81_bias_0_to_fp16 = const()[name = tensor("k_81_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(807969984)))]; + tensor k_81_cast = linear(bias = k_81_bias_0_to_fp16, weight = var_2294_to_fp16, x = var_2279_cast); + tensor var_2298_to_fp16 = const()[name = tensor("op_2298_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(807972608)))]; + tensor var_2299_to_fp16 = const()[name = tensor("op_2299_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(811249472)))]; + tensor v_81_cast = linear(bias = var_2299_to_fp16, weight = var_2298_to_fp16, x = var_2279_cast); + tensor var_2307 = const()[name = tensor("op_2307"), val = tensor([1, 1500, 20, -1])]; + tensor var_2308_cast = reshape(shape = var_2307, x = q_81_cast); + tensor const_264_to_fp16 = const()[name = tensor("const_264_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_83_cast = mul(x = var_2308_cast, y = const_264_to_fp16); + tensor var_2314 = const()[name = tensor("op_2314"), val = tensor([1, 1500, 20, -1])]; + tensor var_2315_cast = reshape(shape = var_2314, x = k_81_cast); + tensor const_265_to_fp16 = const()[name = tensor("const_265_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_83_cast = mul(x = var_2315_cast, y = const_265_to_fp16); + tensor var_2321 = const()[name = tensor("op_2321"), val = tensor([1, 1500, 20, -1])]; + tensor var_2322_cast = reshape(shape = var_2321, x = v_81_cast); + tensor var_2323 = const()[name = tensor("op_2323"), val = tensor([0, 2, 1, 3])]; + tensor qk_41_transpose_x_0 = const()[name = tensor("qk_41_transpose_x_0"), val = tensor(false)]; + tensor qk_41_transpose_y_0 = const()[name = tensor("qk_41_transpose_y_0"), val = tensor(false)]; + tensor transpose_104_perm_0 = const()[name = tensor("transpose_104_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_105_perm_0 = const()[name = tensor("transpose_105_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_173 = transpose(perm = transpose_105_perm_0, x = k_83_cast); + tensor transpose_174 = transpose(perm = transpose_104_perm_0, x = q_83_cast); + tensor qk_41_cast = matmul(transpose_x = qk_41_transpose_x_0, transpose_y = qk_41_transpose_y_0, x = transpose_174, y = transpose_173); + tensor var_2327_cast = softmax(axis = var_2262, x = qk_41_cast); + tensor var_2329_transpose_x_0 = const()[name = tensor("op_2329_transpose_x_0"), val = tensor(false)]; + tensor var_2329_transpose_y_0 = const()[name = tensor("op_2329_transpose_y_0"), val = tensor(false)]; + tensor transpose_175 = transpose(perm = var_2323, x = var_2322_cast); + tensor var_2329_cast = matmul(transpose_x = var_2329_transpose_x_0, transpose_y = var_2329_transpose_y_0, x = var_2327_cast, y = transpose_175); + tensor var_2330 = const()[name = tensor("op_2330"), val = tensor([0, 2, 1, 3])]; + tensor concat_20 = const()[name = tensor("concat_20"), val = tensor([1, 1500, 1280])]; + tensor transpose_172 = transpose(perm = var_2330, x = var_2329_cast); + tensor x_251_cast = reshape(shape = concat_20, x = transpose_172); + tensor var_2335_to_fp16 = const()[name = tensor("op_2335_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(811252096)))]; + tensor var_2336_to_fp16 = const()[name = tensor("op_2336_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814528960)))]; + tensor var_2337_cast = linear(bias = var_2336_to_fp16, weight = var_2335_to_fp16, x = x_251_cast); + tensor x_253_cast = add(x = x_247_cast, y = var_2337_cast); + tensor var_2343_axes_0 = const()[name = tensor("op_2343_axes_0"), val = tensor([-1])]; + tensor blocks_20_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_20_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814531584)))]; + tensor blocks_20_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_20_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814534208)))]; + tensor var_2343_cast = layer_norm(axes = var_2343_axes_0, beta = blocks_20_mlp_ln_bias_to_fp16, epsilon = var_2268_to_fp16, gamma = blocks_20_mlp_ln_weight_to_fp16, x = x_253_cast); + tensor var_2352_to_fp16 = const()[name = tensor("op_2352_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814536832)))]; + tensor var_2353_to_fp16 = const()[name = tensor("op_2353_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(827644096)))]; + tensor input_169_cast = linear(bias = var_2353_to_fp16, weight = var_2352_to_fp16, x = var_2343_cast); + tensor x_257_mode_0 = const()[name = tensor("x_257_mode_0"), val = tensor("EXACT")]; + tensor x_257_cast = gelu(mode = x_257_mode_0, x = input_169_cast); + tensor var_2358_to_fp16 = const()[name = tensor("op_2358_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(827654400)))]; + tensor var_2359_to_fp16 = const()[name = tensor("op_2359_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(840761664)))]; + tensor var_2360_cast = linear(bias = var_2359_to_fp16, weight = var_2358_to_fp16, x = x_257_cast); + tensor x_259_cast = add(x = x_253_cast, y = var_2360_cast); + tensor var_2369 = const()[name = tensor("op_2369"), val = tensor(-1)]; + tensor var_2386_axes_0 = const()[name = tensor("op_2386_axes_0"), val = tensor([-1])]; + tensor blocks_21_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_21_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(840764288)))]; + tensor blocks_21_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_21_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(840766912)))]; + tensor var_2375_to_fp16 = const()[name = tensor("op_2375_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2386_cast = layer_norm(axes = var_2386_axes_0, beta = blocks_21_attn_ln_bias_to_fp16, epsilon = var_2375_to_fp16, gamma = blocks_21_attn_ln_weight_to_fp16, x = x_259_cast); + tensor var_2397_to_fp16 = const()[name = tensor("op_2397_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(840769536)))]; + tensor var_2398_to_fp16 = const()[name = tensor("op_2398_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(844046400)))]; + tensor q_85_cast = linear(bias = var_2398_to_fp16, weight = var_2397_to_fp16, x = var_2386_cast); + tensor var_2401_to_fp16 = const()[name = tensor("op_2401_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(844049024)))]; + tensor k_85_bias_0_to_fp16 = const()[name = tensor("k_85_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(847325888)))]; + tensor k_85_cast = linear(bias = k_85_bias_0_to_fp16, weight = var_2401_to_fp16, x = var_2386_cast); + tensor var_2405_to_fp16 = const()[name = tensor("op_2405_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(847328512)))]; + tensor var_2406_to_fp16 = const()[name = tensor("op_2406_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(850605376)))]; + tensor v_85_cast = linear(bias = var_2406_to_fp16, weight = var_2405_to_fp16, x = var_2386_cast); + tensor var_2414 = const()[name = tensor("op_2414"), val = tensor([1, 1500, 20, -1])]; + tensor var_2415_cast = reshape(shape = var_2414, x = q_85_cast); + tensor const_266_to_fp16 = const()[name = tensor("const_266_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_87_cast = mul(x = var_2415_cast, y = const_266_to_fp16); + tensor var_2421 = const()[name = tensor("op_2421"), val = tensor([1, 1500, 20, -1])]; + tensor var_2422_cast = reshape(shape = var_2421, x = k_85_cast); + tensor const_267_to_fp16 = const()[name = tensor("const_267_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_87_cast = mul(x = var_2422_cast, y = const_267_to_fp16); + tensor var_2428 = const()[name = tensor("op_2428"), val = tensor([1, 1500, 20, -1])]; + tensor var_2429_cast = reshape(shape = var_2428, x = v_85_cast); + tensor var_2430 = const()[name = tensor("op_2430"), val = tensor([0, 2, 1, 3])]; + tensor qk_43_transpose_x_0 = const()[name = tensor("qk_43_transpose_x_0"), val = tensor(false)]; + tensor qk_43_transpose_y_0 = const()[name = tensor("qk_43_transpose_y_0"), val = tensor(false)]; + tensor transpose_106_perm_0 = const()[name = tensor("transpose_106_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_107_perm_0 = const()[name = tensor("transpose_107_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_169 = transpose(perm = transpose_107_perm_0, x = k_87_cast); + tensor transpose_170 = transpose(perm = transpose_106_perm_0, x = q_87_cast); + tensor qk_43_cast = matmul(transpose_x = qk_43_transpose_x_0, transpose_y = qk_43_transpose_y_0, x = transpose_170, y = transpose_169); + tensor var_2434_cast = softmax(axis = var_2369, x = qk_43_cast); + tensor var_2436_transpose_x_0 = const()[name = tensor("op_2436_transpose_x_0"), val = tensor(false)]; + tensor var_2436_transpose_y_0 = const()[name = tensor("op_2436_transpose_y_0"), val = tensor(false)]; + tensor transpose_171 = transpose(perm = var_2430, x = var_2429_cast); + tensor var_2436_cast = matmul(transpose_x = var_2436_transpose_x_0, transpose_y = var_2436_transpose_y_0, x = var_2434_cast, y = transpose_171); + tensor var_2437 = const()[name = tensor("op_2437"), val = tensor([0, 2, 1, 3])]; + tensor concat_21 = const()[name = tensor("concat_21"), val = tensor([1, 1500, 1280])]; + tensor transpose_168 = transpose(perm = var_2437, x = var_2436_cast); + tensor x_263_cast = reshape(shape = concat_21, x = transpose_168); + tensor var_2442_to_fp16 = const()[name = tensor("op_2442_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(850608000)))]; + tensor var_2443_to_fp16 = const()[name = tensor("op_2443_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(853884864)))]; + tensor var_2444_cast = linear(bias = var_2443_to_fp16, weight = var_2442_to_fp16, x = x_263_cast); + tensor x_265_cast = add(x = x_259_cast, y = var_2444_cast); + tensor var_2450_axes_0 = const()[name = tensor("op_2450_axes_0"), val = tensor([-1])]; + tensor blocks_21_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_21_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(853887488)))]; + tensor blocks_21_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_21_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(853890112)))]; + tensor var_2450_cast = layer_norm(axes = var_2450_axes_0, beta = blocks_21_mlp_ln_bias_to_fp16, epsilon = var_2375_to_fp16, gamma = blocks_21_mlp_ln_weight_to_fp16, x = x_265_cast); + tensor var_2459_to_fp16 = const()[name = tensor("op_2459_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(853892736)))]; + tensor var_2460_to_fp16 = const()[name = tensor("op_2460_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(867000000)))]; + tensor input_177_cast = linear(bias = var_2460_to_fp16, weight = var_2459_to_fp16, x = var_2450_cast); + tensor x_269_mode_0 = const()[name = tensor("x_269_mode_0"), val = tensor("EXACT")]; + tensor x_269_cast = gelu(mode = x_269_mode_0, x = input_177_cast); + tensor var_2465_to_fp16 = const()[name = tensor("op_2465_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(867010304)))]; + tensor var_2466_to_fp16 = const()[name = tensor("op_2466_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880117568)))]; + tensor var_2467_cast = linear(bias = var_2466_to_fp16, weight = var_2465_to_fp16, x = x_269_cast); + tensor x_271_cast = add(x = x_265_cast, y = var_2467_cast); + tensor var_2476 = const()[name = tensor("op_2476"), val = tensor(-1)]; + tensor var_2493_axes_0 = const()[name = tensor("op_2493_axes_0"), val = tensor([-1])]; + tensor blocks_22_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_22_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880120192)))]; + tensor blocks_22_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_22_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880122816)))]; + tensor var_2482_to_fp16 = const()[name = tensor("op_2482_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2493_cast = layer_norm(axes = var_2493_axes_0, beta = blocks_22_attn_ln_bias_to_fp16, epsilon = var_2482_to_fp16, gamma = blocks_22_attn_ln_weight_to_fp16, x = x_271_cast); + tensor var_2504_to_fp16 = const()[name = tensor("op_2504_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880125440)))]; + tensor var_2505_to_fp16 = const()[name = tensor("op_2505_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(883402304)))]; + tensor q_89_cast = linear(bias = var_2505_to_fp16, weight = var_2504_to_fp16, x = var_2493_cast); + tensor var_2508_to_fp16 = const()[name = tensor("op_2508_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(883404928)))]; + tensor k_89_bias_0_to_fp16 = const()[name = tensor("k_89_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(886681792)))]; + tensor k_89_cast = linear(bias = k_89_bias_0_to_fp16, weight = var_2508_to_fp16, x = var_2493_cast); + tensor var_2512_to_fp16 = const()[name = tensor("op_2512_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(886684416)))]; + tensor var_2513_to_fp16 = const()[name = tensor("op_2513_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(889961280)))]; + tensor v_89_cast = linear(bias = var_2513_to_fp16, weight = var_2512_to_fp16, x = var_2493_cast); + tensor var_2521 = const()[name = tensor("op_2521"), val = tensor([1, 1500, 20, -1])]; + tensor var_2522_cast = reshape(shape = var_2521, x = q_89_cast); + tensor const_268_to_fp16 = const()[name = tensor("const_268_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_91_cast = mul(x = var_2522_cast, y = const_268_to_fp16); + tensor var_2528 = const()[name = tensor("op_2528"), val = tensor([1, 1500, 20, -1])]; + tensor var_2529_cast = reshape(shape = var_2528, x = k_89_cast); + tensor const_269_to_fp16 = const()[name = tensor("const_269_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_91_cast = mul(x = var_2529_cast, y = const_269_to_fp16); + tensor var_2535 = const()[name = tensor("op_2535"), val = tensor([1, 1500, 20, -1])]; + tensor var_2536_cast = reshape(shape = var_2535, x = v_89_cast); + tensor var_2537 = const()[name = tensor("op_2537"), val = tensor([0, 2, 1, 3])]; + tensor qk_45_transpose_x_0 = const()[name = tensor("qk_45_transpose_x_0"), val = tensor(false)]; + tensor qk_45_transpose_y_0 = const()[name = tensor("qk_45_transpose_y_0"), val = tensor(false)]; + tensor transpose_108_perm_0 = const()[name = tensor("transpose_108_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_109_perm_0 = const()[name = tensor("transpose_109_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_165 = transpose(perm = transpose_109_perm_0, x = k_91_cast); + tensor transpose_166 = transpose(perm = transpose_108_perm_0, x = q_91_cast); + tensor qk_45_cast = matmul(transpose_x = qk_45_transpose_x_0, transpose_y = qk_45_transpose_y_0, x = transpose_166, y = transpose_165); + tensor var_2541_cast = softmax(axis = var_2476, x = qk_45_cast); + tensor var_2543_transpose_x_0 = const()[name = tensor("op_2543_transpose_x_0"), val = tensor(false)]; + tensor var_2543_transpose_y_0 = const()[name = tensor("op_2543_transpose_y_0"), val = tensor(false)]; + tensor transpose_167 = transpose(perm = var_2537, x = var_2536_cast); + tensor var_2543_cast = matmul(transpose_x = var_2543_transpose_x_0, transpose_y = var_2543_transpose_y_0, x = var_2541_cast, y = transpose_167); + tensor var_2544 = const()[name = tensor("op_2544"), val = tensor([0, 2, 1, 3])]; + tensor concat_22 = const()[name = tensor("concat_22"), val = tensor([1, 1500, 1280])]; + tensor transpose_164 = transpose(perm = var_2544, x = var_2543_cast); + tensor x_275_cast = reshape(shape = concat_22, x = transpose_164); + tensor var_2549_to_fp16 = const()[name = tensor("op_2549_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(889963904)))]; + tensor var_2550_to_fp16 = const()[name = tensor("op_2550_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893240768)))]; + tensor var_2551_cast = linear(bias = var_2550_to_fp16, weight = var_2549_to_fp16, x = x_275_cast); + tensor x_277_cast = add(x = x_271_cast, y = var_2551_cast); + tensor var_2557_axes_0 = const()[name = tensor("op_2557_axes_0"), val = tensor([-1])]; + tensor blocks_22_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_22_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893243392)))]; + tensor blocks_22_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_22_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893246016)))]; + tensor var_2557_cast = layer_norm(axes = var_2557_axes_0, beta = blocks_22_mlp_ln_bias_to_fp16, epsilon = var_2482_to_fp16, gamma = blocks_22_mlp_ln_weight_to_fp16, x = x_277_cast); + tensor var_2566_to_fp16 = const()[name = tensor("op_2566_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893248640)))]; + tensor var_2567_to_fp16 = const()[name = tensor("op_2567_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(906355904)))]; + tensor input_185_cast = linear(bias = var_2567_to_fp16, weight = var_2566_to_fp16, x = var_2557_cast); + tensor x_281_mode_0 = const()[name = tensor("x_281_mode_0"), val = tensor("EXACT")]; + tensor x_281_cast = gelu(mode = x_281_mode_0, x = input_185_cast); + tensor var_2572_to_fp16 = const()[name = tensor("op_2572_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(906366208)))]; + tensor var_2573_to_fp16 = const()[name = tensor("op_2573_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(919473472)))]; + tensor var_2574_cast = linear(bias = var_2573_to_fp16, weight = var_2572_to_fp16, x = x_281_cast); + tensor x_283_cast = add(x = x_277_cast, y = var_2574_cast); + tensor var_2583 = const()[name = tensor("op_2583"), val = tensor(-1)]; + tensor var_2600_axes_0 = const()[name = tensor("op_2600_axes_0"), val = tensor([-1])]; + tensor blocks_23_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_23_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(919476096)))]; + tensor blocks_23_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_23_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(919478720)))]; + tensor var_2589_to_fp16 = const()[name = tensor("op_2589_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2600_cast = layer_norm(axes = var_2600_axes_0, beta = blocks_23_attn_ln_bias_to_fp16, epsilon = var_2589_to_fp16, gamma = blocks_23_attn_ln_weight_to_fp16, x = x_283_cast); + tensor var_2611_to_fp16 = const()[name = tensor("op_2611_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(919481344)))]; + tensor var_2612_to_fp16 = const()[name = tensor("op_2612_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(922758208)))]; + tensor q_93_cast = linear(bias = var_2612_to_fp16, weight = var_2611_to_fp16, x = var_2600_cast); + tensor var_2615_to_fp16 = const()[name = tensor("op_2615_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(922760832)))]; + tensor k_93_bias_0_to_fp16 = const()[name = tensor("k_93_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(926037696)))]; + tensor k_93_cast = linear(bias = k_93_bias_0_to_fp16, weight = var_2615_to_fp16, x = var_2600_cast); + tensor var_2619_to_fp16 = const()[name = tensor("op_2619_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(926040320)))]; + tensor var_2620_to_fp16 = const()[name = tensor("op_2620_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(929317184)))]; + tensor v_93_cast = linear(bias = var_2620_to_fp16, weight = var_2619_to_fp16, x = var_2600_cast); + tensor var_2628 = const()[name = tensor("op_2628"), val = tensor([1, 1500, 20, -1])]; + tensor var_2629_cast = reshape(shape = var_2628, x = q_93_cast); + tensor const_270_to_fp16 = const()[name = tensor("const_270_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_95_cast = mul(x = var_2629_cast, y = const_270_to_fp16); + tensor var_2635 = const()[name = tensor("op_2635"), val = tensor([1, 1500, 20, -1])]; + tensor var_2636_cast = reshape(shape = var_2635, x = k_93_cast); + tensor const_271_to_fp16 = const()[name = tensor("const_271_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_95_cast = mul(x = var_2636_cast, y = const_271_to_fp16); + tensor var_2642 = const()[name = tensor("op_2642"), val = tensor([1, 1500, 20, -1])]; + tensor var_2643_cast = reshape(shape = var_2642, x = v_93_cast); + tensor var_2644 = const()[name = tensor("op_2644"), val = tensor([0, 2, 1, 3])]; + tensor qk_47_transpose_x_0 = const()[name = tensor("qk_47_transpose_x_0"), val = tensor(false)]; + tensor qk_47_transpose_y_0 = const()[name = tensor("qk_47_transpose_y_0"), val = tensor(false)]; + tensor transpose_110_perm_0 = const()[name = tensor("transpose_110_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_111_perm_0 = const()[name = tensor("transpose_111_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_161 = transpose(perm = transpose_111_perm_0, x = k_95_cast); + tensor transpose_162 = transpose(perm = transpose_110_perm_0, x = q_95_cast); + tensor qk_47_cast = matmul(transpose_x = qk_47_transpose_x_0, transpose_y = qk_47_transpose_y_0, x = transpose_162, y = transpose_161); + tensor var_2648_cast = softmax(axis = var_2583, x = qk_47_cast); + tensor var_2650_transpose_x_0 = const()[name = tensor("op_2650_transpose_x_0"), val = tensor(false)]; + tensor var_2650_transpose_y_0 = const()[name = tensor("op_2650_transpose_y_0"), val = tensor(false)]; + tensor transpose_163 = transpose(perm = var_2644, x = var_2643_cast); + tensor var_2650_cast = matmul(transpose_x = var_2650_transpose_x_0, transpose_y = var_2650_transpose_y_0, x = var_2648_cast, y = transpose_163); + tensor var_2651 = const()[name = tensor("op_2651"), val = tensor([0, 2, 1, 3])]; + tensor concat_23 = const()[name = tensor("concat_23"), val = tensor([1, 1500, 1280])]; + tensor transpose_160 = transpose(perm = var_2651, x = var_2650_cast); + tensor x_287_cast = reshape(shape = concat_23, x = transpose_160); + tensor var_2656_to_fp16 = const()[name = tensor("op_2656_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(929319808)))]; + tensor var_2657_to_fp16 = const()[name = tensor("op_2657_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(932596672)))]; + tensor var_2658_cast = linear(bias = var_2657_to_fp16, weight = var_2656_to_fp16, x = x_287_cast); + tensor x_289_cast = add(x = x_283_cast, y = var_2658_cast); + tensor var_2664_axes_0 = const()[name = tensor("op_2664_axes_0"), val = tensor([-1])]; + tensor blocks_23_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_23_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(932599296)))]; + tensor blocks_23_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_23_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(932601920)))]; + tensor var_2664_cast = layer_norm(axes = var_2664_axes_0, beta = blocks_23_mlp_ln_bias_to_fp16, epsilon = var_2589_to_fp16, gamma = blocks_23_mlp_ln_weight_to_fp16, x = x_289_cast); + tensor var_2673_to_fp16 = const()[name = tensor("op_2673_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(932604544)))]; + tensor var_2674_to_fp16 = const()[name = tensor("op_2674_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(945711808)))]; + tensor input_193_cast = linear(bias = var_2674_to_fp16, weight = var_2673_to_fp16, x = var_2664_cast); + tensor x_293_mode_0 = const()[name = tensor("x_293_mode_0"), val = tensor("EXACT")]; + tensor x_293_cast = gelu(mode = x_293_mode_0, x = input_193_cast); + tensor var_2679_to_fp16 = const()[name = tensor("op_2679_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(945722112)))]; + tensor var_2680_to_fp16 = const()[name = tensor("op_2680_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(958829376)))]; + tensor var_2681_cast = linear(bias = var_2680_to_fp16, weight = var_2679_to_fp16, x = x_293_cast); + tensor x_295_cast = add(x = x_289_cast, y = var_2681_cast); + tensor var_2690 = const()[name = tensor("op_2690"), val = tensor(-1)]; + tensor var_2707_axes_0 = const()[name = tensor("op_2707_axes_0"), val = tensor([-1])]; + tensor blocks_24_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_24_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(958832000)))]; + tensor blocks_24_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_24_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(958834624)))]; + tensor var_2696_to_fp16 = const()[name = tensor("op_2696_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2707_cast = layer_norm(axes = var_2707_axes_0, beta = blocks_24_attn_ln_bias_to_fp16, epsilon = var_2696_to_fp16, gamma = blocks_24_attn_ln_weight_to_fp16, x = x_295_cast); + tensor var_2718_to_fp16 = const()[name = tensor("op_2718_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(958837248)))]; + tensor var_2719_to_fp16 = const()[name = tensor("op_2719_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(962114112)))]; + tensor q_97_cast = linear(bias = var_2719_to_fp16, weight = var_2718_to_fp16, x = var_2707_cast); + tensor var_2722_to_fp16 = const()[name = tensor("op_2722_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(962116736)))]; + tensor k_97_bias_0_to_fp16 = const()[name = tensor("k_97_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(965393600)))]; + tensor k_97_cast = linear(bias = k_97_bias_0_to_fp16, weight = var_2722_to_fp16, x = var_2707_cast); + tensor var_2726_to_fp16 = const()[name = tensor("op_2726_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(965396224)))]; + tensor var_2727_to_fp16 = const()[name = tensor("op_2727_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(968673088)))]; + tensor v_97_cast = linear(bias = var_2727_to_fp16, weight = var_2726_to_fp16, x = var_2707_cast); + tensor var_2735 = const()[name = tensor("op_2735"), val = tensor([1, 1500, 20, -1])]; + tensor var_2736_cast = reshape(shape = var_2735, x = q_97_cast); + tensor const_272_to_fp16 = const()[name = tensor("const_272_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_99_cast = mul(x = var_2736_cast, y = const_272_to_fp16); + tensor var_2742 = const()[name = tensor("op_2742"), val = tensor([1, 1500, 20, -1])]; + tensor var_2743_cast = reshape(shape = var_2742, x = k_97_cast); + tensor const_273_to_fp16 = const()[name = tensor("const_273_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_99_cast = mul(x = var_2743_cast, y = const_273_to_fp16); + tensor var_2749 = const()[name = tensor("op_2749"), val = tensor([1, 1500, 20, -1])]; + tensor var_2750_cast = reshape(shape = var_2749, x = v_97_cast); + tensor var_2751 = const()[name = tensor("op_2751"), val = tensor([0, 2, 1, 3])]; + tensor qk_49_transpose_x_0 = const()[name = tensor("qk_49_transpose_x_0"), val = tensor(false)]; + tensor qk_49_transpose_y_0 = const()[name = tensor("qk_49_transpose_y_0"), val = tensor(false)]; + tensor transpose_112_perm_0 = const()[name = tensor("transpose_112_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_113_perm_0 = const()[name = tensor("transpose_113_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_157 = transpose(perm = transpose_113_perm_0, x = k_99_cast); + tensor transpose_158 = transpose(perm = transpose_112_perm_0, x = q_99_cast); + tensor qk_49_cast = matmul(transpose_x = qk_49_transpose_x_0, transpose_y = qk_49_transpose_y_0, x = transpose_158, y = transpose_157); + tensor var_2755_cast = softmax(axis = var_2690, x = qk_49_cast); + tensor var_2757_transpose_x_0 = const()[name = tensor("op_2757_transpose_x_0"), val = tensor(false)]; + tensor var_2757_transpose_y_0 = const()[name = tensor("op_2757_transpose_y_0"), val = tensor(false)]; + tensor transpose_159 = transpose(perm = var_2751, x = var_2750_cast); + tensor var_2757_cast = matmul(transpose_x = var_2757_transpose_x_0, transpose_y = var_2757_transpose_y_0, x = var_2755_cast, y = transpose_159); + tensor var_2758 = const()[name = tensor("op_2758"), val = tensor([0, 2, 1, 3])]; + tensor concat_24 = const()[name = tensor("concat_24"), val = tensor([1, 1500, 1280])]; + tensor transpose_156 = transpose(perm = var_2758, x = var_2757_cast); + tensor x_299_cast = reshape(shape = concat_24, x = transpose_156); + tensor var_2763_to_fp16 = const()[name = tensor("op_2763_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(968675712)))]; + tensor var_2764_to_fp16 = const()[name = tensor("op_2764_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(971952576)))]; + tensor var_2765_cast = linear(bias = var_2764_to_fp16, weight = var_2763_to_fp16, x = x_299_cast); + tensor x_301_cast = add(x = x_295_cast, y = var_2765_cast); + tensor var_2771_axes_0 = const()[name = tensor("op_2771_axes_0"), val = tensor([-1])]; + tensor blocks_24_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_24_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(971955200)))]; + tensor blocks_24_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_24_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(971957824)))]; + tensor var_2771_cast = layer_norm(axes = var_2771_axes_0, beta = blocks_24_mlp_ln_bias_to_fp16, epsilon = var_2696_to_fp16, gamma = blocks_24_mlp_ln_weight_to_fp16, x = x_301_cast); + tensor var_2780_to_fp16 = const()[name = tensor("op_2780_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(971960448)))]; + tensor var_2781_to_fp16 = const()[name = tensor("op_2781_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(985067712)))]; + tensor input_201_cast = linear(bias = var_2781_to_fp16, weight = var_2780_to_fp16, x = var_2771_cast); + tensor x_305_mode_0 = const()[name = tensor("x_305_mode_0"), val = tensor("EXACT")]; + tensor x_305_cast = gelu(mode = x_305_mode_0, x = input_201_cast); + tensor var_2786_to_fp16 = const()[name = tensor("op_2786_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(985078016)))]; + tensor var_2787_to_fp16 = const()[name = tensor("op_2787_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(998185280)))]; + tensor var_2788_cast = linear(bias = var_2787_to_fp16, weight = var_2786_to_fp16, x = x_305_cast); + tensor x_307_cast = add(x = x_301_cast, y = var_2788_cast); + tensor var_2797 = const()[name = tensor("op_2797"), val = tensor(-1)]; + tensor var_2814_axes_0 = const()[name = tensor("op_2814_axes_0"), val = tensor([-1])]; + tensor blocks_25_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_25_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(998187904)))]; + tensor blocks_25_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_25_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(998190528)))]; + tensor var_2803_to_fp16 = const()[name = tensor("op_2803_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2814_cast = layer_norm(axes = var_2814_axes_0, beta = blocks_25_attn_ln_bias_to_fp16, epsilon = var_2803_to_fp16, gamma = blocks_25_attn_ln_weight_to_fp16, x = x_307_cast); + tensor var_2825_to_fp16 = const()[name = tensor("op_2825_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(998193152)))]; + tensor var_2826_to_fp16 = const()[name = tensor("op_2826_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1001470016)))]; + tensor q_101_cast = linear(bias = var_2826_to_fp16, weight = var_2825_to_fp16, x = var_2814_cast); + tensor var_2829_to_fp16 = const()[name = tensor("op_2829_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1001472640)))]; + tensor k_101_bias_0_to_fp16 = const()[name = tensor("k_101_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1004749504)))]; + tensor k_101_cast = linear(bias = k_101_bias_0_to_fp16, weight = var_2829_to_fp16, x = var_2814_cast); + tensor var_2833_to_fp16 = const()[name = tensor("op_2833_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1004752128)))]; + tensor var_2834_to_fp16 = const()[name = tensor("op_2834_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1008028992)))]; + tensor v_101_cast = linear(bias = var_2834_to_fp16, weight = var_2833_to_fp16, x = var_2814_cast); + tensor var_2842 = const()[name = tensor("op_2842"), val = tensor([1, 1500, 20, -1])]; + tensor var_2843_cast = reshape(shape = var_2842, x = q_101_cast); + tensor const_274_to_fp16 = const()[name = tensor("const_274_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_103_cast = mul(x = var_2843_cast, y = const_274_to_fp16); + tensor var_2849 = const()[name = tensor("op_2849"), val = tensor([1, 1500, 20, -1])]; + tensor var_2850_cast = reshape(shape = var_2849, x = k_101_cast); + tensor const_275_to_fp16 = const()[name = tensor("const_275_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_103_cast = mul(x = var_2850_cast, y = const_275_to_fp16); + tensor var_2856 = const()[name = tensor("op_2856"), val = tensor([1, 1500, 20, -1])]; + tensor var_2857_cast = reshape(shape = var_2856, x = v_101_cast); + tensor var_2858 = const()[name = tensor("op_2858"), val = tensor([0, 2, 1, 3])]; + tensor qk_51_transpose_x_0 = const()[name = tensor("qk_51_transpose_x_0"), val = tensor(false)]; + tensor qk_51_transpose_y_0 = const()[name = tensor("qk_51_transpose_y_0"), val = tensor(false)]; + tensor transpose_114_perm_0 = const()[name = tensor("transpose_114_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_115_perm_0 = const()[name = tensor("transpose_115_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_153 = transpose(perm = transpose_115_perm_0, x = k_103_cast); + tensor transpose_154 = transpose(perm = transpose_114_perm_0, x = q_103_cast); + tensor qk_51_cast = matmul(transpose_x = qk_51_transpose_x_0, transpose_y = qk_51_transpose_y_0, x = transpose_154, y = transpose_153); + tensor var_2862_cast = softmax(axis = var_2797, x = qk_51_cast); + tensor var_2864_transpose_x_0 = const()[name = tensor("op_2864_transpose_x_0"), val = tensor(false)]; + tensor var_2864_transpose_y_0 = const()[name = tensor("op_2864_transpose_y_0"), val = tensor(false)]; + tensor transpose_155 = transpose(perm = var_2858, x = var_2857_cast); + tensor var_2864_cast = matmul(transpose_x = var_2864_transpose_x_0, transpose_y = var_2864_transpose_y_0, x = var_2862_cast, y = transpose_155); + tensor var_2865 = const()[name = tensor("op_2865"), val = tensor([0, 2, 1, 3])]; + tensor concat_25 = const()[name = tensor("concat_25"), val = tensor([1, 1500, 1280])]; + tensor transpose_152 = transpose(perm = var_2865, x = var_2864_cast); + tensor x_311_cast = reshape(shape = concat_25, x = transpose_152); + tensor var_2870_to_fp16 = const()[name = tensor("op_2870_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1008031616)))]; + tensor var_2871_to_fp16 = const()[name = tensor("op_2871_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1011308480)))]; + tensor var_2872_cast = linear(bias = var_2871_to_fp16, weight = var_2870_to_fp16, x = x_311_cast); + tensor x_313_cast = add(x = x_307_cast, y = var_2872_cast); + tensor var_2878_axes_0 = const()[name = tensor("op_2878_axes_0"), val = tensor([-1])]; + tensor blocks_25_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_25_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1011311104)))]; + tensor blocks_25_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_25_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1011313728)))]; + tensor var_2878_cast = layer_norm(axes = var_2878_axes_0, beta = blocks_25_mlp_ln_bias_to_fp16, epsilon = var_2803_to_fp16, gamma = blocks_25_mlp_ln_weight_to_fp16, x = x_313_cast); + tensor var_2887_to_fp16 = const()[name = tensor("op_2887_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1011316352)))]; + tensor var_2888_to_fp16 = const()[name = tensor("op_2888_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1024423616)))]; + tensor input_209_cast = linear(bias = var_2888_to_fp16, weight = var_2887_to_fp16, x = var_2878_cast); + tensor x_317_mode_0 = const()[name = tensor("x_317_mode_0"), val = tensor("EXACT")]; + tensor x_317_cast = gelu(mode = x_317_mode_0, x = input_209_cast); + tensor var_2893_to_fp16 = const()[name = tensor("op_2893_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1024433920)))]; + tensor var_2894_to_fp16 = const()[name = tensor("op_2894_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1037541184)))]; + tensor var_2895_cast = linear(bias = var_2894_to_fp16, weight = var_2893_to_fp16, x = x_317_cast); + tensor x_319_cast = add(x = x_313_cast, y = var_2895_cast); + tensor var_2904 = const()[name = tensor("op_2904"), val = tensor(-1)]; + tensor var_2921_axes_0 = const()[name = tensor("op_2921_axes_0"), val = tensor([-1])]; + tensor blocks_26_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_26_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1037543808)))]; + tensor blocks_26_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_26_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1037546432)))]; + tensor var_2910_to_fp16 = const()[name = tensor("op_2910_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2921_cast = layer_norm(axes = var_2921_axes_0, beta = blocks_26_attn_ln_bias_to_fp16, epsilon = var_2910_to_fp16, gamma = blocks_26_attn_ln_weight_to_fp16, x = x_319_cast); + tensor var_2932_to_fp16 = const()[name = tensor("op_2932_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1037549056)))]; + tensor var_2933_to_fp16 = const()[name = tensor("op_2933_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1040825920)))]; + tensor q_105_cast = linear(bias = var_2933_to_fp16, weight = var_2932_to_fp16, x = var_2921_cast); + tensor var_2936_to_fp16 = const()[name = tensor("op_2936_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1040828544)))]; + tensor k_105_bias_0_to_fp16 = const()[name = tensor("k_105_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1044105408)))]; + tensor k_105_cast = linear(bias = k_105_bias_0_to_fp16, weight = var_2936_to_fp16, x = var_2921_cast); + tensor var_2940_to_fp16 = const()[name = tensor("op_2940_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1044108032)))]; + tensor var_2941_to_fp16 = const()[name = tensor("op_2941_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1047384896)))]; + tensor v_105_cast = linear(bias = var_2941_to_fp16, weight = var_2940_to_fp16, x = var_2921_cast); + tensor var_2949 = const()[name = tensor("op_2949"), val = tensor([1, 1500, 20, -1])]; + tensor var_2950_cast = reshape(shape = var_2949, x = q_105_cast); + tensor const_276_to_fp16 = const()[name = tensor("const_276_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_107_cast = mul(x = var_2950_cast, y = const_276_to_fp16); + tensor var_2956 = const()[name = tensor("op_2956"), val = tensor([1, 1500, 20, -1])]; + tensor var_2957_cast = reshape(shape = var_2956, x = k_105_cast); + tensor const_277_to_fp16 = const()[name = tensor("const_277_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_107_cast = mul(x = var_2957_cast, y = const_277_to_fp16); + tensor var_2963 = const()[name = tensor("op_2963"), val = tensor([1, 1500, 20, -1])]; + tensor var_2964_cast = reshape(shape = var_2963, x = v_105_cast); + tensor var_2965 = const()[name = tensor("op_2965"), val = tensor([0, 2, 1, 3])]; + tensor qk_53_transpose_x_0 = const()[name = tensor("qk_53_transpose_x_0"), val = tensor(false)]; + tensor qk_53_transpose_y_0 = const()[name = tensor("qk_53_transpose_y_0"), val = tensor(false)]; + tensor transpose_116_perm_0 = const()[name = tensor("transpose_116_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_117_perm_0 = const()[name = tensor("transpose_117_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_149 = transpose(perm = transpose_117_perm_0, x = k_107_cast); + tensor transpose_150 = transpose(perm = transpose_116_perm_0, x = q_107_cast); + tensor qk_53_cast = matmul(transpose_x = qk_53_transpose_x_0, transpose_y = qk_53_transpose_y_0, x = transpose_150, y = transpose_149); + tensor var_2969_cast = softmax(axis = var_2904, x = qk_53_cast); + tensor var_2971_transpose_x_0 = const()[name = tensor("op_2971_transpose_x_0"), val = tensor(false)]; + tensor var_2971_transpose_y_0 = const()[name = tensor("op_2971_transpose_y_0"), val = tensor(false)]; + tensor transpose_151 = transpose(perm = var_2965, x = var_2964_cast); + tensor var_2971_cast = matmul(transpose_x = var_2971_transpose_x_0, transpose_y = var_2971_transpose_y_0, x = var_2969_cast, y = transpose_151); + tensor var_2972 = const()[name = tensor("op_2972"), val = tensor([0, 2, 1, 3])]; + tensor concat_26 = const()[name = tensor("concat_26"), val = tensor([1, 1500, 1280])]; + tensor transpose_148 = transpose(perm = var_2972, x = var_2971_cast); + tensor x_323_cast = reshape(shape = concat_26, x = transpose_148); + tensor var_2977_to_fp16 = const()[name = tensor("op_2977_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1047387520)))]; + tensor var_2978_to_fp16 = const()[name = tensor("op_2978_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1050664384)))]; + tensor var_2979_cast = linear(bias = var_2978_to_fp16, weight = var_2977_to_fp16, x = x_323_cast); + tensor x_325_cast = add(x = x_319_cast, y = var_2979_cast); + tensor var_2985_axes_0 = const()[name = tensor("op_2985_axes_0"), val = tensor([-1])]; + tensor blocks_26_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_26_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1050667008)))]; + tensor blocks_26_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_26_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1050669632)))]; + tensor var_2985_cast = layer_norm(axes = var_2985_axes_0, beta = blocks_26_mlp_ln_bias_to_fp16, epsilon = var_2910_to_fp16, gamma = blocks_26_mlp_ln_weight_to_fp16, x = x_325_cast); + tensor var_2994_to_fp16 = const()[name = tensor("op_2994_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1050672256)))]; + tensor var_2995_to_fp16 = const()[name = tensor("op_2995_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1063779520)))]; + tensor input_217_cast = linear(bias = var_2995_to_fp16, weight = var_2994_to_fp16, x = var_2985_cast); + tensor x_329_mode_0 = const()[name = tensor("x_329_mode_0"), val = tensor("EXACT")]; + tensor x_329_cast = gelu(mode = x_329_mode_0, x = input_217_cast); + tensor var_3000_to_fp16 = const()[name = tensor("op_3000_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1063789824)))]; + tensor var_3001_to_fp16 = const()[name = tensor("op_3001_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1076897088)))]; + tensor var_3002_cast = linear(bias = var_3001_to_fp16, weight = var_3000_to_fp16, x = x_329_cast); + tensor x_331_cast = add(x = x_325_cast, y = var_3002_cast); + tensor var_3011 = const()[name = tensor("op_3011"), val = tensor(-1)]; + tensor var_3028_axes_0 = const()[name = tensor("op_3028_axes_0"), val = tensor([-1])]; + tensor blocks_27_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_27_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1076899712)))]; + tensor blocks_27_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_27_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1076902336)))]; + tensor var_3017_to_fp16 = const()[name = tensor("op_3017_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3028_cast = layer_norm(axes = var_3028_axes_0, beta = blocks_27_attn_ln_bias_to_fp16, epsilon = var_3017_to_fp16, gamma = blocks_27_attn_ln_weight_to_fp16, x = x_331_cast); + tensor var_3039_to_fp16 = const()[name = tensor("op_3039_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1076904960)))]; + tensor var_3040_to_fp16 = const()[name = tensor("op_3040_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1080181824)))]; + tensor q_109_cast = linear(bias = var_3040_to_fp16, weight = var_3039_to_fp16, x = var_3028_cast); + tensor var_3043_to_fp16 = const()[name = tensor("op_3043_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1080184448)))]; + tensor k_109_bias_0_to_fp16 = const()[name = tensor("k_109_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1083461312)))]; + tensor k_109_cast = linear(bias = k_109_bias_0_to_fp16, weight = var_3043_to_fp16, x = var_3028_cast); + tensor var_3047_to_fp16 = const()[name = tensor("op_3047_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1083463936)))]; + tensor var_3048_to_fp16 = const()[name = tensor("op_3048_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1086740800)))]; + tensor v_109_cast = linear(bias = var_3048_to_fp16, weight = var_3047_to_fp16, x = var_3028_cast); + tensor var_3056 = const()[name = tensor("op_3056"), val = tensor([1, 1500, 20, -1])]; + tensor var_3057_cast = reshape(shape = var_3056, x = q_109_cast); + tensor const_278_to_fp16 = const()[name = tensor("const_278_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_111_cast = mul(x = var_3057_cast, y = const_278_to_fp16); + tensor var_3063 = const()[name = tensor("op_3063"), val = tensor([1, 1500, 20, -1])]; + tensor var_3064_cast = reshape(shape = var_3063, x = k_109_cast); + tensor const_279_to_fp16 = const()[name = tensor("const_279_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_111_cast = mul(x = var_3064_cast, y = const_279_to_fp16); + tensor var_3070 = const()[name = tensor("op_3070"), val = tensor([1, 1500, 20, -1])]; + tensor var_3071_cast = reshape(shape = var_3070, x = v_109_cast); + tensor var_3072 = const()[name = tensor("op_3072"), val = tensor([0, 2, 1, 3])]; + tensor qk_55_transpose_x_0 = const()[name = tensor("qk_55_transpose_x_0"), val = tensor(false)]; + tensor qk_55_transpose_y_0 = const()[name = tensor("qk_55_transpose_y_0"), val = tensor(false)]; + tensor transpose_118_perm_0 = const()[name = tensor("transpose_118_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_119_perm_0 = const()[name = tensor("transpose_119_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_145 = transpose(perm = transpose_119_perm_0, x = k_111_cast); + tensor transpose_146 = transpose(perm = transpose_118_perm_0, x = q_111_cast); + tensor qk_55_cast = matmul(transpose_x = qk_55_transpose_x_0, transpose_y = qk_55_transpose_y_0, x = transpose_146, y = transpose_145); + tensor var_3076_cast = softmax(axis = var_3011, x = qk_55_cast); + tensor var_3078_transpose_x_0 = const()[name = tensor("op_3078_transpose_x_0"), val = tensor(false)]; + tensor var_3078_transpose_y_0 = const()[name = tensor("op_3078_transpose_y_0"), val = tensor(false)]; + tensor transpose_147 = transpose(perm = var_3072, x = var_3071_cast); + tensor var_3078_cast = matmul(transpose_x = var_3078_transpose_x_0, transpose_y = var_3078_transpose_y_0, x = var_3076_cast, y = transpose_147); + tensor var_3079 = const()[name = tensor("op_3079"), val = tensor([0, 2, 1, 3])]; + tensor concat_27 = const()[name = tensor("concat_27"), val = tensor([1, 1500, 1280])]; + tensor transpose_144 = transpose(perm = var_3079, x = var_3078_cast); + tensor x_335_cast = reshape(shape = concat_27, x = transpose_144); + tensor var_3084_to_fp16 = const()[name = tensor("op_3084_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1086743424)))]; + tensor var_3085_to_fp16 = const()[name = tensor("op_3085_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1090020288)))]; + tensor var_3086_cast = linear(bias = var_3085_to_fp16, weight = var_3084_to_fp16, x = x_335_cast); + tensor x_337_cast = add(x = x_331_cast, y = var_3086_cast); + tensor var_3092_axes_0 = const()[name = tensor("op_3092_axes_0"), val = tensor([-1])]; + tensor blocks_27_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_27_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1090022912)))]; + tensor blocks_27_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_27_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1090025536)))]; + tensor var_3092_cast = layer_norm(axes = var_3092_axes_0, beta = blocks_27_mlp_ln_bias_to_fp16, epsilon = var_3017_to_fp16, gamma = blocks_27_mlp_ln_weight_to_fp16, x = x_337_cast); + tensor var_3101_to_fp16 = const()[name = tensor("op_3101_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1090028160)))]; + tensor var_3102_to_fp16 = const()[name = tensor("op_3102_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1103135424)))]; + tensor input_225_cast = linear(bias = var_3102_to_fp16, weight = var_3101_to_fp16, x = var_3092_cast); + tensor x_341_mode_0 = const()[name = tensor("x_341_mode_0"), val = tensor("EXACT")]; + tensor x_341_cast = gelu(mode = x_341_mode_0, x = input_225_cast); + tensor var_3107_to_fp16 = const()[name = tensor("op_3107_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1103145728)))]; + tensor var_3108_to_fp16 = const()[name = tensor("op_3108_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116252992)))]; + tensor var_3109_cast = linear(bias = var_3108_to_fp16, weight = var_3107_to_fp16, x = x_341_cast); + tensor x_343_cast = add(x = x_337_cast, y = var_3109_cast); + tensor var_3118 = const()[name = tensor("op_3118"), val = tensor(-1)]; + tensor var_3135_axes_0 = const()[name = tensor("op_3135_axes_0"), val = tensor([-1])]; + tensor blocks_28_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_28_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116255616)))]; + tensor blocks_28_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_28_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116258240)))]; + tensor var_3124_to_fp16 = const()[name = tensor("op_3124_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3135_cast = layer_norm(axes = var_3135_axes_0, beta = blocks_28_attn_ln_bias_to_fp16, epsilon = var_3124_to_fp16, gamma = blocks_28_attn_ln_weight_to_fp16, x = x_343_cast); + tensor var_3146_to_fp16 = const()[name = tensor("op_3146_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116260864)))]; + tensor var_3147_to_fp16 = const()[name = tensor("op_3147_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1119537728)))]; + tensor q_113_cast = linear(bias = var_3147_to_fp16, weight = var_3146_to_fp16, x = var_3135_cast); + tensor var_3150_to_fp16 = const()[name = tensor("op_3150_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1119540352)))]; + tensor k_113_bias_0_to_fp16 = const()[name = tensor("k_113_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1122817216)))]; + tensor k_113_cast = linear(bias = k_113_bias_0_to_fp16, weight = var_3150_to_fp16, x = var_3135_cast); + tensor var_3154_to_fp16 = const()[name = tensor("op_3154_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1122819840)))]; + tensor var_3155_to_fp16 = const()[name = tensor("op_3155_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1126096704)))]; + tensor v_113_cast = linear(bias = var_3155_to_fp16, weight = var_3154_to_fp16, x = var_3135_cast); + tensor var_3163 = const()[name = tensor("op_3163"), val = tensor([1, 1500, 20, -1])]; + tensor var_3164_cast = reshape(shape = var_3163, x = q_113_cast); + tensor const_280_to_fp16 = const()[name = tensor("const_280_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_115_cast = mul(x = var_3164_cast, y = const_280_to_fp16); + tensor var_3170 = const()[name = tensor("op_3170"), val = tensor([1, 1500, 20, -1])]; + tensor var_3171_cast = reshape(shape = var_3170, x = k_113_cast); + tensor const_281_to_fp16 = const()[name = tensor("const_281_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_115_cast = mul(x = var_3171_cast, y = const_281_to_fp16); + tensor var_3177 = const()[name = tensor("op_3177"), val = tensor([1, 1500, 20, -1])]; + tensor var_3178_cast = reshape(shape = var_3177, x = v_113_cast); + tensor var_3179 = const()[name = tensor("op_3179"), val = tensor([0, 2, 1, 3])]; + tensor qk_57_transpose_x_0 = const()[name = tensor("qk_57_transpose_x_0"), val = tensor(false)]; + tensor qk_57_transpose_y_0 = const()[name = tensor("qk_57_transpose_y_0"), val = tensor(false)]; + tensor transpose_120_perm_0 = const()[name = tensor("transpose_120_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_121_perm_0 = const()[name = tensor("transpose_121_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_141 = transpose(perm = transpose_121_perm_0, x = k_115_cast); + tensor transpose_142 = transpose(perm = transpose_120_perm_0, x = q_115_cast); + tensor qk_57_cast = matmul(transpose_x = qk_57_transpose_x_0, transpose_y = qk_57_transpose_y_0, x = transpose_142, y = transpose_141); + tensor var_3183_cast = softmax(axis = var_3118, x = qk_57_cast); + tensor var_3185_transpose_x_0 = const()[name = tensor("op_3185_transpose_x_0"), val = tensor(false)]; + tensor var_3185_transpose_y_0 = const()[name = tensor("op_3185_transpose_y_0"), val = tensor(false)]; + tensor transpose_143 = transpose(perm = var_3179, x = var_3178_cast); + tensor var_3185_cast = matmul(transpose_x = var_3185_transpose_x_0, transpose_y = var_3185_transpose_y_0, x = var_3183_cast, y = transpose_143); + tensor var_3186 = const()[name = tensor("op_3186"), val = tensor([0, 2, 1, 3])]; + tensor concat_28 = const()[name = tensor("concat_28"), val = tensor([1, 1500, 1280])]; + tensor transpose_140 = transpose(perm = var_3186, x = var_3185_cast); + tensor x_347_cast = reshape(shape = concat_28, x = transpose_140); + tensor var_3191_to_fp16 = const()[name = tensor("op_3191_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1126099328)))]; + tensor var_3192_to_fp16 = const()[name = tensor("op_3192_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1129376192)))]; + tensor var_3193_cast = linear(bias = var_3192_to_fp16, weight = var_3191_to_fp16, x = x_347_cast); + tensor x_349_cast = add(x = x_343_cast, y = var_3193_cast); + tensor var_3199_axes_0 = const()[name = tensor("op_3199_axes_0"), val = tensor([-1])]; + tensor blocks_28_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_28_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1129378816)))]; + tensor blocks_28_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_28_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1129381440)))]; + tensor var_3199_cast = layer_norm(axes = var_3199_axes_0, beta = blocks_28_mlp_ln_bias_to_fp16, epsilon = var_3124_to_fp16, gamma = blocks_28_mlp_ln_weight_to_fp16, x = x_349_cast); + tensor var_3208_to_fp16 = const()[name = tensor("op_3208_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1129384064)))]; + tensor var_3209_to_fp16 = const()[name = tensor("op_3209_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1142491328)))]; + tensor input_233_cast = linear(bias = var_3209_to_fp16, weight = var_3208_to_fp16, x = var_3199_cast); + tensor x_353_mode_0 = const()[name = tensor("x_353_mode_0"), val = tensor("EXACT")]; + tensor x_353_cast = gelu(mode = x_353_mode_0, x = input_233_cast); + tensor var_3214_to_fp16 = const()[name = tensor("op_3214_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1142501632)))]; + tensor var_3215_to_fp16 = const()[name = tensor("op_3215_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1155608896)))]; + tensor var_3216_cast = linear(bias = var_3215_to_fp16, weight = var_3214_to_fp16, x = x_353_cast); + tensor x_355_cast = add(x = x_349_cast, y = var_3216_cast); + tensor var_3225 = const()[name = tensor("op_3225"), val = tensor(-1)]; + tensor var_3242_axes_0 = const()[name = tensor("op_3242_axes_0"), val = tensor([-1])]; + tensor blocks_29_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_29_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1155611520)))]; + tensor blocks_29_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_29_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1155614144)))]; + tensor var_3231_to_fp16 = const()[name = tensor("op_3231_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3242_cast = layer_norm(axes = var_3242_axes_0, beta = blocks_29_attn_ln_bias_to_fp16, epsilon = var_3231_to_fp16, gamma = blocks_29_attn_ln_weight_to_fp16, x = x_355_cast); + tensor var_3253_to_fp16 = const()[name = tensor("op_3253_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1155616768)))]; + tensor var_3254_to_fp16 = const()[name = tensor("op_3254_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1158893632)))]; + tensor q_117_cast = linear(bias = var_3254_to_fp16, weight = var_3253_to_fp16, x = var_3242_cast); + tensor var_3257_to_fp16 = const()[name = tensor("op_3257_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1158896256)))]; + tensor k_117_bias_0_to_fp16 = const()[name = tensor("k_117_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1162173120)))]; + tensor k_117_cast = linear(bias = k_117_bias_0_to_fp16, weight = var_3257_to_fp16, x = var_3242_cast); + tensor var_3261_to_fp16 = const()[name = tensor("op_3261_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1162175744)))]; + tensor var_3262_to_fp16 = const()[name = tensor("op_3262_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1165452608)))]; + tensor v_117_cast = linear(bias = var_3262_to_fp16, weight = var_3261_to_fp16, x = var_3242_cast); + tensor var_3270 = const()[name = tensor("op_3270"), val = tensor([1, 1500, 20, -1])]; + tensor var_3271_cast = reshape(shape = var_3270, x = q_117_cast); + tensor const_282_to_fp16 = const()[name = tensor("const_282_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_119_cast = mul(x = var_3271_cast, y = const_282_to_fp16); + tensor var_3277 = const()[name = tensor("op_3277"), val = tensor([1, 1500, 20, -1])]; + tensor var_3278_cast = reshape(shape = var_3277, x = k_117_cast); + tensor const_283_to_fp16 = const()[name = tensor("const_283_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_119_cast = mul(x = var_3278_cast, y = const_283_to_fp16); + tensor var_3284 = const()[name = tensor("op_3284"), val = tensor([1, 1500, 20, -1])]; + tensor var_3285_cast = reshape(shape = var_3284, x = v_117_cast); + tensor var_3286 = const()[name = tensor("op_3286"), val = tensor([0, 2, 1, 3])]; + tensor qk_59_transpose_x_0 = const()[name = tensor("qk_59_transpose_x_0"), val = tensor(false)]; + tensor qk_59_transpose_y_0 = const()[name = tensor("qk_59_transpose_y_0"), val = tensor(false)]; + tensor transpose_122_perm_0 = const()[name = tensor("transpose_122_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_123_perm_0 = const()[name = tensor("transpose_123_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_137 = transpose(perm = transpose_123_perm_0, x = k_119_cast); + tensor transpose_138 = transpose(perm = transpose_122_perm_0, x = q_119_cast); + tensor qk_59_cast = matmul(transpose_x = qk_59_transpose_x_0, transpose_y = qk_59_transpose_y_0, x = transpose_138, y = transpose_137); + tensor var_3290_cast = softmax(axis = var_3225, x = qk_59_cast); + tensor var_3292_transpose_x_0 = const()[name = tensor("op_3292_transpose_x_0"), val = tensor(false)]; + tensor var_3292_transpose_y_0 = const()[name = tensor("op_3292_transpose_y_0"), val = tensor(false)]; + tensor transpose_139 = transpose(perm = var_3286, x = var_3285_cast); + tensor var_3292_cast = matmul(transpose_x = var_3292_transpose_x_0, transpose_y = var_3292_transpose_y_0, x = var_3290_cast, y = transpose_139); + tensor var_3293 = const()[name = tensor("op_3293"), val = tensor([0, 2, 1, 3])]; + tensor concat_29 = const()[name = tensor("concat_29"), val = tensor([1, 1500, 1280])]; + tensor transpose_136 = transpose(perm = var_3293, x = var_3292_cast); + tensor x_359_cast = reshape(shape = concat_29, x = transpose_136); + tensor var_3298_to_fp16 = const()[name = tensor("op_3298_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1165455232)))]; + tensor var_3299_to_fp16 = const()[name = tensor("op_3299_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1168732096)))]; + tensor var_3300_cast = linear(bias = var_3299_to_fp16, weight = var_3298_to_fp16, x = x_359_cast); + tensor x_361_cast = add(x = x_355_cast, y = var_3300_cast); + tensor var_3306_axes_0 = const()[name = tensor("op_3306_axes_0"), val = tensor([-1])]; + tensor blocks_29_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_29_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1168734720)))]; + tensor blocks_29_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_29_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1168737344)))]; + tensor var_3306_cast = layer_norm(axes = var_3306_axes_0, beta = blocks_29_mlp_ln_bias_to_fp16, epsilon = var_3231_to_fp16, gamma = blocks_29_mlp_ln_weight_to_fp16, x = x_361_cast); + tensor var_3315_to_fp16 = const()[name = tensor("op_3315_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1168739968)))]; + tensor var_3316_to_fp16 = const()[name = tensor("op_3316_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1181847232)))]; + tensor input_241_cast = linear(bias = var_3316_to_fp16, weight = var_3315_to_fp16, x = var_3306_cast); + tensor x_365_mode_0 = const()[name = tensor("x_365_mode_0"), val = tensor("EXACT")]; + tensor x_365_cast = gelu(mode = x_365_mode_0, x = input_241_cast); + tensor var_3321_to_fp16 = const()[name = tensor("op_3321_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1181857536)))]; + tensor var_3322_to_fp16 = const()[name = tensor("op_3322_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1194964800)))]; + tensor var_3323_cast = linear(bias = var_3322_to_fp16, weight = var_3321_to_fp16, x = x_365_cast); + tensor x_367_cast = add(x = x_361_cast, y = var_3323_cast); + tensor var_3332 = const()[name = tensor("op_3332"), val = tensor(-1)]; + tensor var_3349_axes_0 = const()[name = tensor("op_3349_axes_0"), val = tensor([-1])]; + tensor blocks_30_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_30_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1194967424)))]; + tensor blocks_30_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_30_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1194970048)))]; + tensor var_3338_to_fp16 = const()[name = tensor("op_3338_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3349_cast = layer_norm(axes = var_3349_axes_0, beta = blocks_30_attn_ln_bias_to_fp16, epsilon = var_3338_to_fp16, gamma = blocks_30_attn_ln_weight_to_fp16, x = x_367_cast); + tensor var_3360_to_fp16 = const()[name = tensor("op_3360_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1194972672)))]; + tensor var_3361_to_fp16 = const()[name = tensor("op_3361_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1198249536)))]; + tensor q_121_cast = linear(bias = var_3361_to_fp16, weight = var_3360_to_fp16, x = var_3349_cast); + tensor var_3364_to_fp16 = const()[name = tensor("op_3364_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1198252160)))]; + tensor k_121_bias_0_to_fp16 = const()[name = tensor("k_121_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1201529024)))]; + tensor k_121_cast = linear(bias = k_121_bias_0_to_fp16, weight = var_3364_to_fp16, x = var_3349_cast); + tensor var_3368_to_fp16 = const()[name = tensor("op_3368_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1201531648)))]; + tensor var_3369_to_fp16 = const()[name = tensor("op_3369_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1204808512)))]; + tensor v_121_cast = linear(bias = var_3369_to_fp16, weight = var_3368_to_fp16, x = var_3349_cast); + tensor var_3377 = const()[name = tensor("op_3377"), val = tensor([1, 1500, 20, -1])]; + tensor var_3378_cast = reshape(shape = var_3377, x = q_121_cast); + tensor const_284_to_fp16 = const()[name = tensor("const_284_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_123_cast = mul(x = var_3378_cast, y = const_284_to_fp16); + tensor var_3384 = const()[name = tensor("op_3384"), val = tensor([1, 1500, 20, -1])]; + tensor var_3385_cast = reshape(shape = var_3384, x = k_121_cast); + tensor const_285_to_fp16 = const()[name = tensor("const_285_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_123_cast = mul(x = var_3385_cast, y = const_285_to_fp16); + tensor var_3391 = const()[name = tensor("op_3391"), val = tensor([1, 1500, 20, -1])]; + tensor var_3392_cast = reshape(shape = var_3391, x = v_121_cast); + tensor var_3393 = const()[name = tensor("op_3393"), val = tensor([0, 2, 1, 3])]; + tensor qk_61_transpose_x_0 = const()[name = tensor("qk_61_transpose_x_0"), val = tensor(false)]; + tensor qk_61_transpose_y_0 = const()[name = tensor("qk_61_transpose_y_0"), val = tensor(false)]; + tensor transpose_124_perm_0 = const()[name = tensor("transpose_124_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_125_perm_0 = const()[name = tensor("transpose_125_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_133 = transpose(perm = transpose_125_perm_0, x = k_123_cast); + tensor transpose_134 = transpose(perm = transpose_124_perm_0, x = q_123_cast); + tensor qk_61_cast = matmul(transpose_x = qk_61_transpose_x_0, transpose_y = qk_61_transpose_y_0, x = transpose_134, y = transpose_133); + tensor var_3397_cast = softmax(axis = var_3332, x = qk_61_cast); + tensor var_3399_transpose_x_0 = const()[name = tensor("op_3399_transpose_x_0"), val = tensor(false)]; + tensor var_3399_transpose_y_0 = const()[name = tensor("op_3399_transpose_y_0"), val = tensor(false)]; + tensor transpose_135 = transpose(perm = var_3393, x = var_3392_cast); + tensor var_3399_cast = matmul(transpose_x = var_3399_transpose_x_0, transpose_y = var_3399_transpose_y_0, x = var_3397_cast, y = transpose_135); + tensor var_3400 = const()[name = tensor("op_3400"), val = tensor([0, 2, 1, 3])]; + tensor concat_30 = const()[name = tensor("concat_30"), val = tensor([1, 1500, 1280])]; + tensor transpose_132 = transpose(perm = var_3400, x = var_3399_cast); + tensor x_371_cast = reshape(shape = concat_30, x = transpose_132); + tensor var_3405_to_fp16 = const()[name = tensor("op_3405_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1204811136)))]; + tensor var_3406_to_fp16 = const()[name = tensor("op_3406_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208088000)))]; + tensor var_3407_cast = linear(bias = var_3406_to_fp16, weight = var_3405_to_fp16, x = x_371_cast); + tensor x_373_cast = add(x = x_367_cast, y = var_3407_cast); + tensor var_3413_axes_0 = const()[name = tensor("op_3413_axes_0"), val = tensor([-1])]; + tensor blocks_30_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_30_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208090624)))]; + tensor blocks_30_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_30_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208093248)))]; + tensor var_3413_cast = layer_norm(axes = var_3413_axes_0, beta = blocks_30_mlp_ln_bias_to_fp16, epsilon = var_3338_to_fp16, gamma = blocks_30_mlp_ln_weight_to_fp16, x = x_373_cast); + tensor var_3422_to_fp16 = const()[name = tensor("op_3422_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208095872)))]; + tensor var_3423_to_fp16 = const()[name = tensor("op_3423_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1221203136)))]; + tensor input_249_cast = linear(bias = var_3423_to_fp16, weight = var_3422_to_fp16, x = var_3413_cast); + tensor x_377_mode_0 = const()[name = tensor("x_377_mode_0"), val = tensor("EXACT")]; + tensor x_377_cast = gelu(mode = x_377_mode_0, x = input_249_cast); + tensor var_3428_to_fp16 = const()[name = tensor("op_3428_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1221213440)))]; + tensor var_3429_to_fp16 = const()[name = tensor("op_3429_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1234320704)))]; + tensor var_3430_cast = linear(bias = var_3429_to_fp16, weight = var_3428_to_fp16, x = x_377_cast); + tensor x_379_cast = add(x = x_373_cast, y = var_3430_cast); + tensor var_3439 = const()[name = tensor("op_3439"), val = tensor(-1)]; + tensor var_3456_axes_0 = const()[name = tensor("op_3456_axes_0"), val = tensor([-1])]; + tensor blocks_31_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_31_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1234323328)))]; + tensor blocks_31_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_31_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1234325952)))]; + tensor var_3445_to_fp16 = const()[name = tensor("op_3445_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3456_cast = layer_norm(axes = var_3456_axes_0, beta = blocks_31_attn_ln_bias_to_fp16, epsilon = var_3445_to_fp16, gamma = blocks_31_attn_ln_weight_to_fp16, x = x_379_cast); + tensor var_3467_to_fp16 = const()[name = tensor("op_3467_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1234328576)))]; + tensor var_3468_to_fp16 = const()[name = tensor("op_3468_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1237605440)))]; + tensor q_125_cast = linear(bias = var_3468_to_fp16, weight = var_3467_to_fp16, x = var_3456_cast); + tensor var_3471_to_fp16 = const()[name = tensor("op_3471_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1237608064)))]; + tensor k_125_bias_0_to_fp16 = const()[name = tensor("k_125_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1240884928)))]; + tensor k_125_cast = linear(bias = k_125_bias_0_to_fp16, weight = var_3471_to_fp16, x = var_3456_cast); + tensor var_3475_to_fp16 = const()[name = tensor("op_3475_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1240887552)))]; + tensor var_3476_to_fp16 = const()[name = tensor("op_3476_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1244164416)))]; + tensor v_125_cast = linear(bias = var_3476_to_fp16, weight = var_3475_to_fp16, x = var_3456_cast); + tensor var_3484 = const()[name = tensor("op_3484"), val = tensor([1, 1500, 20, -1])]; + tensor var_3485_cast = reshape(shape = var_3484, x = q_125_cast); + tensor const_286_to_fp16 = const()[name = tensor("const_286_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_cast = mul(x = var_3485_cast, y = const_286_to_fp16); + tensor var_3491 = const()[name = tensor("op_3491"), val = tensor([1, 1500, 20, -1])]; + tensor var_3492_cast = reshape(shape = var_3491, x = k_125_cast); + tensor const_287_to_fp16 = const()[name = tensor("const_287_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_cast = mul(x = var_3492_cast, y = const_287_to_fp16); + tensor var_3498 = const()[name = tensor("op_3498"), val = tensor([1, 1500, 20, -1])]; + tensor var_3499_cast = reshape(shape = var_3498, x = v_125_cast); + tensor var_3500 = const()[name = tensor("op_3500"), val = tensor([0, 2, 1, 3])]; + tensor qk_transpose_x_0 = const()[name = tensor("qk_transpose_x_0"), val = tensor(false)]; + tensor qk_transpose_y_0 = const()[name = tensor("qk_transpose_y_0"), val = tensor(false)]; + tensor transpose_126_perm_0 = const()[name = tensor("transpose_126_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_127_perm_0 = const()[name = tensor("transpose_127_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_129 = transpose(perm = transpose_127_perm_0, x = k_cast); + tensor transpose_130 = transpose(perm = transpose_126_perm_0, x = q_cast); + tensor qk_cast = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_130, y = transpose_129); + tensor var_3504_cast = softmax(axis = var_3439, x = qk_cast); + tensor var_3506_transpose_x_0 = const()[name = tensor("op_3506_transpose_x_0"), val = tensor(false)]; + tensor var_3506_transpose_y_0 = const()[name = tensor("op_3506_transpose_y_0"), val = tensor(false)]; + tensor transpose_131 = transpose(perm = var_3500, x = var_3499_cast); + tensor var_3506_cast = matmul(transpose_x = var_3506_transpose_x_0, transpose_y = var_3506_transpose_y_0, x = var_3504_cast, y = transpose_131); + tensor var_3507 = const()[name = tensor("op_3507"), val = tensor([0, 2, 1, 3])]; + tensor concat_31 = const()[name = tensor("concat_31"), val = tensor([1, 1500, 1280])]; + tensor transpose_128 = transpose(perm = var_3507, x = var_3506_cast); + tensor x_383_cast = reshape(shape = concat_31, x = transpose_128); + tensor var_3512_to_fp16 = const()[name = tensor("op_3512_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1244167040)))]; + tensor var_3513_to_fp16 = const()[name = tensor("op_3513_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1247443904)))]; + tensor var_3514_cast = linear(bias = var_3513_to_fp16, weight = var_3512_to_fp16, x = x_383_cast); + tensor x_385_cast = add(x = x_379_cast, y = var_3514_cast); + tensor var_3520_axes_0 = const()[name = tensor("op_3520_axes_0"), val = tensor([-1])]; + tensor blocks_31_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_31_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1247446528)))]; + tensor blocks_31_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_31_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1247449152)))]; + tensor var_3520_cast = layer_norm(axes = var_3520_axes_0, beta = blocks_31_mlp_ln_bias_to_fp16, epsilon = var_3445_to_fp16, gamma = blocks_31_mlp_ln_weight_to_fp16, x = x_385_cast); + tensor var_3529_to_fp16 = const()[name = tensor("op_3529_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1247451776)))]; + tensor var_3530_to_fp16 = const()[name = tensor("op_3530_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1260559040)))]; + tensor input_257_cast = linear(bias = var_3530_to_fp16, weight = var_3529_to_fp16, x = var_3520_cast); + tensor x_389_mode_0 = const()[name = tensor("x_389_mode_0"), val = tensor("EXACT")]; + tensor x_389_cast = gelu(mode = x_389_mode_0, x = input_257_cast); + tensor var_3535_to_fp16 = const()[name = tensor("op_3535_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1260569344)))]; + tensor var_3536_to_fp16 = const()[name = tensor("op_3536_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1273676608)))]; + tensor var_3537_cast = linear(bias = var_3536_to_fp16, weight = var_3535_to_fp16, x = x_389_cast); + tensor x_cast = add(x = x_385_cast, y = var_3537_cast); + tensor var_3550_axes_0 = const()[name = tensor("op_3550_axes_0"), val = tensor([-1])]; + tensor ln_post_weight_to_fp16 = const()[name = tensor("ln_post_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1273679232)))]; + tensor ln_post_bias_to_fp16 = const()[name = tensor("ln_post_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1273681856)))]; + tensor var_3541_to_fp16 = const()[name = tensor("op_3541_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3550_cast = layer_norm(axes = var_3550_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_3541_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast); + tensor var_3550_cast_to_fp32_dtype_0 = const()[name = tensor("op_3550_cast_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor output = cast(dtype = var_3550_cast_to_fp32_dtype_0, x = var_3550_cast); + } -> (output); +} \ No newline at end of file diff --git a/ggml-large-v2-encoder.mlmodelc/weights/weight.bin b/ggml-large-v2-encoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..149d83c1f05a98b4ce8c7f39f095d5e0bcf878b8 --- /dev/null +++ b/ggml-large-v2-encoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:42f92759b6f55ec227a3cb7e042c0424219cc781818ed3b8b0e72743714ed5d7 +size 1273684480 diff --git a/ggml-large-v3-encoder.mlmodelc.zip b/ggml-large-v3-encoder.mlmodelc.zip new file mode 100644 index 0000000000000000000000000000000000000000..28bd23baf68396606f11bbf7032274c94027f504 --- /dev/null +++ b/ggml-large-v3-encoder.mlmodelc.zip @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:9be410daad332c1756c34aa119479ef0b0de80c770b094da4ab5cc29ffa417ee +size 1174643677 diff --git a/ggml-large-v3-encoder.mlmodelc/analytics/coremldata.bin b/ggml-large-v3-encoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..75f9ab6ce4798c0794f6210393494b522c7901e4 --- /dev/null +++ b/ggml-large-v3-encoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a9b73481ce4a6af49154956c7f9c72dc02d96448fa69651f5c5a2124edfc8e31 +size 243 diff --git a/ggml-large-v3-encoder.mlmodelc/coremldata.bin b/ggml-large-v3-encoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..c732ed7b8caeac3c15acda1c2836f3e1f525a5a8 --- /dev/null +++ b/ggml-large-v3-encoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:05fe28591b40616fa0c34ad7b853133623f5300923ec812acb11459c411acf3b +size 149 diff --git a/ggml-large-v3-encoder.mlmodelc/metadata.json b/ggml-large-v3-encoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..e448e50ccb4f8475b396f458c2596a5bae361aeb --- /dev/null +++ b/ggml-large-v3-encoder.mlmodelc/metadata.json @@ -0,0 +1,65 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float32", + "formattedType" : "MultiArray (Float32)", + "shortDescription" : "", + "shape" : "[]", + "name" : "output", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 6, + "mlProgramOperationTypeHistogram" : { + "Linear" : 192, + "Matmul" : 64, + "Cast" : 2, + "Conv" : 2, + "Softmax" : 32, + "Add" : 65, + "LayerNorm" : 65, + "Mul" : 64, + "Transpose" : 129, + "Gelu" : 34, + "Reshape" : 128 + }, + "computePrecision" : "Mixed (Float16, Float32, Int32)", + "isUpdatable" : "0", + "availability" : { + "macOS" : "12.0", + "tvOS" : "15.0", + "visionOS" : "1.0", + "watchOS" : "8.0", + "iOS" : "15.0", + "macCatalyst" : "15.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float32", + "formattedType" : "MultiArray (Float32 1 × 80 × 3000)", + "shortDescription" : "", + "shape" : "[1, 80, 3000]", + "name" : "logmel_data", + "type" : "MultiArray" + } + ], + "generatedClassName" : "coreml_encoder_large", + "method" : "predict" + } +] \ No newline at end of file diff --git a/ggml-large-v3-encoder.mlmodelc/model.mil b/ggml-large-v3-encoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..488706ca043b4413a7addd38f7dc67c20e77e409 --- /dev/null +++ b/ggml-large-v3-encoder.mlmodelc/model.mil @@ -0,0 +1,1927 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "5.33.5"}, {"coremlc-version", "1877.40.3"}})] +{ + func main(tensor logmel_data) { + tensor var_72 = const()[name = tensor("op_72"), val = tensor(1)]; + tensor var_80 = const()[name = tensor("op_80"), val = tensor([1])]; + tensor var_82 = const()[name = tensor("op_82"), val = tensor([1])]; + tensor var_84_pad_type_0 = const()[name = tensor("op_84_pad_type_0"), val = tensor("custom")]; + tensor var_84_pad_0 = const()[name = tensor("op_84_pad_0"), val = tensor([1, 1])]; + tensor logmel_data_to_fp16_dtype_0 = const()[name = tensor("logmel_data_to_fp16_dtype_0"), val = tensor("fp16")]; + tensor weight_3_to_fp16 = const()[name = tensor("weight_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor bias_3_to_fp16 = const()[name = tensor("bias_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614528)))]; + tensor cast_967 = cast(dtype = logmel_data_to_fp16_dtype_0, x = logmel_data); + tensor var_84_cast = conv(bias = bias_3_to_fp16, dilations = var_82, groups = var_72, pad = var_84_pad_0, pad_type = var_84_pad_type_0, strides = var_80, weight = weight_3_to_fp16, x = cast_967); + tensor input_1_mode_0 = const()[name = tensor("input_1_mode_0"), val = tensor("EXACT")]; + tensor input_1_cast = gelu(mode = input_1_mode_0, x = var_84_cast); + tensor var_88 = const()[name = tensor("op_88"), val = tensor(1)]; + tensor var_97 = const()[name = tensor("op_97"), val = tensor([2])]; + tensor var_99 = const()[name = tensor("op_99"), val = tensor([1])]; + tensor var_101_pad_type_0 = const()[name = tensor("op_101_pad_type_0"), val = tensor("custom")]; + tensor var_101_pad_0 = const()[name = tensor("op_101_pad_0"), val = tensor([1, 1])]; + tensor weight_7_to_fp16 = const()[name = tensor("weight_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617152)))]; + tensor bias_7_to_fp16 = const()[name = tensor("bias_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10447616)))]; + tensor var_101_cast = conv(bias = bias_7_to_fp16, dilations = var_99, groups = var_88, pad = var_101_pad_0, pad_type = var_101_pad_type_0, strides = var_97, weight = weight_7_to_fp16, x = input_1_cast); + tensor x_3_mode_0 = const()[name = tensor("x_3_mode_0"), val = tensor("EXACT")]; + tensor x_3_cast = gelu(mode = x_3_mode_0, x = var_101_cast); + tensor var_106 = const()[name = tensor("op_106"), val = tensor([0, 2, 1])]; + tensor positional_embedding_to_fp16 = const()[name = tensor("positional_embedding_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10450240)))]; + tensor transpose_256 = transpose(perm = var_106, x = x_3_cast); + tensor var_109_cast = add(x = transpose_256, y = positional_embedding_to_fp16); + tensor var_122 = const()[name = tensor("op_122"), val = tensor(-1)]; + tensor var_139_axes_0 = const()[name = tensor("op_139_axes_0"), val = tensor([-1])]; + tensor blocks_0_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_0_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14290304)))]; + tensor blocks_0_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_0_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14292928)))]; + tensor var_128_to_fp16 = const()[name = tensor("op_128_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_139_cast = layer_norm(axes = var_139_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_128_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_109_cast); + tensor var_150_to_fp16 = const()[name = tensor("op_150_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14295552)))]; + tensor var_151_to_fp16 = const()[name = tensor("op_151_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17572416)))]; + tensor q_1_cast = linear(bias = var_151_to_fp16, weight = var_150_to_fp16, x = var_139_cast); + tensor var_154_to_fp16 = const()[name = tensor("op_154_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17575040)))]; + tensor k_1_bias_0_to_fp16 = const()[name = tensor("k_1_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20851904)))]; + tensor k_1_cast = linear(bias = k_1_bias_0_to_fp16, weight = var_154_to_fp16, x = var_139_cast); + tensor var_158_to_fp16 = const()[name = tensor("op_158_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20854528)))]; + tensor var_159_to_fp16 = const()[name = tensor("op_159_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24131392)))]; + tensor v_1_cast = linear(bias = var_159_to_fp16, weight = var_158_to_fp16, x = var_139_cast); + tensor var_167 = const()[name = tensor("op_167"), val = tensor([1, 1500, 20, -1])]; + tensor var_168_cast = reshape(shape = var_167, x = q_1_cast); + tensor const_224_to_fp16 = const()[name = tensor("const_224_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_3_cast = mul(x = var_168_cast, y = const_224_to_fp16); + tensor var_174 = const()[name = tensor("op_174"), val = tensor([1, 1500, 20, -1])]; + tensor var_175_cast = reshape(shape = var_174, x = k_1_cast); + tensor const_225_to_fp16 = const()[name = tensor("const_225_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_3_cast = mul(x = var_175_cast, y = const_225_to_fp16); + tensor var_181 = const()[name = tensor("op_181"), val = tensor([1, 1500, 20, -1])]; + tensor var_182_cast = reshape(shape = var_181, x = v_1_cast); + tensor var_183 = const()[name = tensor("op_183"), val = tensor([0, 2, 1, 3])]; + tensor qk_1_transpose_x_0 = const()[name = tensor("qk_1_transpose_x_0"), val = tensor(false)]; + tensor qk_1_transpose_y_0 = const()[name = tensor("qk_1_transpose_y_0"), val = tensor(false)]; + tensor transpose_64_perm_0 = const()[name = tensor("transpose_64_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_65_perm_0 = const()[name = tensor("transpose_65_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_253 = transpose(perm = transpose_65_perm_0, x = k_3_cast); + tensor transpose_254 = transpose(perm = transpose_64_perm_0, x = q_3_cast); + tensor qk_1_cast = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_254, y = transpose_253); + tensor var_187_cast = softmax(axis = var_122, x = qk_1_cast); + tensor var_189_transpose_x_0 = const()[name = tensor("op_189_transpose_x_0"), val = tensor(false)]; + tensor var_189_transpose_y_0 = const()[name = tensor("op_189_transpose_y_0"), val = tensor(false)]; + tensor transpose_255 = transpose(perm = var_183, x = var_182_cast); + tensor var_189_cast = matmul(transpose_x = var_189_transpose_x_0, transpose_y = var_189_transpose_y_0, x = var_187_cast, y = transpose_255); + tensor var_190 = const()[name = tensor("op_190"), val = tensor([0, 2, 1, 3])]; + tensor concat_0 = const()[name = tensor("concat_0"), val = tensor([1, 1500, 1280])]; + tensor transpose_252 = transpose(perm = var_190, x = var_189_cast); + tensor x_11_cast = reshape(shape = concat_0, x = transpose_252); + tensor var_195_to_fp16 = const()[name = tensor("op_195_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24134016)))]; + tensor var_196_to_fp16 = const()[name = tensor("op_196_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27410880)))]; + tensor var_197_cast = linear(bias = var_196_to_fp16, weight = var_195_to_fp16, x = x_11_cast); + tensor x_13_cast = add(x = var_109_cast, y = var_197_cast); + tensor var_203_axes_0 = const()[name = tensor("op_203_axes_0"), val = tensor([-1])]; + tensor blocks_0_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27413504)))]; + tensor blocks_0_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27416128)))]; + tensor var_203_cast = layer_norm(axes = var_203_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_128_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast); + tensor var_212_to_fp16 = const()[name = tensor("op_212_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27418752)))]; + tensor var_213_to_fp16 = const()[name = tensor("op_213_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40526016)))]; + tensor input_9_cast = linear(bias = var_213_to_fp16, weight = var_212_to_fp16, x = var_203_cast); + tensor x_17_mode_0 = const()[name = tensor("x_17_mode_0"), val = tensor("EXACT")]; + tensor x_17_cast = gelu(mode = x_17_mode_0, x = input_9_cast); + tensor var_218_to_fp16 = const()[name = tensor("op_218_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40536320)))]; + tensor var_219_to_fp16 = const()[name = tensor("op_219_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53643584)))]; + tensor var_220_cast = linear(bias = var_219_to_fp16, weight = var_218_to_fp16, x = x_17_cast); + tensor x_19_cast = add(x = x_13_cast, y = var_220_cast); + tensor var_229 = const()[name = tensor("op_229"), val = tensor(-1)]; + tensor var_246_axes_0 = const()[name = tensor("op_246_axes_0"), val = tensor([-1])]; + tensor blocks_1_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_1_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53646208)))]; + tensor blocks_1_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_1_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53648832)))]; + tensor var_235_to_fp16 = const()[name = tensor("op_235_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_246_cast = layer_norm(axes = var_246_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_235_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast); + tensor var_257_to_fp16 = const()[name = tensor("op_257_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53651456)))]; + tensor var_258_to_fp16 = const()[name = tensor("op_258_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56928320)))]; + tensor q_5_cast = linear(bias = var_258_to_fp16, weight = var_257_to_fp16, x = var_246_cast); + tensor var_261_to_fp16 = const()[name = tensor("op_261_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56930944)))]; + tensor k_5_bias_0_to_fp16 = const()[name = tensor("k_5_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60207808)))]; + tensor k_5_cast = linear(bias = k_5_bias_0_to_fp16, weight = var_261_to_fp16, x = var_246_cast); + tensor var_265_to_fp16 = const()[name = tensor("op_265_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60210432)))]; + tensor var_266_to_fp16 = const()[name = tensor("op_266_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63487296)))]; + tensor v_5_cast = linear(bias = var_266_to_fp16, weight = var_265_to_fp16, x = var_246_cast); + tensor var_274 = const()[name = tensor("op_274"), val = tensor([1, 1500, 20, -1])]; + tensor var_275_cast = reshape(shape = var_274, x = q_5_cast); + tensor const_226_to_fp16 = const()[name = tensor("const_226_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_7_cast = mul(x = var_275_cast, y = const_226_to_fp16); + tensor var_281 = const()[name = tensor("op_281"), val = tensor([1, 1500, 20, -1])]; + tensor var_282_cast = reshape(shape = var_281, x = k_5_cast); + tensor const_227_to_fp16 = const()[name = tensor("const_227_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_7_cast = mul(x = var_282_cast, y = const_227_to_fp16); + tensor var_288 = const()[name = tensor("op_288"), val = tensor([1, 1500, 20, -1])]; + tensor var_289_cast = reshape(shape = var_288, x = v_5_cast); + tensor var_290 = const()[name = tensor("op_290"), val = tensor([0, 2, 1, 3])]; + tensor qk_3_transpose_x_0 = const()[name = tensor("qk_3_transpose_x_0"), val = tensor(false)]; + tensor qk_3_transpose_y_0 = const()[name = tensor("qk_3_transpose_y_0"), val = tensor(false)]; + tensor transpose_66_perm_0 = const()[name = tensor("transpose_66_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_67_perm_0 = const()[name = tensor("transpose_67_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_249 = transpose(perm = transpose_67_perm_0, x = k_7_cast); + tensor transpose_250 = transpose(perm = transpose_66_perm_0, x = q_7_cast); + tensor qk_3_cast = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_250, y = transpose_249); + tensor var_294_cast = softmax(axis = var_229, x = qk_3_cast); + tensor var_296_transpose_x_0 = const()[name = tensor("op_296_transpose_x_0"), val = tensor(false)]; + tensor var_296_transpose_y_0 = const()[name = tensor("op_296_transpose_y_0"), val = tensor(false)]; + tensor transpose_251 = transpose(perm = var_290, x = var_289_cast); + tensor var_296_cast = matmul(transpose_x = var_296_transpose_x_0, transpose_y = var_296_transpose_y_0, x = var_294_cast, y = transpose_251); + tensor var_297 = const()[name = tensor("op_297"), val = tensor([0, 2, 1, 3])]; + tensor concat_1 = const()[name = tensor("concat_1"), val = tensor([1, 1500, 1280])]; + tensor transpose_248 = transpose(perm = var_297, x = var_296_cast); + tensor x_23_cast = reshape(shape = concat_1, x = transpose_248); + tensor var_302_to_fp16 = const()[name = tensor("op_302_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63489920)))]; + tensor var_303_to_fp16 = const()[name = tensor("op_303_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66766784)))]; + tensor var_304_cast = linear(bias = var_303_to_fp16, weight = var_302_to_fp16, x = x_23_cast); + tensor x_25_cast = add(x = x_19_cast, y = var_304_cast); + tensor var_310_axes_0 = const()[name = tensor("op_310_axes_0"), val = tensor([-1])]; + tensor blocks_1_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66769408)))]; + tensor blocks_1_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66772032)))]; + tensor var_310_cast = layer_norm(axes = var_310_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_235_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast); + tensor var_319_to_fp16 = const()[name = tensor("op_319_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66774656)))]; + tensor var_320_to_fp16 = const()[name = tensor("op_320_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79881920)))]; + tensor input_17_cast = linear(bias = var_320_to_fp16, weight = var_319_to_fp16, x = var_310_cast); + tensor x_29_mode_0 = const()[name = tensor("x_29_mode_0"), val = tensor("EXACT")]; + tensor x_29_cast = gelu(mode = x_29_mode_0, x = input_17_cast); + tensor var_325_to_fp16 = const()[name = tensor("op_325_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79892224)))]; + tensor var_326_to_fp16 = const()[name = tensor("op_326_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92999488)))]; + tensor var_327_cast = linear(bias = var_326_to_fp16, weight = var_325_to_fp16, x = x_29_cast); + tensor x_31_cast = add(x = x_25_cast, y = var_327_cast); + tensor var_336 = const()[name = tensor("op_336"), val = tensor(-1)]; + tensor var_353_axes_0 = const()[name = tensor("op_353_axes_0"), val = tensor([-1])]; + tensor blocks_2_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_2_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93002112)))]; + tensor blocks_2_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_2_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93004736)))]; + tensor var_342_to_fp16 = const()[name = tensor("op_342_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_353_cast = layer_norm(axes = var_353_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_342_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast); + tensor var_364_to_fp16 = const()[name = tensor("op_364_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93007360)))]; + tensor var_365_to_fp16 = const()[name = tensor("op_365_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96284224)))]; + tensor q_9_cast = linear(bias = var_365_to_fp16, weight = var_364_to_fp16, x = var_353_cast); + tensor var_368_to_fp16 = const()[name = tensor("op_368_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96286848)))]; + tensor k_9_bias_0_to_fp16 = const()[name = tensor("k_9_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99563712)))]; + tensor k_9_cast = linear(bias = k_9_bias_0_to_fp16, weight = var_368_to_fp16, x = var_353_cast); + tensor var_372_to_fp16 = const()[name = tensor("op_372_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99566336)))]; + tensor var_373_to_fp16 = const()[name = tensor("op_373_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102843200)))]; + tensor v_9_cast = linear(bias = var_373_to_fp16, weight = var_372_to_fp16, x = var_353_cast); + tensor var_381 = const()[name = tensor("op_381"), val = tensor([1, 1500, 20, -1])]; + tensor var_382_cast = reshape(shape = var_381, x = q_9_cast); + tensor const_228_to_fp16 = const()[name = tensor("const_228_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_11_cast = mul(x = var_382_cast, y = const_228_to_fp16); + tensor var_388 = const()[name = tensor("op_388"), val = tensor([1, 1500, 20, -1])]; + tensor var_389_cast = reshape(shape = var_388, x = k_9_cast); + tensor const_229_to_fp16 = const()[name = tensor("const_229_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_11_cast = mul(x = var_389_cast, y = const_229_to_fp16); + tensor var_395 = const()[name = tensor("op_395"), val = tensor([1, 1500, 20, -1])]; + tensor var_396_cast = reshape(shape = var_395, x = v_9_cast); + tensor var_397 = const()[name = tensor("op_397"), val = tensor([0, 2, 1, 3])]; + tensor qk_5_transpose_x_0 = const()[name = tensor("qk_5_transpose_x_0"), val = tensor(false)]; + tensor qk_5_transpose_y_0 = const()[name = tensor("qk_5_transpose_y_0"), val = tensor(false)]; + tensor transpose_68_perm_0 = const()[name = tensor("transpose_68_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_69_perm_0 = const()[name = tensor("transpose_69_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_245 = transpose(perm = transpose_69_perm_0, x = k_11_cast); + tensor transpose_246 = transpose(perm = transpose_68_perm_0, x = q_11_cast); + tensor qk_5_cast = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_246, y = transpose_245); + tensor var_401_cast = softmax(axis = var_336, x = qk_5_cast); + tensor var_403_transpose_x_0 = const()[name = tensor("op_403_transpose_x_0"), val = tensor(false)]; + tensor var_403_transpose_y_0 = const()[name = tensor("op_403_transpose_y_0"), val = tensor(false)]; + tensor transpose_247 = transpose(perm = var_397, x = var_396_cast); + tensor var_403_cast = matmul(transpose_x = var_403_transpose_x_0, transpose_y = var_403_transpose_y_0, x = var_401_cast, y = transpose_247); + tensor var_404 = const()[name = tensor("op_404"), val = tensor([0, 2, 1, 3])]; + tensor concat_2 = const()[name = tensor("concat_2"), val = tensor([1, 1500, 1280])]; + tensor transpose_244 = transpose(perm = var_404, x = var_403_cast); + tensor x_35_cast = reshape(shape = concat_2, x = transpose_244); + tensor var_409_to_fp16 = const()[name = tensor("op_409_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102845824)))]; + tensor var_410_to_fp16 = const()[name = tensor("op_410_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106122688)))]; + tensor var_411_cast = linear(bias = var_410_to_fp16, weight = var_409_to_fp16, x = x_35_cast); + tensor x_37_cast = add(x = x_31_cast, y = var_411_cast); + tensor var_417_axes_0 = const()[name = tensor("op_417_axes_0"), val = tensor([-1])]; + tensor blocks_2_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106125312)))]; + tensor blocks_2_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106127936)))]; + tensor var_417_cast = layer_norm(axes = var_417_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_342_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast); + tensor var_426_to_fp16 = const()[name = tensor("op_426_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106130560)))]; + tensor var_427_to_fp16 = const()[name = tensor("op_427_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119237824)))]; + tensor input_25_cast = linear(bias = var_427_to_fp16, weight = var_426_to_fp16, x = var_417_cast); + tensor x_41_mode_0 = const()[name = tensor("x_41_mode_0"), val = tensor("EXACT")]; + tensor x_41_cast = gelu(mode = x_41_mode_0, x = input_25_cast); + tensor var_432_to_fp16 = const()[name = tensor("op_432_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119248128)))]; + tensor var_433_to_fp16 = const()[name = tensor("op_433_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132355392)))]; + tensor var_434_cast = linear(bias = var_433_to_fp16, weight = var_432_to_fp16, x = x_41_cast); + tensor x_43_cast = add(x = x_37_cast, y = var_434_cast); + tensor var_443 = const()[name = tensor("op_443"), val = tensor(-1)]; + tensor var_460_axes_0 = const()[name = tensor("op_460_axes_0"), val = tensor([-1])]; + tensor blocks_3_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_3_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132358016)))]; + tensor blocks_3_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_3_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132360640)))]; + tensor var_449_to_fp16 = const()[name = tensor("op_449_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_460_cast = layer_norm(axes = var_460_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_449_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast); + tensor var_471_to_fp16 = const()[name = tensor("op_471_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(132363264)))]; + tensor var_472_to_fp16 = const()[name = tensor("op_472_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135640128)))]; + tensor q_13_cast = linear(bias = var_472_to_fp16, weight = var_471_to_fp16, x = var_460_cast); + tensor var_475_to_fp16 = const()[name = tensor("op_475_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135642752)))]; + tensor k_13_bias_0_to_fp16 = const()[name = tensor("k_13_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138919616)))]; + tensor k_13_cast = linear(bias = k_13_bias_0_to_fp16, weight = var_475_to_fp16, x = var_460_cast); + tensor var_479_to_fp16 = const()[name = tensor("op_479_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138922240)))]; + tensor var_480_to_fp16 = const()[name = tensor("op_480_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142199104)))]; + tensor v_13_cast = linear(bias = var_480_to_fp16, weight = var_479_to_fp16, x = var_460_cast); + tensor var_488 = const()[name = tensor("op_488"), val = tensor([1, 1500, 20, -1])]; + tensor var_489_cast = reshape(shape = var_488, x = q_13_cast); + tensor const_230_to_fp16 = const()[name = tensor("const_230_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_15_cast = mul(x = var_489_cast, y = const_230_to_fp16); + tensor var_495 = const()[name = tensor("op_495"), val = tensor([1, 1500, 20, -1])]; + tensor var_496_cast = reshape(shape = var_495, x = k_13_cast); + tensor const_231_to_fp16 = const()[name = tensor("const_231_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_15_cast = mul(x = var_496_cast, y = const_231_to_fp16); + tensor var_502 = const()[name = tensor("op_502"), val = tensor([1, 1500, 20, -1])]; + tensor var_503_cast = reshape(shape = var_502, x = v_13_cast); + tensor var_504 = const()[name = tensor("op_504"), val = tensor([0, 2, 1, 3])]; + tensor qk_7_transpose_x_0 = const()[name = tensor("qk_7_transpose_x_0"), val = tensor(false)]; + tensor qk_7_transpose_y_0 = const()[name = tensor("qk_7_transpose_y_0"), val = tensor(false)]; + tensor transpose_70_perm_0 = const()[name = tensor("transpose_70_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_71_perm_0 = const()[name = tensor("transpose_71_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_241 = transpose(perm = transpose_71_perm_0, x = k_15_cast); + tensor transpose_242 = transpose(perm = transpose_70_perm_0, x = q_15_cast); + tensor qk_7_cast = matmul(transpose_x = qk_7_transpose_x_0, transpose_y = qk_7_transpose_y_0, x = transpose_242, y = transpose_241); + tensor var_508_cast = softmax(axis = var_443, x = qk_7_cast); + tensor var_510_transpose_x_0 = const()[name = tensor("op_510_transpose_x_0"), val = tensor(false)]; + tensor var_510_transpose_y_0 = const()[name = tensor("op_510_transpose_y_0"), val = tensor(false)]; + tensor transpose_243 = transpose(perm = var_504, x = var_503_cast); + tensor var_510_cast = matmul(transpose_x = var_510_transpose_x_0, transpose_y = var_510_transpose_y_0, x = var_508_cast, y = transpose_243); + tensor var_511 = const()[name = tensor("op_511"), val = tensor([0, 2, 1, 3])]; + tensor concat_3 = const()[name = tensor("concat_3"), val = tensor([1, 1500, 1280])]; + tensor transpose_240 = transpose(perm = var_511, x = var_510_cast); + tensor x_47_cast = reshape(shape = concat_3, x = transpose_240); + tensor var_516_to_fp16 = const()[name = tensor("op_516_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142201728)))]; + tensor var_517_to_fp16 = const()[name = tensor("op_517_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145478592)))]; + tensor var_518_cast = linear(bias = var_517_to_fp16, weight = var_516_to_fp16, x = x_47_cast); + tensor x_49_cast = add(x = x_43_cast, y = var_518_cast); + tensor var_524_axes_0 = const()[name = tensor("op_524_axes_0"), val = tensor([-1])]; + tensor blocks_3_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145481216)))]; + tensor blocks_3_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145483840)))]; + tensor var_524_cast = layer_norm(axes = var_524_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_449_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast); + tensor var_533_to_fp16 = const()[name = tensor("op_533_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(145486464)))]; + tensor var_534_to_fp16 = const()[name = tensor("op_534_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158593728)))]; + tensor input_33_cast = linear(bias = var_534_to_fp16, weight = var_533_to_fp16, x = var_524_cast); + tensor x_53_mode_0 = const()[name = tensor("x_53_mode_0"), val = tensor("EXACT")]; + tensor x_53_cast = gelu(mode = x_53_mode_0, x = input_33_cast); + tensor var_539_to_fp16 = const()[name = tensor("op_539_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(158604032)))]; + tensor var_540_to_fp16 = const()[name = tensor("op_540_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171711296)))]; + tensor var_541_cast = linear(bias = var_540_to_fp16, weight = var_539_to_fp16, x = x_53_cast); + tensor x_55_cast = add(x = x_49_cast, y = var_541_cast); + tensor var_550 = const()[name = tensor("op_550"), val = tensor(-1)]; + tensor var_567_axes_0 = const()[name = tensor("op_567_axes_0"), val = tensor([-1])]; + tensor blocks_4_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_4_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171713920)))]; + tensor blocks_4_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_4_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171716544)))]; + tensor var_556_to_fp16 = const()[name = tensor("op_556_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_567_cast = layer_norm(axes = var_567_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_556_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast); + tensor var_578_to_fp16 = const()[name = tensor("op_578_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171719168)))]; + tensor var_579_to_fp16 = const()[name = tensor("op_579_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174996032)))]; + tensor q_17_cast = linear(bias = var_579_to_fp16, weight = var_578_to_fp16, x = var_567_cast); + tensor var_582_to_fp16 = const()[name = tensor("op_582_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(174998656)))]; + tensor k_17_bias_0_to_fp16 = const()[name = tensor("k_17_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178275520)))]; + tensor k_17_cast = linear(bias = k_17_bias_0_to_fp16, weight = var_582_to_fp16, x = var_567_cast); + tensor var_586_to_fp16 = const()[name = tensor("op_586_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(178278144)))]; + tensor var_587_to_fp16 = const()[name = tensor("op_587_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(181555008)))]; + tensor v_17_cast = linear(bias = var_587_to_fp16, weight = var_586_to_fp16, x = var_567_cast); + tensor var_595 = const()[name = tensor("op_595"), val = tensor([1, 1500, 20, -1])]; + tensor var_596_cast = reshape(shape = var_595, x = q_17_cast); + tensor const_232_to_fp16 = const()[name = tensor("const_232_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_19_cast = mul(x = var_596_cast, y = const_232_to_fp16); + tensor var_602 = const()[name = tensor("op_602"), val = tensor([1, 1500, 20, -1])]; + tensor var_603_cast = reshape(shape = var_602, x = k_17_cast); + tensor const_233_to_fp16 = const()[name = tensor("const_233_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_19_cast = mul(x = var_603_cast, y = const_233_to_fp16); + tensor var_609 = const()[name = tensor("op_609"), val = tensor([1, 1500, 20, -1])]; + tensor var_610_cast = reshape(shape = var_609, x = v_17_cast); + tensor var_611 = const()[name = tensor("op_611"), val = tensor([0, 2, 1, 3])]; + tensor qk_9_transpose_x_0 = const()[name = tensor("qk_9_transpose_x_0"), val = tensor(false)]; + tensor qk_9_transpose_y_0 = const()[name = tensor("qk_9_transpose_y_0"), val = tensor(false)]; + tensor transpose_72_perm_0 = const()[name = tensor("transpose_72_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_73_perm_0 = const()[name = tensor("transpose_73_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_237 = transpose(perm = transpose_73_perm_0, x = k_19_cast); + tensor transpose_238 = transpose(perm = transpose_72_perm_0, x = q_19_cast); + tensor qk_9_cast = matmul(transpose_x = qk_9_transpose_x_0, transpose_y = qk_9_transpose_y_0, x = transpose_238, y = transpose_237); + tensor var_615_cast = softmax(axis = var_550, x = qk_9_cast); + tensor var_617_transpose_x_0 = const()[name = tensor("op_617_transpose_x_0"), val = tensor(false)]; + tensor var_617_transpose_y_0 = const()[name = tensor("op_617_transpose_y_0"), val = tensor(false)]; + tensor transpose_239 = transpose(perm = var_611, x = var_610_cast); + tensor var_617_cast = matmul(transpose_x = var_617_transpose_x_0, transpose_y = var_617_transpose_y_0, x = var_615_cast, y = transpose_239); + tensor var_618 = const()[name = tensor("op_618"), val = tensor([0, 2, 1, 3])]; + tensor concat_4 = const()[name = tensor("concat_4"), val = tensor([1, 1500, 1280])]; + tensor transpose_236 = transpose(perm = var_618, x = var_617_cast); + tensor x_59_cast = reshape(shape = concat_4, x = transpose_236); + tensor var_623_to_fp16 = const()[name = tensor("op_623_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(181557632)))]; + tensor var_624_to_fp16 = const()[name = tensor("op_624_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184834496)))]; + tensor var_625_cast = linear(bias = var_624_to_fp16, weight = var_623_to_fp16, x = x_59_cast); + tensor x_61_cast = add(x = x_55_cast, y = var_625_cast); + tensor var_631_axes_0 = const()[name = tensor("op_631_axes_0"), val = tensor([-1])]; + tensor blocks_4_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184837120)))]; + tensor blocks_4_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184839744)))]; + tensor var_631_cast = layer_norm(axes = var_631_axes_0, beta = blocks_4_mlp_ln_bias_to_fp16, epsilon = var_556_to_fp16, gamma = blocks_4_mlp_ln_weight_to_fp16, x = x_61_cast); + tensor var_640_to_fp16 = const()[name = tensor("op_640_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184842368)))]; + tensor var_641_to_fp16 = const()[name = tensor("op_641_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(197949632)))]; + tensor input_41_cast = linear(bias = var_641_to_fp16, weight = var_640_to_fp16, x = var_631_cast); + tensor x_65_mode_0 = const()[name = tensor("x_65_mode_0"), val = tensor("EXACT")]; + tensor x_65_cast = gelu(mode = x_65_mode_0, x = input_41_cast); + tensor var_646_to_fp16 = const()[name = tensor("op_646_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(197959936)))]; + tensor var_647_to_fp16 = const()[name = tensor("op_647_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211067200)))]; + tensor var_648_cast = linear(bias = var_647_to_fp16, weight = var_646_to_fp16, x = x_65_cast); + tensor x_67_cast = add(x = x_61_cast, y = var_648_cast); + tensor var_657 = const()[name = tensor("op_657"), val = tensor(-1)]; + tensor var_674_axes_0 = const()[name = tensor("op_674_axes_0"), val = tensor([-1])]; + tensor blocks_5_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_5_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211069824)))]; + tensor blocks_5_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_5_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211072448)))]; + tensor var_663_to_fp16 = const()[name = tensor("op_663_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_674_cast = layer_norm(axes = var_674_axes_0, beta = blocks_5_attn_ln_bias_to_fp16, epsilon = var_663_to_fp16, gamma = blocks_5_attn_ln_weight_to_fp16, x = x_67_cast); + tensor var_685_to_fp16 = const()[name = tensor("op_685_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211075072)))]; + tensor var_686_to_fp16 = const()[name = tensor("op_686_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214351936)))]; + tensor q_21_cast = linear(bias = var_686_to_fp16, weight = var_685_to_fp16, x = var_674_cast); + tensor var_689_to_fp16 = const()[name = tensor("op_689_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(214354560)))]; + tensor k_21_bias_0_to_fp16 = const()[name = tensor("k_21_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217631424)))]; + tensor k_21_cast = linear(bias = k_21_bias_0_to_fp16, weight = var_689_to_fp16, x = var_674_cast); + tensor var_693_to_fp16 = const()[name = tensor("op_693_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217634048)))]; + tensor var_694_to_fp16 = const()[name = tensor("op_694_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(220910912)))]; + tensor v_21_cast = linear(bias = var_694_to_fp16, weight = var_693_to_fp16, x = var_674_cast); + tensor var_702 = const()[name = tensor("op_702"), val = tensor([1, 1500, 20, -1])]; + tensor var_703_cast = reshape(shape = var_702, x = q_21_cast); + tensor const_234_to_fp16 = const()[name = tensor("const_234_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_23_cast = mul(x = var_703_cast, y = const_234_to_fp16); + tensor var_709 = const()[name = tensor("op_709"), val = tensor([1, 1500, 20, -1])]; + tensor var_710_cast = reshape(shape = var_709, x = k_21_cast); + tensor const_235_to_fp16 = const()[name = tensor("const_235_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_23_cast = mul(x = var_710_cast, y = const_235_to_fp16); + tensor var_716 = const()[name = tensor("op_716"), val = tensor([1, 1500, 20, -1])]; + tensor var_717_cast = reshape(shape = var_716, x = v_21_cast); + tensor var_718 = const()[name = tensor("op_718"), val = tensor([0, 2, 1, 3])]; + tensor qk_11_transpose_x_0 = const()[name = tensor("qk_11_transpose_x_0"), val = tensor(false)]; + tensor qk_11_transpose_y_0 = const()[name = tensor("qk_11_transpose_y_0"), val = tensor(false)]; + tensor transpose_74_perm_0 = const()[name = tensor("transpose_74_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_75_perm_0 = const()[name = tensor("transpose_75_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_233 = transpose(perm = transpose_75_perm_0, x = k_23_cast); + tensor transpose_234 = transpose(perm = transpose_74_perm_0, x = q_23_cast); + tensor qk_11_cast = matmul(transpose_x = qk_11_transpose_x_0, transpose_y = qk_11_transpose_y_0, x = transpose_234, y = transpose_233); + tensor var_722_cast = softmax(axis = var_657, x = qk_11_cast); + tensor var_724_transpose_x_0 = const()[name = tensor("op_724_transpose_x_0"), val = tensor(false)]; + tensor var_724_transpose_y_0 = const()[name = tensor("op_724_transpose_y_0"), val = tensor(false)]; + tensor transpose_235 = transpose(perm = var_718, x = var_717_cast); + tensor var_724_cast = matmul(transpose_x = var_724_transpose_x_0, transpose_y = var_724_transpose_y_0, x = var_722_cast, y = transpose_235); + tensor var_725 = const()[name = tensor("op_725"), val = tensor([0, 2, 1, 3])]; + tensor concat_5 = const()[name = tensor("concat_5"), val = tensor([1, 1500, 1280])]; + tensor transpose_232 = transpose(perm = var_725, x = var_724_cast); + tensor x_71_cast = reshape(shape = concat_5, x = transpose_232); + tensor var_730_to_fp16 = const()[name = tensor("op_730_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(220913536)))]; + tensor var_731_to_fp16 = const()[name = tensor("op_731_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224190400)))]; + tensor var_732_cast = linear(bias = var_731_to_fp16, weight = var_730_to_fp16, x = x_71_cast); + tensor x_73_cast = add(x = x_67_cast, y = var_732_cast); + tensor var_738_axes_0 = const()[name = tensor("op_738_axes_0"), val = tensor([-1])]; + tensor blocks_5_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224193024)))]; + tensor blocks_5_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224195648)))]; + tensor var_738_cast = layer_norm(axes = var_738_axes_0, beta = blocks_5_mlp_ln_bias_to_fp16, epsilon = var_663_to_fp16, gamma = blocks_5_mlp_ln_weight_to_fp16, x = x_73_cast); + tensor var_747_to_fp16 = const()[name = tensor("op_747_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(224198272)))]; + tensor var_748_to_fp16 = const()[name = tensor("op_748_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237305536)))]; + tensor input_49_cast = linear(bias = var_748_to_fp16, weight = var_747_to_fp16, x = var_738_cast); + tensor x_77_mode_0 = const()[name = tensor("x_77_mode_0"), val = tensor("EXACT")]; + tensor x_77_cast = gelu(mode = x_77_mode_0, x = input_49_cast); + tensor var_753_to_fp16 = const()[name = tensor("op_753_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(237315840)))]; + tensor var_754_to_fp16 = const()[name = tensor("op_754_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250423104)))]; + tensor var_755_cast = linear(bias = var_754_to_fp16, weight = var_753_to_fp16, x = x_77_cast); + tensor x_79_cast = add(x = x_73_cast, y = var_755_cast); + tensor var_764 = const()[name = tensor("op_764"), val = tensor(-1)]; + tensor var_781_axes_0 = const()[name = tensor("op_781_axes_0"), val = tensor([-1])]; + tensor blocks_6_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_6_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250425728)))]; + tensor blocks_6_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_6_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250428352)))]; + tensor var_770_to_fp16 = const()[name = tensor("op_770_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_781_cast = layer_norm(axes = var_781_axes_0, beta = blocks_6_attn_ln_bias_to_fp16, epsilon = var_770_to_fp16, gamma = blocks_6_attn_ln_weight_to_fp16, x = x_79_cast); + tensor var_792_to_fp16 = const()[name = tensor("op_792_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(250430976)))]; + tensor var_793_to_fp16 = const()[name = tensor("op_793_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(253707840)))]; + tensor q_25_cast = linear(bias = var_793_to_fp16, weight = var_792_to_fp16, x = var_781_cast); + tensor var_796_to_fp16 = const()[name = tensor("op_796_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(253710464)))]; + tensor k_25_bias_0_to_fp16 = const()[name = tensor("k_25_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(256987328)))]; + tensor k_25_cast = linear(bias = k_25_bias_0_to_fp16, weight = var_796_to_fp16, x = var_781_cast); + tensor var_800_to_fp16 = const()[name = tensor("op_800_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(256989952)))]; + tensor var_801_to_fp16 = const()[name = tensor("op_801_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260266816)))]; + tensor v_25_cast = linear(bias = var_801_to_fp16, weight = var_800_to_fp16, x = var_781_cast); + tensor var_809 = const()[name = tensor("op_809"), val = tensor([1, 1500, 20, -1])]; + tensor var_810_cast = reshape(shape = var_809, x = q_25_cast); + tensor const_236_to_fp16 = const()[name = tensor("const_236_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_27_cast = mul(x = var_810_cast, y = const_236_to_fp16); + tensor var_816 = const()[name = tensor("op_816"), val = tensor([1, 1500, 20, -1])]; + tensor var_817_cast = reshape(shape = var_816, x = k_25_cast); + tensor const_237_to_fp16 = const()[name = tensor("const_237_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_27_cast = mul(x = var_817_cast, y = const_237_to_fp16); + tensor var_823 = const()[name = tensor("op_823"), val = tensor([1, 1500, 20, -1])]; + tensor var_824_cast = reshape(shape = var_823, x = v_25_cast); + tensor var_825 = const()[name = tensor("op_825"), val = tensor([0, 2, 1, 3])]; + tensor qk_13_transpose_x_0 = const()[name = tensor("qk_13_transpose_x_0"), val = tensor(false)]; + tensor qk_13_transpose_y_0 = const()[name = tensor("qk_13_transpose_y_0"), val = tensor(false)]; + tensor transpose_76_perm_0 = const()[name = tensor("transpose_76_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_77_perm_0 = const()[name = tensor("transpose_77_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_229 = transpose(perm = transpose_77_perm_0, x = k_27_cast); + tensor transpose_230 = transpose(perm = transpose_76_perm_0, x = q_27_cast); + tensor qk_13_cast = matmul(transpose_x = qk_13_transpose_x_0, transpose_y = qk_13_transpose_y_0, x = transpose_230, y = transpose_229); + tensor var_829_cast = softmax(axis = var_764, x = qk_13_cast); + tensor var_831_transpose_x_0 = const()[name = tensor("op_831_transpose_x_0"), val = tensor(false)]; + tensor var_831_transpose_y_0 = const()[name = tensor("op_831_transpose_y_0"), val = tensor(false)]; + tensor transpose_231 = transpose(perm = var_825, x = var_824_cast); + tensor var_831_cast = matmul(transpose_x = var_831_transpose_x_0, transpose_y = var_831_transpose_y_0, x = var_829_cast, y = transpose_231); + tensor var_832 = const()[name = tensor("op_832"), val = tensor([0, 2, 1, 3])]; + tensor concat_6 = const()[name = tensor("concat_6"), val = tensor([1, 1500, 1280])]; + tensor transpose_228 = transpose(perm = var_832, x = var_831_cast); + tensor x_83_cast = reshape(shape = concat_6, x = transpose_228); + tensor var_837_to_fp16 = const()[name = tensor("op_837_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(260269440)))]; + tensor var_838_to_fp16 = const()[name = tensor("op_838_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263546304)))]; + tensor var_839_cast = linear(bias = var_838_to_fp16, weight = var_837_to_fp16, x = x_83_cast); + tensor x_85_cast = add(x = x_79_cast, y = var_839_cast); + tensor var_845_axes_0 = const()[name = tensor("op_845_axes_0"), val = tensor([-1])]; + tensor blocks_6_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_6_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263548928)))]; + tensor blocks_6_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_6_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263551552)))]; + tensor var_845_cast = layer_norm(axes = var_845_axes_0, beta = blocks_6_mlp_ln_bias_to_fp16, epsilon = var_770_to_fp16, gamma = blocks_6_mlp_ln_weight_to_fp16, x = x_85_cast); + tensor var_854_to_fp16 = const()[name = tensor("op_854_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263554176)))]; + tensor var_855_to_fp16 = const()[name = tensor("op_855_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(276661440)))]; + tensor input_57_cast = linear(bias = var_855_to_fp16, weight = var_854_to_fp16, x = var_845_cast); + tensor x_89_mode_0 = const()[name = tensor("x_89_mode_0"), val = tensor("EXACT")]; + tensor x_89_cast = gelu(mode = x_89_mode_0, x = input_57_cast); + tensor var_860_to_fp16 = const()[name = tensor("op_860_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(276671744)))]; + tensor var_861_to_fp16 = const()[name = tensor("op_861_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289779008)))]; + tensor var_862_cast = linear(bias = var_861_to_fp16, weight = var_860_to_fp16, x = x_89_cast); + tensor x_91_cast = add(x = x_85_cast, y = var_862_cast); + tensor var_871 = const()[name = tensor("op_871"), val = tensor(-1)]; + tensor var_888_axes_0 = const()[name = tensor("op_888_axes_0"), val = tensor([-1])]; + tensor blocks_7_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_7_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289781632)))]; + tensor blocks_7_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_7_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289784256)))]; + tensor var_877_to_fp16 = const()[name = tensor("op_877_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_888_cast = layer_norm(axes = var_888_axes_0, beta = blocks_7_attn_ln_bias_to_fp16, epsilon = var_877_to_fp16, gamma = blocks_7_attn_ln_weight_to_fp16, x = x_91_cast); + tensor var_899_to_fp16 = const()[name = tensor("op_899_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289786880)))]; + tensor var_900_to_fp16 = const()[name = tensor("op_900_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293063744)))]; + tensor q_29_cast = linear(bias = var_900_to_fp16, weight = var_899_to_fp16, x = var_888_cast); + tensor var_903_to_fp16 = const()[name = tensor("op_903_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293066368)))]; + tensor k_29_bias_0_to_fp16 = const()[name = tensor("k_29_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(296343232)))]; + tensor k_29_cast = linear(bias = k_29_bias_0_to_fp16, weight = var_903_to_fp16, x = var_888_cast); + tensor var_907_to_fp16 = const()[name = tensor("op_907_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(296345856)))]; + tensor var_908_to_fp16 = const()[name = tensor("op_908_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(299622720)))]; + tensor v_29_cast = linear(bias = var_908_to_fp16, weight = var_907_to_fp16, x = var_888_cast); + tensor var_916 = const()[name = tensor("op_916"), val = tensor([1, 1500, 20, -1])]; + tensor var_917_cast = reshape(shape = var_916, x = q_29_cast); + tensor const_238_to_fp16 = const()[name = tensor("const_238_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_31_cast = mul(x = var_917_cast, y = const_238_to_fp16); + tensor var_923 = const()[name = tensor("op_923"), val = tensor([1, 1500, 20, -1])]; + tensor var_924_cast = reshape(shape = var_923, x = k_29_cast); + tensor const_239_to_fp16 = const()[name = tensor("const_239_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_31_cast = mul(x = var_924_cast, y = const_239_to_fp16); + tensor var_930 = const()[name = tensor("op_930"), val = tensor([1, 1500, 20, -1])]; + tensor var_931_cast = reshape(shape = var_930, x = v_29_cast); + tensor var_932 = const()[name = tensor("op_932"), val = tensor([0, 2, 1, 3])]; + tensor qk_15_transpose_x_0 = const()[name = tensor("qk_15_transpose_x_0"), val = tensor(false)]; + tensor qk_15_transpose_y_0 = const()[name = tensor("qk_15_transpose_y_0"), val = tensor(false)]; + tensor transpose_78_perm_0 = const()[name = tensor("transpose_78_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_79_perm_0 = const()[name = tensor("transpose_79_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_225 = transpose(perm = transpose_79_perm_0, x = k_31_cast); + tensor transpose_226 = transpose(perm = transpose_78_perm_0, x = q_31_cast); + tensor qk_15_cast = matmul(transpose_x = qk_15_transpose_x_0, transpose_y = qk_15_transpose_y_0, x = transpose_226, y = transpose_225); + tensor var_936_cast = softmax(axis = var_871, x = qk_15_cast); + tensor var_938_transpose_x_0 = const()[name = tensor("op_938_transpose_x_0"), val = tensor(false)]; + tensor var_938_transpose_y_0 = const()[name = tensor("op_938_transpose_y_0"), val = tensor(false)]; + tensor transpose_227 = transpose(perm = var_932, x = var_931_cast); + tensor var_938_cast = matmul(transpose_x = var_938_transpose_x_0, transpose_y = var_938_transpose_y_0, x = var_936_cast, y = transpose_227); + tensor var_939 = const()[name = tensor("op_939"), val = tensor([0, 2, 1, 3])]; + tensor concat_7 = const()[name = tensor("concat_7"), val = tensor([1, 1500, 1280])]; + tensor transpose_224 = transpose(perm = var_939, x = var_938_cast); + tensor x_95_cast = reshape(shape = concat_7, x = transpose_224); + tensor var_944_to_fp16 = const()[name = tensor("op_944_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(299625344)))]; + tensor var_945_to_fp16 = const()[name = tensor("op_945_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302902208)))]; + tensor var_946_cast = linear(bias = var_945_to_fp16, weight = var_944_to_fp16, x = x_95_cast); + tensor x_97_cast = add(x = x_91_cast, y = var_946_cast); + tensor var_952_axes_0 = const()[name = tensor("op_952_axes_0"), val = tensor([-1])]; + tensor blocks_7_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_7_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302904832)))]; + tensor blocks_7_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_7_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302907456)))]; + tensor var_952_cast = layer_norm(axes = var_952_axes_0, beta = blocks_7_mlp_ln_bias_to_fp16, epsilon = var_877_to_fp16, gamma = blocks_7_mlp_ln_weight_to_fp16, x = x_97_cast); + tensor var_961_to_fp16 = const()[name = tensor("op_961_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(302910080)))]; + tensor var_962_to_fp16 = const()[name = tensor("op_962_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316017344)))]; + tensor input_65_cast = linear(bias = var_962_to_fp16, weight = var_961_to_fp16, x = var_952_cast); + tensor x_101_mode_0 = const()[name = tensor("x_101_mode_0"), val = tensor("EXACT")]; + tensor x_101_cast = gelu(mode = x_101_mode_0, x = input_65_cast); + tensor var_967_to_fp16 = const()[name = tensor("op_967_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316027648)))]; + tensor var_968_to_fp16 = const()[name = tensor("op_968_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329134912)))]; + tensor var_969_cast = linear(bias = var_968_to_fp16, weight = var_967_to_fp16, x = x_101_cast); + tensor x_103_cast = add(x = x_97_cast, y = var_969_cast); + tensor var_978 = const()[name = tensor("op_978"), val = tensor(-1)]; + tensor var_995_axes_0 = const()[name = tensor("op_995_axes_0"), val = tensor([-1])]; + tensor blocks_8_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_8_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329137536)))]; + tensor blocks_8_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_8_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329140160)))]; + tensor var_984_to_fp16 = const()[name = tensor("op_984_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_995_cast = layer_norm(axes = var_995_axes_0, beta = blocks_8_attn_ln_bias_to_fp16, epsilon = var_984_to_fp16, gamma = blocks_8_attn_ln_weight_to_fp16, x = x_103_cast); + tensor var_1006_to_fp16 = const()[name = tensor("op_1006_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(329142784)))]; + tensor var_1007_to_fp16 = const()[name = tensor("op_1007_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(332419648)))]; + tensor q_33_cast = linear(bias = var_1007_to_fp16, weight = var_1006_to_fp16, x = var_995_cast); + tensor var_1010_to_fp16 = const()[name = tensor("op_1010_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(332422272)))]; + tensor k_33_bias_0_to_fp16 = const()[name = tensor("k_33_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335699136)))]; + tensor k_33_cast = linear(bias = k_33_bias_0_to_fp16, weight = var_1010_to_fp16, x = var_995_cast); + tensor var_1014_to_fp16 = const()[name = tensor("op_1014_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(335701760)))]; + tensor var_1015_to_fp16 = const()[name = tensor("op_1015_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(338978624)))]; + tensor v_33_cast = linear(bias = var_1015_to_fp16, weight = var_1014_to_fp16, x = var_995_cast); + tensor var_1023 = const()[name = tensor("op_1023"), val = tensor([1, 1500, 20, -1])]; + tensor var_1024_cast = reshape(shape = var_1023, x = q_33_cast); + tensor const_240_to_fp16 = const()[name = tensor("const_240_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_35_cast = mul(x = var_1024_cast, y = const_240_to_fp16); + tensor var_1030 = const()[name = tensor("op_1030"), val = tensor([1, 1500, 20, -1])]; + tensor var_1031_cast = reshape(shape = var_1030, x = k_33_cast); + tensor const_241_to_fp16 = const()[name = tensor("const_241_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_35_cast = mul(x = var_1031_cast, y = const_241_to_fp16); + tensor var_1037 = const()[name = tensor("op_1037"), val = tensor([1, 1500, 20, -1])]; + tensor var_1038_cast = reshape(shape = var_1037, x = v_33_cast); + tensor var_1039 = const()[name = tensor("op_1039"), val = tensor([0, 2, 1, 3])]; + tensor qk_17_transpose_x_0 = const()[name = tensor("qk_17_transpose_x_0"), val = tensor(false)]; + tensor qk_17_transpose_y_0 = const()[name = tensor("qk_17_transpose_y_0"), val = tensor(false)]; + tensor transpose_80_perm_0 = const()[name = tensor("transpose_80_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_81_perm_0 = const()[name = tensor("transpose_81_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_221 = transpose(perm = transpose_81_perm_0, x = k_35_cast); + tensor transpose_222 = transpose(perm = transpose_80_perm_0, x = q_35_cast); + tensor qk_17_cast = matmul(transpose_x = qk_17_transpose_x_0, transpose_y = qk_17_transpose_y_0, x = transpose_222, y = transpose_221); + tensor var_1043_cast = softmax(axis = var_978, x = qk_17_cast); + tensor var_1045_transpose_x_0 = const()[name = tensor("op_1045_transpose_x_0"), val = tensor(false)]; + tensor var_1045_transpose_y_0 = const()[name = tensor("op_1045_transpose_y_0"), val = tensor(false)]; + tensor transpose_223 = transpose(perm = var_1039, x = var_1038_cast); + tensor var_1045_cast = matmul(transpose_x = var_1045_transpose_x_0, transpose_y = var_1045_transpose_y_0, x = var_1043_cast, y = transpose_223); + tensor var_1046 = const()[name = tensor("op_1046"), val = tensor([0, 2, 1, 3])]; + tensor concat_8 = const()[name = tensor("concat_8"), val = tensor([1, 1500, 1280])]; + tensor transpose_220 = transpose(perm = var_1046, x = var_1045_cast); + tensor x_107_cast = reshape(shape = concat_8, x = transpose_220); + tensor var_1051_to_fp16 = const()[name = tensor("op_1051_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(338981248)))]; + tensor var_1052_to_fp16 = const()[name = tensor("op_1052_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342258112)))]; + tensor var_1053_cast = linear(bias = var_1052_to_fp16, weight = var_1051_to_fp16, x = x_107_cast); + tensor x_109_cast = add(x = x_103_cast, y = var_1053_cast); + tensor var_1059_axes_0 = const()[name = tensor("op_1059_axes_0"), val = tensor([-1])]; + tensor blocks_8_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_8_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342260736)))]; + tensor blocks_8_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_8_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342263360)))]; + tensor var_1059_cast = layer_norm(axes = var_1059_axes_0, beta = blocks_8_mlp_ln_bias_to_fp16, epsilon = var_984_to_fp16, gamma = blocks_8_mlp_ln_weight_to_fp16, x = x_109_cast); + tensor var_1068_to_fp16 = const()[name = tensor("op_1068_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(342265984)))]; + tensor var_1069_to_fp16 = const()[name = tensor("op_1069_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(355373248)))]; + tensor input_73_cast = linear(bias = var_1069_to_fp16, weight = var_1068_to_fp16, x = var_1059_cast); + tensor x_113_mode_0 = const()[name = tensor("x_113_mode_0"), val = tensor("EXACT")]; + tensor x_113_cast = gelu(mode = x_113_mode_0, x = input_73_cast); + tensor var_1074_to_fp16 = const()[name = tensor("op_1074_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(355383552)))]; + tensor var_1075_to_fp16 = const()[name = tensor("op_1075_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368490816)))]; + tensor var_1076_cast = linear(bias = var_1075_to_fp16, weight = var_1074_to_fp16, x = x_113_cast); + tensor x_115_cast = add(x = x_109_cast, y = var_1076_cast); + tensor var_1085 = const()[name = tensor("op_1085"), val = tensor(-1)]; + tensor var_1102_axes_0 = const()[name = tensor("op_1102_axes_0"), val = tensor([-1])]; + tensor blocks_9_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_9_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368493440)))]; + tensor blocks_9_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_9_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368496064)))]; + tensor var_1091_to_fp16 = const()[name = tensor("op_1091_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1102_cast = layer_norm(axes = var_1102_axes_0, beta = blocks_9_attn_ln_bias_to_fp16, epsilon = var_1091_to_fp16, gamma = blocks_9_attn_ln_weight_to_fp16, x = x_115_cast); + tensor var_1113_to_fp16 = const()[name = tensor("op_1113_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368498688)))]; + tensor var_1114_to_fp16 = const()[name = tensor("op_1114_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(371775552)))]; + tensor q_37_cast = linear(bias = var_1114_to_fp16, weight = var_1113_to_fp16, x = var_1102_cast); + tensor var_1117_to_fp16 = const()[name = tensor("op_1117_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(371778176)))]; + tensor k_37_bias_0_to_fp16 = const()[name = tensor("k_37_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(375055040)))]; + tensor k_37_cast = linear(bias = k_37_bias_0_to_fp16, weight = var_1117_to_fp16, x = var_1102_cast); + tensor var_1121_to_fp16 = const()[name = tensor("op_1121_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(375057664)))]; + tensor var_1122_to_fp16 = const()[name = tensor("op_1122_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(378334528)))]; + tensor v_37_cast = linear(bias = var_1122_to_fp16, weight = var_1121_to_fp16, x = var_1102_cast); + tensor var_1130 = const()[name = tensor("op_1130"), val = tensor([1, 1500, 20, -1])]; + tensor var_1131_cast = reshape(shape = var_1130, x = q_37_cast); + tensor const_242_to_fp16 = const()[name = tensor("const_242_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_39_cast = mul(x = var_1131_cast, y = const_242_to_fp16); + tensor var_1137 = const()[name = tensor("op_1137"), val = tensor([1, 1500, 20, -1])]; + tensor var_1138_cast = reshape(shape = var_1137, x = k_37_cast); + tensor const_243_to_fp16 = const()[name = tensor("const_243_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_39_cast = mul(x = var_1138_cast, y = const_243_to_fp16); + tensor var_1144 = const()[name = tensor("op_1144"), val = tensor([1, 1500, 20, -1])]; + tensor var_1145_cast = reshape(shape = var_1144, x = v_37_cast); + tensor var_1146 = const()[name = tensor("op_1146"), val = tensor([0, 2, 1, 3])]; + tensor qk_19_transpose_x_0 = const()[name = tensor("qk_19_transpose_x_0"), val = tensor(false)]; + tensor qk_19_transpose_y_0 = const()[name = tensor("qk_19_transpose_y_0"), val = tensor(false)]; + tensor transpose_82_perm_0 = const()[name = tensor("transpose_82_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_83_perm_0 = const()[name = tensor("transpose_83_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_217 = transpose(perm = transpose_83_perm_0, x = k_39_cast); + tensor transpose_218 = transpose(perm = transpose_82_perm_0, x = q_39_cast); + tensor qk_19_cast = matmul(transpose_x = qk_19_transpose_x_0, transpose_y = qk_19_transpose_y_0, x = transpose_218, y = transpose_217); + tensor var_1150_cast = softmax(axis = var_1085, x = qk_19_cast); + tensor var_1152_transpose_x_0 = const()[name = tensor("op_1152_transpose_x_0"), val = tensor(false)]; + tensor var_1152_transpose_y_0 = const()[name = tensor("op_1152_transpose_y_0"), val = tensor(false)]; + tensor transpose_219 = transpose(perm = var_1146, x = var_1145_cast); + tensor var_1152_cast = matmul(transpose_x = var_1152_transpose_x_0, transpose_y = var_1152_transpose_y_0, x = var_1150_cast, y = transpose_219); + tensor var_1153 = const()[name = tensor("op_1153"), val = tensor([0, 2, 1, 3])]; + tensor concat_9 = const()[name = tensor("concat_9"), val = tensor([1, 1500, 1280])]; + tensor transpose_216 = transpose(perm = var_1153, x = var_1152_cast); + tensor x_119_cast = reshape(shape = concat_9, x = transpose_216); + tensor var_1158_to_fp16 = const()[name = tensor("op_1158_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(378337152)))]; + tensor var_1159_to_fp16 = const()[name = tensor("op_1159_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381614016)))]; + tensor var_1160_cast = linear(bias = var_1159_to_fp16, weight = var_1158_to_fp16, x = x_119_cast); + tensor x_121_cast = add(x = x_115_cast, y = var_1160_cast); + tensor var_1166_axes_0 = const()[name = tensor("op_1166_axes_0"), val = tensor([-1])]; + tensor blocks_9_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_9_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381616640)))]; + tensor blocks_9_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_9_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381619264)))]; + tensor var_1166_cast = layer_norm(axes = var_1166_axes_0, beta = blocks_9_mlp_ln_bias_to_fp16, epsilon = var_1091_to_fp16, gamma = blocks_9_mlp_ln_weight_to_fp16, x = x_121_cast); + tensor var_1175_to_fp16 = const()[name = tensor("op_1175_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(381621888)))]; + tensor var_1176_to_fp16 = const()[name = tensor("op_1176_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(394729152)))]; + tensor input_81_cast = linear(bias = var_1176_to_fp16, weight = var_1175_to_fp16, x = var_1166_cast); + tensor x_125_mode_0 = const()[name = tensor("x_125_mode_0"), val = tensor("EXACT")]; + tensor x_125_cast = gelu(mode = x_125_mode_0, x = input_81_cast); + tensor var_1181_to_fp16 = const()[name = tensor("op_1181_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(394739456)))]; + tensor var_1182_to_fp16 = const()[name = tensor("op_1182_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407846720)))]; + tensor var_1183_cast = linear(bias = var_1182_to_fp16, weight = var_1181_to_fp16, x = x_125_cast); + tensor x_127_cast = add(x = x_121_cast, y = var_1183_cast); + tensor var_1192 = const()[name = tensor("op_1192"), val = tensor(-1)]; + tensor var_1209_axes_0 = const()[name = tensor("op_1209_axes_0"), val = tensor([-1])]; + tensor blocks_10_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_10_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407849344)))]; + tensor blocks_10_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_10_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407851968)))]; + tensor var_1198_to_fp16 = const()[name = tensor("op_1198_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1209_cast = layer_norm(axes = var_1209_axes_0, beta = blocks_10_attn_ln_bias_to_fp16, epsilon = var_1198_to_fp16, gamma = blocks_10_attn_ln_weight_to_fp16, x = x_127_cast); + tensor var_1220_to_fp16 = const()[name = tensor("op_1220_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(407854592)))]; + tensor var_1221_to_fp16 = const()[name = tensor("op_1221_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411131456)))]; + tensor q_41_cast = linear(bias = var_1221_to_fp16, weight = var_1220_to_fp16, x = var_1209_cast); + tensor var_1224_to_fp16 = const()[name = tensor("op_1224_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(411134080)))]; + tensor k_41_bias_0_to_fp16 = const()[name = tensor("k_41_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(414410944)))]; + tensor k_41_cast = linear(bias = k_41_bias_0_to_fp16, weight = var_1224_to_fp16, x = var_1209_cast); + tensor var_1228_to_fp16 = const()[name = tensor("op_1228_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(414413568)))]; + tensor var_1229_to_fp16 = const()[name = tensor("op_1229_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(417690432)))]; + tensor v_41_cast = linear(bias = var_1229_to_fp16, weight = var_1228_to_fp16, x = var_1209_cast); + tensor var_1237 = const()[name = tensor("op_1237"), val = tensor([1, 1500, 20, -1])]; + tensor var_1238_cast = reshape(shape = var_1237, x = q_41_cast); + tensor const_244_to_fp16 = const()[name = tensor("const_244_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_43_cast = mul(x = var_1238_cast, y = const_244_to_fp16); + tensor var_1244 = const()[name = tensor("op_1244"), val = tensor([1, 1500, 20, -1])]; + tensor var_1245_cast = reshape(shape = var_1244, x = k_41_cast); + tensor const_245_to_fp16 = const()[name = tensor("const_245_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_43_cast = mul(x = var_1245_cast, y = const_245_to_fp16); + tensor var_1251 = const()[name = tensor("op_1251"), val = tensor([1, 1500, 20, -1])]; + tensor var_1252_cast = reshape(shape = var_1251, x = v_41_cast); + tensor var_1253 = const()[name = tensor("op_1253"), val = tensor([0, 2, 1, 3])]; + tensor qk_21_transpose_x_0 = const()[name = tensor("qk_21_transpose_x_0"), val = tensor(false)]; + tensor qk_21_transpose_y_0 = const()[name = tensor("qk_21_transpose_y_0"), val = tensor(false)]; + tensor transpose_84_perm_0 = const()[name = tensor("transpose_84_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_85_perm_0 = const()[name = tensor("transpose_85_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_213 = transpose(perm = transpose_85_perm_0, x = k_43_cast); + tensor transpose_214 = transpose(perm = transpose_84_perm_0, x = q_43_cast); + tensor qk_21_cast = matmul(transpose_x = qk_21_transpose_x_0, transpose_y = qk_21_transpose_y_0, x = transpose_214, y = transpose_213); + tensor var_1257_cast = softmax(axis = var_1192, x = qk_21_cast); + tensor var_1259_transpose_x_0 = const()[name = tensor("op_1259_transpose_x_0"), val = tensor(false)]; + tensor var_1259_transpose_y_0 = const()[name = tensor("op_1259_transpose_y_0"), val = tensor(false)]; + tensor transpose_215 = transpose(perm = var_1253, x = var_1252_cast); + tensor var_1259_cast = matmul(transpose_x = var_1259_transpose_x_0, transpose_y = var_1259_transpose_y_0, x = var_1257_cast, y = transpose_215); + tensor var_1260 = const()[name = tensor("op_1260"), val = tensor([0, 2, 1, 3])]; + tensor concat_10 = const()[name = tensor("concat_10"), val = tensor([1, 1500, 1280])]; + tensor transpose_212 = transpose(perm = var_1260, x = var_1259_cast); + tensor x_131_cast = reshape(shape = concat_10, x = transpose_212); + tensor var_1265_to_fp16 = const()[name = tensor("op_1265_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(417693056)))]; + tensor var_1266_to_fp16 = const()[name = tensor("op_1266_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(420969920)))]; + tensor var_1267_cast = linear(bias = var_1266_to_fp16, weight = var_1265_to_fp16, x = x_131_cast); + tensor x_133_cast = add(x = x_127_cast, y = var_1267_cast); + tensor var_1273_axes_0 = const()[name = tensor("op_1273_axes_0"), val = tensor([-1])]; + tensor blocks_10_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_10_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(420972544)))]; + tensor blocks_10_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_10_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(420975168)))]; + tensor var_1273_cast = layer_norm(axes = var_1273_axes_0, beta = blocks_10_mlp_ln_bias_to_fp16, epsilon = var_1198_to_fp16, gamma = blocks_10_mlp_ln_weight_to_fp16, x = x_133_cast); + tensor var_1282_to_fp16 = const()[name = tensor("op_1282_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(420977792)))]; + tensor var_1283_to_fp16 = const()[name = tensor("op_1283_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(434085056)))]; + tensor input_89_cast = linear(bias = var_1283_to_fp16, weight = var_1282_to_fp16, x = var_1273_cast); + tensor x_137_mode_0 = const()[name = tensor("x_137_mode_0"), val = tensor("EXACT")]; + tensor x_137_cast = gelu(mode = x_137_mode_0, x = input_89_cast); + tensor var_1288_to_fp16 = const()[name = tensor("op_1288_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(434095360)))]; + tensor var_1289_to_fp16 = const()[name = tensor("op_1289_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447202624)))]; + tensor var_1290_cast = linear(bias = var_1289_to_fp16, weight = var_1288_to_fp16, x = x_137_cast); + tensor x_139_cast = add(x = x_133_cast, y = var_1290_cast); + tensor var_1299 = const()[name = tensor("op_1299"), val = tensor(-1)]; + tensor var_1316_axes_0 = const()[name = tensor("op_1316_axes_0"), val = tensor([-1])]; + tensor blocks_11_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_11_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447205248)))]; + tensor blocks_11_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_11_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447207872)))]; + tensor var_1305_to_fp16 = const()[name = tensor("op_1305_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1316_cast = layer_norm(axes = var_1316_axes_0, beta = blocks_11_attn_ln_bias_to_fp16, epsilon = var_1305_to_fp16, gamma = blocks_11_attn_ln_weight_to_fp16, x = x_139_cast); + tensor var_1327_to_fp16 = const()[name = tensor("op_1327_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(447210496)))]; + tensor var_1328_to_fp16 = const()[name = tensor("op_1328_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(450487360)))]; + tensor q_45_cast = linear(bias = var_1328_to_fp16, weight = var_1327_to_fp16, x = var_1316_cast); + tensor var_1331_to_fp16 = const()[name = tensor("op_1331_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(450489984)))]; + tensor k_45_bias_0_to_fp16 = const()[name = tensor("k_45_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(453766848)))]; + tensor k_45_cast = linear(bias = k_45_bias_0_to_fp16, weight = var_1331_to_fp16, x = var_1316_cast); + tensor var_1335_to_fp16 = const()[name = tensor("op_1335_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(453769472)))]; + tensor var_1336_to_fp16 = const()[name = tensor("op_1336_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(457046336)))]; + tensor v_45_cast = linear(bias = var_1336_to_fp16, weight = var_1335_to_fp16, x = var_1316_cast); + tensor var_1344 = const()[name = tensor("op_1344"), val = tensor([1, 1500, 20, -1])]; + tensor var_1345_cast = reshape(shape = var_1344, x = q_45_cast); + tensor const_246_to_fp16 = const()[name = tensor("const_246_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_47_cast = mul(x = var_1345_cast, y = const_246_to_fp16); + tensor var_1351 = const()[name = tensor("op_1351"), val = tensor([1, 1500, 20, -1])]; + tensor var_1352_cast = reshape(shape = var_1351, x = k_45_cast); + tensor const_247_to_fp16 = const()[name = tensor("const_247_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_47_cast = mul(x = var_1352_cast, y = const_247_to_fp16); + tensor var_1358 = const()[name = tensor("op_1358"), val = tensor([1, 1500, 20, -1])]; + tensor var_1359_cast = reshape(shape = var_1358, x = v_45_cast); + tensor var_1360 = const()[name = tensor("op_1360"), val = tensor([0, 2, 1, 3])]; + tensor qk_23_transpose_x_0 = const()[name = tensor("qk_23_transpose_x_0"), val = tensor(false)]; + tensor qk_23_transpose_y_0 = const()[name = tensor("qk_23_transpose_y_0"), val = tensor(false)]; + tensor transpose_86_perm_0 = const()[name = tensor("transpose_86_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_87_perm_0 = const()[name = tensor("transpose_87_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_209 = transpose(perm = transpose_87_perm_0, x = k_47_cast); + tensor transpose_210 = transpose(perm = transpose_86_perm_0, x = q_47_cast); + tensor qk_23_cast = matmul(transpose_x = qk_23_transpose_x_0, transpose_y = qk_23_transpose_y_0, x = transpose_210, y = transpose_209); + tensor var_1364_cast = softmax(axis = var_1299, x = qk_23_cast); + tensor var_1366_transpose_x_0 = const()[name = tensor("op_1366_transpose_x_0"), val = tensor(false)]; + tensor var_1366_transpose_y_0 = const()[name = tensor("op_1366_transpose_y_0"), val = tensor(false)]; + tensor transpose_211 = transpose(perm = var_1360, x = var_1359_cast); + tensor var_1366_cast = matmul(transpose_x = var_1366_transpose_x_0, transpose_y = var_1366_transpose_y_0, x = var_1364_cast, y = transpose_211); + tensor var_1367 = const()[name = tensor("op_1367"), val = tensor([0, 2, 1, 3])]; + tensor concat_11 = const()[name = tensor("concat_11"), val = tensor([1, 1500, 1280])]; + tensor transpose_208 = transpose(perm = var_1367, x = var_1366_cast); + tensor x_143_cast = reshape(shape = concat_11, x = transpose_208); + tensor var_1372_to_fp16 = const()[name = tensor("op_1372_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(457048960)))]; + tensor var_1373_to_fp16 = const()[name = tensor("op_1373_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460325824)))]; + tensor var_1374_cast = linear(bias = var_1373_to_fp16, weight = var_1372_to_fp16, x = x_143_cast); + tensor x_145_cast = add(x = x_139_cast, y = var_1374_cast); + tensor var_1380_axes_0 = const()[name = tensor("op_1380_axes_0"), val = tensor([-1])]; + tensor blocks_11_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_11_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460328448)))]; + tensor blocks_11_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_11_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460331072)))]; + tensor var_1380_cast = layer_norm(axes = var_1380_axes_0, beta = blocks_11_mlp_ln_bias_to_fp16, epsilon = var_1305_to_fp16, gamma = blocks_11_mlp_ln_weight_to_fp16, x = x_145_cast); + tensor var_1389_to_fp16 = const()[name = tensor("op_1389_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(460333696)))]; + tensor var_1390_to_fp16 = const()[name = tensor("op_1390_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473440960)))]; + tensor input_97_cast = linear(bias = var_1390_to_fp16, weight = var_1389_to_fp16, x = var_1380_cast); + tensor x_149_mode_0 = const()[name = tensor("x_149_mode_0"), val = tensor("EXACT")]; + tensor x_149_cast = gelu(mode = x_149_mode_0, x = input_97_cast); + tensor var_1395_to_fp16 = const()[name = tensor("op_1395_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(473451264)))]; + tensor var_1396_to_fp16 = const()[name = tensor("op_1396_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486558528)))]; + tensor var_1397_cast = linear(bias = var_1396_to_fp16, weight = var_1395_to_fp16, x = x_149_cast); + tensor x_151_cast = add(x = x_145_cast, y = var_1397_cast); + tensor var_1406 = const()[name = tensor("op_1406"), val = tensor(-1)]; + tensor var_1423_axes_0 = const()[name = tensor("op_1423_axes_0"), val = tensor([-1])]; + tensor blocks_12_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_12_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486561152)))]; + tensor blocks_12_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_12_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486563776)))]; + tensor var_1412_to_fp16 = const()[name = tensor("op_1412_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1423_cast = layer_norm(axes = var_1423_axes_0, beta = blocks_12_attn_ln_bias_to_fp16, epsilon = var_1412_to_fp16, gamma = blocks_12_attn_ln_weight_to_fp16, x = x_151_cast); + tensor var_1434_to_fp16 = const()[name = tensor("op_1434_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(486566400)))]; + tensor var_1435_to_fp16 = const()[name = tensor("op_1435_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(489843264)))]; + tensor q_49_cast = linear(bias = var_1435_to_fp16, weight = var_1434_to_fp16, x = var_1423_cast); + tensor var_1438_to_fp16 = const()[name = tensor("op_1438_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(489845888)))]; + tensor k_49_bias_0_to_fp16 = const()[name = tensor("k_49_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(493122752)))]; + tensor k_49_cast = linear(bias = k_49_bias_0_to_fp16, weight = var_1438_to_fp16, x = var_1423_cast); + tensor var_1442_to_fp16 = const()[name = tensor("op_1442_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(493125376)))]; + tensor var_1443_to_fp16 = const()[name = tensor("op_1443_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496402240)))]; + tensor v_49_cast = linear(bias = var_1443_to_fp16, weight = var_1442_to_fp16, x = var_1423_cast); + tensor var_1451 = const()[name = tensor("op_1451"), val = tensor([1, 1500, 20, -1])]; + tensor var_1452_cast = reshape(shape = var_1451, x = q_49_cast); + tensor const_248_to_fp16 = const()[name = tensor("const_248_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_51_cast = mul(x = var_1452_cast, y = const_248_to_fp16); + tensor var_1458 = const()[name = tensor("op_1458"), val = tensor([1, 1500, 20, -1])]; + tensor var_1459_cast = reshape(shape = var_1458, x = k_49_cast); + tensor const_249_to_fp16 = const()[name = tensor("const_249_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_51_cast = mul(x = var_1459_cast, y = const_249_to_fp16); + tensor var_1465 = const()[name = tensor("op_1465"), val = tensor([1, 1500, 20, -1])]; + tensor var_1466_cast = reshape(shape = var_1465, x = v_49_cast); + tensor var_1467 = const()[name = tensor("op_1467"), val = tensor([0, 2, 1, 3])]; + tensor qk_25_transpose_x_0 = const()[name = tensor("qk_25_transpose_x_0"), val = tensor(false)]; + tensor qk_25_transpose_y_0 = const()[name = tensor("qk_25_transpose_y_0"), val = tensor(false)]; + tensor transpose_88_perm_0 = const()[name = tensor("transpose_88_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_89_perm_0 = const()[name = tensor("transpose_89_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_205 = transpose(perm = transpose_89_perm_0, x = k_51_cast); + tensor transpose_206 = transpose(perm = transpose_88_perm_0, x = q_51_cast); + tensor qk_25_cast = matmul(transpose_x = qk_25_transpose_x_0, transpose_y = qk_25_transpose_y_0, x = transpose_206, y = transpose_205); + tensor var_1471_cast = softmax(axis = var_1406, x = qk_25_cast); + tensor var_1473_transpose_x_0 = const()[name = tensor("op_1473_transpose_x_0"), val = tensor(false)]; + tensor var_1473_transpose_y_0 = const()[name = tensor("op_1473_transpose_y_0"), val = tensor(false)]; + tensor transpose_207 = transpose(perm = var_1467, x = var_1466_cast); + tensor var_1473_cast = matmul(transpose_x = var_1473_transpose_x_0, transpose_y = var_1473_transpose_y_0, x = var_1471_cast, y = transpose_207); + tensor var_1474 = const()[name = tensor("op_1474"), val = tensor([0, 2, 1, 3])]; + tensor concat_12 = const()[name = tensor("concat_12"), val = tensor([1, 1500, 1280])]; + tensor transpose_204 = transpose(perm = var_1474, x = var_1473_cast); + tensor x_155_cast = reshape(shape = concat_12, x = transpose_204); + tensor var_1479_to_fp16 = const()[name = tensor("op_1479_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496404864)))]; + tensor var_1480_to_fp16 = const()[name = tensor("op_1480_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(499681728)))]; + tensor var_1481_cast = linear(bias = var_1480_to_fp16, weight = var_1479_to_fp16, x = x_155_cast); + tensor x_157_cast = add(x = x_151_cast, y = var_1481_cast); + tensor var_1487_axes_0 = const()[name = tensor("op_1487_axes_0"), val = tensor([-1])]; + tensor blocks_12_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_12_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(499684352)))]; + tensor blocks_12_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_12_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(499686976)))]; + tensor var_1487_cast = layer_norm(axes = var_1487_axes_0, beta = blocks_12_mlp_ln_bias_to_fp16, epsilon = var_1412_to_fp16, gamma = blocks_12_mlp_ln_weight_to_fp16, x = x_157_cast); + tensor var_1496_to_fp16 = const()[name = tensor("op_1496_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(499689600)))]; + tensor var_1497_to_fp16 = const()[name = tensor("op_1497_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(512796864)))]; + tensor input_105_cast = linear(bias = var_1497_to_fp16, weight = var_1496_to_fp16, x = var_1487_cast); + tensor x_161_mode_0 = const()[name = tensor("x_161_mode_0"), val = tensor("EXACT")]; + tensor x_161_cast = gelu(mode = x_161_mode_0, x = input_105_cast); + tensor var_1502_to_fp16 = const()[name = tensor("op_1502_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(512807168)))]; + tensor var_1503_to_fp16 = const()[name = tensor("op_1503_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(525914432)))]; + tensor var_1504_cast = linear(bias = var_1503_to_fp16, weight = var_1502_to_fp16, x = x_161_cast); + tensor x_163_cast = add(x = x_157_cast, y = var_1504_cast); + tensor var_1513 = const()[name = tensor("op_1513"), val = tensor(-1)]; + tensor var_1530_axes_0 = const()[name = tensor("op_1530_axes_0"), val = tensor([-1])]; + tensor blocks_13_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_13_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(525917056)))]; + tensor blocks_13_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_13_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(525919680)))]; + tensor var_1519_to_fp16 = const()[name = tensor("op_1519_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1530_cast = layer_norm(axes = var_1530_axes_0, beta = blocks_13_attn_ln_bias_to_fp16, epsilon = var_1519_to_fp16, gamma = blocks_13_attn_ln_weight_to_fp16, x = x_163_cast); + tensor var_1541_to_fp16 = const()[name = tensor("op_1541_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(525922304)))]; + tensor var_1542_to_fp16 = const()[name = tensor("op_1542_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529199168)))]; + tensor q_53_cast = linear(bias = var_1542_to_fp16, weight = var_1541_to_fp16, x = var_1530_cast); + tensor var_1545_to_fp16 = const()[name = tensor("op_1545_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(529201792)))]; + tensor k_53_bias_0_to_fp16 = const()[name = tensor("k_53_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(532478656)))]; + tensor k_53_cast = linear(bias = k_53_bias_0_to_fp16, weight = var_1545_to_fp16, x = var_1530_cast); + tensor var_1549_to_fp16 = const()[name = tensor("op_1549_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(532481280)))]; + tensor var_1550_to_fp16 = const()[name = tensor("op_1550_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(535758144)))]; + tensor v_53_cast = linear(bias = var_1550_to_fp16, weight = var_1549_to_fp16, x = var_1530_cast); + tensor var_1558 = const()[name = tensor("op_1558"), val = tensor([1, 1500, 20, -1])]; + tensor var_1559_cast = reshape(shape = var_1558, x = q_53_cast); + tensor const_250_to_fp16 = const()[name = tensor("const_250_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_55_cast = mul(x = var_1559_cast, y = const_250_to_fp16); + tensor var_1565 = const()[name = tensor("op_1565"), val = tensor([1, 1500, 20, -1])]; + tensor var_1566_cast = reshape(shape = var_1565, x = k_53_cast); + tensor const_251_to_fp16 = const()[name = tensor("const_251_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_55_cast = mul(x = var_1566_cast, y = const_251_to_fp16); + tensor var_1572 = const()[name = tensor("op_1572"), val = tensor([1, 1500, 20, -1])]; + tensor var_1573_cast = reshape(shape = var_1572, x = v_53_cast); + tensor var_1574 = const()[name = tensor("op_1574"), val = tensor([0, 2, 1, 3])]; + tensor qk_27_transpose_x_0 = const()[name = tensor("qk_27_transpose_x_0"), val = tensor(false)]; + tensor qk_27_transpose_y_0 = const()[name = tensor("qk_27_transpose_y_0"), val = tensor(false)]; + tensor transpose_90_perm_0 = const()[name = tensor("transpose_90_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_91_perm_0 = const()[name = tensor("transpose_91_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_201 = transpose(perm = transpose_91_perm_0, x = k_55_cast); + tensor transpose_202 = transpose(perm = transpose_90_perm_0, x = q_55_cast); + tensor qk_27_cast = matmul(transpose_x = qk_27_transpose_x_0, transpose_y = qk_27_transpose_y_0, x = transpose_202, y = transpose_201); + tensor var_1578_cast = softmax(axis = var_1513, x = qk_27_cast); + tensor var_1580_transpose_x_0 = const()[name = tensor("op_1580_transpose_x_0"), val = tensor(false)]; + tensor var_1580_transpose_y_0 = const()[name = tensor("op_1580_transpose_y_0"), val = tensor(false)]; + tensor transpose_203 = transpose(perm = var_1574, x = var_1573_cast); + tensor var_1580_cast = matmul(transpose_x = var_1580_transpose_x_0, transpose_y = var_1580_transpose_y_0, x = var_1578_cast, y = transpose_203); + tensor var_1581 = const()[name = tensor("op_1581"), val = tensor([0, 2, 1, 3])]; + tensor concat_13 = const()[name = tensor("concat_13"), val = tensor([1, 1500, 1280])]; + tensor transpose_200 = transpose(perm = var_1581, x = var_1580_cast); + tensor x_167_cast = reshape(shape = concat_13, x = transpose_200); + tensor var_1586_to_fp16 = const()[name = tensor("op_1586_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(535760768)))]; + tensor var_1587_to_fp16 = const()[name = tensor("op_1587_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(539037632)))]; + tensor var_1588_cast = linear(bias = var_1587_to_fp16, weight = var_1586_to_fp16, x = x_167_cast); + tensor x_169_cast = add(x = x_163_cast, y = var_1588_cast); + tensor var_1594_axes_0 = const()[name = tensor("op_1594_axes_0"), val = tensor([-1])]; + tensor blocks_13_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_13_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(539040256)))]; + tensor blocks_13_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_13_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(539042880)))]; + tensor var_1594_cast = layer_norm(axes = var_1594_axes_0, beta = blocks_13_mlp_ln_bias_to_fp16, epsilon = var_1519_to_fp16, gamma = blocks_13_mlp_ln_weight_to_fp16, x = x_169_cast); + tensor var_1603_to_fp16 = const()[name = tensor("op_1603_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(539045504)))]; + tensor var_1604_to_fp16 = const()[name = tensor("op_1604_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552152768)))]; + tensor input_113_cast = linear(bias = var_1604_to_fp16, weight = var_1603_to_fp16, x = var_1594_cast); + tensor x_173_mode_0 = const()[name = tensor("x_173_mode_0"), val = tensor("EXACT")]; + tensor x_173_cast = gelu(mode = x_173_mode_0, x = input_113_cast); + tensor var_1609_to_fp16 = const()[name = tensor("op_1609_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(552163072)))]; + tensor var_1610_to_fp16 = const()[name = tensor("op_1610_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565270336)))]; + tensor var_1611_cast = linear(bias = var_1610_to_fp16, weight = var_1609_to_fp16, x = x_173_cast); + tensor x_175_cast = add(x = x_169_cast, y = var_1611_cast); + tensor var_1620 = const()[name = tensor("op_1620"), val = tensor(-1)]; + tensor var_1637_axes_0 = const()[name = tensor("op_1637_axes_0"), val = tensor([-1])]; + tensor blocks_14_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_14_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565272960)))]; + tensor blocks_14_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_14_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565275584)))]; + tensor var_1626_to_fp16 = const()[name = tensor("op_1626_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1637_cast = layer_norm(axes = var_1637_axes_0, beta = blocks_14_attn_ln_bias_to_fp16, epsilon = var_1626_to_fp16, gamma = blocks_14_attn_ln_weight_to_fp16, x = x_175_cast); + tensor var_1648_to_fp16 = const()[name = tensor("op_1648_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(565278208)))]; + tensor var_1649_to_fp16 = const()[name = tensor("op_1649_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568555072)))]; + tensor q_57_cast = linear(bias = var_1649_to_fp16, weight = var_1648_to_fp16, x = var_1637_cast); + tensor var_1652_to_fp16 = const()[name = tensor("op_1652_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568557696)))]; + tensor k_57_bias_0_to_fp16 = const()[name = tensor("k_57_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(571834560)))]; + tensor k_57_cast = linear(bias = k_57_bias_0_to_fp16, weight = var_1652_to_fp16, x = var_1637_cast); + tensor var_1656_to_fp16 = const()[name = tensor("op_1656_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(571837184)))]; + tensor var_1657_to_fp16 = const()[name = tensor("op_1657_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(575114048)))]; + tensor v_57_cast = linear(bias = var_1657_to_fp16, weight = var_1656_to_fp16, x = var_1637_cast); + tensor var_1665 = const()[name = tensor("op_1665"), val = tensor([1, 1500, 20, -1])]; + tensor var_1666_cast = reshape(shape = var_1665, x = q_57_cast); + tensor const_252_to_fp16 = const()[name = tensor("const_252_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_59_cast = mul(x = var_1666_cast, y = const_252_to_fp16); + tensor var_1672 = const()[name = tensor("op_1672"), val = tensor([1, 1500, 20, -1])]; + tensor var_1673_cast = reshape(shape = var_1672, x = k_57_cast); + tensor const_253_to_fp16 = const()[name = tensor("const_253_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_59_cast = mul(x = var_1673_cast, y = const_253_to_fp16); + tensor var_1679 = const()[name = tensor("op_1679"), val = tensor([1, 1500, 20, -1])]; + tensor var_1680_cast = reshape(shape = var_1679, x = v_57_cast); + tensor var_1681 = const()[name = tensor("op_1681"), val = tensor([0, 2, 1, 3])]; + tensor qk_29_transpose_x_0 = const()[name = tensor("qk_29_transpose_x_0"), val = tensor(false)]; + tensor qk_29_transpose_y_0 = const()[name = tensor("qk_29_transpose_y_0"), val = tensor(false)]; + tensor transpose_92_perm_0 = const()[name = tensor("transpose_92_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_93_perm_0 = const()[name = tensor("transpose_93_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_197 = transpose(perm = transpose_93_perm_0, x = k_59_cast); + tensor transpose_198 = transpose(perm = transpose_92_perm_0, x = q_59_cast); + tensor qk_29_cast = matmul(transpose_x = qk_29_transpose_x_0, transpose_y = qk_29_transpose_y_0, x = transpose_198, y = transpose_197); + tensor var_1685_cast = softmax(axis = var_1620, x = qk_29_cast); + tensor var_1687_transpose_x_0 = const()[name = tensor("op_1687_transpose_x_0"), val = tensor(false)]; + tensor var_1687_transpose_y_0 = const()[name = tensor("op_1687_transpose_y_0"), val = tensor(false)]; + tensor transpose_199 = transpose(perm = var_1681, x = var_1680_cast); + tensor var_1687_cast = matmul(transpose_x = var_1687_transpose_x_0, transpose_y = var_1687_transpose_y_0, x = var_1685_cast, y = transpose_199); + tensor var_1688 = const()[name = tensor("op_1688"), val = tensor([0, 2, 1, 3])]; + tensor concat_14 = const()[name = tensor("concat_14"), val = tensor([1, 1500, 1280])]; + tensor transpose_196 = transpose(perm = var_1688, x = var_1687_cast); + tensor x_179_cast = reshape(shape = concat_14, x = transpose_196); + tensor var_1693_to_fp16 = const()[name = tensor("op_1693_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(575116672)))]; + tensor var_1694_to_fp16 = const()[name = tensor("op_1694_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578393536)))]; + tensor var_1695_cast = linear(bias = var_1694_to_fp16, weight = var_1693_to_fp16, x = x_179_cast); + tensor x_181_cast = add(x = x_175_cast, y = var_1695_cast); + tensor var_1701_axes_0 = const()[name = tensor("op_1701_axes_0"), val = tensor([-1])]; + tensor blocks_14_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_14_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578396160)))]; + tensor blocks_14_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_14_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578398784)))]; + tensor var_1701_cast = layer_norm(axes = var_1701_axes_0, beta = blocks_14_mlp_ln_bias_to_fp16, epsilon = var_1626_to_fp16, gamma = blocks_14_mlp_ln_weight_to_fp16, x = x_181_cast); + tensor var_1710_to_fp16 = const()[name = tensor("op_1710_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(578401408)))]; + tensor var_1711_to_fp16 = const()[name = tensor("op_1711_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(591508672)))]; + tensor input_121_cast = linear(bias = var_1711_to_fp16, weight = var_1710_to_fp16, x = var_1701_cast); + tensor x_185_mode_0 = const()[name = tensor("x_185_mode_0"), val = tensor("EXACT")]; + tensor x_185_cast = gelu(mode = x_185_mode_0, x = input_121_cast); + tensor var_1716_to_fp16 = const()[name = tensor("op_1716_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(591518976)))]; + tensor var_1717_to_fp16 = const()[name = tensor("op_1717_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604626240)))]; + tensor var_1718_cast = linear(bias = var_1717_to_fp16, weight = var_1716_to_fp16, x = x_185_cast); + tensor x_187_cast = add(x = x_181_cast, y = var_1718_cast); + tensor var_1727 = const()[name = tensor("op_1727"), val = tensor(-1)]; + tensor var_1744_axes_0 = const()[name = tensor("op_1744_axes_0"), val = tensor([-1])]; + tensor blocks_15_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_15_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604628864)))]; + tensor blocks_15_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_15_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604631488)))]; + tensor var_1733_to_fp16 = const()[name = tensor("op_1733_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1744_cast = layer_norm(axes = var_1744_axes_0, beta = blocks_15_attn_ln_bias_to_fp16, epsilon = var_1733_to_fp16, gamma = blocks_15_attn_ln_weight_to_fp16, x = x_187_cast); + tensor var_1755_to_fp16 = const()[name = tensor("op_1755_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(604634112)))]; + tensor var_1756_to_fp16 = const()[name = tensor("op_1756_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(607910976)))]; + tensor q_61_cast = linear(bias = var_1756_to_fp16, weight = var_1755_to_fp16, x = var_1744_cast); + tensor var_1759_to_fp16 = const()[name = tensor("op_1759_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(607913600)))]; + tensor k_61_bias_0_to_fp16 = const()[name = tensor("k_61_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(611190464)))]; + tensor k_61_cast = linear(bias = k_61_bias_0_to_fp16, weight = var_1759_to_fp16, x = var_1744_cast); + tensor var_1763_to_fp16 = const()[name = tensor("op_1763_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(611193088)))]; + tensor var_1764_to_fp16 = const()[name = tensor("op_1764_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614469952)))]; + tensor v_61_cast = linear(bias = var_1764_to_fp16, weight = var_1763_to_fp16, x = var_1744_cast); + tensor var_1772 = const()[name = tensor("op_1772"), val = tensor([1, 1500, 20, -1])]; + tensor var_1773_cast = reshape(shape = var_1772, x = q_61_cast); + tensor const_254_to_fp16 = const()[name = tensor("const_254_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_63_cast = mul(x = var_1773_cast, y = const_254_to_fp16); + tensor var_1779 = const()[name = tensor("op_1779"), val = tensor([1, 1500, 20, -1])]; + tensor var_1780_cast = reshape(shape = var_1779, x = k_61_cast); + tensor const_255_to_fp16 = const()[name = tensor("const_255_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_63_cast = mul(x = var_1780_cast, y = const_255_to_fp16); + tensor var_1786 = const()[name = tensor("op_1786"), val = tensor([1, 1500, 20, -1])]; + tensor var_1787_cast = reshape(shape = var_1786, x = v_61_cast); + tensor var_1788 = const()[name = tensor("op_1788"), val = tensor([0, 2, 1, 3])]; + tensor qk_31_transpose_x_0 = const()[name = tensor("qk_31_transpose_x_0"), val = tensor(false)]; + tensor qk_31_transpose_y_0 = const()[name = tensor("qk_31_transpose_y_0"), val = tensor(false)]; + tensor transpose_94_perm_0 = const()[name = tensor("transpose_94_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_95_perm_0 = const()[name = tensor("transpose_95_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_193 = transpose(perm = transpose_95_perm_0, x = k_63_cast); + tensor transpose_194 = transpose(perm = transpose_94_perm_0, x = q_63_cast); + tensor qk_31_cast = matmul(transpose_x = qk_31_transpose_x_0, transpose_y = qk_31_transpose_y_0, x = transpose_194, y = transpose_193); + tensor var_1792_cast = softmax(axis = var_1727, x = qk_31_cast); + tensor var_1794_transpose_x_0 = const()[name = tensor("op_1794_transpose_x_0"), val = tensor(false)]; + tensor var_1794_transpose_y_0 = const()[name = tensor("op_1794_transpose_y_0"), val = tensor(false)]; + tensor transpose_195 = transpose(perm = var_1788, x = var_1787_cast); + tensor var_1794_cast = matmul(transpose_x = var_1794_transpose_x_0, transpose_y = var_1794_transpose_y_0, x = var_1792_cast, y = transpose_195); + tensor var_1795 = const()[name = tensor("op_1795"), val = tensor([0, 2, 1, 3])]; + tensor concat_15 = const()[name = tensor("concat_15"), val = tensor([1, 1500, 1280])]; + tensor transpose_192 = transpose(perm = var_1795, x = var_1794_cast); + tensor x_191_cast = reshape(shape = concat_15, x = transpose_192); + tensor var_1800_to_fp16 = const()[name = tensor("op_1800_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614472576)))]; + tensor var_1801_to_fp16 = const()[name = tensor("op_1801_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617749440)))]; + tensor var_1802_cast = linear(bias = var_1801_to_fp16, weight = var_1800_to_fp16, x = x_191_cast); + tensor x_193_cast = add(x = x_187_cast, y = var_1802_cast); + tensor var_1808_axes_0 = const()[name = tensor("op_1808_axes_0"), val = tensor([-1])]; + tensor blocks_15_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_15_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617752064)))]; + tensor blocks_15_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_15_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617754688)))]; + tensor var_1808_cast = layer_norm(axes = var_1808_axes_0, beta = blocks_15_mlp_ln_bias_to_fp16, epsilon = var_1733_to_fp16, gamma = blocks_15_mlp_ln_weight_to_fp16, x = x_193_cast); + tensor var_1817_to_fp16 = const()[name = tensor("op_1817_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(617757312)))]; + tensor var_1818_to_fp16 = const()[name = tensor("op_1818_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(630864576)))]; + tensor input_129_cast = linear(bias = var_1818_to_fp16, weight = var_1817_to_fp16, x = var_1808_cast); + tensor x_197_mode_0 = const()[name = tensor("x_197_mode_0"), val = tensor("EXACT")]; + tensor x_197_cast = gelu(mode = x_197_mode_0, x = input_129_cast); + tensor var_1823_to_fp16 = const()[name = tensor("op_1823_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(630874880)))]; + tensor var_1824_to_fp16 = const()[name = tensor("op_1824_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643982144)))]; + tensor var_1825_cast = linear(bias = var_1824_to_fp16, weight = var_1823_to_fp16, x = x_197_cast); + tensor x_199_cast = add(x = x_193_cast, y = var_1825_cast); + tensor var_1834 = const()[name = tensor("op_1834"), val = tensor(-1)]; + tensor var_1851_axes_0 = const()[name = tensor("op_1851_axes_0"), val = tensor([-1])]; + tensor blocks_16_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_16_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643984768)))]; + tensor blocks_16_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_16_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643987392)))]; + tensor var_1840_to_fp16 = const()[name = tensor("op_1840_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1851_cast = layer_norm(axes = var_1851_axes_0, beta = blocks_16_attn_ln_bias_to_fp16, epsilon = var_1840_to_fp16, gamma = blocks_16_attn_ln_weight_to_fp16, x = x_199_cast); + tensor var_1862_to_fp16 = const()[name = tensor("op_1862_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(643990016)))]; + tensor var_1863_to_fp16 = const()[name = tensor("op_1863_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647266880)))]; + tensor q_65_cast = linear(bias = var_1863_to_fp16, weight = var_1862_to_fp16, x = var_1851_cast); + tensor var_1866_to_fp16 = const()[name = tensor("op_1866_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(647269504)))]; + tensor k_65_bias_0_to_fp16 = const()[name = tensor("k_65_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(650546368)))]; + tensor k_65_cast = linear(bias = k_65_bias_0_to_fp16, weight = var_1866_to_fp16, x = var_1851_cast); + tensor var_1870_to_fp16 = const()[name = tensor("op_1870_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(650548992)))]; + tensor var_1871_to_fp16 = const()[name = tensor("op_1871_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(653825856)))]; + tensor v_65_cast = linear(bias = var_1871_to_fp16, weight = var_1870_to_fp16, x = var_1851_cast); + tensor var_1879 = const()[name = tensor("op_1879"), val = tensor([1, 1500, 20, -1])]; + tensor var_1880_cast = reshape(shape = var_1879, x = q_65_cast); + tensor const_256_to_fp16 = const()[name = tensor("const_256_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_67_cast = mul(x = var_1880_cast, y = const_256_to_fp16); + tensor var_1886 = const()[name = tensor("op_1886"), val = tensor([1, 1500, 20, -1])]; + tensor var_1887_cast = reshape(shape = var_1886, x = k_65_cast); + tensor const_257_to_fp16 = const()[name = tensor("const_257_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_67_cast = mul(x = var_1887_cast, y = const_257_to_fp16); + tensor var_1893 = const()[name = tensor("op_1893"), val = tensor([1, 1500, 20, -1])]; + tensor var_1894_cast = reshape(shape = var_1893, x = v_65_cast); + tensor var_1895 = const()[name = tensor("op_1895"), val = tensor([0, 2, 1, 3])]; + tensor qk_33_transpose_x_0 = const()[name = tensor("qk_33_transpose_x_0"), val = tensor(false)]; + tensor qk_33_transpose_y_0 = const()[name = tensor("qk_33_transpose_y_0"), val = tensor(false)]; + tensor transpose_96_perm_0 = const()[name = tensor("transpose_96_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_97_perm_0 = const()[name = tensor("transpose_97_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_189 = transpose(perm = transpose_97_perm_0, x = k_67_cast); + tensor transpose_190 = transpose(perm = transpose_96_perm_0, x = q_67_cast); + tensor qk_33_cast = matmul(transpose_x = qk_33_transpose_x_0, transpose_y = qk_33_transpose_y_0, x = transpose_190, y = transpose_189); + tensor var_1899_cast = softmax(axis = var_1834, x = qk_33_cast); + tensor var_1901_transpose_x_0 = const()[name = tensor("op_1901_transpose_x_0"), val = tensor(false)]; + tensor var_1901_transpose_y_0 = const()[name = tensor("op_1901_transpose_y_0"), val = tensor(false)]; + tensor transpose_191 = transpose(perm = var_1895, x = var_1894_cast); + tensor var_1901_cast = matmul(transpose_x = var_1901_transpose_x_0, transpose_y = var_1901_transpose_y_0, x = var_1899_cast, y = transpose_191); + tensor var_1902 = const()[name = tensor("op_1902"), val = tensor([0, 2, 1, 3])]; + tensor concat_16 = const()[name = tensor("concat_16"), val = tensor([1, 1500, 1280])]; + tensor transpose_188 = transpose(perm = var_1902, x = var_1901_cast); + tensor x_203_cast = reshape(shape = concat_16, x = transpose_188); + tensor var_1907_to_fp16 = const()[name = tensor("op_1907_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(653828480)))]; + tensor var_1908_to_fp16 = const()[name = tensor("op_1908_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657105344)))]; + tensor var_1909_cast = linear(bias = var_1908_to_fp16, weight = var_1907_to_fp16, x = x_203_cast); + tensor x_205_cast = add(x = x_199_cast, y = var_1909_cast); + tensor var_1915_axes_0 = const()[name = tensor("op_1915_axes_0"), val = tensor([-1])]; + tensor blocks_16_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_16_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657107968)))]; + tensor blocks_16_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_16_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657110592)))]; + tensor var_1915_cast = layer_norm(axes = var_1915_axes_0, beta = blocks_16_mlp_ln_bias_to_fp16, epsilon = var_1840_to_fp16, gamma = blocks_16_mlp_ln_weight_to_fp16, x = x_205_cast); + tensor var_1924_to_fp16 = const()[name = tensor("op_1924_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(657113216)))]; + tensor var_1925_to_fp16 = const()[name = tensor("op_1925_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(670220480)))]; + tensor input_137_cast = linear(bias = var_1925_to_fp16, weight = var_1924_to_fp16, x = var_1915_cast); + tensor x_209_mode_0 = const()[name = tensor("x_209_mode_0"), val = tensor("EXACT")]; + tensor x_209_cast = gelu(mode = x_209_mode_0, x = input_137_cast); + tensor var_1930_to_fp16 = const()[name = tensor("op_1930_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(670230784)))]; + tensor var_1931_to_fp16 = const()[name = tensor("op_1931_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(683338048)))]; + tensor var_1932_cast = linear(bias = var_1931_to_fp16, weight = var_1930_to_fp16, x = x_209_cast); + tensor x_211_cast = add(x = x_205_cast, y = var_1932_cast); + tensor var_1941 = const()[name = tensor("op_1941"), val = tensor(-1)]; + tensor var_1958_axes_0 = const()[name = tensor("op_1958_axes_0"), val = tensor([-1])]; + tensor blocks_17_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_17_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(683340672)))]; + tensor blocks_17_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_17_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(683343296)))]; + tensor var_1947_to_fp16 = const()[name = tensor("op_1947_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1958_cast = layer_norm(axes = var_1958_axes_0, beta = blocks_17_attn_ln_bias_to_fp16, epsilon = var_1947_to_fp16, gamma = blocks_17_attn_ln_weight_to_fp16, x = x_211_cast); + tensor var_1969_to_fp16 = const()[name = tensor("op_1969_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(683345920)))]; + tensor var_1970_to_fp16 = const()[name = tensor("op_1970_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(686622784)))]; + tensor q_69_cast = linear(bias = var_1970_to_fp16, weight = var_1969_to_fp16, x = var_1958_cast); + tensor var_1973_to_fp16 = const()[name = tensor("op_1973_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(686625408)))]; + tensor k_69_bias_0_to_fp16 = const()[name = tensor("k_69_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(689902272)))]; + tensor k_69_cast = linear(bias = k_69_bias_0_to_fp16, weight = var_1973_to_fp16, x = var_1958_cast); + tensor var_1977_to_fp16 = const()[name = tensor("op_1977_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(689904896)))]; + tensor var_1978_to_fp16 = const()[name = tensor("op_1978_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(693181760)))]; + tensor v_69_cast = linear(bias = var_1978_to_fp16, weight = var_1977_to_fp16, x = var_1958_cast); + tensor var_1986 = const()[name = tensor("op_1986"), val = tensor([1, 1500, 20, -1])]; + tensor var_1987_cast = reshape(shape = var_1986, x = q_69_cast); + tensor const_258_to_fp16 = const()[name = tensor("const_258_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_71_cast = mul(x = var_1987_cast, y = const_258_to_fp16); + tensor var_1993 = const()[name = tensor("op_1993"), val = tensor([1, 1500, 20, -1])]; + tensor var_1994_cast = reshape(shape = var_1993, x = k_69_cast); + tensor const_259_to_fp16 = const()[name = tensor("const_259_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_71_cast = mul(x = var_1994_cast, y = const_259_to_fp16); + tensor var_2000 = const()[name = tensor("op_2000"), val = tensor([1, 1500, 20, -1])]; + tensor var_2001_cast = reshape(shape = var_2000, x = v_69_cast); + tensor var_2002 = const()[name = tensor("op_2002"), val = tensor([0, 2, 1, 3])]; + tensor qk_35_transpose_x_0 = const()[name = tensor("qk_35_transpose_x_0"), val = tensor(false)]; + tensor qk_35_transpose_y_0 = const()[name = tensor("qk_35_transpose_y_0"), val = tensor(false)]; + tensor transpose_98_perm_0 = const()[name = tensor("transpose_98_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_99_perm_0 = const()[name = tensor("transpose_99_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_185 = transpose(perm = transpose_99_perm_0, x = k_71_cast); + tensor transpose_186 = transpose(perm = transpose_98_perm_0, x = q_71_cast); + tensor qk_35_cast = matmul(transpose_x = qk_35_transpose_x_0, transpose_y = qk_35_transpose_y_0, x = transpose_186, y = transpose_185); + tensor var_2006_cast = softmax(axis = var_1941, x = qk_35_cast); + tensor var_2008_transpose_x_0 = const()[name = tensor("op_2008_transpose_x_0"), val = tensor(false)]; + tensor var_2008_transpose_y_0 = const()[name = tensor("op_2008_transpose_y_0"), val = tensor(false)]; + tensor transpose_187 = transpose(perm = var_2002, x = var_2001_cast); + tensor var_2008_cast = matmul(transpose_x = var_2008_transpose_x_0, transpose_y = var_2008_transpose_y_0, x = var_2006_cast, y = transpose_187); + tensor var_2009 = const()[name = tensor("op_2009"), val = tensor([0, 2, 1, 3])]; + tensor concat_17 = const()[name = tensor("concat_17"), val = tensor([1, 1500, 1280])]; + tensor transpose_184 = transpose(perm = var_2009, x = var_2008_cast); + tensor x_215_cast = reshape(shape = concat_17, x = transpose_184); + tensor var_2014_to_fp16 = const()[name = tensor("op_2014_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(693184384)))]; + tensor var_2015_to_fp16 = const()[name = tensor("op_2015_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696461248)))]; + tensor var_2016_cast = linear(bias = var_2015_to_fp16, weight = var_2014_to_fp16, x = x_215_cast); + tensor x_217_cast = add(x = x_211_cast, y = var_2016_cast); + tensor var_2022_axes_0 = const()[name = tensor("op_2022_axes_0"), val = tensor([-1])]; + tensor blocks_17_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_17_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696463872)))]; + tensor blocks_17_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_17_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696466496)))]; + tensor var_2022_cast = layer_norm(axes = var_2022_axes_0, beta = blocks_17_mlp_ln_bias_to_fp16, epsilon = var_1947_to_fp16, gamma = blocks_17_mlp_ln_weight_to_fp16, x = x_217_cast); + tensor var_2031_to_fp16 = const()[name = tensor("op_2031_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(696469120)))]; + tensor var_2032_to_fp16 = const()[name = tensor("op_2032_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(709576384)))]; + tensor input_145_cast = linear(bias = var_2032_to_fp16, weight = var_2031_to_fp16, x = var_2022_cast); + tensor x_221_mode_0 = const()[name = tensor("x_221_mode_0"), val = tensor("EXACT")]; + tensor x_221_cast = gelu(mode = x_221_mode_0, x = input_145_cast); + tensor var_2037_to_fp16 = const()[name = tensor("op_2037_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(709586688)))]; + tensor var_2038_to_fp16 = const()[name = tensor("op_2038_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(722693952)))]; + tensor var_2039_cast = linear(bias = var_2038_to_fp16, weight = var_2037_to_fp16, x = x_221_cast); + tensor x_223_cast = add(x = x_217_cast, y = var_2039_cast); + tensor var_2048 = const()[name = tensor("op_2048"), val = tensor(-1)]; + tensor var_2065_axes_0 = const()[name = tensor("op_2065_axes_0"), val = tensor([-1])]; + tensor blocks_18_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_18_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(722696576)))]; + tensor blocks_18_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_18_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(722699200)))]; + tensor var_2054_to_fp16 = const()[name = tensor("op_2054_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2065_cast = layer_norm(axes = var_2065_axes_0, beta = blocks_18_attn_ln_bias_to_fp16, epsilon = var_2054_to_fp16, gamma = blocks_18_attn_ln_weight_to_fp16, x = x_223_cast); + tensor var_2076_to_fp16 = const()[name = tensor("op_2076_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(722701824)))]; + tensor var_2077_to_fp16 = const()[name = tensor("op_2077_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(725978688)))]; + tensor q_73_cast = linear(bias = var_2077_to_fp16, weight = var_2076_to_fp16, x = var_2065_cast); + tensor var_2080_to_fp16 = const()[name = tensor("op_2080_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(725981312)))]; + tensor k_73_bias_0_to_fp16 = const()[name = tensor("k_73_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(729258176)))]; + tensor k_73_cast = linear(bias = k_73_bias_0_to_fp16, weight = var_2080_to_fp16, x = var_2065_cast); + tensor var_2084_to_fp16 = const()[name = tensor("op_2084_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(729260800)))]; + tensor var_2085_to_fp16 = const()[name = tensor("op_2085_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(732537664)))]; + tensor v_73_cast = linear(bias = var_2085_to_fp16, weight = var_2084_to_fp16, x = var_2065_cast); + tensor var_2093 = const()[name = tensor("op_2093"), val = tensor([1, 1500, 20, -1])]; + tensor var_2094_cast = reshape(shape = var_2093, x = q_73_cast); + tensor const_260_to_fp16 = const()[name = tensor("const_260_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_75_cast = mul(x = var_2094_cast, y = const_260_to_fp16); + tensor var_2100 = const()[name = tensor("op_2100"), val = tensor([1, 1500, 20, -1])]; + tensor var_2101_cast = reshape(shape = var_2100, x = k_73_cast); + tensor const_261_to_fp16 = const()[name = tensor("const_261_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_75_cast = mul(x = var_2101_cast, y = const_261_to_fp16); + tensor var_2107 = const()[name = tensor("op_2107"), val = tensor([1, 1500, 20, -1])]; + tensor var_2108_cast = reshape(shape = var_2107, x = v_73_cast); + tensor var_2109 = const()[name = tensor("op_2109"), val = tensor([0, 2, 1, 3])]; + tensor qk_37_transpose_x_0 = const()[name = tensor("qk_37_transpose_x_0"), val = tensor(false)]; + tensor qk_37_transpose_y_0 = const()[name = tensor("qk_37_transpose_y_0"), val = tensor(false)]; + tensor transpose_100_perm_0 = const()[name = tensor("transpose_100_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_101_perm_0 = const()[name = tensor("transpose_101_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_181 = transpose(perm = transpose_101_perm_0, x = k_75_cast); + tensor transpose_182 = transpose(perm = transpose_100_perm_0, x = q_75_cast); + tensor qk_37_cast = matmul(transpose_x = qk_37_transpose_x_0, transpose_y = qk_37_transpose_y_0, x = transpose_182, y = transpose_181); + tensor var_2113_cast = softmax(axis = var_2048, x = qk_37_cast); + tensor var_2115_transpose_x_0 = const()[name = tensor("op_2115_transpose_x_0"), val = tensor(false)]; + tensor var_2115_transpose_y_0 = const()[name = tensor("op_2115_transpose_y_0"), val = tensor(false)]; + tensor transpose_183 = transpose(perm = var_2109, x = var_2108_cast); + tensor var_2115_cast = matmul(transpose_x = var_2115_transpose_x_0, transpose_y = var_2115_transpose_y_0, x = var_2113_cast, y = transpose_183); + tensor var_2116 = const()[name = tensor("op_2116"), val = tensor([0, 2, 1, 3])]; + tensor concat_18 = const()[name = tensor("concat_18"), val = tensor([1, 1500, 1280])]; + tensor transpose_180 = transpose(perm = var_2116, x = var_2115_cast); + tensor x_227_cast = reshape(shape = concat_18, x = transpose_180); + tensor var_2121_to_fp16 = const()[name = tensor("op_2121_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(732540288)))]; + tensor var_2122_to_fp16 = const()[name = tensor("op_2122_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(735817152)))]; + tensor var_2123_cast = linear(bias = var_2122_to_fp16, weight = var_2121_to_fp16, x = x_227_cast); + tensor x_229_cast = add(x = x_223_cast, y = var_2123_cast); + tensor var_2129_axes_0 = const()[name = tensor("op_2129_axes_0"), val = tensor([-1])]; + tensor blocks_18_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_18_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(735819776)))]; + tensor blocks_18_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_18_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(735822400)))]; + tensor var_2129_cast = layer_norm(axes = var_2129_axes_0, beta = blocks_18_mlp_ln_bias_to_fp16, epsilon = var_2054_to_fp16, gamma = blocks_18_mlp_ln_weight_to_fp16, x = x_229_cast); + tensor var_2138_to_fp16 = const()[name = tensor("op_2138_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(735825024)))]; + tensor var_2139_to_fp16 = const()[name = tensor("op_2139_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(748932288)))]; + tensor input_153_cast = linear(bias = var_2139_to_fp16, weight = var_2138_to_fp16, x = var_2129_cast); + tensor x_233_mode_0 = const()[name = tensor("x_233_mode_0"), val = tensor("EXACT")]; + tensor x_233_cast = gelu(mode = x_233_mode_0, x = input_153_cast); + tensor var_2144_to_fp16 = const()[name = tensor("op_2144_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(748942592)))]; + tensor var_2145_to_fp16 = const()[name = tensor("op_2145_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762049856)))]; + tensor var_2146_cast = linear(bias = var_2145_to_fp16, weight = var_2144_to_fp16, x = x_233_cast); + tensor x_235_cast = add(x = x_229_cast, y = var_2146_cast); + tensor var_2155 = const()[name = tensor("op_2155"), val = tensor(-1)]; + tensor var_2172_axes_0 = const()[name = tensor("op_2172_axes_0"), val = tensor([-1])]; + tensor blocks_19_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_19_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762052480)))]; + tensor blocks_19_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_19_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762055104)))]; + tensor var_2161_to_fp16 = const()[name = tensor("op_2161_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2172_cast = layer_norm(axes = var_2172_axes_0, beta = blocks_19_attn_ln_bias_to_fp16, epsilon = var_2161_to_fp16, gamma = blocks_19_attn_ln_weight_to_fp16, x = x_235_cast); + tensor var_2183_to_fp16 = const()[name = tensor("op_2183_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(762057728)))]; + tensor var_2184_to_fp16 = const()[name = tensor("op_2184_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(765334592)))]; + tensor q_77_cast = linear(bias = var_2184_to_fp16, weight = var_2183_to_fp16, x = var_2172_cast); + tensor var_2187_to_fp16 = const()[name = tensor("op_2187_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(765337216)))]; + tensor k_77_bias_0_to_fp16 = const()[name = tensor("k_77_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(768614080)))]; + tensor k_77_cast = linear(bias = k_77_bias_0_to_fp16, weight = var_2187_to_fp16, x = var_2172_cast); + tensor var_2191_to_fp16 = const()[name = tensor("op_2191_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(768616704)))]; + tensor var_2192_to_fp16 = const()[name = tensor("op_2192_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(771893568)))]; + tensor v_77_cast = linear(bias = var_2192_to_fp16, weight = var_2191_to_fp16, x = var_2172_cast); + tensor var_2200 = const()[name = tensor("op_2200"), val = tensor([1, 1500, 20, -1])]; + tensor var_2201_cast = reshape(shape = var_2200, x = q_77_cast); + tensor const_262_to_fp16 = const()[name = tensor("const_262_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_79_cast = mul(x = var_2201_cast, y = const_262_to_fp16); + tensor var_2207 = const()[name = tensor("op_2207"), val = tensor([1, 1500, 20, -1])]; + tensor var_2208_cast = reshape(shape = var_2207, x = k_77_cast); + tensor const_263_to_fp16 = const()[name = tensor("const_263_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_79_cast = mul(x = var_2208_cast, y = const_263_to_fp16); + tensor var_2214 = const()[name = tensor("op_2214"), val = tensor([1, 1500, 20, -1])]; + tensor var_2215_cast = reshape(shape = var_2214, x = v_77_cast); + tensor var_2216 = const()[name = tensor("op_2216"), val = tensor([0, 2, 1, 3])]; + tensor qk_39_transpose_x_0 = const()[name = tensor("qk_39_transpose_x_0"), val = tensor(false)]; + tensor qk_39_transpose_y_0 = const()[name = tensor("qk_39_transpose_y_0"), val = tensor(false)]; + tensor transpose_102_perm_0 = const()[name = tensor("transpose_102_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_103_perm_0 = const()[name = tensor("transpose_103_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_177 = transpose(perm = transpose_103_perm_0, x = k_79_cast); + tensor transpose_178 = transpose(perm = transpose_102_perm_0, x = q_79_cast); + tensor qk_39_cast = matmul(transpose_x = qk_39_transpose_x_0, transpose_y = qk_39_transpose_y_0, x = transpose_178, y = transpose_177); + tensor var_2220_cast = softmax(axis = var_2155, x = qk_39_cast); + tensor var_2222_transpose_x_0 = const()[name = tensor("op_2222_transpose_x_0"), val = tensor(false)]; + tensor var_2222_transpose_y_0 = const()[name = tensor("op_2222_transpose_y_0"), val = tensor(false)]; + tensor transpose_179 = transpose(perm = var_2216, x = var_2215_cast); + tensor var_2222_cast = matmul(transpose_x = var_2222_transpose_x_0, transpose_y = var_2222_transpose_y_0, x = var_2220_cast, y = transpose_179); + tensor var_2223 = const()[name = tensor("op_2223"), val = tensor([0, 2, 1, 3])]; + tensor concat_19 = const()[name = tensor("concat_19"), val = tensor([1, 1500, 1280])]; + tensor transpose_176 = transpose(perm = var_2223, x = var_2222_cast); + tensor x_239_cast = reshape(shape = concat_19, x = transpose_176); + tensor var_2228_to_fp16 = const()[name = tensor("op_2228_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(771896192)))]; + tensor var_2229_to_fp16 = const()[name = tensor("op_2229_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775173056)))]; + tensor var_2230_cast = linear(bias = var_2229_to_fp16, weight = var_2228_to_fp16, x = x_239_cast); + tensor x_241_cast = add(x = x_235_cast, y = var_2230_cast); + tensor var_2236_axes_0 = const()[name = tensor("op_2236_axes_0"), val = tensor([-1])]; + tensor blocks_19_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_19_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775175680)))]; + tensor blocks_19_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_19_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775178304)))]; + tensor var_2236_cast = layer_norm(axes = var_2236_axes_0, beta = blocks_19_mlp_ln_bias_to_fp16, epsilon = var_2161_to_fp16, gamma = blocks_19_mlp_ln_weight_to_fp16, x = x_241_cast); + tensor var_2245_to_fp16 = const()[name = tensor("op_2245_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(775180928)))]; + tensor var_2246_to_fp16 = const()[name = tensor("op_2246_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(788288192)))]; + tensor input_161_cast = linear(bias = var_2246_to_fp16, weight = var_2245_to_fp16, x = var_2236_cast); + tensor x_245_mode_0 = const()[name = tensor("x_245_mode_0"), val = tensor("EXACT")]; + tensor x_245_cast = gelu(mode = x_245_mode_0, x = input_161_cast); + tensor var_2251_to_fp16 = const()[name = tensor("op_2251_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(788298496)))]; + tensor var_2252_to_fp16 = const()[name = tensor("op_2252_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(801405760)))]; + tensor var_2253_cast = linear(bias = var_2252_to_fp16, weight = var_2251_to_fp16, x = x_245_cast); + tensor x_247_cast = add(x = x_241_cast, y = var_2253_cast); + tensor var_2262 = const()[name = tensor("op_2262"), val = tensor(-1)]; + tensor var_2279_axes_0 = const()[name = tensor("op_2279_axes_0"), val = tensor([-1])]; + tensor blocks_20_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_20_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(801408384)))]; + tensor blocks_20_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_20_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(801411008)))]; + tensor var_2268_to_fp16 = const()[name = tensor("op_2268_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2279_cast = layer_norm(axes = var_2279_axes_0, beta = blocks_20_attn_ln_bias_to_fp16, epsilon = var_2268_to_fp16, gamma = blocks_20_attn_ln_weight_to_fp16, x = x_247_cast); + tensor var_2290_to_fp16 = const()[name = tensor("op_2290_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(801413632)))]; + tensor var_2291_to_fp16 = const()[name = tensor("op_2291_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(804690496)))]; + tensor q_81_cast = linear(bias = var_2291_to_fp16, weight = var_2290_to_fp16, x = var_2279_cast); + tensor var_2294_to_fp16 = const()[name = tensor("op_2294_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(804693120)))]; + tensor k_81_bias_0_to_fp16 = const()[name = tensor("k_81_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(807969984)))]; + tensor k_81_cast = linear(bias = k_81_bias_0_to_fp16, weight = var_2294_to_fp16, x = var_2279_cast); + tensor var_2298_to_fp16 = const()[name = tensor("op_2298_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(807972608)))]; + tensor var_2299_to_fp16 = const()[name = tensor("op_2299_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(811249472)))]; + tensor v_81_cast = linear(bias = var_2299_to_fp16, weight = var_2298_to_fp16, x = var_2279_cast); + tensor var_2307 = const()[name = tensor("op_2307"), val = tensor([1, 1500, 20, -1])]; + tensor var_2308_cast = reshape(shape = var_2307, x = q_81_cast); + tensor const_264_to_fp16 = const()[name = tensor("const_264_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_83_cast = mul(x = var_2308_cast, y = const_264_to_fp16); + tensor var_2314 = const()[name = tensor("op_2314"), val = tensor([1, 1500, 20, -1])]; + tensor var_2315_cast = reshape(shape = var_2314, x = k_81_cast); + tensor const_265_to_fp16 = const()[name = tensor("const_265_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_83_cast = mul(x = var_2315_cast, y = const_265_to_fp16); + tensor var_2321 = const()[name = tensor("op_2321"), val = tensor([1, 1500, 20, -1])]; + tensor var_2322_cast = reshape(shape = var_2321, x = v_81_cast); + tensor var_2323 = const()[name = tensor("op_2323"), val = tensor([0, 2, 1, 3])]; + tensor qk_41_transpose_x_0 = const()[name = tensor("qk_41_transpose_x_0"), val = tensor(false)]; + tensor qk_41_transpose_y_0 = const()[name = tensor("qk_41_transpose_y_0"), val = tensor(false)]; + tensor transpose_104_perm_0 = const()[name = tensor("transpose_104_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_105_perm_0 = const()[name = tensor("transpose_105_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_173 = transpose(perm = transpose_105_perm_0, x = k_83_cast); + tensor transpose_174 = transpose(perm = transpose_104_perm_0, x = q_83_cast); + tensor qk_41_cast = matmul(transpose_x = qk_41_transpose_x_0, transpose_y = qk_41_transpose_y_0, x = transpose_174, y = transpose_173); + tensor var_2327_cast = softmax(axis = var_2262, x = qk_41_cast); + tensor var_2329_transpose_x_0 = const()[name = tensor("op_2329_transpose_x_0"), val = tensor(false)]; + tensor var_2329_transpose_y_0 = const()[name = tensor("op_2329_transpose_y_0"), val = tensor(false)]; + tensor transpose_175 = transpose(perm = var_2323, x = var_2322_cast); + tensor var_2329_cast = matmul(transpose_x = var_2329_transpose_x_0, transpose_y = var_2329_transpose_y_0, x = var_2327_cast, y = transpose_175); + tensor var_2330 = const()[name = tensor("op_2330"), val = tensor([0, 2, 1, 3])]; + tensor concat_20 = const()[name = tensor("concat_20"), val = tensor([1, 1500, 1280])]; + tensor transpose_172 = transpose(perm = var_2330, x = var_2329_cast); + tensor x_251_cast = reshape(shape = concat_20, x = transpose_172); + tensor var_2335_to_fp16 = const()[name = tensor("op_2335_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(811252096)))]; + tensor var_2336_to_fp16 = const()[name = tensor("op_2336_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814528960)))]; + tensor var_2337_cast = linear(bias = var_2336_to_fp16, weight = var_2335_to_fp16, x = x_251_cast); + tensor x_253_cast = add(x = x_247_cast, y = var_2337_cast); + tensor var_2343_axes_0 = const()[name = tensor("op_2343_axes_0"), val = tensor([-1])]; + tensor blocks_20_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_20_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814531584)))]; + tensor blocks_20_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_20_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814534208)))]; + tensor var_2343_cast = layer_norm(axes = var_2343_axes_0, beta = blocks_20_mlp_ln_bias_to_fp16, epsilon = var_2268_to_fp16, gamma = blocks_20_mlp_ln_weight_to_fp16, x = x_253_cast); + tensor var_2352_to_fp16 = const()[name = tensor("op_2352_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(814536832)))]; + tensor var_2353_to_fp16 = const()[name = tensor("op_2353_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(827644096)))]; + tensor input_169_cast = linear(bias = var_2353_to_fp16, weight = var_2352_to_fp16, x = var_2343_cast); + tensor x_257_mode_0 = const()[name = tensor("x_257_mode_0"), val = tensor("EXACT")]; + tensor x_257_cast = gelu(mode = x_257_mode_0, x = input_169_cast); + tensor var_2358_to_fp16 = const()[name = tensor("op_2358_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(827654400)))]; + tensor var_2359_to_fp16 = const()[name = tensor("op_2359_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(840761664)))]; + tensor var_2360_cast = linear(bias = var_2359_to_fp16, weight = var_2358_to_fp16, x = x_257_cast); + tensor x_259_cast = add(x = x_253_cast, y = var_2360_cast); + tensor var_2369 = const()[name = tensor("op_2369"), val = tensor(-1)]; + tensor var_2386_axes_0 = const()[name = tensor("op_2386_axes_0"), val = tensor([-1])]; + tensor blocks_21_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_21_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(840764288)))]; + tensor blocks_21_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_21_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(840766912)))]; + tensor var_2375_to_fp16 = const()[name = tensor("op_2375_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2386_cast = layer_norm(axes = var_2386_axes_0, beta = blocks_21_attn_ln_bias_to_fp16, epsilon = var_2375_to_fp16, gamma = blocks_21_attn_ln_weight_to_fp16, x = x_259_cast); + tensor var_2397_to_fp16 = const()[name = tensor("op_2397_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(840769536)))]; + tensor var_2398_to_fp16 = const()[name = tensor("op_2398_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(844046400)))]; + tensor q_85_cast = linear(bias = var_2398_to_fp16, weight = var_2397_to_fp16, x = var_2386_cast); + tensor var_2401_to_fp16 = const()[name = tensor("op_2401_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(844049024)))]; + tensor k_85_bias_0_to_fp16 = const()[name = tensor("k_85_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(847325888)))]; + tensor k_85_cast = linear(bias = k_85_bias_0_to_fp16, weight = var_2401_to_fp16, x = var_2386_cast); + tensor var_2405_to_fp16 = const()[name = tensor("op_2405_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(847328512)))]; + tensor var_2406_to_fp16 = const()[name = tensor("op_2406_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(850605376)))]; + tensor v_85_cast = linear(bias = var_2406_to_fp16, weight = var_2405_to_fp16, x = var_2386_cast); + tensor var_2414 = const()[name = tensor("op_2414"), val = tensor([1, 1500, 20, -1])]; + tensor var_2415_cast = reshape(shape = var_2414, x = q_85_cast); + tensor const_266_to_fp16 = const()[name = tensor("const_266_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_87_cast = mul(x = var_2415_cast, y = const_266_to_fp16); + tensor var_2421 = const()[name = tensor("op_2421"), val = tensor([1, 1500, 20, -1])]; + tensor var_2422_cast = reshape(shape = var_2421, x = k_85_cast); + tensor const_267_to_fp16 = const()[name = tensor("const_267_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_87_cast = mul(x = var_2422_cast, y = const_267_to_fp16); + tensor var_2428 = const()[name = tensor("op_2428"), val = tensor([1, 1500, 20, -1])]; + tensor var_2429_cast = reshape(shape = var_2428, x = v_85_cast); + tensor var_2430 = const()[name = tensor("op_2430"), val = tensor([0, 2, 1, 3])]; + tensor qk_43_transpose_x_0 = const()[name = tensor("qk_43_transpose_x_0"), val = tensor(false)]; + tensor qk_43_transpose_y_0 = const()[name = tensor("qk_43_transpose_y_0"), val = tensor(false)]; + tensor transpose_106_perm_0 = const()[name = tensor("transpose_106_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_107_perm_0 = const()[name = tensor("transpose_107_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_169 = transpose(perm = transpose_107_perm_0, x = k_87_cast); + tensor transpose_170 = transpose(perm = transpose_106_perm_0, x = q_87_cast); + tensor qk_43_cast = matmul(transpose_x = qk_43_transpose_x_0, transpose_y = qk_43_transpose_y_0, x = transpose_170, y = transpose_169); + tensor var_2434_cast = softmax(axis = var_2369, x = qk_43_cast); + tensor var_2436_transpose_x_0 = const()[name = tensor("op_2436_transpose_x_0"), val = tensor(false)]; + tensor var_2436_transpose_y_0 = const()[name = tensor("op_2436_transpose_y_0"), val = tensor(false)]; + tensor transpose_171 = transpose(perm = var_2430, x = var_2429_cast); + tensor var_2436_cast = matmul(transpose_x = var_2436_transpose_x_0, transpose_y = var_2436_transpose_y_0, x = var_2434_cast, y = transpose_171); + tensor var_2437 = const()[name = tensor("op_2437"), val = tensor([0, 2, 1, 3])]; + tensor concat_21 = const()[name = tensor("concat_21"), val = tensor([1, 1500, 1280])]; + tensor transpose_168 = transpose(perm = var_2437, x = var_2436_cast); + tensor x_263_cast = reshape(shape = concat_21, x = transpose_168); + tensor var_2442_to_fp16 = const()[name = tensor("op_2442_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(850608000)))]; + tensor var_2443_to_fp16 = const()[name = tensor("op_2443_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(853884864)))]; + tensor var_2444_cast = linear(bias = var_2443_to_fp16, weight = var_2442_to_fp16, x = x_263_cast); + tensor x_265_cast = add(x = x_259_cast, y = var_2444_cast); + tensor var_2450_axes_0 = const()[name = tensor("op_2450_axes_0"), val = tensor([-1])]; + tensor blocks_21_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_21_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(853887488)))]; + tensor blocks_21_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_21_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(853890112)))]; + tensor var_2450_cast = layer_norm(axes = var_2450_axes_0, beta = blocks_21_mlp_ln_bias_to_fp16, epsilon = var_2375_to_fp16, gamma = blocks_21_mlp_ln_weight_to_fp16, x = x_265_cast); + tensor var_2459_to_fp16 = const()[name = tensor("op_2459_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(853892736)))]; + tensor var_2460_to_fp16 = const()[name = tensor("op_2460_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(867000000)))]; + tensor input_177_cast = linear(bias = var_2460_to_fp16, weight = var_2459_to_fp16, x = var_2450_cast); + tensor x_269_mode_0 = const()[name = tensor("x_269_mode_0"), val = tensor("EXACT")]; + tensor x_269_cast = gelu(mode = x_269_mode_0, x = input_177_cast); + tensor var_2465_to_fp16 = const()[name = tensor("op_2465_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(867010304)))]; + tensor var_2466_to_fp16 = const()[name = tensor("op_2466_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880117568)))]; + tensor var_2467_cast = linear(bias = var_2466_to_fp16, weight = var_2465_to_fp16, x = x_269_cast); + tensor x_271_cast = add(x = x_265_cast, y = var_2467_cast); + tensor var_2476 = const()[name = tensor("op_2476"), val = tensor(-1)]; + tensor var_2493_axes_0 = const()[name = tensor("op_2493_axes_0"), val = tensor([-1])]; + tensor blocks_22_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_22_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880120192)))]; + tensor blocks_22_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_22_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880122816)))]; + tensor var_2482_to_fp16 = const()[name = tensor("op_2482_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2493_cast = layer_norm(axes = var_2493_axes_0, beta = blocks_22_attn_ln_bias_to_fp16, epsilon = var_2482_to_fp16, gamma = blocks_22_attn_ln_weight_to_fp16, x = x_271_cast); + tensor var_2504_to_fp16 = const()[name = tensor("op_2504_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(880125440)))]; + tensor var_2505_to_fp16 = const()[name = tensor("op_2505_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(883402304)))]; + tensor q_89_cast = linear(bias = var_2505_to_fp16, weight = var_2504_to_fp16, x = var_2493_cast); + tensor var_2508_to_fp16 = const()[name = tensor("op_2508_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(883404928)))]; + tensor k_89_bias_0_to_fp16 = const()[name = tensor("k_89_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(886681792)))]; + tensor k_89_cast = linear(bias = k_89_bias_0_to_fp16, weight = var_2508_to_fp16, x = var_2493_cast); + tensor var_2512_to_fp16 = const()[name = tensor("op_2512_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(886684416)))]; + tensor var_2513_to_fp16 = const()[name = tensor("op_2513_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(889961280)))]; + tensor v_89_cast = linear(bias = var_2513_to_fp16, weight = var_2512_to_fp16, x = var_2493_cast); + tensor var_2521 = const()[name = tensor("op_2521"), val = tensor([1, 1500, 20, -1])]; + tensor var_2522_cast = reshape(shape = var_2521, x = q_89_cast); + tensor const_268_to_fp16 = const()[name = tensor("const_268_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_91_cast = mul(x = var_2522_cast, y = const_268_to_fp16); + tensor var_2528 = const()[name = tensor("op_2528"), val = tensor([1, 1500, 20, -1])]; + tensor var_2529_cast = reshape(shape = var_2528, x = k_89_cast); + tensor const_269_to_fp16 = const()[name = tensor("const_269_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_91_cast = mul(x = var_2529_cast, y = const_269_to_fp16); + tensor var_2535 = const()[name = tensor("op_2535"), val = tensor([1, 1500, 20, -1])]; + tensor var_2536_cast = reshape(shape = var_2535, x = v_89_cast); + tensor var_2537 = const()[name = tensor("op_2537"), val = tensor([0, 2, 1, 3])]; + tensor qk_45_transpose_x_0 = const()[name = tensor("qk_45_transpose_x_0"), val = tensor(false)]; + tensor qk_45_transpose_y_0 = const()[name = tensor("qk_45_transpose_y_0"), val = tensor(false)]; + tensor transpose_108_perm_0 = const()[name = tensor("transpose_108_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_109_perm_0 = const()[name = tensor("transpose_109_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_165 = transpose(perm = transpose_109_perm_0, x = k_91_cast); + tensor transpose_166 = transpose(perm = transpose_108_perm_0, x = q_91_cast); + tensor qk_45_cast = matmul(transpose_x = qk_45_transpose_x_0, transpose_y = qk_45_transpose_y_0, x = transpose_166, y = transpose_165); + tensor var_2541_cast = softmax(axis = var_2476, x = qk_45_cast); + tensor var_2543_transpose_x_0 = const()[name = tensor("op_2543_transpose_x_0"), val = tensor(false)]; + tensor var_2543_transpose_y_0 = const()[name = tensor("op_2543_transpose_y_0"), val = tensor(false)]; + tensor transpose_167 = transpose(perm = var_2537, x = var_2536_cast); + tensor var_2543_cast = matmul(transpose_x = var_2543_transpose_x_0, transpose_y = var_2543_transpose_y_0, x = var_2541_cast, y = transpose_167); + tensor var_2544 = const()[name = tensor("op_2544"), val = tensor([0, 2, 1, 3])]; + tensor concat_22 = const()[name = tensor("concat_22"), val = tensor([1, 1500, 1280])]; + tensor transpose_164 = transpose(perm = var_2544, x = var_2543_cast); + tensor x_275_cast = reshape(shape = concat_22, x = transpose_164); + tensor var_2549_to_fp16 = const()[name = tensor("op_2549_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(889963904)))]; + tensor var_2550_to_fp16 = const()[name = tensor("op_2550_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893240768)))]; + tensor var_2551_cast = linear(bias = var_2550_to_fp16, weight = var_2549_to_fp16, x = x_275_cast); + tensor x_277_cast = add(x = x_271_cast, y = var_2551_cast); + tensor var_2557_axes_0 = const()[name = tensor("op_2557_axes_0"), val = tensor([-1])]; + tensor blocks_22_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_22_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893243392)))]; + tensor blocks_22_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_22_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893246016)))]; + tensor var_2557_cast = layer_norm(axes = var_2557_axes_0, beta = blocks_22_mlp_ln_bias_to_fp16, epsilon = var_2482_to_fp16, gamma = blocks_22_mlp_ln_weight_to_fp16, x = x_277_cast); + tensor var_2566_to_fp16 = const()[name = tensor("op_2566_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(893248640)))]; + tensor var_2567_to_fp16 = const()[name = tensor("op_2567_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(906355904)))]; + tensor input_185_cast = linear(bias = var_2567_to_fp16, weight = var_2566_to_fp16, x = var_2557_cast); + tensor x_281_mode_0 = const()[name = tensor("x_281_mode_0"), val = tensor("EXACT")]; + tensor x_281_cast = gelu(mode = x_281_mode_0, x = input_185_cast); + tensor var_2572_to_fp16 = const()[name = tensor("op_2572_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(906366208)))]; + tensor var_2573_to_fp16 = const()[name = tensor("op_2573_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(919473472)))]; + tensor var_2574_cast = linear(bias = var_2573_to_fp16, weight = var_2572_to_fp16, x = x_281_cast); + tensor x_283_cast = add(x = x_277_cast, y = var_2574_cast); + tensor var_2583 = const()[name = tensor("op_2583"), val = tensor(-1)]; + tensor var_2600_axes_0 = const()[name = tensor("op_2600_axes_0"), val = tensor([-1])]; + tensor blocks_23_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_23_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(919476096)))]; + tensor blocks_23_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_23_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(919478720)))]; + tensor var_2589_to_fp16 = const()[name = tensor("op_2589_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2600_cast = layer_norm(axes = var_2600_axes_0, beta = blocks_23_attn_ln_bias_to_fp16, epsilon = var_2589_to_fp16, gamma = blocks_23_attn_ln_weight_to_fp16, x = x_283_cast); + tensor var_2611_to_fp16 = const()[name = tensor("op_2611_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(919481344)))]; + tensor var_2612_to_fp16 = const()[name = tensor("op_2612_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(922758208)))]; + tensor q_93_cast = linear(bias = var_2612_to_fp16, weight = var_2611_to_fp16, x = var_2600_cast); + tensor var_2615_to_fp16 = const()[name = tensor("op_2615_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(922760832)))]; + tensor k_93_bias_0_to_fp16 = const()[name = tensor("k_93_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(926037696)))]; + tensor k_93_cast = linear(bias = k_93_bias_0_to_fp16, weight = var_2615_to_fp16, x = var_2600_cast); + tensor var_2619_to_fp16 = const()[name = tensor("op_2619_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(926040320)))]; + tensor var_2620_to_fp16 = const()[name = tensor("op_2620_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(929317184)))]; + tensor v_93_cast = linear(bias = var_2620_to_fp16, weight = var_2619_to_fp16, x = var_2600_cast); + tensor var_2628 = const()[name = tensor("op_2628"), val = tensor([1, 1500, 20, -1])]; + tensor var_2629_cast = reshape(shape = var_2628, x = q_93_cast); + tensor const_270_to_fp16 = const()[name = tensor("const_270_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_95_cast = mul(x = var_2629_cast, y = const_270_to_fp16); + tensor var_2635 = const()[name = tensor("op_2635"), val = tensor([1, 1500, 20, -1])]; + tensor var_2636_cast = reshape(shape = var_2635, x = k_93_cast); + tensor const_271_to_fp16 = const()[name = tensor("const_271_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_95_cast = mul(x = var_2636_cast, y = const_271_to_fp16); + tensor var_2642 = const()[name = tensor("op_2642"), val = tensor([1, 1500, 20, -1])]; + tensor var_2643_cast = reshape(shape = var_2642, x = v_93_cast); + tensor var_2644 = const()[name = tensor("op_2644"), val = tensor([0, 2, 1, 3])]; + tensor qk_47_transpose_x_0 = const()[name = tensor("qk_47_transpose_x_0"), val = tensor(false)]; + tensor qk_47_transpose_y_0 = const()[name = tensor("qk_47_transpose_y_0"), val = tensor(false)]; + tensor transpose_110_perm_0 = const()[name = tensor("transpose_110_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_111_perm_0 = const()[name = tensor("transpose_111_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_161 = transpose(perm = transpose_111_perm_0, x = k_95_cast); + tensor transpose_162 = transpose(perm = transpose_110_perm_0, x = q_95_cast); + tensor qk_47_cast = matmul(transpose_x = qk_47_transpose_x_0, transpose_y = qk_47_transpose_y_0, x = transpose_162, y = transpose_161); + tensor var_2648_cast = softmax(axis = var_2583, x = qk_47_cast); + tensor var_2650_transpose_x_0 = const()[name = tensor("op_2650_transpose_x_0"), val = tensor(false)]; + tensor var_2650_transpose_y_0 = const()[name = tensor("op_2650_transpose_y_0"), val = tensor(false)]; + tensor transpose_163 = transpose(perm = var_2644, x = var_2643_cast); + tensor var_2650_cast = matmul(transpose_x = var_2650_transpose_x_0, transpose_y = var_2650_transpose_y_0, x = var_2648_cast, y = transpose_163); + tensor var_2651 = const()[name = tensor("op_2651"), val = tensor([0, 2, 1, 3])]; + tensor concat_23 = const()[name = tensor("concat_23"), val = tensor([1, 1500, 1280])]; + tensor transpose_160 = transpose(perm = var_2651, x = var_2650_cast); + tensor x_287_cast = reshape(shape = concat_23, x = transpose_160); + tensor var_2656_to_fp16 = const()[name = tensor("op_2656_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(929319808)))]; + tensor var_2657_to_fp16 = const()[name = tensor("op_2657_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(932596672)))]; + tensor var_2658_cast = linear(bias = var_2657_to_fp16, weight = var_2656_to_fp16, x = x_287_cast); + tensor x_289_cast = add(x = x_283_cast, y = var_2658_cast); + tensor var_2664_axes_0 = const()[name = tensor("op_2664_axes_0"), val = tensor([-1])]; + tensor blocks_23_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_23_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(932599296)))]; + tensor blocks_23_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_23_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(932601920)))]; + tensor var_2664_cast = layer_norm(axes = var_2664_axes_0, beta = blocks_23_mlp_ln_bias_to_fp16, epsilon = var_2589_to_fp16, gamma = blocks_23_mlp_ln_weight_to_fp16, x = x_289_cast); + tensor var_2673_to_fp16 = const()[name = tensor("op_2673_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(932604544)))]; + tensor var_2674_to_fp16 = const()[name = tensor("op_2674_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(945711808)))]; + tensor input_193_cast = linear(bias = var_2674_to_fp16, weight = var_2673_to_fp16, x = var_2664_cast); + tensor x_293_mode_0 = const()[name = tensor("x_293_mode_0"), val = tensor("EXACT")]; + tensor x_293_cast = gelu(mode = x_293_mode_0, x = input_193_cast); + tensor var_2679_to_fp16 = const()[name = tensor("op_2679_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(945722112)))]; + tensor var_2680_to_fp16 = const()[name = tensor("op_2680_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(958829376)))]; + tensor var_2681_cast = linear(bias = var_2680_to_fp16, weight = var_2679_to_fp16, x = x_293_cast); + tensor x_295_cast = add(x = x_289_cast, y = var_2681_cast); + tensor var_2690 = const()[name = tensor("op_2690"), val = tensor(-1)]; + tensor var_2707_axes_0 = const()[name = tensor("op_2707_axes_0"), val = tensor([-1])]; + tensor blocks_24_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_24_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(958832000)))]; + tensor blocks_24_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_24_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(958834624)))]; + tensor var_2696_to_fp16 = const()[name = tensor("op_2696_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2707_cast = layer_norm(axes = var_2707_axes_0, beta = blocks_24_attn_ln_bias_to_fp16, epsilon = var_2696_to_fp16, gamma = blocks_24_attn_ln_weight_to_fp16, x = x_295_cast); + tensor var_2718_to_fp16 = const()[name = tensor("op_2718_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(958837248)))]; + tensor var_2719_to_fp16 = const()[name = tensor("op_2719_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(962114112)))]; + tensor q_97_cast = linear(bias = var_2719_to_fp16, weight = var_2718_to_fp16, x = var_2707_cast); + tensor var_2722_to_fp16 = const()[name = tensor("op_2722_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(962116736)))]; + tensor k_97_bias_0_to_fp16 = const()[name = tensor("k_97_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(965393600)))]; + tensor k_97_cast = linear(bias = k_97_bias_0_to_fp16, weight = var_2722_to_fp16, x = var_2707_cast); + tensor var_2726_to_fp16 = const()[name = tensor("op_2726_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(965396224)))]; + tensor var_2727_to_fp16 = const()[name = tensor("op_2727_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(968673088)))]; + tensor v_97_cast = linear(bias = var_2727_to_fp16, weight = var_2726_to_fp16, x = var_2707_cast); + tensor var_2735 = const()[name = tensor("op_2735"), val = tensor([1, 1500, 20, -1])]; + tensor var_2736_cast = reshape(shape = var_2735, x = q_97_cast); + tensor const_272_to_fp16 = const()[name = tensor("const_272_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_99_cast = mul(x = var_2736_cast, y = const_272_to_fp16); + tensor var_2742 = const()[name = tensor("op_2742"), val = tensor([1, 1500, 20, -1])]; + tensor var_2743_cast = reshape(shape = var_2742, x = k_97_cast); + tensor const_273_to_fp16 = const()[name = tensor("const_273_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_99_cast = mul(x = var_2743_cast, y = const_273_to_fp16); + tensor var_2749 = const()[name = tensor("op_2749"), val = tensor([1, 1500, 20, -1])]; + tensor var_2750_cast = reshape(shape = var_2749, x = v_97_cast); + tensor var_2751 = const()[name = tensor("op_2751"), val = tensor([0, 2, 1, 3])]; + tensor qk_49_transpose_x_0 = const()[name = tensor("qk_49_transpose_x_0"), val = tensor(false)]; + tensor qk_49_transpose_y_0 = const()[name = tensor("qk_49_transpose_y_0"), val = tensor(false)]; + tensor transpose_112_perm_0 = const()[name = tensor("transpose_112_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_113_perm_0 = const()[name = tensor("transpose_113_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_157 = transpose(perm = transpose_113_perm_0, x = k_99_cast); + tensor transpose_158 = transpose(perm = transpose_112_perm_0, x = q_99_cast); + tensor qk_49_cast = matmul(transpose_x = qk_49_transpose_x_0, transpose_y = qk_49_transpose_y_0, x = transpose_158, y = transpose_157); + tensor var_2755_cast = softmax(axis = var_2690, x = qk_49_cast); + tensor var_2757_transpose_x_0 = const()[name = tensor("op_2757_transpose_x_0"), val = tensor(false)]; + tensor var_2757_transpose_y_0 = const()[name = tensor("op_2757_transpose_y_0"), val = tensor(false)]; + tensor transpose_159 = transpose(perm = var_2751, x = var_2750_cast); + tensor var_2757_cast = matmul(transpose_x = var_2757_transpose_x_0, transpose_y = var_2757_transpose_y_0, x = var_2755_cast, y = transpose_159); + tensor var_2758 = const()[name = tensor("op_2758"), val = tensor([0, 2, 1, 3])]; + tensor concat_24 = const()[name = tensor("concat_24"), val = tensor([1, 1500, 1280])]; + tensor transpose_156 = transpose(perm = var_2758, x = var_2757_cast); + tensor x_299_cast = reshape(shape = concat_24, x = transpose_156); + tensor var_2763_to_fp16 = const()[name = tensor("op_2763_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(968675712)))]; + tensor var_2764_to_fp16 = const()[name = tensor("op_2764_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(971952576)))]; + tensor var_2765_cast = linear(bias = var_2764_to_fp16, weight = var_2763_to_fp16, x = x_299_cast); + tensor x_301_cast = add(x = x_295_cast, y = var_2765_cast); + tensor var_2771_axes_0 = const()[name = tensor("op_2771_axes_0"), val = tensor([-1])]; + tensor blocks_24_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_24_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(971955200)))]; + tensor blocks_24_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_24_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(971957824)))]; + tensor var_2771_cast = layer_norm(axes = var_2771_axes_0, beta = blocks_24_mlp_ln_bias_to_fp16, epsilon = var_2696_to_fp16, gamma = blocks_24_mlp_ln_weight_to_fp16, x = x_301_cast); + tensor var_2780_to_fp16 = const()[name = tensor("op_2780_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(971960448)))]; + tensor var_2781_to_fp16 = const()[name = tensor("op_2781_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(985067712)))]; + tensor input_201_cast = linear(bias = var_2781_to_fp16, weight = var_2780_to_fp16, x = var_2771_cast); + tensor x_305_mode_0 = const()[name = tensor("x_305_mode_0"), val = tensor("EXACT")]; + tensor x_305_cast = gelu(mode = x_305_mode_0, x = input_201_cast); + tensor var_2786_to_fp16 = const()[name = tensor("op_2786_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(985078016)))]; + tensor var_2787_to_fp16 = const()[name = tensor("op_2787_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(998185280)))]; + tensor var_2788_cast = linear(bias = var_2787_to_fp16, weight = var_2786_to_fp16, x = x_305_cast); + tensor x_307_cast = add(x = x_301_cast, y = var_2788_cast); + tensor var_2797 = const()[name = tensor("op_2797"), val = tensor(-1)]; + tensor var_2814_axes_0 = const()[name = tensor("op_2814_axes_0"), val = tensor([-1])]; + tensor blocks_25_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_25_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(998187904)))]; + tensor blocks_25_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_25_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(998190528)))]; + tensor var_2803_to_fp16 = const()[name = tensor("op_2803_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2814_cast = layer_norm(axes = var_2814_axes_0, beta = blocks_25_attn_ln_bias_to_fp16, epsilon = var_2803_to_fp16, gamma = blocks_25_attn_ln_weight_to_fp16, x = x_307_cast); + tensor var_2825_to_fp16 = const()[name = tensor("op_2825_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(998193152)))]; + tensor var_2826_to_fp16 = const()[name = tensor("op_2826_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1001470016)))]; + tensor q_101_cast = linear(bias = var_2826_to_fp16, weight = var_2825_to_fp16, x = var_2814_cast); + tensor var_2829_to_fp16 = const()[name = tensor("op_2829_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1001472640)))]; + tensor k_101_bias_0_to_fp16 = const()[name = tensor("k_101_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1004749504)))]; + tensor k_101_cast = linear(bias = k_101_bias_0_to_fp16, weight = var_2829_to_fp16, x = var_2814_cast); + tensor var_2833_to_fp16 = const()[name = tensor("op_2833_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1004752128)))]; + tensor var_2834_to_fp16 = const()[name = tensor("op_2834_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1008028992)))]; + tensor v_101_cast = linear(bias = var_2834_to_fp16, weight = var_2833_to_fp16, x = var_2814_cast); + tensor var_2842 = const()[name = tensor("op_2842"), val = tensor([1, 1500, 20, -1])]; + tensor var_2843_cast = reshape(shape = var_2842, x = q_101_cast); + tensor const_274_to_fp16 = const()[name = tensor("const_274_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_103_cast = mul(x = var_2843_cast, y = const_274_to_fp16); + tensor var_2849 = const()[name = tensor("op_2849"), val = tensor([1, 1500, 20, -1])]; + tensor var_2850_cast = reshape(shape = var_2849, x = k_101_cast); + tensor const_275_to_fp16 = const()[name = tensor("const_275_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_103_cast = mul(x = var_2850_cast, y = const_275_to_fp16); + tensor var_2856 = const()[name = tensor("op_2856"), val = tensor([1, 1500, 20, -1])]; + tensor var_2857_cast = reshape(shape = var_2856, x = v_101_cast); + tensor var_2858 = const()[name = tensor("op_2858"), val = tensor([0, 2, 1, 3])]; + tensor qk_51_transpose_x_0 = const()[name = tensor("qk_51_transpose_x_0"), val = tensor(false)]; + tensor qk_51_transpose_y_0 = const()[name = tensor("qk_51_transpose_y_0"), val = tensor(false)]; + tensor transpose_114_perm_0 = const()[name = tensor("transpose_114_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_115_perm_0 = const()[name = tensor("transpose_115_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_153 = transpose(perm = transpose_115_perm_0, x = k_103_cast); + tensor transpose_154 = transpose(perm = transpose_114_perm_0, x = q_103_cast); + tensor qk_51_cast = matmul(transpose_x = qk_51_transpose_x_0, transpose_y = qk_51_transpose_y_0, x = transpose_154, y = transpose_153); + tensor var_2862_cast = softmax(axis = var_2797, x = qk_51_cast); + tensor var_2864_transpose_x_0 = const()[name = tensor("op_2864_transpose_x_0"), val = tensor(false)]; + tensor var_2864_transpose_y_0 = const()[name = tensor("op_2864_transpose_y_0"), val = tensor(false)]; + tensor transpose_155 = transpose(perm = var_2858, x = var_2857_cast); + tensor var_2864_cast = matmul(transpose_x = var_2864_transpose_x_0, transpose_y = var_2864_transpose_y_0, x = var_2862_cast, y = transpose_155); + tensor var_2865 = const()[name = tensor("op_2865"), val = tensor([0, 2, 1, 3])]; + tensor concat_25 = const()[name = tensor("concat_25"), val = tensor([1, 1500, 1280])]; + tensor transpose_152 = transpose(perm = var_2865, x = var_2864_cast); + tensor x_311_cast = reshape(shape = concat_25, x = transpose_152); + tensor var_2870_to_fp16 = const()[name = tensor("op_2870_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1008031616)))]; + tensor var_2871_to_fp16 = const()[name = tensor("op_2871_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1011308480)))]; + tensor var_2872_cast = linear(bias = var_2871_to_fp16, weight = var_2870_to_fp16, x = x_311_cast); + tensor x_313_cast = add(x = x_307_cast, y = var_2872_cast); + tensor var_2878_axes_0 = const()[name = tensor("op_2878_axes_0"), val = tensor([-1])]; + tensor blocks_25_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_25_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1011311104)))]; + tensor blocks_25_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_25_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1011313728)))]; + tensor var_2878_cast = layer_norm(axes = var_2878_axes_0, beta = blocks_25_mlp_ln_bias_to_fp16, epsilon = var_2803_to_fp16, gamma = blocks_25_mlp_ln_weight_to_fp16, x = x_313_cast); + tensor var_2887_to_fp16 = const()[name = tensor("op_2887_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1011316352)))]; + tensor var_2888_to_fp16 = const()[name = tensor("op_2888_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1024423616)))]; + tensor input_209_cast = linear(bias = var_2888_to_fp16, weight = var_2887_to_fp16, x = var_2878_cast); + tensor x_317_mode_0 = const()[name = tensor("x_317_mode_0"), val = tensor("EXACT")]; + tensor x_317_cast = gelu(mode = x_317_mode_0, x = input_209_cast); + tensor var_2893_to_fp16 = const()[name = tensor("op_2893_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1024433920)))]; + tensor var_2894_to_fp16 = const()[name = tensor("op_2894_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1037541184)))]; + tensor var_2895_cast = linear(bias = var_2894_to_fp16, weight = var_2893_to_fp16, x = x_317_cast); + tensor x_319_cast = add(x = x_313_cast, y = var_2895_cast); + tensor var_2904 = const()[name = tensor("op_2904"), val = tensor(-1)]; + tensor var_2921_axes_0 = const()[name = tensor("op_2921_axes_0"), val = tensor([-1])]; + tensor blocks_26_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_26_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1037543808)))]; + tensor blocks_26_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_26_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1037546432)))]; + tensor var_2910_to_fp16 = const()[name = tensor("op_2910_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2921_cast = layer_norm(axes = var_2921_axes_0, beta = blocks_26_attn_ln_bias_to_fp16, epsilon = var_2910_to_fp16, gamma = blocks_26_attn_ln_weight_to_fp16, x = x_319_cast); + tensor var_2932_to_fp16 = const()[name = tensor("op_2932_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1037549056)))]; + tensor var_2933_to_fp16 = const()[name = tensor("op_2933_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1040825920)))]; + tensor q_105_cast = linear(bias = var_2933_to_fp16, weight = var_2932_to_fp16, x = var_2921_cast); + tensor var_2936_to_fp16 = const()[name = tensor("op_2936_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1040828544)))]; + tensor k_105_bias_0_to_fp16 = const()[name = tensor("k_105_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1044105408)))]; + tensor k_105_cast = linear(bias = k_105_bias_0_to_fp16, weight = var_2936_to_fp16, x = var_2921_cast); + tensor var_2940_to_fp16 = const()[name = tensor("op_2940_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1044108032)))]; + tensor var_2941_to_fp16 = const()[name = tensor("op_2941_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1047384896)))]; + tensor v_105_cast = linear(bias = var_2941_to_fp16, weight = var_2940_to_fp16, x = var_2921_cast); + tensor var_2949 = const()[name = tensor("op_2949"), val = tensor([1, 1500, 20, -1])]; + tensor var_2950_cast = reshape(shape = var_2949, x = q_105_cast); + tensor const_276_to_fp16 = const()[name = tensor("const_276_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_107_cast = mul(x = var_2950_cast, y = const_276_to_fp16); + tensor var_2956 = const()[name = tensor("op_2956"), val = tensor([1, 1500, 20, -1])]; + tensor var_2957_cast = reshape(shape = var_2956, x = k_105_cast); + tensor const_277_to_fp16 = const()[name = tensor("const_277_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_107_cast = mul(x = var_2957_cast, y = const_277_to_fp16); + tensor var_2963 = const()[name = tensor("op_2963"), val = tensor([1, 1500, 20, -1])]; + tensor var_2964_cast = reshape(shape = var_2963, x = v_105_cast); + tensor var_2965 = const()[name = tensor("op_2965"), val = tensor([0, 2, 1, 3])]; + tensor qk_53_transpose_x_0 = const()[name = tensor("qk_53_transpose_x_0"), val = tensor(false)]; + tensor qk_53_transpose_y_0 = const()[name = tensor("qk_53_transpose_y_0"), val = tensor(false)]; + tensor transpose_116_perm_0 = const()[name = tensor("transpose_116_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_117_perm_0 = const()[name = tensor("transpose_117_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_149 = transpose(perm = transpose_117_perm_0, x = k_107_cast); + tensor transpose_150 = transpose(perm = transpose_116_perm_0, x = q_107_cast); + tensor qk_53_cast = matmul(transpose_x = qk_53_transpose_x_0, transpose_y = qk_53_transpose_y_0, x = transpose_150, y = transpose_149); + tensor var_2969_cast = softmax(axis = var_2904, x = qk_53_cast); + tensor var_2971_transpose_x_0 = const()[name = tensor("op_2971_transpose_x_0"), val = tensor(false)]; + tensor var_2971_transpose_y_0 = const()[name = tensor("op_2971_transpose_y_0"), val = tensor(false)]; + tensor transpose_151 = transpose(perm = var_2965, x = var_2964_cast); + tensor var_2971_cast = matmul(transpose_x = var_2971_transpose_x_0, transpose_y = var_2971_transpose_y_0, x = var_2969_cast, y = transpose_151); + tensor var_2972 = const()[name = tensor("op_2972"), val = tensor([0, 2, 1, 3])]; + tensor concat_26 = const()[name = tensor("concat_26"), val = tensor([1, 1500, 1280])]; + tensor transpose_148 = transpose(perm = var_2972, x = var_2971_cast); + tensor x_323_cast = reshape(shape = concat_26, x = transpose_148); + tensor var_2977_to_fp16 = const()[name = tensor("op_2977_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1047387520)))]; + tensor var_2978_to_fp16 = const()[name = tensor("op_2978_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1050664384)))]; + tensor var_2979_cast = linear(bias = var_2978_to_fp16, weight = var_2977_to_fp16, x = x_323_cast); + tensor x_325_cast = add(x = x_319_cast, y = var_2979_cast); + tensor var_2985_axes_0 = const()[name = tensor("op_2985_axes_0"), val = tensor([-1])]; + tensor blocks_26_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_26_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1050667008)))]; + tensor blocks_26_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_26_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1050669632)))]; + tensor var_2985_cast = layer_norm(axes = var_2985_axes_0, beta = blocks_26_mlp_ln_bias_to_fp16, epsilon = var_2910_to_fp16, gamma = blocks_26_mlp_ln_weight_to_fp16, x = x_325_cast); + tensor var_2994_to_fp16 = const()[name = tensor("op_2994_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1050672256)))]; + tensor var_2995_to_fp16 = const()[name = tensor("op_2995_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1063779520)))]; + tensor input_217_cast = linear(bias = var_2995_to_fp16, weight = var_2994_to_fp16, x = var_2985_cast); + tensor x_329_mode_0 = const()[name = tensor("x_329_mode_0"), val = tensor("EXACT")]; + tensor x_329_cast = gelu(mode = x_329_mode_0, x = input_217_cast); + tensor var_3000_to_fp16 = const()[name = tensor("op_3000_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1063789824)))]; + tensor var_3001_to_fp16 = const()[name = tensor("op_3001_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1076897088)))]; + tensor var_3002_cast = linear(bias = var_3001_to_fp16, weight = var_3000_to_fp16, x = x_329_cast); + tensor x_331_cast = add(x = x_325_cast, y = var_3002_cast); + tensor var_3011 = const()[name = tensor("op_3011"), val = tensor(-1)]; + tensor var_3028_axes_0 = const()[name = tensor("op_3028_axes_0"), val = tensor([-1])]; + tensor blocks_27_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_27_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1076899712)))]; + tensor blocks_27_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_27_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1076902336)))]; + tensor var_3017_to_fp16 = const()[name = tensor("op_3017_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3028_cast = layer_norm(axes = var_3028_axes_0, beta = blocks_27_attn_ln_bias_to_fp16, epsilon = var_3017_to_fp16, gamma = blocks_27_attn_ln_weight_to_fp16, x = x_331_cast); + tensor var_3039_to_fp16 = const()[name = tensor("op_3039_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1076904960)))]; + tensor var_3040_to_fp16 = const()[name = tensor("op_3040_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1080181824)))]; + tensor q_109_cast = linear(bias = var_3040_to_fp16, weight = var_3039_to_fp16, x = var_3028_cast); + tensor var_3043_to_fp16 = const()[name = tensor("op_3043_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1080184448)))]; + tensor k_109_bias_0_to_fp16 = const()[name = tensor("k_109_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1083461312)))]; + tensor k_109_cast = linear(bias = k_109_bias_0_to_fp16, weight = var_3043_to_fp16, x = var_3028_cast); + tensor var_3047_to_fp16 = const()[name = tensor("op_3047_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1083463936)))]; + tensor var_3048_to_fp16 = const()[name = tensor("op_3048_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1086740800)))]; + tensor v_109_cast = linear(bias = var_3048_to_fp16, weight = var_3047_to_fp16, x = var_3028_cast); + tensor var_3056 = const()[name = tensor("op_3056"), val = tensor([1, 1500, 20, -1])]; + tensor var_3057_cast = reshape(shape = var_3056, x = q_109_cast); + tensor const_278_to_fp16 = const()[name = tensor("const_278_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_111_cast = mul(x = var_3057_cast, y = const_278_to_fp16); + tensor var_3063 = const()[name = tensor("op_3063"), val = tensor([1, 1500, 20, -1])]; + tensor var_3064_cast = reshape(shape = var_3063, x = k_109_cast); + tensor const_279_to_fp16 = const()[name = tensor("const_279_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_111_cast = mul(x = var_3064_cast, y = const_279_to_fp16); + tensor var_3070 = const()[name = tensor("op_3070"), val = tensor([1, 1500, 20, -1])]; + tensor var_3071_cast = reshape(shape = var_3070, x = v_109_cast); + tensor var_3072 = const()[name = tensor("op_3072"), val = tensor([0, 2, 1, 3])]; + tensor qk_55_transpose_x_0 = const()[name = tensor("qk_55_transpose_x_0"), val = tensor(false)]; + tensor qk_55_transpose_y_0 = const()[name = tensor("qk_55_transpose_y_0"), val = tensor(false)]; + tensor transpose_118_perm_0 = const()[name = tensor("transpose_118_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_119_perm_0 = const()[name = tensor("transpose_119_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_145 = transpose(perm = transpose_119_perm_0, x = k_111_cast); + tensor transpose_146 = transpose(perm = transpose_118_perm_0, x = q_111_cast); + tensor qk_55_cast = matmul(transpose_x = qk_55_transpose_x_0, transpose_y = qk_55_transpose_y_0, x = transpose_146, y = transpose_145); + tensor var_3076_cast = softmax(axis = var_3011, x = qk_55_cast); + tensor var_3078_transpose_x_0 = const()[name = tensor("op_3078_transpose_x_0"), val = tensor(false)]; + tensor var_3078_transpose_y_0 = const()[name = tensor("op_3078_transpose_y_0"), val = tensor(false)]; + tensor transpose_147 = transpose(perm = var_3072, x = var_3071_cast); + tensor var_3078_cast = matmul(transpose_x = var_3078_transpose_x_0, transpose_y = var_3078_transpose_y_0, x = var_3076_cast, y = transpose_147); + tensor var_3079 = const()[name = tensor("op_3079"), val = tensor([0, 2, 1, 3])]; + tensor concat_27 = const()[name = tensor("concat_27"), val = tensor([1, 1500, 1280])]; + tensor transpose_144 = transpose(perm = var_3079, x = var_3078_cast); + tensor x_335_cast = reshape(shape = concat_27, x = transpose_144); + tensor var_3084_to_fp16 = const()[name = tensor("op_3084_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1086743424)))]; + tensor var_3085_to_fp16 = const()[name = tensor("op_3085_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1090020288)))]; + tensor var_3086_cast = linear(bias = var_3085_to_fp16, weight = var_3084_to_fp16, x = x_335_cast); + tensor x_337_cast = add(x = x_331_cast, y = var_3086_cast); + tensor var_3092_axes_0 = const()[name = tensor("op_3092_axes_0"), val = tensor([-1])]; + tensor blocks_27_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_27_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1090022912)))]; + tensor blocks_27_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_27_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1090025536)))]; + tensor var_3092_cast = layer_norm(axes = var_3092_axes_0, beta = blocks_27_mlp_ln_bias_to_fp16, epsilon = var_3017_to_fp16, gamma = blocks_27_mlp_ln_weight_to_fp16, x = x_337_cast); + tensor var_3101_to_fp16 = const()[name = tensor("op_3101_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1090028160)))]; + tensor var_3102_to_fp16 = const()[name = tensor("op_3102_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1103135424)))]; + tensor input_225_cast = linear(bias = var_3102_to_fp16, weight = var_3101_to_fp16, x = var_3092_cast); + tensor x_341_mode_0 = const()[name = tensor("x_341_mode_0"), val = tensor("EXACT")]; + tensor x_341_cast = gelu(mode = x_341_mode_0, x = input_225_cast); + tensor var_3107_to_fp16 = const()[name = tensor("op_3107_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1103145728)))]; + tensor var_3108_to_fp16 = const()[name = tensor("op_3108_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116252992)))]; + tensor var_3109_cast = linear(bias = var_3108_to_fp16, weight = var_3107_to_fp16, x = x_341_cast); + tensor x_343_cast = add(x = x_337_cast, y = var_3109_cast); + tensor var_3118 = const()[name = tensor("op_3118"), val = tensor(-1)]; + tensor var_3135_axes_0 = const()[name = tensor("op_3135_axes_0"), val = tensor([-1])]; + tensor blocks_28_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_28_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116255616)))]; + tensor blocks_28_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_28_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116258240)))]; + tensor var_3124_to_fp16 = const()[name = tensor("op_3124_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3135_cast = layer_norm(axes = var_3135_axes_0, beta = blocks_28_attn_ln_bias_to_fp16, epsilon = var_3124_to_fp16, gamma = blocks_28_attn_ln_weight_to_fp16, x = x_343_cast); + tensor var_3146_to_fp16 = const()[name = tensor("op_3146_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1116260864)))]; + tensor var_3147_to_fp16 = const()[name = tensor("op_3147_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1119537728)))]; + tensor q_113_cast = linear(bias = var_3147_to_fp16, weight = var_3146_to_fp16, x = var_3135_cast); + tensor var_3150_to_fp16 = const()[name = tensor("op_3150_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1119540352)))]; + tensor k_113_bias_0_to_fp16 = const()[name = tensor("k_113_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1122817216)))]; + tensor k_113_cast = linear(bias = k_113_bias_0_to_fp16, weight = var_3150_to_fp16, x = var_3135_cast); + tensor var_3154_to_fp16 = const()[name = tensor("op_3154_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1122819840)))]; + tensor var_3155_to_fp16 = const()[name = tensor("op_3155_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1126096704)))]; + tensor v_113_cast = linear(bias = var_3155_to_fp16, weight = var_3154_to_fp16, x = var_3135_cast); + tensor var_3163 = const()[name = tensor("op_3163"), val = tensor([1, 1500, 20, -1])]; + tensor var_3164_cast = reshape(shape = var_3163, x = q_113_cast); + tensor const_280_to_fp16 = const()[name = tensor("const_280_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_115_cast = mul(x = var_3164_cast, y = const_280_to_fp16); + tensor var_3170 = const()[name = tensor("op_3170"), val = tensor([1, 1500, 20, -1])]; + tensor var_3171_cast = reshape(shape = var_3170, x = k_113_cast); + tensor const_281_to_fp16 = const()[name = tensor("const_281_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_115_cast = mul(x = var_3171_cast, y = const_281_to_fp16); + tensor var_3177 = const()[name = tensor("op_3177"), val = tensor([1, 1500, 20, -1])]; + tensor var_3178_cast = reshape(shape = var_3177, x = v_113_cast); + tensor var_3179 = const()[name = tensor("op_3179"), val = tensor([0, 2, 1, 3])]; + tensor qk_57_transpose_x_0 = const()[name = tensor("qk_57_transpose_x_0"), val = tensor(false)]; + tensor qk_57_transpose_y_0 = const()[name = tensor("qk_57_transpose_y_0"), val = tensor(false)]; + tensor transpose_120_perm_0 = const()[name = tensor("transpose_120_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_121_perm_0 = const()[name = tensor("transpose_121_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_141 = transpose(perm = transpose_121_perm_0, x = k_115_cast); + tensor transpose_142 = transpose(perm = transpose_120_perm_0, x = q_115_cast); + tensor qk_57_cast = matmul(transpose_x = qk_57_transpose_x_0, transpose_y = qk_57_transpose_y_0, x = transpose_142, y = transpose_141); + tensor var_3183_cast = softmax(axis = var_3118, x = qk_57_cast); + tensor var_3185_transpose_x_0 = const()[name = tensor("op_3185_transpose_x_0"), val = tensor(false)]; + tensor var_3185_transpose_y_0 = const()[name = tensor("op_3185_transpose_y_0"), val = tensor(false)]; + tensor transpose_143 = transpose(perm = var_3179, x = var_3178_cast); + tensor var_3185_cast = matmul(transpose_x = var_3185_transpose_x_0, transpose_y = var_3185_transpose_y_0, x = var_3183_cast, y = transpose_143); + tensor var_3186 = const()[name = tensor("op_3186"), val = tensor([0, 2, 1, 3])]; + tensor concat_28 = const()[name = tensor("concat_28"), val = tensor([1, 1500, 1280])]; + tensor transpose_140 = transpose(perm = var_3186, x = var_3185_cast); + tensor x_347_cast = reshape(shape = concat_28, x = transpose_140); + tensor var_3191_to_fp16 = const()[name = tensor("op_3191_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1126099328)))]; + tensor var_3192_to_fp16 = const()[name = tensor("op_3192_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1129376192)))]; + tensor var_3193_cast = linear(bias = var_3192_to_fp16, weight = var_3191_to_fp16, x = x_347_cast); + tensor x_349_cast = add(x = x_343_cast, y = var_3193_cast); + tensor var_3199_axes_0 = const()[name = tensor("op_3199_axes_0"), val = tensor([-1])]; + tensor blocks_28_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_28_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1129378816)))]; + tensor blocks_28_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_28_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1129381440)))]; + tensor var_3199_cast = layer_norm(axes = var_3199_axes_0, beta = blocks_28_mlp_ln_bias_to_fp16, epsilon = var_3124_to_fp16, gamma = blocks_28_mlp_ln_weight_to_fp16, x = x_349_cast); + tensor var_3208_to_fp16 = const()[name = tensor("op_3208_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1129384064)))]; + tensor var_3209_to_fp16 = const()[name = tensor("op_3209_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1142491328)))]; + tensor input_233_cast = linear(bias = var_3209_to_fp16, weight = var_3208_to_fp16, x = var_3199_cast); + tensor x_353_mode_0 = const()[name = tensor("x_353_mode_0"), val = tensor("EXACT")]; + tensor x_353_cast = gelu(mode = x_353_mode_0, x = input_233_cast); + tensor var_3214_to_fp16 = const()[name = tensor("op_3214_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1142501632)))]; + tensor var_3215_to_fp16 = const()[name = tensor("op_3215_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1155608896)))]; + tensor var_3216_cast = linear(bias = var_3215_to_fp16, weight = var_3214_to_fp16, x = x_353_cast); + tensor x_355_cast = add(x = x_349_cast, y = var_3216_cast); + tensor var_3225 = const()[name = tensor("op_3225"), val = tensor(-1)]; + tensor var_3242_axes_0 = const()[name = tensor("op_3242_axes_0"), val = tensor([-1])]; + tensor blocks_29_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_29_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1155611520)))]; + tensor blocks_29_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_29_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1155614144)))]; + tensor var_3231_to_fp16 = const()[name = tensor("op_3231_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3242_cast = layer_norm(axes = var_3242_axes_0, beta = blocks_29_attn_ln_bias_to_fp16, epsilon = var_3231_to_fp16, gamma = blocks_29_attn_ln_weight_to_fp16, x = x_355_cast); + tensor var_3253_to_fp16 = const()[name = tensor("op_3253_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1155616768)))]; + tensor var_3254_to_fp16 = const()[name = tensor("op_3254_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1158893632)))]; + tensor q_117_cast = linear(bias = var_3254_to_fp16, weight = var_3253_to_fp16, x = var_3242_cast); + tensor var_3257_to_fp16 = const()[name = tensor("op_3257_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1158896256)))]; + tensor k_117_bias_0_to_fp16 = const()[name = tensor("k_117_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1162173120)))]; + tensor k_117_cast = linear(bias = k_117_bias_0_to_fp16, weight = var_3257_to_fp16, x = var_3242_cast); + tensor var_3261_to_fp16 = const()[name = tensor("op_3261_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1162175744)))]; + tensor var_3262_to_fp16 = const()[name = tensor("op_3262_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1165452608)))]; + tensor v_117_cast = linear(bias = var_3262_to_fp16, weight = var_3261_to_fp16, x = var_3242_cast); + tensor var_3270 = const()[name = tensor("op_3270"), val = tensor([1, 1500, 20, -1])]; + tensor var_3271_cast = reshape(shape = var_3270, x = q_117_cast); + tensor const_282_to_fp16 = const()[name = tensor("const_282_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_119_cast = mul(x = var_3271_cast, y = const_282_to_fp16); + tensor var_3277 = const()[name = tensor("op_3277"), val = tensor([1, 1500, 20, -1])]; + tensor var_3278_cast = reshape(shape = var_3277, x = k_117_cast); + tensor const_283_to_fp16 = const()[name = tensor("const_283_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_119_cast = mul(x = var_3278_cast, y = const_283_to_fp16); + tensor var_3284 = const()[name = tensor("op_3284"), val = tensor([1, 1500, 20, -1])]; + tensor var_3285_cast = reshape(shape = var_3284, x = v_117_cast); + tensor var_3286 = const()[name = tensor("op_3286"), val = tensor([0, 2, 1, 3])]; + tensor qk_59_transpose_x_0 = const()[name = tensor("qk_59_transpose_x_0"), val = tensor(false)]; + tensor qk_59_transpose_y_0 = const()[name = tensor("qk_59_transpose_y_0"), val = tensor(false)]; + tensor transpose_122_perm_0 = const()[name = tensor("transpose_122_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_123_perm_0 = const()[name = tensor("transpose_123_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_137 = transpose(perm = transpose_123_perm_0, x = k_119_cast); + tensor transpose_138 = transpose(perm = transpose_122_perm_0, x = q_119_cast); + tensor qk_59_cast = matmul(transpose_x = qk_59_transpose_x_0, transpose_y = qk_59_transpose_y_0, x = transpose_138, y = transpose_137); + tensor var_3290_cast = softmax(axis = var_3225, x = qk_59_cast); + tensor var_3292_transpose_x_0 = const()[name = tensor("op_3292_transpose_x_0"), val = tensor(false)]; + tensor var_3292_transpose_y_0 = const()[name = tensor("op_3292_transpose_y_0"), val = tensor(false)]; + tensor transpose_139 = transpose(perm = var_3286, x = var_3285_cast); + tensor var_3292_cast = matmul(transpose_x = var_3292_transpose_x_0, transpose_y = var_3292_transpose_y_0, x = var_3290_cast, y = transpose_139); + tensor var_3293 = const()[name = tensor("op_3293"), val = tensor([0, 2, 1, 3])]; + tensor concat_29 = const()[name = tensor("concat_29"), val = tensor([1, 1500, 1280])]; + tensor transpose_136 = transpose(perm = var_3293, x = var_3292_cast); + tensor x_359_cast = reshape(shape = concat_29, x = transpose_136); + tensor var_3298_to_fp16 = const()[name = tensor("op_3298_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1165455232)))]; + tensor var_3299_to_fp16 = const()[name = tensor("op_3299_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1168732096)))]; + tensor var_3300_cast = linear(bias = var_3299_to_fp16, weight = var_3298_to_fp16, x = x_359_cast); + tensor x_361_cast = add(x = x_355_cast, y = var_3300_cast); + tensor var_3306_axes_0 = const()[name = tensor("op_3306_axes_0"), val = tensor([-1])]; + tensor blocks_29_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_29_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1168734720)))]; + tensor blocks_29_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_29_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1168737344)))]; + tensor var_3306_cast = layer_norm(axes = var_3306_axes_0, beta = blocks_29_mlp_ln_bias_to_fp16, epsilon = var_3231_to_fp16, gamma = blocks_29_mlp_ln_weight_to_fp16, x = x_361_cast); + tensor var_3315_to_fp16 = const()[name = tensor("op_3315_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1168739968)))]; + tensor var_3316_to_fp16 = const()[name = tensor("op_3316_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1181847232)))]; + tensor input_241_cast = linear(bias = var_3316_to_fp16, weight = var_3315_to_fp16, x = var_3306_cast); + tensor x_365_mode_0 = const()[name = tensor("x_365_mode_0"), val = tensor("EXACT")]; + tensor x_365_cast = gelu(mode = x_365_mode_0, x = input_241_cast); + tensor var_3321_to_fp16 = const()[name = tensor("op_3321_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1181857536)))]; + tensor var_3322_to_fp16 = const()[name = tensor("op_3322_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1194964800)))]; + tensor var_3323_cast = linear(bias = var_3322_to_fp16, weight = var_3321_to_fp16, x = x_365_cast); + tensor x_367_cast = add(x = x_361_cast, y = var_3323_cast); + tensor var_3332 = const()[name = tensor("op_3332"), val = tensor(-1)]; + tensor var_3349_axes_0 = const()[name = tensor("op_3349_axes_0"), val = tensor([-1])]; + tensor blocks_30_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_30_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1194967424)))]; + tensor blocks_30_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_30_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1194970048)))]; + tensor var_3338_to_fp16 = const()[name = tensor("op_3338_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3349_cast = layer_norm(axes = var_3349_axes_0, beta = blocks_30_attn_ln_bias_to_fp16, epsilon = var_3338_to_fp16, gamma = blocks_30_attn_ln_weight_to_fp16, x = x_367_cast); + tensor var_3360_to_fp16 = const()[name = tensor("op_3360_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1194972672)))]; + tensor var_3361_to_fp16 = const()[name = tensor("op_3361_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1198249536)))]; + tensor q_121_cast = linear(bias = var_3361_to_fp16, weight = var_3360_to_fp16, x = var_3349_cast); + tensor var_3364_to_fp16 = const()[name = tensor("op_3364_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1198252160)))]; + tensor k_121_bias_0_to_fp16 = const()[name = tensor("k_121_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1201529024)))]; + tensor k_121_cast = linear(bias = k_121_bias_0_to_fp16, weight = var_3364_to_fp16, x = var_3349_cast); + tensor var_3368_to_fp16 = const()[name = tensor("op_3368_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1201531648)))]; + tensor var_3369_to_fp16 = const()[name = tensor("op_3369_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1204808512)))]; + tensor v_121_cast = linear(bias = var_3369_to_fp16, weight = var_3368_to_fp16, x = var_3349_cast); + tensor var_3377 = const()[name = tensor("op_3377"), val = tensor([1, 1500, 20, -1])]; + tensor var_3378_cast = reshape(shape = var_3377, x = q_121_cast); + tensor const_284_to_fp16 = const()[name = tensor("const_284_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_123_cast = mul(x = var_3378_cast, y = const_284_to_fp16); + tensor var_3384 = const()[name = tensor("op_3384"), val = tensor([1, 1500, 20, -1])]; + tensor var_3385_cast = reshape(shape = var_3384, x = k_121_cast); + tensor const_285_to_fp16 = const()[name = tensor("const_285_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_123_cast = mul(x = var_3385_cast, y = const_285_to_fp16); + tensor var_3391 = const()[name = tensor("op_3391"), val = tensor([1, 1500, 20, -1])]; + tensor var_3392_cast = reshape(shape = var_3391, x = v_121_cast); + tensor var_3393 = const()[name = tensor("op_3393"), val = tensor([0, 2, 1, 3])]; + tensor qk_61_transpose_x_0 = const()[name = tensor("qk_61_transpose_x_0"), val = tensor(false)]; + tensor qk_61_transpose_y_0 = const()[name = tensor("qk_61_transpose_y_0"), val = tensor(false)]; + tensor transpose_124_perm_0 = const()[name = tensor("transpose_124_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_125_perm_0 = const()[name = tensor("transpose_125_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_133 = transpose(perm = transpose_125_perm_0, x = k_123_cast); + tensor transpose_134 = transpose(perm = transpose_124_perm_0, x = q_123_cast); + tensor qk_61_cast = matmul(transpose_x = qk_61_transpose_x_0, transpose_y = qk_61_transpose_y_0, x = transpose_134, y = transpose_133); + tensor var_3397_cast = softmax(axis = var_3332, x = qk_61_cast); + tensor var_3399_transpose_x_0 = const()[name = tensor("op_3399_transpose_x_0"), val = tensor(false)]; + tensor var_3399_transpose_y_0 = const()[name = tensor("op_3399_transpose_y_0"), val = tensor(false)]; + tensor transpose_135 = transpose(perm = var_3393, x = var_3392_cast); + tensor var_3399_cast = matmul(transpose_x = var_3399_transpose_x_0, transpose_y = var_3399_transpose_y_0, x = var_3397_cast, y = transpose_135); + tensor var_3400 = const()[name = tensor("op_3400"), val = tensor([0, 2, 1, 3])]; + tensor concat_30 = const()[name = tensor("concat_30"), val = tensor([1, 1500, 1280])]; + tensor transpose_132 = transpose(perm = var_3400, x = var_3399_cast); + tensor x_371_cast = reshape(shape = concat_30, x = transpose_132); + tensor var_3405_to_fp16 = const()[name = tensor("op_3405_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1204811136)))]; + tensor var_3406_to_fp16 = const()[name = tensor("op_3406_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208088000)))]; + tensor var_3407_cast = linear(bias = var_3406_to_fp16, weight = var_3405_to_fp16, x = x_371_cast); + tensor x_373_cast = add(x = x_367_cast, y = var_3407_cast); + tensor var_3413_axes_0 = const()[name = tensor("op_3413_axes_0"), val = tensor([-1])]; + tensor blocks_30_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_30_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208090624)))]; + tensor blocks_30_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_30_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208093248)))]; + tensor var_3413_cast = layer_norm(axes = var_3413_axes_0, beta = blocks_30_mlp_ln_bias_to_fp16, epsilon = var_3338_to_fp16, gamma = blocks_30_mlp_ln_weight_to_fp16, x = x_373_cast); + tensor var_3422_to_fp16 = const()[name = tensor("op_3422_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1208095872)))]; + tensor var_3423_to_fp16 = const()[name = tensor("op_3423_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1221203136)))]; + tensor input_249_cast = linear(bias = var_3423_to_fp16, weight = var_3422_to_fp16, x = var_3413_cast); + tensor x_377_mode_0 = const()[name = tensor("x_377_mode_0"), val = tensor("EXACT")]; + tensor x_377_cast = gelu(mode = x_377_mode_0, x = input_249_cast); + tensor var_3428_to_fp16 = const()[name = tensor("op_3428_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1221213440)))]; + tensor var_3429_to_fp16 = const()[name = tensor("op_3429_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1234320704)))]; + tensor var_3430_cast = linear(bias = var_3429_to_fp16, weight = var_3428_to_fp16, x = x_377_cast); + tensor x_379_cast = add(x = x_373_cast, y = var_3430_cast); + tensor var_3439 = const()[name = tensor("op_3439"), val = tensor(-1)]; + tensor var_3456_axes_0 = const()[name = tensor("op_3456_axes_0"), val = tensor([-1])]; + tensor blocks_31_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_31_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1234323328)))]; + tensor blocks_31_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_31_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1234325952)))]; + tensor var_3445_to_fp16 = const()[name = tensor("op_3445_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3456_cast = layer_norm(axes = var_3456_axes_0, beta = blocks_31_attn_ln_bias_to_fp16, epsilon = var_3445_to_fp16, gamma = blocks_31_attn_ln_weight_to_fp16, x = x_379_cast); + tensor var_3467_to_fp16 = const()[name = tensor("op_3467_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1234328576)))]; + tensor var_3468_to_fp16 = const()[name = tensor("op_3468_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1237605440)))]; + tensor q_125_cast = linear(bias = var_3468_to_fp16, weight = var_3467_to_fp16, x = var_3456_cast); + tensor var_3471_to_fp16 = const()[name = tensor("op_3471_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1237608064)))]; + tensor k_125_bias_0_to_fp16 = const()[name = tensor("k_125_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1240884928)))]; + tensor k_125_cast = linear(bias = k_125_bias_0_to_fp16, weight = var_3471_to_fp16, x = var_3456_cast); + tensor var_3475_to_fp16 = const()[name = tensor("op_3475_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1240887552)))]; + tensor var_3476_to_fp16 = const()[name = tensor("op_3476_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1244164416)))]; + tensor v_125_cast = linear(bias = var_3476_to_fp16, weight = var_3475_to_fp16, x = var_3456_cast); + tensor var_3484 = const()[name = tensor("op_3484"), val = tensor([1, 1500, 20, -1])]; + tensor var_3485_cast = reshape(shape = var_3484, x = q_125_cast); + tensor const_286_to_fp16 = const()[name = tensor("const_286_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_cast = mul(x = var_3485_cast, y = const_286_to_fp16); + tensor var_3491 = const()[name = tensor("op_3491"), val = tensor([1, 1500, 20, -1])]; + tensor var_3492_cast = reshape(shape = var_3491, x = k_125_cast); + tensor const_287_to_fp16 = const()[name = tensor("const_287_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_cast = mul(x = var_3492_cast, y = const_287_to_fp16); + tensor var_3498 = const()[name = tensor("op_3498"), val = tensor([1, 1500, 20, -1])]; + tensor var_3499_cast = reshape(shape = var_3498, x = v_125_cast); + tensor var_3500 = const()[name = tensor("op_3500"), val = tensor([0, 2, 1, 3])]; + tensor qk_transpose_x_0 = const()[name = tensor("qk_transpose_x_0"), val = tensor(false)]; + tensor qk_transpose_y_0 = const()[name = tensor("qk_transpose_y_0"), val = tensor(false)]; + tensor transpose_126_perm_0 = const()[name = tensor("transpose_126_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_127_perm_0 = const()[name = tensor("transpose_127_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_129 = transpose(perm = transpose_127_perm_0, x = k_cast); + tensor transpose_130 = transpose(perm = transpose_126_perm_0, x = q_cast); + tensor qk_cast = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_130, y = transpose_129); + tensor var_3504_cast = softmax(axis = var_3439, x = qk_cast); + tensor var_3506_transpose_x_0 = const()[name = tensor("op_3506_transpose_x_0"), val = tensor(false)]; + tensor var_3506_transpose_y_0 = const()[name = tensor("op_3506_transpose_y_0"), val = tensor(false)]; + tensor transpose_131 = transpose(perm = var_3500, x = var_3499_cast); + tensor var_3506_cast = matmul(transpose_x = var_3506_transpose_x_0, transpose_y = var_3506_transpose_y_0, x = var_3504_cast, y = transpose_131); + tensor var_3507 = const()[name = tensor("op_3507"), val = tensor([0, 2, 1, 3])]; + tensor concat_31 = const()[name = tensor("concat_31"), val = tensor([1, 1500, 1280])]; + tensor transpose_128 = transpose(perm = var_3507, x = var_3506_cast); + tensor x_383_cast = reshape(shape = concat_31, x = transpose_128); + tensor var_3512_to_fp16 = const()[name = tensor("op_3512_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1244167040)))]; + tensor var_3513_to_fp16 = const()[name = tensor("op_3513_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1247443904)))]; + tensor var_3514_cast = linear(bias = var_3513_to_fp16, weight = var_3512_to_fp16, x = x_383_cast); + tensor x_385_cast = add(x = x_379_cast, y = var_3514_cast); + tensor var_3520_axes_0 = const()[name = tensor("op_3520_axes_0"), val = tensor([-1])]; + tensor blocks_31_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_31_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1247446528)))]; + tensor blocks_31_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_31_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1247449152)))]; + tensor var_3520_cast = layer_norm(axes = var_3520_axes_0, beta = blocks_31_mlp_ln_bias_to_fp16, epsilon = var_3445_to_fp16, gamma = blocks_31_mlp_ln_weight_to_fp16, x = x_385_cast); + tensor var_3529_to_fp16 = const()[name = tensor("op_3529_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1247451776)))]; + tensor var_3530_to_fp16 = const()[name = tensor("op_3530_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1260559040)))]; + tensor input_257_cast = linear(bias = var_3530_to_fp16, weight = var_3529_to_fp16, x = var_3520_cast); + tensor x_389_mode_0 = const()[name = tensor("x_389_mode_0"), val = tensor("EXACT")]; + tensor x_389_cast = gelu(mode = x_389_mode_0, x = input_257_cast); + tensor var_3535_to_fp16 = const()[name = tensor("op_3535_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1260569344)))]; + tensor var_3536_to_fp16 = const()[name = tensor("op_3536_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1273676608)))]; + tensor var_3537_cast = linear(bias = var_3536_to_fp16, weight = var_3535_to_fp16, x = x_389_cast); + tensor x_cast = add(x = x_385_cast, y = var_3537_cast); + tensor var_3550_axes_0 = const()[name = tensor("op_3550_axes_0"), val = tensor([-1])]; + tensor ln_post_weight_to_fp16 = const()[name = tensor("ln_post_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1273679232)))]; + tensor ln_post_bias_to_fp16 = const()[name = tensor("ln_post_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1273681856)))]; + tensor var_3541_to_fp16 = const()[name = tensor("op_3541_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_3550_cast = layer_norm(axes = var_3550_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_3541_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast); + tensor var_3550_cast_to_fp32_dtype_0 = const()[name = tensor("op_3550_cast_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor output = cast(dtype = var_3550_cast_to_fp32_dtype_0, x = var_3550_cast); + } -> (output); +} \ No newline at end of file diff --git a/ggml-large-v3-encoder.mlmodelc/weights/weight.bin b/ggml-large-v3-encoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..05ddd420672b6a09550eba68c4b9ad064066e67e --- /dev/null +++ b/ggml-large-v3-encoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e3c79ca8b3de455956bb7e2af31f525413431950048a54b2e4c72c39d8c5acc7 +size 1273684480 diff --git a/ggml-large.bin b/ggml-large.bin deleted file mode 100644 index 649aafd67e30021d0140c24342ee2ffb947f4bde..0000000000000000000000000000000000000000 --- a/ggml-large.bin +++ /dev/null @@ -1,3 +0,0 @@ -version https://git-lfs.github.com/spec/v1 -oid sha256:9a423fe4d40c82774b6af34115b8b935f34152246eb19e80e376071d3f999487 -size 3094623691 diff --git a/ggml-medium-encoder.mlmodelc.zip b/ggml-medium-encoder.mlmodelc.zip index a5e970d86b7f2a7478ba0a0f5356c95b0e6353ab..ca1b0f280326b415f1ac667472bfa3c99796ef47 100644 --- a/ggml-medium-encoder.mlmodelc.zip +++ b/ggml-medium-encoder.mlmodelc.zip @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:79b0b8d436d47d3f24dd3afc91f19447dd686a4f37521b2f6d9c30a642133fbd +oid sha256:991295c77a7f73bd04b9310e8ca2aa279826acb7c643f7973cd27fa892487377 size 567829413 diff --git a/ggml-medium-encoder.mlmodelc/analytics/coremldata.bin b/ggml-medium-encoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..cfebf363e7217ce13431621718bc293f474504b9 --- /dev/null +++ b/ggml-medium-encoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:adbe456375e7eb3407732a426ecb65bbda86860e4aa801f3a696b70b8a533cdd +size 207 diff --git a/ggml-medium-encoder.mlmodelc/coremldata.bin b/ggml-medium-encoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..c732ed7b8caeac3c15acda1c2836f3e1f525a5a8 --- /dev/null +++ b/ggml-medium-encoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:05fe28591b40616fa0c34ad7b853133623f5300923ec812acb11459c411acf3b +size 149 diff --git a/ggml-medium-encoder.mlmodelc/metadata.json b/ggml-medium-encoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..5e8a17765c9a9adbc7ce77736778897c9543219c --- /dev/null +++ b/ggml-medium-encoder.mlmodelc/metadata.json @@ -0,0 +1,64 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float32", + "formattedType" : "MultiArray (Float32)", + "shortDescription" : "", + "shape" : "[]", + "name" : "output", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 6, + "mlProgramOperationTypeHistogram" : { + "Linear" : 144, + "Matmul" : 48, + "Cast" : 2, + "Conv" : 2, + "Softmax" : 24, + "Add" : 49, + "LayerNorm" : 49, + "Mul" : 48, + "Transpose" : 97, + "Gelu" : 26, + "Reshape" : 96 + }, + "computePrecision" : "Mixed (Float16, Float32, Int32)", + "isUpdatable" : "0", + "availability" : { + "macOS" : "12.0", + "tvOS" : "15.0", + "watchOS" : "8.0", + "iOS" : "15.0", + "macCatalyst" : "15.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float32", + "formattedType" : "MultiArray (Float32 1 × 80 × 3000)", + "shortDescription" : "", + "shape" : "[1, 80, 3000]", + "name" : "logmel_data", + "type" : "MultiArray" + } + ], + "generatedClassName" : "coreml_encoder_medium", + "method" : "predict" + } +] \ No newline at end of file diff --git a/ggml-medium-encoder.mlmodelc/model.mil b/ggml-medium-encoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..4771e10615202abab5bfc7b155ba0872539682da --- /dev/null +++ b/ggml-medium-encoder.mlmodelc/model.mil @@ -0,0 +1,1455 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "4.28.4"}, {"coremlc-version", "1436.100.10"}})] +{ + func main(tensor logmel_data) { + tensor var_56 = const()[name = tensor("op_56"), val = tensor(1)]; + tensor var_64 = const()[name = tensor("op_64"), val = tensor([1])]; + tensor var_66 = const()[name = tensor("op_66"), val = tensor([1])]; + tensor var_68_pad_type_0 = const()[name = tensor("op_68_pad_type_0"), val = tensor("custom")]; + tensor var_68_pad_0 = const()[name = tensor("op_68_pad_0"), val = tensor([1, 1])]; + tensor logmel_data_to_fp16_dtype_0 = const()[name = tensor("logmel_data_to_fp16_dtype_0"), val = tensor("fp16")]; + tensor weight_3_to_fp16 = const()[name = tensor("weight_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor bias_3_to_fp16 = const()[name = tensor("bias_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(491648)))]; + tensor cast_727 = cast(dtype = logmel_data_to_fp16_dtype_0, x = logmel_data); + tensor var_68_cast = conv(bias = bias_3_to_fp16, dilations = var_66, groups = var_56, pad = var_68_pad_0, pad_type = var_68_pad_type_0, strides = var_64, weight = weight_3_to_fp16, x = cast_727); + tensor input_1_mode_0 = const()[name = tensor("input_1_mode_0"), val = tensor("EXACT")]; + tensor input_1_cast = gelu(mode = input_1_mode_0, x = var_68_cast); + tensor var_72 = const()[name = tensor("op_72"), val = tensor(1)]; + tensor var_81 = const()[name = tensor("op_81"), val = tensor([2])]; + tensor var_83 = const()[name = tensor("op_83"), val = tensor([1])]; + tensor var_85_pad_type_0 = const()[name = tensor("op_85_pad_type_0"), val = tensor("custom")]; + tensor var_85_pad_0 = const()[name = tensor("op_85_pad_0"), val = tensor([1, 1])]; + tensor weight_7_to_fp16 = const()[name = tensor("weight_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(493760)))]; + tensor bias_7_to_fp16 = const()[name = tensor("bias_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6785280)))]; + tensor var_85_cast = conv(bias = bias_7_to_fp16, dilations = var_83, groups = var_72, pad = var_85_pad_0, pad_type = var_85_pad_type_0, strides = var_81, weight = weight_7_to_fp16, x = input_1_cast); + tensor x_3_mode_0 = const()[name = tensor("x_3_mode_0"), val = tensor("EXACT")]; + tensor x_3_cast = gelu(mode = x_3_mode_0, x = var_85_cast); + tensor var_90 = const()[name = tensor("op_90"), val = tensor([0, 2, 1])]; + tensor positional_embedding_to_fp16 = const()[name = tensor("positional_embedding_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6787392)))]; + tensor transpose_192 = transpose(perm = var_90, x = x_3_cast); + tensor var_93_cast = add(x = transpose_192, y = positional_embedding_to_fp16); + tensor var_106 = const()[name = tensor("op_106"), val = tensor(-1)]; + tensor var_123_axes_0 = const()[name = tensor("op_123_axes_0"), val = tensor([-1])]; + tensor blocks_0_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_0_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9859456)))]; + tensor blocks_0_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_0_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9861568)))]; + tensor var_112_to_fp16 = const()[name = tensor("op_112_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_123_cast = layer_norm(axes = var_123_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_112_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_93_cast); + tensor var_134_to_fp16 = const()[name = tensor("op_134_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9863680)))]; + tensor var_135_to_fp16 = const()[name = tensor("op_135_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11960896)))]; + tensor q_1_cast = linear(bias = var_135_to_fp16, weight = var_134_to_fp16, x = var_123_cast); + tensor var_138_to_fp16 = const()[name = tensor("op_138_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11963008)))]; + tensor k_1_bias_0_to_fp16 = const()[name = tensor("k_1_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14060224)))]; + tensor k_1_cast = linear(bias = k_1_bias_0_to_fp16, weight = var_138_to_fp16, x = var_123_cast); + tensor var_142_to_fp16 = const()[name = tensor("op_142_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14062336)))]; + tensor var_143_to_fp16 = const()[name = tensor("op_143_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16159552)))]; + tensor v_1_cast = linear(bias = var_143_to_fp16, weight = var_142_to_fp16, x = var_123_cast); + tensor var_151 = const()[name = tensor("op_151"), val = tensor([1, 1500, 16, -1])]; + tensor var_152_cast = reshape(shape = var_151, x = q_1_cast); + tensor const_168_to_fp16 = const()[name = tensor("const_168_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_3_cast = mul(x = var_152_cast, y = const_168_to_fp16); + tensor var_158 = const()[name = tensor("op_158"), val = tensor([1, 1500, 16, -1])]; + tensor var_159_cast = reshape(shape = var_158, x = k_1_cast); + tensor const_169_to_fp16 = const()[name = tensor("const_169_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_3_cast = mul(x = var_159_cast, y = const_169_to_fp16); + tensor var_165 = const()[name = tensor("op_165"), val = tensor([1, 1500, 16, -1])]; + tensor var_166_cast = reshape(shape = var_165, x = v_1_cast); + tensor var_167 = const()[name = tensor("op_167"), val = tensor([0, 2, 1, 3])]; + tensor qk_1_transpose_x_0 = const()[name = tensor("qk_1_transpose_x_0"), val = tensor(false)]; + tensor qk_1_transpose_y_0 = const()[name = tensor("qk_1_transpose_y_0"), val = tensor(false)]; + tensor transpose_48_perm_0 = const()[name = tensor("transpose_48_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_49_perm_0 = const()[name = tensor("transpose_49_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_189 = transpose(perm = transpose_49_perm_0, x = k_3_cast); + tensor transpose_190 = transpose(perm = transpose_48_perm_0, x = q_3_cast); + tensor qk_1_cast = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_190, y = transpose_189); + tensor var_171_cast = softmax(axis = var_106, x = qk_1_cast); + tensor var_173_transpose_x_0 = const()[name = tensor("op_173_transpose_x_0"), val = tensor(false)]; + tensor var_173_transpose_y_0 = const()[name = tensor("op_173_transpose_y_0"), val = tensor(false)]; + tensor transpose_191 = transpose(perm = var_167, x = var_166_cast); + tensor var_173_cast = matmul(transpose_x = var_173_transpose_x_0, transpose_y = var_173_transpose_y_0, x = var_171_cast, y = transpose_191); + tensor var_174 = const()[name = tensor("op_174"), val = tensor([0, 2, 1, 3])]; + tensor concat_0 = const()[name = tensor("concat_0"), val = tensor([1, 1500, 1024])]; + tensor transpose_188 = transpose(perm = var_174, x = var_173_cast); + tensor x_11_cast = reshape(shape = concat_0, x = transpose_188); + tensor var_179_to_fp16 = const()[name = tensor("op_179_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16161664)))]; + tensor var_180_to_fp16 = const()[name = tensor("op_180_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18258880)))]; + tensor var_181_cast = linear(bias = var_180_to_fp16, weight = var_179_to_fp16, x = x_11_cast); + tensor x_13_cast = add(x = var_93_cast, y = var_181_cast); + tensor var_187_axes_0 = const()[name = tensor("op_187_axes_0"), val = tensor([-1])]; + tensor blocks_0_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18260992)))]; + tensor blocks_0_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18263104)))]; + tensor var_187_cast = layer_norm(axes = var_187_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_112_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast); + tensor var_196_to_fp16 = const()[name = tensor("op_196_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18265216)))]; + tensor var_197_to_fp16 = const()[name = tensor("op_197_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26653888)))]; + tensor input_9_cast = linear(bias = var_197_to_fp16, weight = var_196_to_fp16, x = var_187_cast); + tensor x_17_mode_0 = const()[name = tensor("x_17_mode_0"), val = tensor("EXACT")]; + tensor x_17_cast = gelu(mode = x_17_mode_0, x = input_9_cast); + tensor var_202_to_fp16 = const()[name = tensor("op_202_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26662144)))]; + tensor var_203_to_fp16 = const()[name = tensor("op_203_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35050816)))]; + tensor var_204_cast = linear(bias = var_203_to_fp16, weight = var_202_to_fp16, x = x_17_cast); + tensor x_19_cast = add(x = x_13_cast, y = var_204_cast); + tensor var_213 = const()[name = tensor("op_213"), val = tensor(-1)]; + tensor var_230_axes_0 = const()[name = tensor("op_230_axes_0"), val = tensor([-1])]; + tensor blocks_1_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_1_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35052928)))]; + tensor blocks_1_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_1_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35055040)))]; + tensor var_219_to_fp16 = const()[name = tensor("op_219_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_230_cast = layer_norm(axes = var_230_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_219_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast); + tensor var_241_to_fp16 = const()[name = tensor("op_241_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35057152)))]; + tensor var_242_to_fp16 = const()[name = tensor("op_242_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37154368)))]; + tensor q_5_cast = linear(bias = var_242_to_fp16, weight = var_241_to_fp16, x = var_230_cast); + tensor var_245_to_fp16 = const()[name = tensor("op_245_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37156480)))]; + tensor k_5_bias_0_to_fp16 = const()[name = tensor("k_5_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39253696)))]; + tensor k_5_cast = linear(bias = k_5_bias_0_to_fp16, weight = var_245_to_fp16, x = var_230_cast); + tensor var_249_to_fp16 = const()[name = tensor("op_249_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39255808)))]; + tensor var_250_to_fp16 = const()[name = tensor("op_250_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(41353024)))]; + tensor v_5_cast = linear(bias = var_250_to_fp16, weight = var_249_to_fp16, x = var_230_cast); + tensor var_258 = const()[name = tensor("op_258"), val = tensor([1, 1500, 16, -1])]; + tensor var_259_cast = reshape(shape = var_258, x = q_5_cast); + tensor const_170_to_fp16 = const()[name = tensor("const_170_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_7_cast = mul(x = var_259_cast, y = const_170_to_fp16); + tensor var_265 = const()[name = tensor("op_265"), val = tensor([1, 1500, 16, -1])]; + tensor var_266_cast = reshape(shape = var_265, x = k_5_cast); + tensor const_171_to_fp16 = const()[name = tensor("const_171_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_7_cast = mul(x = var_266_cast, y = const_171_to_fp16); + tensor var_272 = const()[name = tensor("op_272"), val = tensor([1, 1500, 16, -1])]; + tensor var_273_cast = reshape(shape = var_272, x = v_5_cast); + tensor var_274 = const()[name = tensor("op_274"), val = tensor([0, 2, 1, 3])]; + tensor qk_3_transpose_x_0 = const()[name = tensor("qk_3_transpose_x_0"), val = tensor(false)]; + tensor qk_3_transpose_y_0 = const()[name = tensor("qk_3_transpose_y_0"), val = tensor(false)]; + tensor transpose_50_perm_0 = const()[name = tensor("transpose_50_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_51_perm_0 = const()[name = tensor("transpose_51_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_185 = transpose(perm = transpose_51_perm_0, x = k_7_cast); + tensor transpose_186 = transpose(perm = transpose_50_perm_0, x = q_7_cast); + tensor qk_3_cast = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_186, y = transpose_185); + tensor var_278_cast = softmax(axis = var_213, x = qk_3_cast); + tensor var_280_transpose_x_0 = const()[name = tensor("op_280_transpose_x_0"), val = tensor(false)]; + tensor var_280_transpose_y_0 = const()[name = tensor("op_280_transpose_y_0"), val = tensor(false)]; + tensor transpose_187 = transpose(perm = var_274, x = var_273_cast); + tensor var_280_cast = matmul(transpose_x = var_280_transpose_x_0, transpose_y = var_280_transpose_y_0, x = var_278_cast, y = transpose_187); + tensor var_281 = const()[name = tensor("op_281"), val = tensor([0, 2, 1, 3])]; + tensor concat_1 = const()[name = tensor("concat_1"), val = tensor([1, 1500, 1024])]; + tensor transpose_184 = transpose(perm = var_281, x = var_280_cast); + tensor x_23_cast = reshape(shape = concat_1, x = transpose_184); + tensor var_286_to_fp16 = const()[name = tensor("op_286_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(41355136)))]; + tensor var_287_to_fp16 = const()[name = tensor("op_287_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43452352)))]; + tensor var_288_cast = linear(bias = var_287_to_fp16, weight = var_286_to_fp16, x = x_23_cast); + tensor x_25_cast = add(x = x_19_cast, y = var_288_cast); + tensor var_294_axes_0 = const()[name = tensor("op_294_axes_0"), val = tensor([-1])]; + tensor blocks_1_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43454464)))]; + tensor blocks_1_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43456576)))]; + tensor var_294_cast = layer_norm(axes = var_294_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_219_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast); + tensor var_303_to_fp16 = const()[name = tensor("op_303_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43458688)))]; + tensor var_304_to_fp16 = const()[name = tensor("op_304_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51847360)))]; + tensor input_17_cast = linear(bias = var_304_to_fp16, weight = var_303_to_fp16, x = var_294_cast); + tensor x_29_mode_0 = const()[name = tensor("x_29_mode_0"), val = tensor("EXACT")]; + tensor x_29_cast = gelu(mode = x_29_mode_0, x = input_17_cast); + tensor var_309_to_fp16 = const()[name = tensor("op_309_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51855616)))]; + tensor var_310_to_fp16 = const()[name = tensor("op_310_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60244288)))]; + tensor var_311_cast = linear(bias = var_310_to_fp16, weight = var_309_to_fp16, x = x_29_cast); + tensor x_31_cast = add(x = x_25_cast, y = var_311_cast); + tensor var_320 = const()[name = tensor("op_320"), val = tensor(-1)]; + tensor var_337_axes_0 = const()[name = tensor("op_337_axes_0"), val = tensor([-1])]; + tensor blocks_2_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_2_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60246400)))]; + tensor blocks_2_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_2_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60248512)))]; + tensor var_326_to_fp16 = const()[name = tensor("op_326_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_337_cast = layer_norm(axes = var_337_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_326_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast); + tensor var_348_to_fp16 = const()[name = tensor("op_348_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60250624)))]; + tensor var_349_to_fp16 = const()[name = tensor("op_349_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62347840)))]; + tensor q_9_cast = linear(bias = var_349_to_fp16, weight = var_348_to_fp16, x = var_337_cast); + tensor var_352_to_fp16 = const()[name = tensor("op_352_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62349952)))]; + tensor k_9_bias_0_to_fp16 = const()[name = tensor("k_9_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64447168)))]; + tensor k_9_cast = linear(bias = k_9_bias_0_to_fp16, weight = var_352_to_fp16, x = var_337_cast); + tensor var_356_to_fp16 = const()[name = tensor("op_356_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64449280)))]; + tensor var_357_to_fp16 = const()[name = tensor("op_357_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66546496)))]; + tensor v_9_cast = linear(bias = var_357_to_fp16, weight = var_356_to_fp16, x = var_337_cast); + tensor var_365 = const()[name = tensor("op_365"), val = tensor([1, 1500, 16, -1])]; + tensor var_366_cast = reshape(shape = var_365, x = q_9_cast); + tensor const_172_to_fp16 = const()[name = tensor("const_172_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_11_cast = mul(x = var_366_cast, y = const_172_to_fp16); + tensor var_372 = const()[name = tensor("op_372"), val = tensor([1, 1500, 16, -1])]; + tensor var_373_cast = reshape(shape = var_372, x = k_9_cast); + tensor const_173_to_fp16 = const()[name = tensor("const_173_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_11_cast = mul(x = var_373_cast, y = const_173_to_fp16); + tensor var_379 = const()[name = tensor("op_379"), val = tensor([1, 1500, 16, -1])]; + tensor var_380_cast = reshape(shape = var_379, x = v_9_cast); + tensor var_381 = const()[name = tensor("op_381"), val = tensor([0, 2, 1, 3])]; + tensor qk_5_transpose_x_0 = const()[name = tensor("qk_5_transpose_x_0"), val = tensor(false)]; + tensor qk_5_transpose_y_0 = const()[name = tensor("qk_5_transpose_y_0"), val = tensor(false)]; + tensor transpose_52_perm_0 = const()[name = tensor("transpose_52_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_53_perm_0 = const()[name = tensor("transpose_53_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_181 = transpose(perm = transpose_53_perm_0, x = k_11_cast); + tensor transpose_182 = transpose(perm = transpose_52_perm_0, x = q_11_cast); + tensor qk_5_cast = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_182, y = transpose_181); + tensor var_385_cast = softmax(axis = var_320, x = qk_5_cast); + tensor var_387_transpose_x_0 = const()[name = tensor("op_387_transpose_x_0"), val = tensor(false)]; + tensor var_387_transpose_y_0 = const()[name = tensor("op_387_transpose_y_0"), val = tensor(false)]; + tensor transpose_183 = transpose(perm = var_381, x = var_380_cast); + tensor var_387_cast = matmul(transpose_x = var_387_transpose_x_0, transpose_y = var_387_transpose_y_0, x = var_385_cast, y = transpose_183); + tensor var_388 = const()[name = tensor("op_388"), val = tensor([0, 2, 1, 3])]; + tensor concat_2 = const()[name = tensor("concat_2"), val = tensor([1, 1500, 1024])]; + tensor transpose_180 = transpose(perm = var_388, x = var_387_cast); + tensor x_35_cast = reshape(shape = concat_2, x = transpose_180); + tensor var_393_to_fp16 = const()[name = tensor("op_393_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66548608)))]; + tensor var_394_to_fp16 = const()[name = tensor("op_394_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68645824)))]; + tensor var_395_cast = linear(bias = var_394_to_fp16, weight = var_393_to_fp16, x = x_35_cast); + tensor x_37_cast = add(x = x_31_cast, y = var_395_cast); + tensor var_401_axes_0 = const()[name = tensor("op_401_axes_0"), val = tensor([-1])]; + tensor blocks_2_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68647936)))]; + tensor blocks_2_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68650048)))]; + tensor var_401_cast = layer_norm(axes = var_401_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_326_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast); + tensor var_410_to_fp16 = const()[name = tensor("op_410_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68652160)))]; + tensor var_411_to_fp16 = const()[name = tensor("op_411_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77040832)))]; + tensor input_25_cast = linear(bias = var_411_to_fp16, weight = var_410_to_fp16, x = var_401_cast); + tensor x_41_mode_0 = const()[name = tensor("x_41_mode_0"), val = tensor("EXACT")]; + tensor x_41_cast = gelu(mode = x_41_mode_0, x = input_25_cast); + tensor var_416_to_fp16 = const()[name = tensor("op_416_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77049088)))]; + tensor var_417_to_fp16 = const()[name = tensor("op_417_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85437760)))]; + tensor var_418_cast = linear(bias = var_417_to_fp16, weight = var_416_to_fp16, x = x_41_cast); + tensor x_43_cast = add(x = x_37_cast, y = var_418_cast); + tensor var_427 = const()[name = tensor("op_427"), val = tensor(-1)]; + tensor var_444_axes_0 = const()[name = tensor("op_444_axes_0"), val = tensor([-1])]; + tensor blocks_3_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_3_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85439872)))]; + tensor blocks_3_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_3_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85441984)))]; + tensor var_433_to_fp16 = const()[name = tensor("op_433_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_444_cast = layer_norm(axes = var_444_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_433_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast); + tensor var_455_to_fp16 = const()[name = tensor("op_455_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85444096)))]; + tensor var_456_to_fp16 = const()[name = tensor("op_456_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87541312)))]; + tensor q_13_cast = linear(bias = var_456_to_fp16, weight = var_455_to_fp16, x = var_444_cast); + tensor var_459_to_fp16 = const()[name = tensor("op_459_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87543424)))]; + tensor k_13_bias_0_to_fp16 = const()[name = tensor("k_13_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89640640)))]; + tensor k_13_cast = linear(bias = k_13_bias_0_to_fp16, weight = var_459_to_fp16, x = var_444_cast); + tensor var_463_to_fp16 = const()[name = tensor("op_463_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89642752)))]; + tensor var_464_to_fp16 = const()[name = tensor("op_464_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91739968)))]; + tensor v_13_cast = linear(bias = var_464_to_fp16, weight = var_463_to_fp16, x = var_444_cast); + tensor var_472 = const()[name = tensor("op_472"), val = tensor([1, 1500, 16, -1])]; + tensor var_473_cast = reshape(shape = var_472, x = q_13_cast); + tensor const_174_to_fp16 = const()[name = tensor("const_174_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_15_cast = mul(x = var_473_cast, y = const_174_to_fp16); + tensor var_479 = const()[name = tensor("op_479"), val = tensor([1, 1500, 16, -1])]; + tensor var_480_cast = reshape(shape = var_479, x = k_13_cast); + tensor const_175_to_fp16 = const()[name = tensor("const_175_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_15_cast = mul(x = var_480_cast, y = const_175_to_fp16); + tensor var_486 = const()[name = tensor("op_486"), val = tensor([1, 1500, 16, -1])]; + tensor var_487_cast = reshape(shape = var_486, x = v_13_cast); + tensor var_488 = const()[name = tensor("op_488"), val = tensor([0, 2, 1, 3])]; + tensor qk_7_transpose_x_0 = const()[name = tensor("qk_7_transpose_x_0"), val = tensor(false)]; + tensor qk_7_transpose_y_0 = const()[name = tensor("qk_7_transpose_y_0"), val = tensor(false)]; + tensor transpose_54_perm_0 = const()[name = tensor("transpose_54_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_55_perm_0 = const()[name = tensor("transpose_55_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_177 = transpose(perm = transpose_55_perm_0, x = k_15_cast); + tensor transpose_178 = transpose(perm = transpose_54_perm_0, x = q_15_cast); + tensor qk_7_cast = matmul(transpose_x = qk_7_transpose_x_0, transpose_y = qk_7_transpose_y_0, x = transpose_178, y = transpose_177); + tensor var_492_cast = softmax(axis = var_427, x = qk_7_cast); + tensor var_494_transpose_x_0 = const()[name = tensor("op_494_transpose_x_0"), val = tensor(false)]; + tensor var_494_transpose_y_0 = const()[name = tensor("op_494_transpose_y_0"), val = tensor(false)]; + tensor transpose_179 = transpose(perm = var_488, x = var_487_cast); + tensor var_494_cast = matmul(transpose_x = var_494_transpose_x_0, transpose_y = var_494_transpose_y_0, x = var_492_cast, y = transpose_179); + tensor var_495 = const()[name = tensor("op_495"), val = tensor([0, 2, 1, 3])]; + tensor concat_3 = const()[name = tensor("concat_3"), val = tensor([1, 1500, 1024])]; + tensor transpose_176 = transpose(perm = var_495, x = var_494_cast); + tensor x_47_cast = reshape(shape = concat_3, x = transpose_176); + tensor var_500_to_fp16 = const()[name = tensor("op_500_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91742080)))]; + tensor var_501_to_fp16 = const()[name = tensor("op_501_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93839296)))]; + tensor var_502_cast = linear(bias = var_501_to_fp16, weight = var_500_to_fp16, x = x_47_cast); + tensor x_49_cast = add(x = x_43_cast, y = var_502_cast); + tensor var_508_axes_0 = const()[name = tensor("op_508_axes_0"), val = tensor([-1])]; + tensor blocks_3_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93841408)))]; + tensor blocks_3_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93843520)))]; + tensor var_508_cast = layer_norm(axes = var_508_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_433_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast); + tensor var_517_to_fp16 = const()[name = tensor("op_517_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93845632)))]; + tensor var_518_to_fp16 = const()[name = tensor("op_518_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102234304)))]; + tensor input_33_cast = linear(bias = var_518_to_fp16, weight = var_517_to_fp16, x = var_508_cast); + tensor x_53_mode_0 = const()[name = tensor("x_53_mode_0"), val = tensor("EXACT")]; + tensor x_53_cast = gelu(mode = x_53_mode_0, x = input_33_cast); + tensor var_523_to_fp16 = const()[name = tensor("op_523_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102242560)))]; + tensor var_524_to_fp16 = const()[name = tensor("op_524_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110631232)))]; + tensor var_525_cast = linear(bias = var_524_to_fp16, weight = var_523_to_fp16, x = x_53_cast); + tensor x_55_cast = add(x = x_49_cast, y = var_525_cast); + tensor var_534 = const()[name = tensor("op_534"), val = tensor(-1)]; + tensor var_551_axes_0 = const()[name = tensor("op_551_axes_0"), val = tensor([-1])]; + tensor blocks_4_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_4_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110633344)))]; + tensor blocks_4_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_4_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110635456)))]; + tensor var_540_to_fp16 = const()[name = tensor("op_540_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_551_cast = layer_norm(axes = var_551_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_540_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast); + tensor var_562_to_fp16 = const()[name = tensor("op_562_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110637568)))]; + tensor var_563_to_fp16 = const()[name = tensor("op_563_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112734784)))]; + tensor q_17_cast = linear(bias = var_563_to_fp16, weight = var_562_to_fp16, x = var_551_cast); + tensor var_566_to_fp16 = const()[name = tensor("op_566_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112736896)))]; + tensor k_17_bias_0_to_fp16 = const()[name = tensor("k_17_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114834112)))]; + tensor k_17_cast = linear(bias = k_17_bias_0_to_fp16, weight = var_566_to_fp16, x = var_551_cast); + tensor var_570_to_fp16 = const()[name = tensor("op_570_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114836224)))]; + tensor var_571_to_fp16 = const()[name = tensor("op_571_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116933440)))]; + tensor v_17_cast = linear(bias = var_571_to_fp16, weight = var_570_to_fp16, x = var_551_cast); + tensor var_579 = const()[name = tensor("op_579"), val = tensor([1, 1500, 16, -1])]; + tensor var_580_cast = reshape(shape = var_579, x = q_17_cast); + tensor const_176_to_fp16 = const()[name = tensor("const_176_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_19_cast = mul(x = var_580_cast, y = const_176_to_fp16); + tensor var_586 = const()[name = tensor("op_586"), val = tensor([1, 1500, 16, -1])]; + tensor var_587_cast = reshape(shape = var_586, x = k_17_cast); + tensor const_177_to_fp16 = const()[name = tensor("const_177_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_19_cast = mul(x = var_587_cast, y = const_177_to_fp16); + tensor var_593 = const()[name = tensor("op_593"), val = tensor([1, 1500, 16, -1])]; + tensor var_594_cast = reshape(shape = var_593, x = v_17_cast); + tensor var_595 = const()[name = tensor("op_595"), val = tensor([0, 2, 1, 3])]; + tensor qk_9_transpose_x_0 = const()[name = tensor("qk_9_transpose_x_0"), val = tensor(false)]; + tensor qk_9_transpose_y_0 = const()[name = tensor("qk_9_transpose_y_0"), val = tensor(false)]; + tensor transpose_56_perm_0 = const()[name = tensor("transpose_56_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_57_perm_0 = const()[name = tensor("transpose_57_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_173 = transpose(perm = transpose_57_perm_0, x = k_19_cast); + tensor transpose_174 = transpose(perm = transpose_56_perm_0, x = q_19_cast); + tensor qk_9_cast = matmul(transpose_x = qk_9_transpose_x_0, transpose_y = qk_9_transpose_y_0, x = transpose_174, y = transpose_173); + tensor var_599_cast = softmax(axis = var_534, x = qk_9_cast); + tensor var_601_transpose_x_0 = const()[name = tensor("op_601_transpose_x_0"), val = tensor(false)]; + tensor var_601_transpose_y_0 = const()[name = tensor("op_601_transpose_y_0"), val = tensor(false)]; + tensor transpose_175 = transpose(perm = var_595, x = var_594_cast); + tensor var_601_cast = matmul(transpose_x = var_601_transpose_x_0, transpose_y = var_601_transpose_y_0, x = var_599_cast, y = transpose_175); + tensor var_602 = const()[name = tensor("op_602"), val = tensor([0, 2, 1, 3])]; + tensor concat_4 = const()[name = tensor("concat_4"), val = tensor([1, 1500, 1024])]; + tensor transpose_172 = transpose(perm = var_602, x = var_601_cast); + tensor x_59_cast = reshape(shape = concat_4, x = transpose_172); + tensor var_607_to_fp16 = const()[name = tensor("op_607_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116935552)))]; + tensor var_608_to_fp16 = const()[name = tensor("op_608_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119032768)))]; + tensor var_609_cast = linear(bias = var_608_to_fp16, weight = var_607_to_fp16, x = x_59_cast); + tensor x_61_cast = add(x = x_55_cast, y = var_609_cast); + tensor var_615_axes_0 = const()[name = tensor("op_615_axes_0"), val = tensor([-1])]; + tensor blocks_4_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119034880)))]; + tensor blocks_4_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119036992)))]; + tensor var_615_cast = layer_norm(axes = var_615_axes_0, beta = blocks_4_mlp_ln_bias_to_fp16, epsilon = var_540_to_fp16, gamma = blocks_4_mlp_ln_weight_to_fp16, x = x_61_cast); + tensor var_624_to_fp16 = const()[name = tensor("op_624_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119039104)))]; + tensor var_625_to_fp16 = const()[name = tensor("op_625_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127427776)))]; + tensor input_41_cast = linear(bias = var_625_to_fp16, weight = var_624_to_fp16, x = var_615_cast); + tensor x_65_mode_0 = const()[name = tensor("x_65_mode_0"), val = tensor("EXACT")]; + tensor x_65_cast = gelu(mode = x_65_mode_0, x = input_41_cast); + tensor var_630_to_fp16 = const()[name = tensor("op_630_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127436032)))]; + tensor var_631_to_fp16 = const()[name = tensor("op_631_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135824704)))]; + tensor var_632_cast = linear(bias = var_631_to_fp16, weight = var_630_to_fp16, x = x_65_cast); + tensor x_67_cast = add(x = x_61_cast, y = var_632_cast); + tensor var_641 = const()[name = tensor("op_641"), val = tensor(-1)]; + tensor var_658_axes_0 = const()[name = tensor("op_658_axes_0"), val = tensor([-1])]; + tensor blocks_5_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_5_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135826816)))]; + tensor blocks_5_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_5_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135828928)))]; + tensor var_647_to_fp16 = const()[name = tensor("op_647_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_658_cast = layer_norm(axes = var_658_axes_0, beta = blocks_5_attn_ln_bias_to_fp16, epsilon = var_647_to_fp16, gamma = blocks_5_attn_ln_weight_to_fp16, x = x_67_cast); + tensor var_669_to_fp16 = const()[name = tensor("op_669_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135831040)))]; + tensor var_670_to_fp16 = const()[name = tensor("op_670_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137928256)))]; + tensor q_21_cast = linear(bias = var_670_to_fp16, weight = var_669_to_fp16, x = var_658_cast); + tensor var_673_to_fp16 = const()[name = tensor("op_673_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137930368)))]; + tensor k_21_bias_0_to_fp16 = const()[name = tensor("k_21_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(140027584)))]; + tensor k_21_cast = linear(bias = k_21_bias_0_to_fp16, weight = var_673_to_fp16, x = var_658_cast); + tensor var_677_to_fp16 = const()[name = tensor("op_677_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(140029696)))]; + tensor var_678_to_fp16 = const()[name = tensor("op_678_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142126912)))]; + tensor v_21_cast = linear(bias = var_678_to_fp16, weight = var_677_to_fp16, x = var_658_cast); + tensor var_686 = const()[name = tensor("op_686"), val = tensor([1, 1500, 16, -1])]; + tensor var_687_cast = reshape(shape = var_686, x = q_21_cast); + tensor const_178_to_fp16 = const()[name = tensor("const_178_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_23_cast = mul(x = var_687_cast, y = const_178_to_fp16); + tensor var_693 = const()[name = tensor("op_693"), val = tensor([1, 1500, 16, -1])]; + tensor var_694_cast = reshape(shape = var_693, x = k_21_cast); + tensor const_179_to_fp16 = const()[name = tensor("const_179_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_23_cast = mul(x = var_694_cast, y = const_179_to_fp16); + tensor var_700 = const()[name = tensor("op_700"), val = tensor([1, 1500, 16, -1])]; + tensor var_701_cast = reshape(shape = var_700, x = v_21_cast); + tensor var_702 = const()[name = tensor("op_702"), val = tensor([0, 2, 1, 3])]; + tensor qk_11_transpose_x_0 = const()[name = tensor("qk_11_transpose_x_0"), val = tensor(false)]; + tensor qk_11_transpose_y_0 = const()[name = tensor("qk_11_transpose_y_0"), val = tensor(false)]; + tensor transpose_58_perm_0 = const()[name = tensor("transpose_58_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_59_perm_0 = const()[name = tensor("transpose_59_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_169 = transpose(perm = transpose_59_perm_0, x = k_23_cast); + tensor transpose_170 = transpose(perm = transpose_58_perm_0, x = q_23_cast); + tensor qk_11_cast = matmul(transpose_x = qk_11_transpose_x_0, transpose_y = qk_11_transpose_y_0, x = transpose_170, y = transpose_169); + tensor var_706_cast = softmax(axis = var_641, x = qk_11_cast); + tensor var_708_transpose_x_0 = const()[name = tensor("op_708_transpose_x_0"), val = tensor(false)]; + tensor var_708_transpose_y_0 = const()[name = tensor("op_708_transpose_y_0"), val = tensor(false)]; + tensor transpose_171 = transpose(perm = var_702, x = var_701_cast); + tensor var_708_cast = matmul(transpose_x = var_708_transpose_x_0, transpose_y = var_708_transpose_y_0, x = var_706_cast, y = transpose_171); + tensor var_709 = const()[name = tensor("op_709"), val = tensor([0, 2, 1, 3])]; + tensor concat_5 = const()[name = tensor("concat_5"), val = tensor([1, 1500, 1024])]; + tensor transpose_168 = transpose(perm = var_709, x = var_708_cast); + tensor x_71_cast = reshape(shape = concat_5, x = transpose_168); + tensor var_714_to_fp16 = const()[name = tensor("op_714_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142129024)))]; + tensor var_715_to_fp16 = const()[name = tensor("op_715_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(144226240)))]; + tensor var_716_cast = linear(bias = var_715_to_fp16, weight = var_714_to_fp16, x = x_71_cast); + tensor x_73_cast = add(x = x_67_cast, y = var_716_cast); + tensor var_722_axes_0 = const()[name = tensor("op_722_axes_0"), val = tensor([-1])]; + tensor blocks_5_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(144228352)))]; + tensor blocks_5_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(144230464)))]; + tensor var_722_cast = layer_norm(axes = var_722_axes_0, beta = blocks_5_mlp_ln_bias_to_fp16, epsilon = var_647_to_fp16, gamma = blocks_5_mlp_ln_weight_to_fp16, x = x_73_cast); + tensor var_731_to_fp16 = const()[name = tensor("op_731_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(144232576)))]; + tensor var_732_to_fp16 = const()[name = tensor("op_732_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152621248)))]; + tensor input_49_cast = linear(bias = var_732_to_fp16, weight = var_731_to_fp16, x = var_722_cast); + tensor x_77_mode_0 = const()[name = tensor("x_77_mode_0"), val = tensor("EXACT")]; + tensor x_77_cast = gelu(mode = x_77_mode_0, x = input_49_cast); + tensor var_737_to_fp16 = const()[name = tensor("op_737_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152629504)))]; + tensor var_738_to_fp16 = const()[name = tensor("op_738_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161018176)))]; + tensor var_739_cast = linear(bias = var_738_to_fp16, weight = var_737_to_fp16, x = x_77_cast); + tensor x_79_cast = add(x = x_73_cast, y = var_739_cast); + tensor var_748 = const()[name = tensor("op_748"), val = tensor(-1)]; + tensor var_765_axes_0 = const()[name = tensor("op_765_axes_0"), val = tensor([-1])]; + tensor blocks_6_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_6_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161020288)))]; + tensor blocks_6_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_6_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161022400)))]; + tensor var_754_to_fp16 = const()[name = tensor("op_754_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_765_cast = layer_norm(axes = var_765_axes_0, beta = blocks_6_attn_ln_bias_to_fp16, epsilon = var_754_to_fp16, gamma = blocks_6_attn_ln_weight_to_fp16, x = x_79_cast); + tensor var_776_to_fp16 = const()[name = tensor("op_776_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161024512)))]; + tensor var_777_to_fp16 = const()[name = tensor("op_777_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163121728)))]; + tensor q_25_cast = linear(bias = var_777_to_fp16, weight = var_776_to_fp16, x = var_765_cast); + tensor var_780_to_fp16 = const()[name = tensor("op_780_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163123840)))]; + tensor k_25_bias_0_to_fp16 = const()[name = tensor("k_25_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165221056)))]; + tensor k_25_cast = linear(bias = k_25_bias_0_to_fp16, weight = var_780_to_fp16, x = var_765_cast); + tensor var_784_to_fp16 = const()[name = tensor("op_784_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165223168)))]; + tensor var_785_to_fp16 = const()[name = tensor("op_785_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(167320384)))]; + tensor v_25_cast = linear(bias = var_785_to_fp16, weight = var_784_to_fp16, x = var_765_cast); + tensor var_793 = const()[name = tensor("op_793"), val = tensor([1, 1500, 16, -1])]; + tensor var_794_cast = reshape(shape = var_793, x = q_25_cast); + tensor const_180_to_fp16 = const()[name = tensor("const_180_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_27_cast = mul(x = var_794_cast, y = const_180_to_fp16); + tensor var_800 = const()[name = tensor("op_800"), val = tensor([1, 1500, 16, -1])]; + tensor var_801_cast = reshape(shape = var_800, x = k_25_cast); + tensor const_181_to_fp16 = const()[name = tensor("const_181_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_27_cast = mul(x = var_801_cast, y = const_181_to_fp16); + tensor var_807 = const()[name = tensor("op_807"), val = tensor([1, 1500, 16, -1])]; + tensor var_808_cast = reshape(shape = var_807, x = v_25_cast); + tensor var_809 = const()[name = tensor("op_809"), val = tensor([0, 2, 1, 3])]; + tensor qk_13_transpose_x_0 = const()[name = tensor("qk_13_transpose_x_0"), val = tensor(false)]; + tensor qk_13_transpose_y_0 = const()[name = tensor("qk_13_transpose_y_0"), val = tensor(false)]; + tensor transpose_60_perm_0 = const()[name = tensor("transpose_60_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_61_perm_0 = const()[name = tensor("transpose_61_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_165 = transpose(perm = transpose_61_perm_0, x = k_27_cast); + tensor transpose_166 = transpose(perm = transpose_60_perm_0, x = q_27_cast); + tensor qk_13_cast = matmul(transpose_x = qk_13_transpose_x_0, transpose_y = qk_13_transpose_y_0, x = transpose_166, y = transpose_165); + tensor var_813_cast = softmax(axis = var_748, x = qk_13_cast); + tensor var_815_transpose_x_0 = const()[name = tensor("op_815_transpose_x_0"), val = tensor(false)]; + tensor var_815_transpose_y_0 = const()[name = tensor("op_815_transpose_y_0"), val = tensor(false)]; + tensor transpose_167 = transpose(perm = var_809, x = var_808_cast); + tensor var_815_cast = matmul(transpose_x = var_815_transpose_x_0, transpose_y = var_815_transpose_y_0, x = var_813_cast, y = transpose_167); + tensor var_816 = const()[name = tensor("op_816"), val = tensor([0, 2, 1, 3])]; + tensor concat_6 = const()[name = tensor("concat_6"), val = tensor([1, 1500, 1024])]; + tensor transpose_164 = transpose(perm = var_816, x = var_815_cast); + tensor x_83_cast = reshape(shape = concat_6, x = transpose_164); + tensor var_821_to_fp16 = const()[name = tensor("op_821_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(167322496)))]; + tensor var_822_to_fp16 = const()[name = tensor("op_822_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(169419712)))]; + tensor var_823_cast = linear(bias = var_822_to_fp16, weight = var_821_to_fp16, x = x_83_cast); + tensor x_85_cast = add(x = x_79_cast, y = var_823_cast); + tensor var_829_axes_0 = const()[name = tensor("op_829_axes_0"), val = tensor([-1])]; + tensor blocks_6_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_6_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(169421824)))]; + tensor blocks_6_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_6_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(169423936)))]; + tensor var_829_cast = layer_norm(axes = var_829_axes_0, beta = blocks_6_mlp_ln_bias_to_fp16, epsilon = var_754_to_fp16, gamma = blocks_6_mlp_ln_weight_to_fp16, x = x_85_cast); + tensor var_838_to_fp16 = const()[name = tensor("op_838_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(169426048)))]; + tensor var_839_to_fp16 = const()[name = tensor("op_839_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(177814720)))]; + tensor input_57_cast = linear(bias = var_839_to_fp16, weight = var_838_to_fp16, x = var_829_cast); + tensor x_89_mode_0 = const()[name = tensor("x_89_mode_0"), val = tensor("EXACT")]; + tensor x_89_cast = gelu(mode = x_89_mode_0, x = input_57_cast); + tensor var_844_to_fp16 = const()[name = tensor("op_844_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(177822976)))]; + tensor var_845_to_fp16 = const()[name = tensor("op_845_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186211648)))]; + tensor var_846_cast = linear(bias = var_845_to_fp16, weight = var_844_to_fp16, x = x_89_cast); + tensor x_91_cast = add(x = x_85_cast, y = var_846_cast); + tensor var_855 = const()[name = tensor("op_855"), val = tensor(-1)]; + tensor var_872_axes_0 = const()[name = tensor("op_872_axes_0"), val = tensor([-1])]; + tensor blocks_7_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_7_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186213760)))]; + tensor blocks_7_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_7_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186215872)))]; + tensor var_861_to_fp16 = const()[name = tensor("op_861_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_872_cast = layer_norm(axes = var_872_axes_0, beta = blocks_7_attn_ln_bias_to_fp16, epsilon = var_861_to_fp16, gamma = blocks_7_attn_ln_weight_to_fp16, x = x_91_cast); + tensor var_883_to_fp16 = const()[name = tensor("op_883_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186217984)))]; + tensor var_884_to_fp16 = const()[name = tensor("op_884_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(188315200)))]; + tensor q_29_cast = linear(bias = var_884_to_fp16, weight = var_883_to_fp16, x = var_872_cast); + tensor var_887_to_fp16 = const()[name = tensor("op_887_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(188317312)))]; + tensor k_29_bias_0_to_fp16 = const()[name = tensor("k_29_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190414528)))]; + tensor k_29_cast = linear(bias = k_29_bias_0_to_fp16, weight = var_887_to_fp16, x = var_872_cast); + tensor var_891_to_fp16 = const()[name = tensor("op_891_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190416640)))]; + tensor var_892_to_fp16 = const()[name = tensor("op_892_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(192513856)))]; + tensor v_29_cast = linear(bias = var_892_to_fp16, weight = var_891_to_fp16, x = var_872_cast); + tensor var_900 = const()[name = tensor("op_900"), val = tensor([1, 1500, 16, -1])]; + tensor var_901_cast = reshape(shape = var_900, x = q_29_cast); + tensor const_182_to_fp16 = const()[name = tensor("const_182_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_31_cast = mul(x = var_901_cast, y = const_182_to_fp16); + tensor var_907 = const()[name = tensor("op_907"), val = tensor([1, 1500, 16, -1])]; + tensor var_908_cast = reshape(shape = var_907, x = k_29_cast); + tensor const_183_to_fp16 = const()[name = tensor("const_183_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_31_cast = mul(x = var_908_cast, y = const_183_to_fp16); + tensor var_914 = const()[name = tensor("op_914"), val = tensor([1, 1500, 16, -1])]; + tensor var_915_cast = reshape(shape = var_914, x = v_29_cast); + tensor var_916 = const()[name = tensor("op_916"), val = tensor([0, 2, 1, 3])]; + tensor qk_15_transpose_x_0 = const()[name = tensor("qk_15_transpose_x_0"), val = tensor(false)]; + tensor qk_15_transpose_y_0 = const()[name = tensor("qk_15_transpose_y_0"), val = tensor(false)]; + tensor transpose_62_perm_0 = const()[name = tensor("transpose_62_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_63_perm_0 = const()[name = tensor("transpose_63_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_161 = transpose(perm = transpose_63_perm_0, x = k_31_cast); + tensor transpose_162 = transpose(perm = transpose_62_perm_0, x = q_31_cast); + tensor qk_15_cast = matmul(transpose_x = qk_15_transpose_x_0, transpose_y = qk_15_transpose_y_0, x = transpose_162, y = transpose_161); + tensor var_920_cast = softmax(axis = var_855, x = qk_15_cast); + tensor var_922_transpose_x_0 = const()[name = tensor("op_922_transpose_x_0"), val = tensor(false)]; + tensor var_922_transpose_y_0 = const()[name = tensor("op_922_transpose_y_0"), val = tensor(false)]; + tensor transpose_163 = transpose(perm = var_916, x = var_915_cast); + tensor var_922_cast = matmul(transpose_x = var_922_transpose_x_0, transpose_y = var_922_transpose_y_0, x = var_920_cast, y = transpose_163); + tensor var_923 = const()[name = tensor("op_923"), val = tensor([0, 2, 1, 3])]; + tensor concat_7 = const()[name = tensor("concat_7"), val = tensor([1, 1500, 1024])]; + tensor transpose_160 = transpose(perm = var_923, x = var_922_cast); + tensor x_95_cast = reshape(shape = concat_7, x = transpose_160); + tensor var_928_to_fp16 = const()[name = tensor("op_928_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(192515968)))]; + tensor var_929_to_fp16 = const()[name = tensor("op_929_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194613184)))]; + tensor var_930_cast = linear(bias = var_929_to_fp16, weight = var_928_to_fp16, x = x_95_cast); + tensor x_97_cast = add(x = x_91_cast, y = var_930_cast); + tensor var_936_axes_0 = const()[name = tensor("op_936_axes_0"), val = tensor([-1])]; + tensor blocks_7_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_7_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194615296)))]; + tensor blocks_7_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_7_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194617408)))]; + tensor var_936_cast = layer_norm(axes = var_936_axes_0, beta = blocks_7_mlp_ln_bias_to_fp16, epsilon = var_861_to_fp16, gamma = blocks_7_mlp_ln_weight_to_fp16, x = x_97_cast); + tensor var_945_to_fp16 = const()[name = tensor("op_945_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194619520)))]; + tensor var_946_to_fp16 = const()[name = tensor("op_946_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(203008192)))]; + tensor input_65_cast = linear(bias = var_946_to_fp16, weight = var_945_to_fp16, x = var_936_cast); + tensor x_101_mode_0 = const()[name = tensor("x_101_mode_0"), val = tensor("EXACT")]; + tensor x_101_cast = gelu(mode = x_101_mode_0, x = input_65_cast); + tensor var_951_to_fp16 = const()[name = tensor("op_951_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(203016448)))]; + tensor var_952_to_fp16 = const()[name = tensor("op_952_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211405120)))]; + tensor var_953_cast = linear(bias = var_952_to_fp16, weight = var_951_to_fp16, x = x_101_cast); + tensor x_103_cast = add(x = x_97_cast, y = var_953_cast); + tensor var_962 = const()[name = tensor("op_962"), val = tensor(-1)]; + tensor var_979_axes_0 = const()[name = tensor("op_979_axes_0"), val = tensor([-1])]; + tensor blocks_8_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_8_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211407232)))]; + tensor blocks_8_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_8_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211409344)))]; + tensor var_968_to_fp16 = const()[name = tensor("op_968_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_979_cast = layer_norm(axes = var_979_axes_0, beta = blocks_8_attn_ln_bias_to_fp16, epsilon = var_968_to_fp16, gamma = blocks_8_attn_ln_weight_to_fp16, x = x_103_cast); + tensor var_990_to_fp16 = const()[name = tensor("op_990_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211411456)))]; + tensor var_991_to_fp16 = const()[name = tensor("op_991_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(213508672)))]; + tensor q_33_cast = linear(bias = var_991_to_fp16, weight = var_990_to_fp16, x = var_979_cast); + tensor var_994_to_fp16 = const()[name = tensor("op_994_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(213510784)))]; + tensor k_33_bias_0_to_fp16 = const()[name = tensor("k_33_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215608000)))]; + tensor k_33_cast = linear(bias = k_33_bias_0_to_fp16, weight = var_994_to_fp16, x = var_979_cast); + tensor var_998_to_fp16 = const()[name = tensor("op_998_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215610112)))]; + tensor var_999_to_fp16 = const()[name = tensor("op_999_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217707328)))]; + tensor v_33_cast = linear(bias = var_999_to_fp16, weight = var_998_to_fp16, x = var_979_cast); + tensor var_1007 = const()[name = tensor("op_1007"), val = tensor([1, 1500, 16, -1])]; + tensor var_1008_cast = reshape(shape = var_1007, x = q_33_cast); + tensor const_184_to_fp16 = const()[name = tensor("const_184_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_35_cast = mul(x = var_1008_cast, y = const_184_to_fp16); + tensor var_1014 = const()[name = tensor("op_1014"), val = tensor([1, 1500, 16, -1])]; + tensor var_1015_cast = reshape(shape = var_1014, x = k_33_cast); + tensor const_185_to_fp16 = const()[name = tensor("const_185_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_35_cast = mul(x = var_1015_cast, y = const_185_to_fp16); + tensor var_1021 = const()[name = tensor("op_1021"), val = tensor([1, 1500, 16, -1])]; + tensor var_1022_cast = reshape(shape = var_1021, x = v_33_cast); + tensor var_1023 = const()[name = tensor("op_1023"), val = tensor([0, 2, 1, 3])]; + tensor qk_17_transpose_x_0 = const()[name = tensor("qk_17_transpose_x_0"), val = tensor(false)]; + tensor qk_17_transpose_y_0 = const()[name = tensor("qk_17_transpose_y_0"), val = tensor(false)]; + tensor transpose_64_perm_0 = const()[name = tensor("transpose_64_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_65_perm_0 = const()[name = tensor("transpose_65_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_157 = transpose(perm = transpose_65_perm_0, x = k_35_cast); + tensor transpose_158 = transpose(perm = transpose_64_perm_0, x = q_35_cast); + tensor qk_17_cast = matmul(transpose_x = qk_17_transpose_x_0, transpose_y = qk_17_transpose_y_0, x = transpose_158, y = transpose_157); + tensor var_1027_cast = softmax(axis = var_962, x = qk_17_cast); + tensor var_1029_transpose_x_0 = const()[name = tensor("op_1029_transpose_x_0"), val = tensor(false)]; + tensor var_1029_transpose_y_0 = const()[name = tensor("op_1029_transpose_y_0"), val = tensor(false)]; + tensor transpose_159 = transpose(perm = var_1023, x = var_1022_cast); + tensor var_1029_cast = matmul(transpose_x = var_1029_transpose_x_0, transpose_y = var_1029_transpose_y_0, x = var_1027_cast, y = transpose_159); + tensor var_1030 = const()[name = tensor("op_1030"), val = tensor([0, 2, 1, 3])]; + tensor concat_8 = const()[name = tensor("concat_8"), val = tensor([1, 1500, 1024])]; + tensor transpose_156 = transpose(perm = var_1030, x = var_1029_cast); + tensor x_107_cast = reshape(shape = concat_8, x = transpose_156); + tensor var_1035_to_fp16 = const()[name = tensor("op_1035_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217709440)))]; + tensor var_1036_to_fp16 = const()[name = tensor("op_1036_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219806656)))]; + tensor var_1037_cast = linear(bias = var_1036_to_fp16, weight = var_1035_to_fp16, x = x_107_cast); + tensor x_109_cast = add(x = x_103_cast, y = var_1037_cast); + tensor var_1043_axes_0 = const()[name = tensor("op_1043_axes_0"), val = tensor([-1])]; + tensor blocks_8_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_8_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219808768)))]; + tensor blocks_8_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_8_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219810880)))]; + tensor var_1043_cast = layer_norm(axes = var_1043_axes_0, beta = blocks_8_mlp_ln_bias_to_fp16, epsilon = var_968_to_fp16, gamma = blocks_8_mlp_ln_weight_to_fp16, x = x_109_cast); + tensor var_1052_to_fp16 = const()[name = tensor("op_1052_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219812992)))]; + tensor var_1053_to_fp16 = const()[name = tensor("op_1053_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228201664)))]; + tensor input_73_cast = linear(bias = var_1053_to_fp16, weight = var_1052_to_fp16, x = var_1043_cast); + tensor x_113_mode_0 = const()[name = tensor("x_113_mode_0"), val = tensor("EXACT")]; + tensor x_113_cast = gelu(mode = x_113_mode_0, x = input_73_cast); + tensor var_1058_to_fp16 = const()[name = tensor("op_1058_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228209920)))]; + tensor var_1059_to_fp16 = const()[name = tensor("op_1059_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236598592)))]; + tensor var_1060_cast = linear(bias = var_1059_to_fp16, weight = var_1058_to_fp16, x = x_113_cast); + tensor x_115_cast = add(x = x_109_cast, y = var_1060_cast); + tensor var_1069 = const()[name = tensor("op_1069"), val = tensor(-1)]; + tensor var_1086_axes_0 = const()[name = tensor("op_1086_axes_0"), val = tensor([-1])]; + tensor blocks_9_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_9_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236600704)))]; + tensor blocks_9_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_9_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236602816)))]; + tensor var_1075_to_fp16 = const()[name = tensor("op_1075_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1086_cast = layer_norm(axes = var_1086_axes_0, beta = blocks_9_attn_ln_bias_to_fp16, epsilon = var_1075_to_fp16, gamma = blocks_9_attn_ln_weight_to_fp16, x = x_115_cast); + tensor var_1097_to_fp16 = const()[name = tensor("op_1097_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236604928)))]; + tensor var_1098_to_fp16 = const()[name = tensor("op_1098_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(238702144)))]; + tensor q_37_cast = linear(bias = var_1098_to_fp16, weight = var_1097_to_fp16, x = var_1086_cast); + tensor var_1101_to_fp16 = const()[name = tensor("op_1101_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(238704256)))]; + tensor k_37_bias_0_to_fp16 = const()[name = tensor("k_37_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240801472)))]; + tensor k_37_cast = linear(bias = k_37_bias_0_to_fp16, weight = var_1101_to_fp16, x = var_1086_cast); + tensor var_1105_to_fp16 = const()[name = tensor("op_1105_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240803584)))]; + tensor var_1106_to_fp16 = const()[name = tensor("op_1106_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242900800)))]; + tensor v_37_cast = linear(bias = var_1106_to_fp16, weight = var_1105_to_fp16, x = var_1086_cast); + tensor var_1114 = const()[name = tensor("op_1114"), val = tensor([1, 1500, 16, -1])]; + tensor var_1115_cast = reshape(shape = var_1114, x = q_37_cast); + tensor const_186_to_fp16 = const()[name = tensor("const_186_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_39_cast = mul(x = var_1115_cast, y = const_186_to_fp16); + tensor var_1121 = const()[name = tensor("op_1121"), val = tensor([1, 1500, 16, -1])]; + tensor var_1122_cast = reshape(shape = var_1121, x = k_37_cast); + tensor const_187_to_fp16 = const()[name = tensor("const_187_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_39_cast = mul(x = var_1122_cast, y = const_187_to_fp16); + tensor var_1128 = const()[name = tensor("op_1128"), val = tensor([1, 1500, 16, -1])]; + tensor var_1129_cast = reshape(shape = var_1128, x = v_37_cast); + tensor var_1130 = const()[name = tensor("op_1130"), val = tensor([0, 2, 1, 3])]; + tensor qk_19_transpose_x_0 = const()[name = tensor("qk_19_transpose_x_0"), val = tensor(false)]; + tensor qk_19_transpose_y_0 = const()[name = tensor("qk_19_transpose_y_0"), val = tensor(false)]; + tensor transpose_66_perm_0 = const()[name = tensor("transpose_66_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_67_perm_0 = const()[name = tensor("transpose_67_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_153 = transpose(perm = transpose_67_perm_0, x = k_39_cast); + tensor transpose_154 = transpose(perm = transpose_66_perm_0, x = q_39_cast); + tensor qk_19_cast = matmul(transpose_x = qk_19_transpose_x_0, transpose_y = qk_19_transpose_y_0, x = transpose_154, y = transpose_153); + tensor var_1134_cast = softmax(axis = var_1069, x = qk_19_cast); + tensor var_1136_transpose_x_0 = const()[name = tensor("op_1136_transpose_x_0"), val = tensor(false)]; + tensor var_1136_transpose_y_0 = const()[name = tensor("op_1136_transpose_y_0"), val = tensor(false)]; + tensor transpose_155 = transpose(perm = var_1130, x = var_1129_cast); + tensor var_1136_cast = matmul(transpose_x = var_1136_transpose_x_0, transpose_y = var_1136_transpose_y_0, x = var_1134_cast, y = transpose_155); + tensor var_1137 = const()[name = tensor("op_1137"), val = tensor([0, 2, 1, 3])]; + tensor concat_9 = const()[name = tensor("concat_9"), val = tensor([1, 1500, 1024])]; + tensor transpose_152 = transpose(perm = var_1137, x = var_1136_cast); + tensor x_119_cast = reshape(shape = concat_9, x = transpose_152); + tensor var_1142_to_fp16 = const()[name = tensor("op_1142_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242902912)))]; + tensor var_1143_to_fp16 = const()[name = tensor("op_1143_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(245000128)))]; + tensor var_1144_cast = linear(bias = var_1143_to_fp16, weight = var_1142_to_fp16, x = x_119_cast); + tensor x_121_cast = add(x = x_115_cast, y = var_1144_cast); + tensor var_1150_axes_0 = const()[name = tensor("op_1150_axes_0"), val = tensor([-1])]; + tensor blocks_9_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_9_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(245002240)))]; + tensor blocks_9_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_9_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(245004352)))]; + tensor var_1150_cast = layer_norm(axes = var_1150_axes_0, beta = blocks_9_mlp_ln_bias_to_fp16, epsilon = var_1075_to_fp16, gamma = blocks_9_mlp_ln_weight_to_fp16, x = x_121_cast); + tensor var_1159_to_fp16 = const()[name = tensor("op_1159_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(245006464)))]; + tensor var_1160_to_fp16 = const()[name = tensor("op_1160_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(253395136)))]; + tensor input_81_cast = linear(bias = var_1160_to_fp16, weight = var_1159_to_fp16, x = var_1150_cast); + tensor x_125_mode_0 = const()[name = tensor("x_125_mode_0"), val = tensor("EXACT")]; + tensor x_125_cast = gelu(mode = x_125_mode_0, x = input_81_cast); + tensor var_1165_to_fp16 = const()[name = tensor("op_1165_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(253403392)))]; + tensor var_1166_to_fp16 = const()[name = tensor("op_1166_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261792064)))]; + tensor var_1167_cast = linear(bias = var_1166_to_fp16, weight = var_1165_to_fp16, x = x_125_cast); + tensor x_127_cast = add(x = x_121_cast, y = var_1167_cast); + tensor var_1176 = const()[name = tensor("op_1176"), val = tensor(-1)]; + tensor var_1193_axes_0 = const()[name = tensor("op_1193_axes_0"), val = tensor([-1])]; + tensor blocks_10_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_10_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261794176)))]; + tensor blocks_10_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_10_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261796288)))]; + tensor var_1182_to_fp16 = const()[name = tensor("op_1182_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1193_cast = layer_norm(axes = var_1193_axes_0, beta = blocks_10_attn_ln_bias_to_fp16, epsilon = var_1182_to_fp16, gamma = blocks_10_attn_ln_weight_to_fp16, x = x_127_cast); + tensor var_1204_to_fp16 = const()[name = tensor("op_1204_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261798400)))]; + tensor var_1205_to_fp16 = const()[name = tensor("op_1205_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263895616)))]; + tensor q_41_cast = linear(bias = var_1205_to_fp16, weight = var_1204_to_fp16, x = var_1193_cast); + tensor var_1208_to_fp16 = const()[name = tensor("op_1208_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263897728)))]; + tensor k_41_bias_0_to_fp16 = const()[name = tensor("k_41_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265994944)))]; + tensor k_41_cast = linear(bias = k_41_bias_0_to_fp16, weight = var_1208_to_fp16, x = var_1193_cast); + tensor var_1212_to_fp16 = const()[name = tensor("op_1212_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265997056)))]; + tensor var_1213_to_fp16 = const()[name = tensor("op_1213_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268094272)))]; + tensor v_41_cast = linear(bias = var_1213_to_fp16, weight = var_1212_to_fp16, x = var_1193_cast); + tensor var_1221 = const()[name = tensor("op_1221"), val = tensor([1, 1500, 16, -1])]; + tensor var_1222_cast = reshape(shape = var_1221, x = q_41_cast); + tensor const_188_to_fp16 = const()[name = tensor("const_188_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_43_cast = mul(x = var_1222_cast, y = const_188_to_fp16); + tensor var_1228 = const()[name = tensor("op_1228"), val = tensor([1, 1500, 16, -1])]; + tensor var_1229_cast = reshape(shape = var_1228, x = k_41_cast); + tensor const_189_to_fp16 = const()[name = tensor("const_189_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_43_cast = mul(x = var_1229_cast, y = const_189_to_fp16); + tensor var_1235 = const()[name = tensor("op_1235"), val = tensor([1, 1500, 16, -1])]; + tensor var_1236_cast = reshape(shape = var_1235, x = v_41_cast); + tensor var_1237 = const()[name = tensor("op_1237"), val = tensor([0, 2, 1, 3])]; + tensor qk_21_transpose_x_0 = const()[name = tensor("qk_21_transpose_x_0"), val = tensor(false)]; + tensor qk_21_transpose_y_0 = const()[name = tensor("qk_21_transpose_y_0"), val = tensor(false)]; + tensor transpose_68_perm_0 = const()[name = tensor("transpose_68_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_69_perm_0 = const()[name = tensor("transpose_69_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_149 = transpose(perm = transpose_69_perm_0, x = k_43_cast); + tensor transpose_150 = transpose(perm = transpose_68_perm_0, x = q_43_cast); + tensor qk_21_cast = matmul(transpose_x = qk_21_transpose_x_0, transpose_y = qk_21_transpose_y_0, x = transpose_150, y = transpose_149); + tensor var_1241_cast = softmax(axis = var_1176, x = qk_21_cast); + tensor var_1243_transpose_x_0 = const()[name = tensor("op_1243_transpose_x_0"), val = tensor(false)]; + tensor var_1243_transpose_y_0 = const()[name = tensor("op_1243_transpose_y_0"), val = tensor(false)]; + tensor transpose_151 = transpose(perm = var_1237, x = var_1236_cast); + tensor var_1243_cast = matmul(transpose_x = var_1243_transpose_x_0, transpose_y = var_1243_transpose_y_0, x = var_1241_cast, y = transpose_151); + tensor var_1244 = const()[name = tensor("op_1244"), val = tensor([0, 2, 1, 3])]; + tensor concat_10 = const()[name = tensor("concat_10"), val = tensor([1, 1500, 1024])]; + tensor transpose_148 = transpose(perm = var_1244, x = var_1243_cast); + tensor x_131_cast = reshape(shape = concat_10, x = transpose_148); + tensor var_1249_to_fp16 = const()[name = tensor("op_1249_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268096384)))]; + tensor var_1250_to_fp16 = const()[name = tensor("op_1250_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(270193600)))]; + tensor var_1251_cast = linear(bias = var_1250_to_fp16, weight = var_1249_to_fp16, x = x_131_cast); + tensor x_133_cast = add(x = x_127_cast, y = var_1251_cast); + tensor var_1257_axes_0 = const()[name = tensor("op_1257_axes_0"), val = tensor([-1])]; + tensor blocks_10_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_10_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(270195712)))]; + tensor blocks_10_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_10_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(270197824)))]; + tensor var_1257_cast = layer_norm(axes = var_1257_axes_0, beta = blocks_10_mlp_ln_bias_to_fp16, epsilon = var_1182_to_fp16, gamma = blocks_10_mlp_ln_weight_to_fp16, x = x_133_cast); + tensor var_1266_to_fp16 = const()[name = tensor("op_1266_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(270199936)))]; + tensor var_1267_to_fp16 = const()[name = tensor("op_1267_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(278588608)))]; + tensor input_89_cast = linear(bias = var_1267_to_fp16, weight = var_1266_to_fp16, x = var_1257_cast); + tensor x_137_mode_0 = const()[name = tensor("x_137_mode_0"), val = tensor("EXACT")]; + tensor x_137_cast = gelu(mode = x_137_mode_0, x = input_89_cast); + tensor var_1272_to_fp16 = const()[name = tensor("op_1272_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(278596864)))]; + tensor var_1273_to_fp16 = const()[name = tensor("op_1273_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286985536)))]; + tensor var_1274_cast = linear(bias = var_1273_to_fp16, weight = var_1272_to_fp16, x = x_137_cast); + tensor x_139_cast = add(x = x_133_cast, y = var_1274_cast); + tensor var_1283 = const()[name = tensor("op_1283"), val = tensor(-1)]; + tensor var_1300_axes_0 = const()[name = tensor("op_1300_axes_0"), val = tensor([-1])]; + tensor blocks_11_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_11_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286987648)))]; + tensor blocks_11_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_11_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286989760)))]; + tensor var_1289_to_fp16 = const()[name = tensor("op_1289_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1300_cast = layer_norm(axes = var_1300_axes_0, beta = blocks_11_attn_ln_bias_to_fp16, epsilon = var_1289_to_fp16, gamma = blocks_11_attn_ln_weight_to_fp16, x = x_139_cast); + tensor var_1311_to_fp16 = const()[name = tensor("op_1311_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286991872)))]; + tensor var_1312_to_fp16 = const()[name = tensor("op_1312_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289089088)))]; + tensor q_45_cast = linear(bias = var_1312_to_fp16, weight = var_1311_to_fp16, x = var_1300_cast); + tensor var_1315_to_fp16 = const()[name = tensor("op_1315_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289091200)))]; + tensor k_45_bias_0_to_fp16 = const()[name = tensor("k_45_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291188416)))]; + tensor k_45_cast = linear(bias = k_45_bias_0_to_fp16, weight = var_1315_to_fp16, x = var_1300_cast); + tensor var_1319_to_fp16 = const()[name = tensor("op_1319_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291190528)))]; + tensor var_1320_to_fp16 = const()[name = tensor("op_1320_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293287744)))]; + tensor v_45_cast = linear(bias = var_1320_to_fp16, weight = var_1319_to_fp16, x = var_1300_cast); + tensor var_1328 = const()[name = tensor("op_1328"), val = tensor([1, 1500, 16, -1])]; + tensor var_1329_cast = reshape(shape = var_1328, x = q_45_cast); + tensor const_190_to_fp16 = const()[name = tensor("const_190_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_47_cast = mul(x = var_1329_cast, y = const_190_to_fp16); + tensor var_1335 = const()[name = tensor("op_1335"), val = tensor([1, 1500, 16, -1])]; + tensor var_1336_cast = reshape(shape = var_1335, x = k_45_cast); + tensor const_191_to_fp16 = const()[name = tensor("const_191_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_47_cast = mul(x = var_1336_cast, y = const_191_to_fp16); + tensor var_1342 = const()[name = tensor("op_1342"), val = tensor([1, 1500, 16, -1])]; + tensor var_1343_cast = reshape(shape = var_1342, x = v_45_cast); + tensor var_1344 = const()[name = tensor("op_1344"), val = tensor([0, 2, 1, 3])]; + tensor qk_23_transpose_x_0 = const()[name = tensor("qk_23_transpose_x_0"), val = tensor(false)]; + tensor qk_23_transpose_y_0 = const()[name = tensor("qk_23_transpose_y_0"), val = tensor(false)]; + tensor transpose_70_perm_0 = const()[name = tensor("transpose_70_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_71_perm_0 = const()[name = tensor("transpose_71_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_145 = transpose(perm = transpose_71_perm_0, x = k_47_cast); + tensor transpose_146 = transpose(perm = transpose_70_perm_0, x = q_47_cast); + tensor qk_23_cast = matmul(transpose_x = qk_23_transpose_x_0, transpose_y = qk_23_transpose_y_0, x = transpose_146, y = transpose_145); + tensor var_1348_cast = softmax(axis = var_1283, x = qk_23_cast); + tensor var_1350_transpose_x_0 = const()[name = tensor("op_1350_transpose_x_0"), val = tensor(false)]; + tensor var_1350_transpose_y_0 = const()[name = tensor("op_1350_transpose_y_0"), val = tensor(false)]; + tensor transpose_147 = transpose(perm = var_1344, x = var_1343_cast); + tensor var_1350_cast = matmul(transpose_x = var_1350_transpose_x_0, transpose_y = var_1350_transpose_y_0, x = var_1348_cast, y = transpose_147); + tensor var_1351 = const()[name = tensor("op_1351"), val = tensor([0, 2, 1, 3])]; + tensor concat_11 = const()[name = tensor("concat_11"), val = tensor([1, 1500, 1024])]; + tensor transpose_144 = transpose(perm = var_1351, x = var_1350_cast); + tensor x_143_cast = reshape(shape = concat_11, x = transpose_144); + tensor var_1356_to_fp16 = const()[name = tensor("op_1356_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293289856)))]; + tensor var_1357_to_fp16 = const()[name = tensor("op_1357_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(295387072)))]; + tensor var_1358_cast = linear(bias = var_1357_to_fp16, weight = var_1356_to_fp16, x = x_143_cast); + tensor x_145_cast = add(x = x_139_cast, y = var_1358_cast); + tensor var_1364_axes_0 = const()[name = tensor("op_1364_axes_0"), val = tensor([-1])]; + tensor blocks_11_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_11_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(295389184)))]; + tensor blocks_11_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_11_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(295391296)))]; + tensor var_1364_cast = layer_norm(axes = var_1364_axes_0, beta = blocks_11_mlp_ln_bias_to_fp16, epsilon = var_1289_to_fp16, gamma = blocks_11_mlp_ln_weight_to_fp16, x = x_145_cast); + tensor var_1373_to_fp16 = const()[name = tensor("op_1373_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(295393408)))]; + tensor var_1374_to_fp16 = const()[name = tensor("op_1374_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303782080)))]; + tensor input_97_cast = linear(bias = var_1374_to_fp16, weight = var_1373_to_fp16, x = var_1364_cast); + tensor x_149_mode_0 = const()[name = tensor("x_149_mode_0"), val = tensor("EXACT")]; + tensor x_149_cast = gelu(mode = x_149_mode_0, x = input_97_cast); + tensor var_1379_to_fp16 = const()[name = tensor("op_1379_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303790336)))]; + tensor var_1380_to_fp16 = const()[name = tensor("op_1380_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312179008)))]; + tensor var_1381_cast = linear(bias = var_1380_to_fp16, weight = var_1379_to_fp16, x = x_149_cast); + tensor x_151_cast = add(x = x_145_cast, y = var_1381_cast); + tensor var_1390 = const()[name = tensor("op_1390"), val = tensor(-1)]; + tensor var_1407_axes_0 = const()[name = tensor("op_1407_axes_0"), val = tensor([-1])]; + tensor blocks_12_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_12_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312181120)))]; + tensor blocks_12_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_12_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312183232)))]; + tensor var_1396_to_fp16 = const()[name = tensor("op_1396_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1407_cast = layer_norm(axes = var_1407_axes_0, beta = blocks_12_attn_ln_bias_to_fp16, epsilon = var_1396_to_fp16, gamma = blocks_12_attn_ln_weight_to_fp16, x = x_151_cast); + tensor var_1418_to_fp16 = const()[name = tensor("op_1418_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312185344)))]; + tensor var_1419_to_fp16 = const()[name = tensor("op_1419_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314282560)))]; + tensor q_49_cast = linear(bias = var_1419_to_fp16, weight = var_1418_to_fp16, x = var_1407_cast); + tensor var_1422_to_fp16 = const()[name = tensor("op_1422_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314284672)))]; + tensor k_49_bias_0_to_fp16 = const()[name = tensor("k_49_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316381888)))]; + tensor k_49_cast = linear(bias = k_49_bias_0_to_fp16, weight = var_1422_to_fp16, x = var_1407_cast); + tensor var_1426_to_fp16 = const()[name = tensor("op_1426_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316384000)))]; + tensor var_1427_to_fp16 = const()[name = tensor("op_1427_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(318481216)))]; + tensor v_49_cast = linear(bias = var_1427_to_fp16, weight = var_1426_to_fp16, x = var_1407_cast); + tensor var_1435 = const()[name = tensor("op_1435"), val = tensor([1, 1500, 16, -1])]; + tensor var_1436_cast = reshape(shape = var_1435, x = q_49_cast); + tensor const_192_to_fp16 = const()[name = tensor("const_192_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_51_cast = mul(x = var_1436_cast, y = const_192_to_fp16); + tensor var_1442 = const()[name = tensor("op_1442"), val = tensor([1, 1500, 16, -1])]; + tensor var_1443_cast = reshape(shape = var_1442, x = k_49_cast); + tensor const_193_to_fp16 = const()[name = tensor("const_193_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_51_cast = mul(x = var_1443_cast, y = const_193_to_fp16); + tensor var_1449 = const()[name = tensor("op_1449"), val = tensor([1, 1500, 16, -1])]; + tensor var_1450_cast = reshape(shape = var_1449, x = v_49_cast); + tensor var_1451 = const()[name = tensor("op_1451"), val = tensor([0, 2, 1, 3])]; + tensor qk_25_transpose_x_0 = const()[name = tensor("qk_25_transpose_x_0"), val = tensor(false)]; + tensor qk_25_transpose_y_0 = const()[name = tensor("qk_25_transpose_y_0"), val = tensor(false)]; + tensor transpose_72_perm_0 = const()[name = tensor("transpose_72_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_73_perm_0 = const()[name = tensor("transpose_73_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_141 = transpose(perm = transpose_73_perm_0, x = k_51_cast); + tensor transpose_142 = transpose(perm = transpose_72_perm_0, x = q_51_cast); + tensor qk_25_cast = matmul(transpose_x = qk_25_transpose_x_0, transpose_y = qk_25_transpose_y_0, x = transpose_142, y = transpose_141); + tensor var_1455_cast = softmax(axis = var_1390, x = qk_25_cast); + tensor var_1457_transpose_x_0 = const()[name = tensor("op_1457_transpose_x_0"), val = tensor(false)]; + tensor var_1457_transpose_y_0 = const()[name = tensor("op_1457_transpose_y_0"), val = tensor(false)]; + tensor transpose_143 = transpose(perm = var_1451, x = var_1450_cast); + tensor var_1457_cast = matmul(transpose_x = var_1457_transpose_x_0, transpose_y = var_1457_transpose_y_0, x = var_1455_cast, y = transpose_143); + tensor var_1458 = const()[name = tensor("op_1458"), val = tensor([0, 2, 1, 3])]; + tensor concat_12 = const()[name = tensor("concat_12"), val = tensor([1, 1500, 1024])]; + tensor transpose_140 = transpose(perm = var_1458, x = var_1457_cast); + tensor x_155_cast = reshape(shape = concat_12, x = transpose_140); + tensor var_1463_to_fp16 = const()[name = tensor("op_1463_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(318483328)))]; + tensor var_1464_to_fp16 = const()[name = tensor("op_1464_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(320580544)))]; + tensor var_1465_cast = linear(bias = var_1464_to_fp16, weight = var_1463_to_fp16, x = x_155_cast); + tensor x_157_cast = add(x = x_151_cast, y = var_1465_cast); + tensor var_1471_axes_0 = const()[name = tensor("op_1471_axes_0"), val = tensor([-1])]; + tensor blocks_12_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_12_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(320582656)))]; + tensor blocks_12_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_12_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(320584768)))]; + tensor var_1471_cast = layer_norm(axes = var_1471_axes_0, beta = blocks_12_mlp_ln_bias_to_fp16, epsilon = var_1396_to_fp16, gamma = blocks_12_mlp_ln_weight_to_fp16, x = x_157_cast); + tensor var_1480_to_fp16 = const()[name = tensor("op_1480_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(320586880)))]; + tensor var_1481_to_fp16 = const()[name = tensor("op_1481_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(328975552)))]; + tensor input_105_cast = linear(bias = var_1481_to_fp16, weight = var_1480_to_fp16, x = var_1471_cast); + tensor x_161_mode_0 = const()[name = tensor("x_161_mode_0"), val = tensor("EXACT")]; + tensor x_161_cast = gelu(mode = x_161_mode_0, x = input_105_cast); + tensor var_1486_to_fp16 = const()[name = tensor("op_1486_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(328983808)))]; + tensor var_1487_to_fp16 = const()[name = tensor("op_1487_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337372480)))]; + tensor var_1488_cast = linear(bias = var_1487_to_fp16, weight = var_1486_to_fp16, x = x_161_cast); + tensor x_163_cast = add(x = x_157_cast, y = var_1488_cast); + tensor var_1497 = const()[name = tensor("op_1497"), val = tensor(-1)]; + tensor var_1514_axes_0 = const()[name = tensor("op_1514_axes_0"), val = tensor([-1])]; + tensor blocks_13_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_13_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337374592)))]; + tensor blocks_13_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_13_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337376704)))]; + tensor var_1503_to_fp16 = const()[name = tensor("op_1503_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1514_cast = layer_norm(axes = var_1514_axes_0, beta = blocks_13_attn_ln_bias_to_fp16, epsilon = var_1503_to_fp16, gamma = blocks_13_attn_ln_weight_to_fp16, x = x_163_cast); + tensor var_1525_to_fp16 = const()[name = tensor("op_1525_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337378816)))]; + tensor var_1526_to_fp16 = const()[name = tensor("op_1526_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339476032)))]; + tensor q_53_cast = linear(bias = var_1526_to_fp16, weight = var_1525_to_fp16, x = var_1514_cast); + tensor var_1529_to_fp16 = const()[name = tensor("op_1529_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339478144)))]; + tensor k_53_bias_0_to_fp16 = const()[name = tensor("k_53_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(341575360)))]; + tensor k_53_cast = linear(bias = k_53_bias_0_to_fp16, weight = var_1529_to_fp16, x = var_1514_cast); + tensor var_1533_to_fp16 = const()[name = tensor("op_1533_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(341577472)))]; + tensor var_1534_to_fp16 = const()[name = tensor("op_1534_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(343674688)))]; + tensor v_53_cast = linear(bias = var_1534_to_fp16, weight = var_1533_to_fp16, x = var_1514_cast); + tensor var_1542 = const()[name = tensor("op_1542"), val = tensor([1, 1500, 16, -1])]; + tensor var_1543_cast = reshape(shape = var_1542, x = q_53_cast); + tensor const_194_to_fp16 = const()[name = tensor("const_194_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_55_cast = mul(x = var_1543_cast, y = const_194_to_fp16); + tensor var_1549 = const()[name = tensor("op_1549"), val = tensor([1, 1500, 16, -1])]; + tensor var_1550_cast = reshape(shape = var_1549, x = k_53_cast); + tensor const_195_to_fp16 = const()[name = tensor("const_195_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_55_cast = mul(x = var_1550_cast, y = const_195_to_fp16); + tensor var_1556 = const()[name = tensor("op_1556"), val = tensor([1, 1500, 16, -1])]; + tensor var_1557_cast = reshape(shape = var_1556, x = v_53_cast); + tensor var_1558 = const()[name = tensor("op_1558"), val = tensor([0, 2, 1, 3])]; + tensor qk_27_transpose_x_0 = const()[name = tensor("qk_27_transpose_x_0"), val = tensor(false)]; + tensor qk_27_transpose_y_0 = const()[name = tensor("qk_27_transpose_y_0"), val = tensor(false)]; + tensor transpose_74_perm_0 = const()[name = tensor("transpose_74_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_75_perm_0 = const()[name = tensor("transpose_75_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_137 = transpose(perm = transpose_75_perm_0, x = k_55_cast); + tensor transpose_138 = transpose(perm = transpose_74_perm_0, x = q_55_cast); + tensor qk_27_cast = matmul(transpose_x = qk_27_transpose_x_0, transpose_y = qk_27_transpose_y_0, x = transpose_138, y = transpose_137); + tensor var_1562_cast = softmax(axis = var_1497, x = qk_27_cast); + tensor var_1564_transpose_x_0 = const()[name = tensor("op_1564_transpose_x_0"), val = tensor(false)]; + tensor var_1564_transpose_y_0 = const()[name = tensor("op_1564_transpose_y_0"), val = tensor(false)]; + tensor transpose_139 = transpose(perm = var_1558, x = var_1557_cast); + tensor var_1564_cast = matmul(transpose_x = var_1564_transpose_x_0, transpose_y = var_1564_transpose_y_0, x = var_1562_cast, y = transpose_139); + tensor var_1565 = const()[name = tensor("op_1565"), val = tensor([0, 2, 1, 3])]; + tensor concat_13 = const()[name = tensor("concat_13"), val = tensor([1, 1500, 1024])]; + tensor transpose_136 = transpose(perm = var_1565, x = var_1564_cast); + tensor x_167_cast = reshape(shape = concat_13, x = transpose_136); + tensor var_1570_to_fp16 = const()[name = tensor("op_1570_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(343676800)))]; + tensor var_1571_to_fp16 = const()[name = tensor("op_1571_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(345774016)))]; + tensor var_1572_cast = linear(bias = var_1571_to_fp16, weight = var_1570_to_fp16, x = x_167_cast); + tensor x_169_cast = add(x = x_163_cast, y = var_1572_cast); + tensor var_1578_axes_0 = const()[name = tensor("op_1578_axes_0"), val = tensor([-1])]; + tensor blocks_13_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_13_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(345776128)))]; + tensor blocks_13_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_13_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(345778240)))]; + tensor var_1578_cast = layer_norm(axes = var_1578_axes_0, beta = blocks_13_mlp_ln_bias_to_fp16, epsilon = var_1503_to_fp16, gamma = blocks_13_mlp_ln_weight_to_fp16, x = x_169_cast); + tensor var_1587_to_fp16 = const()[name = tensor("op_1587_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(345780352)))]; + tensor var_1588_to_fp16 = const()[name = tensor("op_1588_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(354169024)))]; + tensor input_113_cast = linear(bias = var_1588_to_fp16, weight = var_1587_to_fp16, x = var_1578_cast); + tensor x_173_mode_0 = const()[name = tensor("x_173_mode_0"), val = tensor("EXACT")]; + tensor x_173_cast = gelu(mode = x_173_mode_0, x = input_113_cast); + tensor var_1593_to_fp16 = const()[name = tensor("op_1593_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(354177280)))]; + tensor var_1594_to_fp16 = const()[name = tensor("op_1594_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362565952)))]; + tensor var_1595_cast = linear(bias = var_1594_to_fp16, weight = var_1593_to_fp16, x = x_173_cast); + tensor x_175_cast = add(x = x_169_cast, y = var_1595_cast); + tensor var_1604 = const()[name = tensor("op_1604"), val = tensor(-1)]; + tensor var_1621_axes_0 = const()[name = tensor("op_1621_axes_0"), val = tensor([-1])]; + tensor blocks_14_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_14_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362568064)))]; + tensor blocks_14_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_14_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362570176)))]; + tensor var_1610_to_fp16 = const()[name = tensor("op_1610_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1621_cast = layer_norm(axes = var_1621_axes_0, beta = blocks_14_attn_ln_bias_to_fp16, epsilon = var_1610_to_fp16, gamma = blocks_14_attn_ln_weight_to_fp16, x = x_175_cast); + tensor var_1632_to_fp16 = const()[name = tensor("op_1632_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362572288)))]; + tensor var_1633_to_fp16 = const()[name = tensor("op_1633_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364669504)))]; + tensor q_57_cast = linear(bias = var_1633_to_fp16, weight = var_1632_to_fp16, x = var_1621_cast); + tensor var_1636_to_fp16 = const()[name = tensor("op_1636_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364671616)))]; + tensor k_57_bias_0_to_fp16 = const()[name = tensor("k_57_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(366768832)))]; + tensor k_57_cast = linear(bias = k_57_bias_0_to_fp16, weight = var_1636_to_fp16, x = var_1621_cast); + tensor var_1640_to_fp16 = const()[name = tensor("op_1640_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(366770944)))]; + tensor var_1641_to_fp16 = const()[name = tensor("op_1641_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368868160)))]; + tensor v_57_cast = linear(bias = var_1641_to_fp16, weight = var_1640_to_fp16, x = var_1621_cast); + tensor var_1649 = const()[name = tensor("op_1649"), val = tensor([1, 1500, 16, -1])]; + tensor var_1650_cast = reshape(shape = var_1649, x = q_57_cast); + tensor const_196_to_fp16 = const()[name = tensor("const_196_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_59_cast = mul(x = var_1650_cast, y = const_196_to_fp16); + tensor var_1656 = const()[name = tensor("op_1656"), val = tensor([1, 1500, 16, -1])]; + tensor var_1657_cast = reshape(shape = var_1656, x = k_57_cast); + tensor const_197_to_fp16 = const()[name = tensor("const_197_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_59_cast = mul(x = var_1657_cast, y = const_197_to_fp16); + tensor var_1663 = const()[name = tensor("op_1663"), val = tensor([1, 1500, 16, -1])]; + tensor var_1664_cast = reshape(shape = var_1663, x = v_57_cast); + tensor var_1665 = const()[name = tensor("op_1665"), val = tensor([0, 2, 1, 3])]; + tensor qk_29_transpose_x_0 = const()[name = tensor("qk_29_transpose_x_0"), val = tensor(false)]; + tensor qk_29_transpose_y_0 = const()[name = tensor("qk_29_transpose_y_0"), val = tensor(false)]; + tensor transpose_76_perm_0 = const()[name = tensor("transpose_76_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_77_perm_0 = const()[name = tensor("transpose_77_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_133 = transpose(perm = transpose_77_perm_0, x = k_59_cast); + tensor transpose_134 = transpose(perm = transpose_76_perm_0, x = q_59_cast); + tensor qk_29_cast = matmul(transpose_x = qk_29_transpose_x_0, transpose_y = qk_29_transpose_y_0, x = transpose_134, y = transpose_133); + tensor var_1669_cast = softmax(axis = var_1604, x = qk_29_cast); + tensor var_1671_transpose_x_0 = const()[name = tensor("op_1671_transpose_x_0"), val = tensor(false)]; + tensor var_1671_transpose_y_0 = const()[name = tensor("op_1671_transpose_y_0"), val = tensor(false)]; + tensor transpose_135 = transpose(perm = var_1665, x = var_1664_cast); + tensor var_1671_cast = matmul(transpose_x = var_1671_transpose_x_0, transpose_y = var_1671_transpose_y_0, x = var_1669_cast, y = transpose_135); + tensor var_1672 = const()[name = tensor("op_1672"), val = tensor([0, 2, 1, 3])]; + tensor concat_14 = const()[name = tensor("concat_14"), val = tensor([1, 1500, 1024])]; + tensor transpose_132 = transpose(perm = var_1672, x = var_1671_cast); + tensor x_179_cast = reshape(shape = concat_14, x = transpose_132); + tensor var_1677_to_fp16 = const()[name = tensor("op_1677_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368870272)))]; + tensor var_1678_to_fp16 = const()[name = tensor("op_1678_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(370967488)))]; + tensor var_1679_cast = linear(bias = var_1678_to_fp16, weight = var_1677_to_fp16, x = x_179_cast); + tensor x_181_cast = add(x = x_175_cast, y = var_1679_cast); + tensor var_1685_axes_0 = const()[name = tensor("op_1685_axes_0"), val = tensor([-1])]; + tensor blocks_14_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_14_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(370969600)))]; + tensor blocks_14_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_14_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(370971712)))]; + tensor var_1685_cast = layer_norm(axes = var_1685_axes_0, beta = blocks_14_mlp_ln_bias_to_fp16, epsilon = var_1610_to_fp16, gamma = blocks_14_mlp_ln_weight_to_fp16, x = x_181_cast); + tensor var_1694_to_fp16 = const()[name = tensor("op_1694_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(370973824)))]; + tensor var_1695_to_fp16 = const()[name = tensor("op_1695_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(379362496)))]; + tensor input_121_cast = linear(bias = var_1695_to_fp16, weight = var_1694_to_fp16, x = var_1685_cast); + tensor x_185_mode_0 = const()[name = tensor("x_185_mode_0"), val = tensor("EXACT")]; + tensor x_185_cast = gelu(mode = x_185_mode_0, x = input_121_cast); + tensor var_1700_to_fp16 = const()[name = tensor("op_1700_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(379370752)))]; + tensor var_1701_to_fp16 = const()[name = tensor("op_1701_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387759424)))]; + tensor var_1702_cast = linear(bias = var_1701_to_fp16, weight = var_1700_to_fp16, x = x_185_cast); + tensor x_187_cast = add(x = x_181_cast, y = var_1702_cast); + tensor var_1711 = const()[name = tensor("op_1711"), val = tensor(-1)]; + tensor var_1728_axes_0 = const()[name = tensor("op_1728_axes_0"), val = tensor([-1])]; + tensor blocks_15_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_15_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387761536)))]; + tensor blocks_15_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_15_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387763648)))]; + tensor var_1717_to_fp16 = const()[name = tensor("op_1717_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1728_cast = layer_norm(axes = var_1728_axes_0, beta = blocks_15_attn_ln_bias_to_fp16, epsilon = var_1717_to_fp16, gamma = blocks_15_attn_ln_weight_to_fp16, x = x_187_cast); + tensor var_1739_to_fp16 = const()[name = tensor("op_1739_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387765760)))]; + tensor var_1740_to_fp16 = const()[name = tensor("op_1740_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(389862976)))]; + tensor q_61_cast = linear(bias = var_1740_to_fp16, weight = var_1739_to_fp16, x = var_1728_cast); + tensor var_1743_to_fp16 = const()[name = tensor("op_1743_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(389865088)))]; + tensor k_61_bias_0_to_fp16 = const()[name = tensor("k_61_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(391962304)))]; + tensor k_61_cast = linear(bias = k_61_bias_0_to_fp16, weight = var_1743_to_fp16, x = var_1728_cast); + tensor var_1747_to_fp16 = const()[name = tensor("op_1747_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(391964416)))]; + tensor var_1748_to_fp16 = const()[name = tensor("op_1748_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(394061632)))]; + tensor v_61_cast = linear(bias = var_1748_to_fp16, weight = var_1747_to_fp16, x = var_1728_cast); + tensor var_1756 = const()[name = tensor("op_1756"), val = tensor([1, 1500, 16, -1])]; + tensor var_1757_cast = reshape(shape = var_1756, x = q_61_cast); + tensor const_198_to_fp16 = const()[name = tensor("const_198_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_63_cast = mul(x = var_1757_cast, y = const_198_to_fp16); + tensor var_1763 = const()[name = tensor("op_1763"), val = tensor([1, 1500, 16, -1])]; + tensor var_1764_cast = reshape(shape = var_1763, x = k_61_cast); + tensor const_199_to_fp16 = const()[name = tensor("const_199_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_63_cast = mul(x = var_1764_cast, y = const_199_to_fp16); + tensor var_1770 = const()[name = tensor("op_1770"), val = tensor([1, 1500, 16, -1])]; + tensor var_1771_cast = reshape(shape = var_1770, x = v_61_cast); + tensor var_1772 = const()[name = tensor("op_1772"), val = tensor([0, 2, 1, 3])]; + tensor qk_31_transpose_x_0 = const()[name = tensor("qk_31_transpose_x_0"), val = tensor(false)]; + tensor qk_31_transpose_y_0 = const()[name = tensor("qk_31_transpose_y_0"), val = tensor(false)]; + tensor transpose_78_perm_0 = const()[name = tensor("transpose_78_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_79_perm_0 = const()[name = tensor("transpose_79_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_129 = transpose(perm = transpose_79_perm_0, x = k_63_cast); + tensor transpose_130 = transpose(perm = transpose_78_perm_0, x = q_63_cast); + tensor qk_31_cast = matmul(transpose_x = qk_31_transpose_x_0, transpose_y = qk_31_transpose_y_0, x = transpose_130, y = transpose_129); + tensor var_1776_cast = softmax(axis = var_1711, x = qk_31_cast); + tensor var_1778_transpose_x_0 = const()[name = tensor("op_1778_transpose_x_0"), val = tensor(false)]; + tensor var_1778_transpose_y_0 = const()[name = tensor("op_1778_transpose_y_0"), val = tensor(false)]; + tensor transpose_131 = transpose(perm = var_1772, x = var_1771_cast); + tensor var_1778_cast = matmul(transpose_x = var_1778_transpose_x_0, transpose_y = var_1778_transpose_y_0, x = var_1776_cast, y = transpose_131); + tensor var_1779 = const()[name = tensor("op_1779"), val = tensor([0, 2, 1, 3])]; + tensor concat_15 = const()[name = tensor("concat_15"), val = tensor([1, 1500, 1024])]; + tensor transpose_128 = transpose(perm = var_1779, x = var_1778_cast); + tensor x_191_cast = reshape(shape = concat_15, x = transpose_128); + tensor var_1784_to_fp16 = const()[name = tensor("op_1784_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(394063744)))]; + tensor var_1785_to_fp16 = const()[name = tensor("op_1785_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(396160960)))]; + tensor var_1786_cast = linear(bias = var_1785_to_fp16, weight = var_1784_to_fp16, x = x_191_cast); + tensor x_193_cast = add(x = x_187_cast, y = var_1786_cast); + tensor var_1792_axes_0 = const()[name = tensor("op_1792_axes_0"), val = tensor([-1])]; + tensor blocks_15_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_15_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(396163072)))]; + tensor blocks_15_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_15_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(396165184)))]; + tensor var_1792_cast = layer_norm(axes = var_1792_axes_0, beta = blocks_15_mlp_ln_bias_to_fp16, epsilon = var_1717_to_fp16, gamma = blocks_15_mlp_ln_weight_to_fp16, x = x_193_cast); + tensor var_1801_to_fp16 = const()[name = tensor("op_1801_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(396167296)))]; + tensor var_1802_to_fp16 = const()[name = tensor("op_1802_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(404555968)))]; + tensor input_129_cast = linear(bias = var_1802_to_fp16, weight = var_1801_to_fp16, x = var_1792_cast); + tensor x_197_mode_0 = const()[name = tensor("x_197_mode_0"), val = tensor("EXACT")]; + tensor x_197_cast = gelu(mode = x_197_mode_0, x = input_129_cast); + tensor var_1807_to_fp16 = const()[name = tensor("op_1807_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(404564224)))]; + tensor var_1808_to_fp16 = const()[name = tensor("op_1808_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(412952896)))]; + tensor var_1809_cast = linear(bias = var_1808_to_fp16, weight = var_1807_to_fp16, x = x_197_cast); + tensor x_199_cast = add(x = x_193_cast, y = var_1809_cast); + tensor var_1818 = const()[name = tensor("op_1818"), val = tensor(-1)]; + tensor var_1835_axes_0 = const()[name = tensor("op_1835_axes_0"), val = tensor([-1])]; + tensor blocks_16_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_16_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(412955008)))]; + tensor blocks_16_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_16_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(412957120)))]; + tensor var_1824_to_fp16 = const()[name = tensor("op_1824_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1835_cast = layer_norm(axes = var_1835_axes_0, beta = blocks_16_attn_ln_bias_to_fp16, epsilon = var_1824_to_fp16, gamma = blocks_16_attn_ln_weight_to_fp16, x = x_199_cast); + tensor var_1846_to_fp16 = const()[name = tensor("op_1846_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(412959232)))]; + tensor var_1847_to_fp16 = const()[name = tensor("op_1847_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(415056448)))]; + tensor q_65_cast = linear(bias = var_1847_to_fp16, weight = var_1846_to_fp16, x = var_1835_cast); + tensor var_1850_to_fp16 = const()[name = tensor("op_1850_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(415058560)))]; + tensor k_65_bias_0_to_fp16 = const()[name = tensor("k_65_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(417155776)))]; + tensor k_65_cast = linear(bias = k_65_bias_0_to_fp16, weight = var_1850_to_fp16, x = var_1835_cast); + tensor var_1854_to_fp16 = const()[name = tensor("op_1854_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(417157888)))]; + tensor var_1855_to_fp16 = const()[name = tensor("op_1855_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(419255104)))]; + tensor v_65_cast = linear(bias = var_1855_to_fp16, weight = var_1854_to_fp16, x = var_1835_cast); + tensor var_1863 = const()[name = tensor("op_1863"), val = tensor([1, 1500, 16, -1])]; + tensor var_1864_cast = reshape(shape = var_1863, x = q_65_cast); + tensor const_200_to_fp16 = const()[name = tensor("const_200_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_67_cast = mul(x = var_1864_cast, y = const_200_to_fp16); + tensor var_1870 = const()[name = tensor("op_1870"), val = tensor([1, 1500, 16, -1])]; + tensor var_1871_cast = reshape(shape = var_1870, x = k_65_cast); + tensor const_201_to_fp16 = const()[name = tensor("const_201_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_67_cast = mul(x = var_1871_cast, y = const_201_to_fp16); + tensor var_1877 = const()[name = tensor("op_1877"), val = tensor([1, 1500, 16, -1])]; + tensor var_1878_cast = reshape(shape = var_1877, x = v_65_cast); + tensor var_1879 = const()[name = tensor("op_1879"), val = tensor([0, 2, 1, 3])]; + tensor qk_33_transpose_x_0 = const()[name = tensor("qk_33_transpose_x_0"), val = tensor(false)]; + tensor qk_33_transpose_y_0 = const()[name = tensor("qk_33_transpose_y_0"), val = tensor(false)]; + tensor transpose_80_perm_0 = const()[name = tensor("transpose_80_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_81_perm_0 = const()[name = tensor("transpose_81_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_125 = transpose(perm = transpose_81_perm_0, x = k_67_cast); + tensor transpose_126 = transpose(perm = transpose_80_perm_0, x = q_67_cast); + tensor qk_33_cast = matmul(transpose_x = qk_33_transpose_x_0, transpose_y = qk_33_transpose_y_0, x = transpose_126, y = transpose_125); + tensor var_1883_cast = softmax(axis = var_1818, x = qk_33_cast); + tensor var_1885_transpose_x_0 = const()[name = tensor("op_1885_transpose_x_0"), val = tensor(false)]; + tensor var_1885_transpose_y_0 = const()[name = tensor("op_1885_transpose_y_0"), val = tensor(false)]; + tensor transpose_127 = transpose(perm = var_1879, x = var_1878_cast); + tensor var_1885_cast = matmul(transpose_x = var_1885_transpose_x_0, transpose_y = var_1885_transpose_y_0, x = var_1883_cast, y = transpose_127); + tensor var_1886 = const()[name = tensor("op_1886"), val = tensor([0, 2, 1, 3])]; + tensor concat_16 = const()[name = tensor("concat_16"), val = tensor([1, 1500, 1024])]; + tensor transpose_124 = transpose(perm = var_1886, x = var_1885_cast); + tensor x_203_cast = reshape(shape = concat_16, x = transpose_124); + tensor var_1891_to_fp16 = const()[name = tensor("op_1891_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(419257216)))]; + tensor var_1892_to_fp16 = const()[name = tensor("op_1892_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(421354432)))]; + tensor var_1893_cast = linear(bias = var_1892_to_fp16, weight = var_1891_to_fp16, x = x_203_cast); + tensor x_205_cast = add(x = x_199_cast, y = var_1893_cast); + tensor var_1899_axes_0 = const()[name = tensor("op_1899_axes_0"), val = tensor([-1])]; + tensor blocks_16_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_16_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(421356544)))]; + tensor blocks_16_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_16_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(421358656)))]; + tensor var_1899_cast = layer_norm(axes = var_1899_axes_0, beta = blocks_16_mlp_ln_bias_to_fp16, epsilon = var_1824_to_fp16, gamma = blocks_16_mlp_ln_weight_to_fp16, x = x_205_cast); + tensor var_1908_to_fp16 = const()[name = tensor("op_1908_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(421360768)))]; + tensor var_1909_to_fp16 = const()[name = tensor("op_1909_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(429749440)))]; + tensor input_137_cast = linear(bias = var_1909_to_fp16, weight = var_1908_to_fp16, x = var_1899_cast); + tensor x_209_mode_0 = const()[name = tensor("x_209_mode_0"), val = tensor("EXACT")]; + tensor x_209_cast = gelu(mode = x_209_mode_0, x = input_137_cast); + tensor var_1914_to_fp16 = const()[name = tensor("op_1914_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(429757696)))]; + tensor var_1915_to_fp16 = const()[name = tensor("op_1915_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(438146368)))]; + tensor var_1916_cast = linear(bias = var_1915_to_fp16, weight = var_1914_to_fp16, x = x_209_cast); + tensor x_211_cast = add(x = x_205_cast, y = var_1916_cast); + tensor var_1925 = const()[name = tensor("op_1925"), val = tensor(-1)]; + tensor var_1942_axes_0 = const()[name = tensor("op_1942_axes_0"), val = tensor([-1])]; + tensor blocks_17_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_17_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(438148480)))]; + tensor blocks_17_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_17_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(438150592)))]; + tensor var_1931_to_fp16 = const()[name = tensor("op_1931_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1942_cast = layer_norm(axes = var_1942_axes_0, beta = blocks_17_attn_ln_bias_to_fp16, epsilon = var_1931_to_fp16, gamma = blocks_17_attn_ln_weight_to_fp16, x = x_211_cast); + tensor var_1953_to_fp16 = const()[name = tensor("op_1953_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(438152704)))]; + tensor var_1954_to_fp16 = const()[name = tensor("op_1954_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(440249920)))]; + tensor q_69_cast = linear(bias = var_1954_to_fp16, weight = var_1953_to_fp16, x = var_1942_cast); + tensor var_1957_to_fp16 = const()[name = tensor("op_1957_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(440252032)))]; + tensor k_69_bias_0_to_fp16 = const()[name = tensor("k_69_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(442349248)))]; + tensor k_69_cast = linear(bias = k_69_bias_0_to_fp16, weight = var_1957_to_fp16, x = var_1942_cast); + tensor var_1961_to_fp16 = const()[name = tensor("op_1961_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(442351360)))]; + tensor var_1962_to_fp16 = const()[name = tensor("op_1962_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(444448576)))]; + tensor v_69_cast = linear(bias = var_1962_to_fp16, weight = var_1961_to_fp16, x = var_1942_cast); + tensor var_1970 = const()[name = tensor("op_1970"), val = tensor([1, 1500, 16, -1])]; + tensor var_1971_cast = reshape(shape = var_1970, x = q_69_cast); + tensor const_202_to_fp16 = const()[name = tensor("const_202_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_71_cast = mul(x = var_1971_cast, y = const_202_to_fp16); + tensor var_1977 = const()[name = tensor("op_1977"), val = tensor([1, 1500, 16, -1])]; + tensor var_1978_cast = reshape(shape = var_1977, x = k_69_cast); + tensor const_203_to_fp16 = const()[name = tensor("const_203_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_71_cast = mul(x = var_1978_cast, y = const_203_to_fp16); + tensor var_1984 = const()[name = tensor("op_1984"), val = tensor([1, 1500, 16, -1])]; + tensor var_1985_cast = reshape(shape = var_1984, x = v_69_cast); + tensor var_1986 = const()[name = tensor("op_1986"), val = tensor([0, 2, 1, 3])]; + tensor qk_35_transpose_x_0 = const()[name = tensor("qk_35_transpose_x_0"), val = tensor(false)]; + tensor qk_35_transpose_y_0 = const()[name = tensor("qk_35_transpose_y_0"), val = tensor(false)]; + tensor transpose_82_perm_0 = const()[name = tensor("transpose_82_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_83_perm_0 = const()[name = tensor("transpose_83_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_121 = transpose(perm = transpose_83_perm_0, x = k_71_cast); + tensor transpose_122 = transpose(perm = transpose_82_perm_0, x = q_71_cast); + tensor qk_35_cast = matmul(transpose_x = qk_35_transpose_x_0, transpose_y = qk_35_transpose_y_0, x = transpose_122, y = transpose_121); + tensor var_1990_cast = softmax(axis = var_1925, x = qk_35_cast); + tensor var_1992_transpose_x_0 = const()[name = tensor("op_1992_transpose_x_0"), val = tensor(false)]; + tensor var_1992_transpose_y_0 = const()[name = tensor("op_1992_transpose_y_0"), val = tensor(false)]; + tensor transpose_123 = transpose(perm = var_1986, x = var_1985_cast); + tensor var_1992_cast = matmul(transpose_x = var_1992_transpose_x_0, transpose_y = var_1992_transpose_y_0, x = var_1990_cast, y = transpose_123); + tensor var_1993 = const()[name = tensor("op_1993"), val = tensor([0, 2, 1, 3])]; + tensor concat_17 = const()[name = tensor("concat_17"), val = tensor([1, 1500, 1024])]; + tensor transpose_120 = transpose(perm = var_1993, x = var_1992_cast); + tensor x_215_cast = reshape(shape = concat_17, x = transpose_120); + tensor var_1998_to_fp16 = const()[name = tensor("op_1998_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(444450688)))]; + tensor var_1999_to_fp16 = const()[name = tensor("op_1999_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(446547904)))]; + tensor var_2000_cast = linear(bias = var_1999_to_fp16, weight = var_1998_to_fp16, x = x_215_cast); + tensor x_217_cast = add(x = x_211_cast, y = var_2000_cast); + tensor var_2006_axes_0 = const()[name = tensor("op_2006_axes_0"), val = tensor([-1])]; + tensor blocks_17_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_17_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(446550016)))]; + tensor blocks_17_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_17_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(446552128)))]; + tensor var_2006_cast = layer_norm(axes = var_2006_axes_0, beta = blocks_17_mlp_ln_bias_to_fp16, epsilon = var_1931_to_fp16, gamma = blocks_17_mlp_ln_weight_to_fp16, x = x_217_cast); + tensor var_2015_to_fp16 = const()[name = tensor("op_2015_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(446554240)))]; + tensor var_2016_to_fp16 = const()[name = tensor("op_2016_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(454942912)))]; + tensor input_145_cast = linear(bias = var_2016_to_fp16, weight = var_2015_to_fp16, x = var_2006_cast); + tensor x_221_mode_0 = const()[name = tensor("x_221_mode_0"), val = tensor("EXACT")]; + tensor x_221_cast = gelu(mode = x_221_mode_0, x = input_145_cast); + tensor var_2021_to_fp16 = const()[name = tensor("op_2021_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(454951168)))]; + tensor var_2022_to_fp16 = const()[name = tensor("op_2022_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(463339840)))]; + tensor var_2023_cast = linear(bias = var_2022_to_fp16, weight = var_2021_to_fp16, x = x_221_cast); + tensor x_223_cast = add(x = x_217_cast, y = var_2023_cast); + tensor var_2032 = const()[name = tensor("op_2032"), val = tensor(-1)]; + tensor var_2049_axes_0 = const()[name = tensor("op_2049_axes_0"), val = tensor([-1])]; + tensor blocks_18_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_18_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(463341952)))]; + tensor blocks_18_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_18_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(463344064)))]; + tensor var_2038_to_fp16 = const()[name = tensor("op_2038_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2049_cast = layer_norm(axes = var_2049_axes_0, beta = blocks_18_attn_ln_bias_to_fp16, epsilon = var_2038_to_fp16, gamma = blocks_18_attn_ln_weight_to_fp16, x = x_223_cast); + tensor var_2060_to_fp16 = const()[name = tensor("op_2060_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(463346176)))]; + tensor var_2061_to_fp16 = const()[name = tensor("op_2061_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(465443392)))]; + tensor q_73_cast = linear(bias = var_2061_to_fp16, weight = var_2060_to_fp16, x = var_2049_cast); + tensor var_2064_to_fp16 = const()[name = tensor("op_2064_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(465445504)))]; + tensor k_73_bias_0_to_fp16 = const()[name = tensor("k_73_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(467542720)))]; + tensor k_73_cast = linear(bias = k_73_bias_0_to_fp16, weight = var_2064_to_fp16, x = var_2049_cast); + tensor var_2068_to_fp16 = const()[name = tensor("op_2068_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(467544832)))]; + tensor var_2069_to_fp16 = const()[name = tensor("op_2069_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(469642048)))]; + tensor v_73_cast = linear(bias = var_2069_to_fp16, weight = var_2068_to_fp16, x = var_2049_cast); + tensor var_2077 = const()[name = tensor("op_2077"), val = tensor([1, 1500, 16, -1])]; + tensor var_2078_cast = reshape(shape = var_2077, x = q_73_cast); + tensor const_204_to_fp16 = const()[name = tensor("const_204_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_75_cast = mul(x = var_2078_cast, y = const_204_to_fp16); + tensor var_2084 = const()[name = tensor("op_2084"), val = tensor([1, 1500, 16, -1])]; + tensor var_2085_cast = reshape(shape = var_2084, x = k_73_cast); + tensor const_205_to_fp16 = const()[name = tensor("const_205_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_75_cast = mul(x = var_2085_cast, y = const_205_to_fp16); + tensor var_2091 = const()[name = tensor("op_2091"), val = tensor([1, 1500, 16, -1])]; + tensor var_2092_cast = reshape(shape = var_2091, x = v_73_cast); + tensor var_2093 = const()[name = tensor("op_2093"), val = tensor([0, 2, 1, 3])]; + tensor qk_37_transpose_x_0 = const()[name = tensor("qk_37_transpose_x_0"), val = tensor(false)]; + tensor qk_37_transpose_y_0 = const()[name = tensor("qk_37_transpose_y_0"), val = tensor(false)]; + tensor transpose_84_perm_0 = const()[name = tensor("transpose_84_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_85_perm_0 = const()[name = tensor("transpose_85_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_117 = transpose(perm = transpose_85_perm_0, x = k_75_cast); + tensor transpose_118 = transpose(perm = transpose_84_perm_0, x = q_75_cast); + tensor qk_37_cast = matmul(transpose_x = qk_37_transpose_x_0, transpose_y = qk_37_transpose_y_0, x = transpose_118, y = transpose_117); + tensor var_2097_cast = softmax(axis = var_2032, x = qk_37_cast); + tensor var_2099_transpose_x_0 = const()[name = tensor("op_2099_transpose_x_0"), val = tensor(false)]; + tensor var_2099_transpose_y_0 = const()[name = tensor("op_2099_transpose_y_0"), val = tensor(false)]; + tensor transpose_119 = transpose(perm = var_2093, x = var_2092_cast); + tensor var_2099_cast = matmul(transpose_x = var_2099_transpose_x_0, transpose_y = var_2099_transpose_y_0, x = var_2097_cast, y = transpose_119); + tensor var_2100 = const()[name = tensor("op_2100"), val = tensor([0, 2, 1, 3])]; + tensor concat_18 = const()[name = tensor("concat_18"), val = tensor([1, 1500, 1024])]; + tensor transpose_116 = transpose(perm = var_2100, x = var_2099_cast); + tensor x_227_cast = reshape(shape = concat_18, x = transpose_116); + tensor var_2105_to_fp16 = const()[name = tensor("op_2105_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(469644160)))]; + tensor var_2106_to_fp16 = const()[name = tensor("op_2106_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(471741376)))]; + tensor var_2107_cast = linear(bias = var_2106_to_fp16, weight = var_2105_to_fp16, x = x_227_cast); + tensor x_229_cast = add(x = x_223_cast, y = var_2107_cast); + tensor var_2113_axes_0 = const()[name = tensor("op_2113_axes_0"), val = tensor([-1])]; + tensor blocks_18_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_18_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(471743488)))]; + tensor blocks_18_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_18_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(471745600)))]; + tensor var_2113_cast = layer_norm(axes = var_2113_axes_0, beta = blocks_18_mlp_ln_bias_to_fp16, epsilon = var_2038_to_fp16, gamma = blocks_18_mlp_ln_weight_to_fp16, x = x_229_cast); + tensor var_2122_to_fp16 = const()[name = tensor("op_2122_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(471747712)))]; + tensor var_2123_to_fp16 = const()[name = tensor("op_2123_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(480136384)))]; + tensor input_153_cast = linear(bias = var_2123_to_fp16, weight = var_2122_to_fp16, x = var_2113_cast); + tensor x_233_mode_0 = const()[name = tensor("x_233_mode_0"), val = tensor("EXACT")]; + tensor x_233_cast = gelu(mode = x_233_mode_0, x = input_153_cast); + tensor var_2128_to_fp16 = const()[name = tensor("op_2128_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(480144640)))]; + tensor var_2129_to_fp16 = const()[name = tensor("op_2129_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(488533312)))]; + tensor var_2130_cast = linear(bias = var_2129_to_fp16, weight = var_2128_to_fp16, x = x_233_cast); + tensor x_235_cast = add(x = x_229_cast, y = var_2130_cast); + tensor var_2139 = const()[name = tensor("op_2139"), val = tensor(-1)]; + tensor var_2156_axes_0 = const()[name = tensor("op_2156_axes_0"), val = tensor([-1])]; + tensor blocks_19_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_19_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(488535424)))]; + tensor blocks_19_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_19_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(488537536)))]; + tensor var_2145_to_fp16 = const()[name = tensor("op_2145_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2156_cast = layer_norm(axes = var_2156_axes_0, beta = blocks_19_attn_ln_bias_to_fp16, epsilon = var_2145_to_fp16, gamma = blocks_19_attn_ln_weight_to_fp16, x = x_235_cast); + tensor var_2167_to_fp16 = const()[name = tensor("op_2167_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(488539648)))]; + tensor var_2168_to_fp16 = const()[name = tensor("op_2168_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(490636864)))]; + tensor q_77_cast = linear(bias = var_2168_to_fp16, weight = var_2167_to_fp16, x = var_2156_cast); + tensor var_2171_to_fp16 = const()[name = tensor("op_2171_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(490638976)))]; + tensor k_77_bias_0_to_fp16 = const()[name = tensor("k_77_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(492736192)))]; + tensor k_77_cast = linear(bias = k_77_bias_0_to_fp16, weight = var_2171_to_fp16, x = var_2156_cast); + tensor var_2175_to_fp16 = const()[name = tensor("op_2175_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(492738304)))]; + tensor var_2176_to_fp16 = const()[name = tensor("op_2176_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(494835520)))]; + tensor v_77_cast = linear(bias = var_2176_to_fp16, weight = var_2175_to_fp16, x = var_2156_cast); + tensor var_2184 = const()[name = tensor("op_2184"), val = tensor([1, 1500, 16, -1])]; + tensor var_2185_cast = reshape(shape = var_2184, x = q_77_cast); + tensor const_206_to_fp16 = const()[name = tensor("const_206_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_79_cast = mul(x = var_2185_cast, y = const_206_to_fp16); + tensor var_2191 = const()[name = tensor("op_2191"), val = tensor([1, 1500, 16, -1])]; + tensor var_2192_cast = reshape(shape = var_2191, x = k_77_cast); + tensor const_207_to_fp16 = const()[name = tensor("const_207_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_79_cast = mul(x = var_2192_cast, y = const_207_to_fp16); + tensor var_2198 = const()[name = tensor("op_2198"), val = tensor([1, 1500, 16, -1])]; + tensor var_2199_cast = reshape(shape = var_2198, x = v_77_cast); + tensor var_2200 = const()[name = tensor("op_2200"), val = tensor([0, 2, 1, 3])]; + tensor qk_39_transpose_x_0 = const()[name = tensor("qk_39_transpose_x_0"), val = tensor(false)]; + tensor qk_39_transpose_y_0 = const()[name = tensor("qk_39_transpose_y_0"), val = tensor(false)]; + tensor transpose_86_perm_0 = const()[name = tensor("transpose_86_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_87_perm_0 = const()[name = tensor("transpose_87_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_113 = transpose(perm = transpose_87_perm_0, x = k_79_cast); + tensor transpose_114 = transpose(perm = transpose_86_perm_0, x = q_79_cast); + tensor qk_39_cast = matmul(transpose_x = qk_39_transpose_x_0, transpose_y = qk_39_transpose_y_0, x = transpose_114, y = transpose_113); + tensor var_2204_cast = softmax(axis = var_2139, x = qk_39_cast); + tensor var_2206_transpose_x_0 = const()[name = tensor("op_2206_transpose_x_0"), val = tensor(false)]; + tensor var_2206_transpose_y_0 = const()[name = tensor("op_2206_transpose_y_0"), val = tensor(false)]; + tensor transpose_115 = transpose(perm = var_2200, x = var_2199_cast); + tensor var_2206_cast = matmul(transpose_x = var_2206_transpose_x_0, transpose_y = var_2206_transpose_y_0, x = var_2204_cast, y = transpose_115); + tensor var_2207 = const()[name = tensor("op_2207"), val = tensor([0, 2, 1, 3])]; + tensor concat_19 = const()[name = tensor("concat_19"), val = tensor([1, 1500, 1024])]; + tensor transpose_112 = transpose(perm = var_2207, x = var_2206_cast); + tensor x_239_cast = reshape(shape = concat_19, x = transpose_112); + tensor var_2212_to_fp16 = const()[name = tensor("op_2212_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(494837632)))]; + tensor var_2213_to_fp16 = const()[name = tensor("op_2213_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496934848)))]; + tensor var_2214_cast = linear(bias = var_2213_to_fp16, weight = var_2212_to_fp16, x = x_239_cast); + tensor x_241_cast = add(x = x_235_cast, y = var_2214_cast); + tensor var_2220_axes_0 = const()[name = tensor("op_2220_axes_0"), val = tensor([-1])]; + tensor blocks_19_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_19_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496936960)))]; + tensor blocks_19_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_19_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496939072)))]; + tensor var_2220_cast = layer_norm(axes = var_2220_axes_0, beta = blocks_19_mlp_ln_bias_to_fp16, epsilon = var_2145_to_fp16, gamma = blocks_19_mlp_ln_weight_to_fp16, x = x_241_cast); + tensor var_2229_to_fp16 = const()[name = tensor("op_2229_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496941184)))]; + tensor var_2230_to_fp16 = const()[name = tensor("op_2230_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(505329856)))]; + tensor input_161_cast = linear(bias = var_2230_to_fp16, weight = var_2229_to_fp16, x = var_2220_cast); + tensor x_245_mode_0 = const()[name = tensor("x_245_mode_0"), val = tensor("EXACT")]; + tensor x_245_cast = gelu(mode = x_245_mode_0, x = input_161_cast); + tensor var_2235_to_fp16 = const()[name = tensor("op_2235_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(505338112)))]; + tensor var_2236_to_fp16 = const()[name = tensor("op_2236_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(513726784)))]; + tensor var_2237_cast = linear(bias = var_2236_to_fp16, weight = var_2235_to_fp16, x = x_245_cast); + tensor x_247_cast = add(x = x_241_cast, y = var_2237_cast); + tensor var_2246 = const()[name = tensor("op_2246"), val = tensor(-1)]; + tensor var_2263_axes_0 = const()[name = tensor("op_2263_axes_0"), val = tensor([-1])]; + tensor blocks_20_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_20_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(513728896)))]; + tensor blocks_20_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_20_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(513731008)))]; + tensor var_2252_to_fp16 = const()[name = tensor("op_2252_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2263_cast = layer_norm(axes = var_2263_axes_0, beta = blocks_20_attn_ln_bias_to_fp16, epsilon = var_2252_to_fp16, gamma = blocks_20_attn_ln_weight_to_fp16, x = x_247_cast); + tensor var_2274_to_fp16 = const()[name = tensor("op_2274_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(513733120)))]; + tensor var_2275_to_fp16 = const()[name = tensor("op_2275_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(515830336)))]; + tensor q_81_cast = linear(bias = var_2275_to_fp16, weight = var_2274_to_fp16, x = var_2263_cast); + tensor var_2278_to_fp16 = const()[name = tensor("op_2278_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(515832448)))]; + tensor k_81_bias_0_to_fp16 = const()[name = tensor("k_81_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(517929664)))]; + tensor k_81_cast = linear(bias = k_81_bias_0_to_fp16, weight = var_2278_to_fp16, x = var_2263_cast); + tensor var_2282_to_fp16 = const()[name = tensor("op_2282_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(517931776)))]; + tensor var_2283_to_fp16 = const()[name = tensor("op_2283_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(520028992)))]; + tensor v_81_cast = linear(bias = var_2283_to_fp16, weight = var_2282_to_fp16, x = var_2263_cast); + tensor var_2291 = const()[name = tensor("op_2291"), val = tensor([1, 1500, 16, -1])]; + tensor var_2292_cast = reshape(shape = var_2291, x = q_81_cast); + tensor const_208_to_fp16 = const()[name = tensor("const_208_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_83_cast = mul(x = var_2292_cast, y = const_208_to_fp16); + tensor var_2298 = const()[name = tensor("op_2298"), val = tensor([1, 1500, 16, -1])]; + tensor var_2299_cast = reshape(shape = var_2298, x = k_81_cast); + tensor const_209_to_fp16 = const()[name = tensor("const_209_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_83_cast = mul(x = var_2299_cast, y = const_209_to_fp16); + tensor var_2305 = const()[name = tensor("op_2305"), val = tensor([1, 1500, 16, -1])]; + tensor var_2306_cast = reshape(shape = var_2305, x = v_81_cast); + tensor var_2307 = const()[name = tensor("op_2307"), val = tensor([0, 2, 1, 3])]; + tensor qk_41_transpose_x_0 = const()[name = tensor("qk_41_transpose_x_0"), val = tensor(false)]; + tensor qk_41_transpose_y_0 = const()[name = tensor("qk_41_transpose_y_0"), val = tensor(false)]; + tensor transpose_88_perm_0 = const()[name = tensor("transpose_88_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_89_perm_0 = const()[name = tensor("transpose_89_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_109 = transpose(perm = transpose_89_perm_0, x = k_83_cast); + tensor transpose_110 = transpose(perm = transpose_88_perm_0, x = q_83_cast); + tensor qk_41_cast = matmul(transpose_x = qk_41_transpose_x_0, transpose_y = qk_41_transpose_y_0, x = transpose_110, y = transpose_109); + tensor var_2311_cast = softmax(axis = var_2246, x = qk_41_cast); + tensor var_2313_transpose_x_0 = const()[name = tensor("op_2313_transpose_x_0"), val = tensor(false)]; + tensor var_2313_transpose_y_0 = const()[name = tensor("op_2313_transpose_y_0"), val = tensor(false)]; + tensor transpose_111 = transpose(perm = var_2307, x = var_2306_cast); + tensor var_2313_cast = matmul(transpose_x = var_2313_transpose_x_0, transpose_y = var_2313_transpose_y_0, x = var_2311_cast, y = transpose_111); + tensor var_2314 = const()[name = tensor("op_2314"), val = tensor([0, 2, 1, 3])]; + tensor concat_20 = const()[name = tensor("concat_20"), val = tensor([1, 1500, 1024])]; + tensor transpose_108 = transpose(perm = var_2314, x = var_2313_cast); + tensor x_251_cast = reshape(shape = concat_20, x = transpose_108); + tensor var_2319_to_fp16 = const()[name = tensor("op_2319_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(520031104)))]; + tensor var_2320_to_fp16 = const()[name = tensor("op_2320_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(522128320)))]; + tensor var_2321_cast = linear(bias = var_2320_to_fp16, weight = var_2319_to_fp16, x = x_251_cast); + tensor x_253_cast = add(x = x_247_cast, y = var_2321_cast); + tensor var_2327_axes_0 = const()[name = tensor("op_2327_axes_0"), val = tensor([-1])]; + tensor blocks_20_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_20_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(522130432)))]; + tensor blocks_20_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_20_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(522132544)))]; + tensor var_2327_cast = layer_norm(axes = var_2327_axes_0, beta = blocks_20_mlp_ln_bias_to_fp16, epsilon = var_2252_to_fp16, gamma = blocks_20_mlp_ln_weight_to_fp16, x = x_253_cast); + tensor var_2336_to_fp16 = const()[name = tensor("op_2336_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(522134656)))]; + tensor var_2337_to_fp16 = const()[name = tensor("op_2337_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(530523328)))]; + tensor input_169_cast = linear(bias = var_2337_to_fp16, weight = var_2336_to_fp16, x = var_2327_cast); + tensor x_257_mode_0 = const()[name = tensor("x_257_mode_0"), val = tensor("EXACT")]; + tensor x_257_cast = gelu(mode = x_257_mode_0, x = input_169_cast); + tensor var_2342_to_fp16 = const()[name = tensor("op_2342_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(530531584)))]; + tensor var_2343_to_fp16 = const()[name = tensor("op_2343_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(538920256)))]; + tensor var_2344_cast = linear(bias = var_2343_to_fp16, weight = var_2342_to_fp16, x = x_257_cast); + tensor x_259_cast = add(x = x_253_cast, y = var_2344_cast); + tensor var_2353 = const()[name = tensor("op_2353"), val = tensor(-1)]; + tensor var_2370_axes_0 = const()[name = tensor("op_2370_axes_0"), val = tensor([-1])]; + tensor blocks_21_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_21_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(538922368)))]; + tensor blocks_21_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_21_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(538924480)))]; + tensor var_2359_to_fp16 = const()[name = tensor("op_2359_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2370_cast = layer_norm(axes = var_2370_axes_0, beta = blocks_21_attn_ln_bias_to_fp16, epsilon = var_2359_to_fp16, gamma = blocks_21_attn_ln_weight_to_fp16, x = x_259_cast); + tensor var_2381_to_fp16 = const()[name = tensor("op_2381_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(538926592)))]; + tensor var_2382_to_fp16 = const()[name = tensor("op_2382_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(541023808)))]; + tensor q_85_cast = linear(bias = var_2382_to_fp16, weight = var_2381_to_fp16, x = var_2370_cast); + tensor var_2385_to_fp16 = const()[name = tensor("op_2385_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(541025920)))]; + tensor k_85_bias_0_to_fp16 = const()[name = tensor("k_85_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(543123136)))]; + tensor k_85_cast = linear(bias = k_85_bias_0_to_fp16, weight = var_2385_to_fp16, x = var_2370_cast); + tensor var_2389_to_fp16 = const()[name = tensor("op_2389_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(543125248)))]; + tensor var_2390_to_fp16 = const()[name = tensor("op_2390_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(545222464)))]; + tensor v_85_cast = linear(bias = var_2390_to_fp16, weight = var_2389_to_fp16, x = var_2370_cast); + tensor var_2398 = const()[name = tensor("op_2398"), val = tensor([1, 1500, 16, -1])]; + tensor var_2399_cast = reshape(shape = var_2398, x = q_85_cast); + tensor const_210_to_fp16 = const()[name = tensor("const_210_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_87_cast = mul(x = var_2399_cast, y = const_210_to_fp16); + tensor var_2405 = const()[name = tensor("op_2405"), val = tensor([1, 1500, 16, -1])]; + tensor var_2406_cast = reshape(shape = var_2405, x = k_85_cast); + tensor const_211_to_fp16 = const()[name = tensor("const_211_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_87_cast = mul(x = var_2406_cast, y = const_211_to_fp16); + tensor var_2412 = const()[name = tensor("op_2412"), val = tensor([1, 1500, 16, -1])]; + tensor var_2413_cast = reshape(shape = var_2412, x = v_85_cast); + tensor var_2414 = const()[name = tensor("op_2414"), val = tensor([0, 2, 1, 3])]; + tensor qk_43_transpose_x_0 = const()[name = tensor("qk_43_transpose_x_0"), val = tensor(false)]; + tensor qk_43_transpose_y_0 = const()[name = tensor("qk_43_transpose_y_0"), val = tensor(false)]; + tensor transpose_90_perm_0 = const()[name = tensor("transpose_90_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_91_perm_0 = const()[name = tensor("transpose_91_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_105 = transpose(perm = transpose_91_perm_0, x = k_87_cast); + tensor transpose_106 = transpose(perm = transpose_90_perm_0, x = q_87_cast); + tensor qk_43_cast = matmul(transpose_x = qk_43_transpose_x_0, transpose_y = qk_43_transpose_y_0, x = transpose_106, y = transpose_105); + tensor var_2418_cast = softmax(axis = var_2353, x = qk_43_cast); + tensor var_2420_transpose_x_0 = const()[name = tensor("op_2420_transpose_x_0"), val = tensor(false)]; + tensor var_2420_transpose_y_0 = const()[name = tensor("op_2420_transpose_y_0"), val = tensor(false)]; + tensor transpose_107 = transpose(perm = var_2414, x = var_2413_cast); + tensor var_2420_cast = matmul(transpose_x = var_2420_transpose_x_0, transpose_y = var_2420_transpose_y_0, x = var_2418_cast, y = transpose_107); + tensor var_2421 = const()[name = tensor("op_2421"), val = tensor([0, 2, 1, 3])]; + tensor concat_21 = const()[name = tensor("concat_21"), val = tensor([1, 1500, 1024])]; + tensor transpose_104 = transpose(perm = var_2421, x = var_2420_cast); + tensor x_263_cast = reshape(shape = concat_21, x = transpose_104); + tensor var_2426_to_fp16 = const()[name = tensor("op_2426_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(545224576)))]; + tensor var_2427_to_fp16 = const()[name = tensor("op_2427_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(547321792)))]; + tensor var_2428_cast = linear(bias = var_2427_to_fp16, weight = var_2426_to_fp16, x = x_263_cast); + tensor x_265_cast = add(x = x_259_cast, y = var_2428_cast); + tensor var_2434_axes_0 = const()[name = tensor("op_2434_axes_0"), val = tensor([-1])]; + tensor blocks_21_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_21_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(547323904)))]; + tensor blocks_21_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_21_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(547326016)))]; + tensor var_2434_cast = layer_norm(axes = var_2434_axes_0, beta = blocks_21_mlp_ln_bias_to_fp16, epsilon = var_2359_to_fp16, gamma = blocks_21_mlp_ln_weight_to_fp16, x = x_265_cast); + tensor var_2443_to_fp16 = const()[name = tensor("op_2443_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(547328128)))]; + tensor var_2444_to_fp16 = const()[name = tensor("op_2444_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(555716800)))]; + tensor input_177_cast = linear(bias = var_2444_to_fp16, weight = var_2443_to_fp16, x = var_2434_cast); + tensor x_269_mode_0 = const()[name = tensor("x_269_mode_0"), val = tensor("EXACT")]; + tensor x_269_cast = gelu(mode = x_269_mode_0, x = input_177_cast); + tensor var_2449_to_fp16 = const()[name = tensor("op_2449_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(555725056)))]; + tensor var_2450_to_fp16 = const()[name = tensor("op_2450_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(564113728)))]; + tensor var_2451_cast = linear(bias = var_2450_to_fp16, weight = var_2449_to_fp16, x = x_269_cast); + tensor x_271_cast = add(x = x_265_cast, y = var_2451_cast); + tensor var_2460 = const()[name = tensor("op_2460"), val = tensor(-1)]; + tensor var_2477_axes_0 = const()[name = tensor("op_2477_axes_0"), val = tensor([-1])]; + tensor blocks_22_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_22_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(564115840)))]; + tensor blocks_22_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_22_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(564117952)))]; + tensor var_2466_to_fp16 = const()[name = tensor("op_2466_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2477_cast = layer_norm(axes = var_2477_axes_0, beta = blocks_22_attn_ln_bias_to_fp16, epsilon = var_2466_to_fp16, gamma = blocks_22_attn_ln_weight_to_fp16, x = x_271_cast); + tensor var_2488_to_fp16 = const()[name = tensor("op_2488_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(564120064)))]; + tensor var_2489_to_fp16 = const()[name = tensor("op_2489_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(566217280)))]; + tensor q_89_cast = linear(bias = var_2489_to_fp16, weight = var_2488_to_fp16, x = var_2477_cast); + tensor var_2492_to_fp16 = const()[name = tensor("op_2492_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(566219392)))]; + tensor k_89_bias_0_to_fp16 = const()[name = tensor("k_89_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568316608)))]; + tensor k_89_cast = linear(bias = k_89_bias_0_to_fp16, weight = var_2492_to_fp16, x = var_2477_cast); + tensor var_2496_to_fp16 = const()[name = tensor("op_2496_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568318720)))]; + tensor var_2497_to_fp16 = const()[name = tensor("op_2497_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(570415936)))]; + tensor v_89_cast = linear(bias = var_2497_to_fp16, weight = var_2496_to_fp16, x = var_2477_cast); + tensor var_2505 = const()[name = tensor("op_2505"), val = tensor([1, 1500, 16, -1])]; + tensor var_2506_cast = reshape(shape = var_2505, x = q_89_cast); + tensor const_212_to_fp16 = const()[name = tensor("const_212_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_91_cast = mul(x = var_2506_cast, y = const_212_to_fp16); + tensor var_2512 = const()[name = tensor("op_2512"), val = tensor([1, 1500, 16, -1])]; + tensor var_2513_cast = reshape(shape = var_2512, x = k_89_cast); + tensor const_213_to_fp16 = const()[name = tensor("const_213_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_91_cast = mul(x = var_2513_cast, y = const_213_to_fp16); + tensor var_2519 = const()[name = tensor("op_2519"), val = tensor([1, 1500, 16, -1])]; + tensor var_2520_cast = reshape(shape = var_2519, x = v_89_cast); + tensor var_2521 = const()[name = tensor("op_2521"), val = tensor([0, 2, 1, 3])]; + tensor qk_45_transpose_x_0 = const()[name = tensor("qk_45_transpose_x_0"), val = tensor(false)]; + tensor qk_45_transpose_y_0 = const()[name = tensor("qk_45_transpose_y_0"), val = tensor(false)]; + tensor transpose_92_perm_0 = const()[name = tensor("transpose_92_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_93_perm_0 = const()[name = tensor("transpose_93_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_101 = transpose(perm = transpose_93_perm_0, x = k_91_cast); + tensor transpose_102 = transpose(perm = transpose_92_perm_0, x = q_91_cast); + tensor qk_45_cast = matmul(transpose_x = qk_45_transpose_x_0, transpose_y = qk_45_transpose_y_0, x = transpose_102, y = transpose_101); + tensor var_2525_cast = softmax(axis = var_2460, x = qk_45_cast); + tensor var_2527_transpose_x_0 = const()[name = tensor("op_2527_transpose_x_0"), val = tensor(false)]; + tensor var_2527_transpose_y_0 = const()[name = tensor("op_2527_transpose_y_0"), val = tensor(false)]; + tensor transpose_103 = transpose(perm = var_2521, x = var_2520_cast); + tensor var_2527_cast = matmul(transpose_x = var_2527_transpose_x_0, transpose_y = var_2527_transpose_y_0, x = var_2525_cast, y = transpose_103); + tensor var_2528 = const()[name = tensor("op_2528"), val = tensor([0, 2, 1, 3])]; + tensor concat_22 = const()[name = tensor("concat_22"), val = tensor([1, 1500, 1024])]; + tensor transpose_100 = transpose(perm = var_2528, x = var_2527_cast); + tensor x_275_cast = reshape(shape = concat_22, x = transpose_100); + tensor var_2533_to_fp16 = const()[name = tensor("op_2533_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(570418048)))]; + tensor var_2534_to_fp16 = const()[name = tensor("op_2534_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(572515264)))]; + tensor var_2535_cast = linear(bias = var_2534_to_fp16, weight = var_2533_to_fp16, x = x_275_cast); + tensor x_277_cast = add(x = x_271_cast, y = var_2535_cast); + tensor var_2541_axes_0 = const()[name = tensor("op_2541_axes_0"), val = tensor([-1])]; + tensor blocks_22_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_22_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(572517376)))]; + tensor blocks_22_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_22_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(572519488)))]; + tensor var_2541_cast = layer_norm(axes = var_2541_axes_0, beta = blocks_22_mlp_ln_bias_to_fp16, epsilon = var_2466_to_fp16, gamma = blocks_22_mlp_ln_weight_to_fp16, x = x_277_cast); + tensor var_2550_to_fp16 = const()[name = tensor("op_2550_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(572521600)))]; + tensor var_2551_to_fp16 = const()[name = tensor("op_2551_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(580910272)))]; + tensor input_185_cast = linear(bias = var_2551_to_fp16, weight = var_2550_to_fp16, x = var_2541_cast); + tensor x_281_mode_0 = const()[name = tensor("x_281_mode_0"), val = tensor("EXACT")]; + tensor x_281_cast = gelu(mode = x_281_mode_0, x = input_185_cast); + tensor var_2556_to_fp16 = const()[name = tensor("op_2556_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(580918528)))]; + tensor var_2557_to_fp16 = const()[name = tensor("op_2557_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(589307200)))]; + tensor var_2558_cast = linear(bias = var_2557_to_fp16, weight = var_2556_to_fp16, x = x_281_cast); + tensor x_283_cast = add(x = x_277_cast, y = var_2558_cast); + tensor var_2567 = const()[name = tensor("op_2567"), val = tensor(-1)]; + tensor var_2584_axes_0 = const()[name = tensor("op_2584_axes_0"), val = tensor([-1])]; + tensor blocks_23_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_23_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(589309312)))]; + tensor blocks_23_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_23_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(589311424)))]; + tensor var_2573_to_fp16 = const()[name = tensor("op_2573_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2584_cast = layer_norm(axes = var_2584_axes_0, beta = blocks_23_attn_ln_bias_to_fp16, epsilon = var_2573_to_fp16, gamma = blocks_23_attn_ln_weight_to_fp16, x = x_283_cast); + tensor var_2595_to_fp16 = const()[name = tensor("op_2595_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(589313536)))]; + tensor var_2596_to_fp16 = const()[name = tensor("op_2596_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(591410752)))]; + tensor q_93_cast = linear(bias = var_2596_to_fp16, weight = var_2595_to_fp16, x = var_2584_cast); + tensor var_2599_to_fp16 = const()[name = tensor("op_2599_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(591412864)))]; + tensor k_93_bias_0_to_fp16 = const()[name = tensor("k_93_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(593510080)))]; + tensor k_93_cast = linear(bias = k_93_bias_0_to_fp16, weight = var_2599_to_fp16, x = var_2584_cast); + tensor var_2603_to_fp16 = const()[name = tensor("op_2603_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(593512192)))]; + tensor var_2604_to_fp16 = const()[name = tensor("op_2604_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(595609408)))]; + tensor v_93_cast = linear(bias = var_2604_to_fp16, weight = var_2603_to_fp16, x = var_2584_cast); + tensor var_2612 = const()[name = tensor("op_2612"), val = tensor([1, 1500, 16, -1])]; + tensor var_2613_cast = reshape(shape = var_2612, x = q_93_cast); + tensor const_214_to_fp16 = const()[name = tensor("const_214_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_cast = mul(x = var_2613_cast, y = const_214_to_fp16); + tensor var_2619 = const()[name = tensor("op_2619"), val = tensor([1, 1500, 16, -1])]; + tensor var_2620_cast = reshape(shape = var_2619, x = k_93_cast); + tensor const_215_to_fp16 = const()[name = tensor("const_215_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_cast = mul(x = var_2620_cast, y = const_215_to_fp16); + tensor var_2626 = const()[name = tensor("op_2626"), val = tensor([1, 1500, 16, -1])]; + tensor var_2627_cast = reshape(shape = var_2626, x = v_93_cast); + tensor var_2628 = const()[name = tensor("op_2628"), val = tensor([0, 2, 1, 3])]; + tensor qk_transpose_x_0 = const()[name = tensor("qk_transpose_x_0"), val = tensor(false)]; + tensor qk_transpose_y_0 = const()[name = tensor("qk_transpose_y_0"), val = tensor(false)]; + tensor transpose_94_perm_0 = const()[name = tensor("transpose_94_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_95_perm_0 = const()[name = tensor("transpose_95_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_97 = transpose(perm = transpose_95_perm_0, x = k_cast); + tensor transpose_98 = transpose(perm = transpose_94_perm_0, x = q_cast); + tensor qk_cast = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_98, y = transpose_97); + tensor var_2632_cast = softmax(axis = var_2567, x = qk_cast); + tensor var_2634_transpose_x_0 = const()[name = tensor("op_2634_transpose_x_0"), val = tensor(false)]; + tensor var_2634_transpose_y_0 = const()[name = tensor("op_2634_transpose_y_0"), val = tensor(false)]; + tensor transpose_99 = transpose(perm = var_2628, x = var_2627_cast); + tensor var_2634_cast = matmul(transpose_x = var_2634_transpose_x_0, transpose_y = var_2634_transpose_y_0, x = var_2632_cast, y = transpose_99); + tensor var_2635 = const()[name = tensor("op_2635"), val = tensor([0, 2, 1, 3])]; + tensor concat_23 = const()[name = tensor("concat_23"), val = tensor([1, 1500, 1024])]; + tensor transpose_96 = transpose(perm = var_2635, x = var_2634_cast); + tensor x_287_cast = reshape(shape = concat_23, x = transpose_96); + tensor var_2640_to_fp16 = const()[name = tensor("op_2640_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(595611520)))]; + tensor var_2641_to_fp16 = const()[name = tensor("op_2641_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(597708736)))]; + tensor var_2642_cast = linear(bias = var_2641_to_fp16, weight = var_2640_to_fp16, x = x_287_cast); + tensor x_289_cast = add(x = x_283_cast, y = var_2642_cast); + tensor var_2648_axes_0 = const()[name = tensor("op_2648_axes_0"), val = tensor([-1])]; + tensor blocks_23_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_23_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(597710848)))]; + tensor blocks_23_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_23_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(597712960)))]; + tensor var_2648_cast = layer_norm(axes = var_2648_axes_0, beta = blocks_23_mlp_ln_bias_to_fp16, epsilon = var_2573_to_fp16, gamma = blocks_23_mlp_ln_weight_to_fp16, x = x_289_cast); + tensor var_2657_to_fp16 = const()[name = tensor("op_2657_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(597715072)))]; + tensor var_2658_to_fp16 = const()[name = tensor("op_2658_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(606103744)))]; + tensor input_193_cast = linear(bias = var_2658_to_fp16, weight = var_2657_to_fp16, x = var_2648_cast); + tensor x_293_mode_0 = const()[name = tensor("x_293_mode_0"), val = tensor("EXACT")]; + tensor x_293_cast = gelu(mode = x_293_mode_0, x = input_193_cast); + tensor var_2663_to_fp16 = const()[name = tensor("op_2663_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(606112000)))]; + tensor var_2664_to_fp16 = const()[name = tensor("op_2664_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614500672)))]; + tensor var_2665_cast = linear(bias = var_2664_to_fp16, weight = var_2663_to_fp16, x = x_293_cast); + tensor x_cast = add(x = x_289_cast, y = var_2665_cast); + tensor var_2678_axes_0 = const()[name = tensor("op_2678_axes_0"), val = tensor([-1])]; + tensor ln_post_weight_to_fp16 = const()[name = tensor("ln_post_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614502784)))]; + tensor ln_post_bias_to_fp16 = const()[name = tensor("ln_post_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614504896)))]; + tensor var_2669_to_fp16 = const()[name = tensor("op_2669_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2678_cast = layer_norm(axes = var_2678_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_2669_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast); + tensor var_2678_cast_to_fp32_dtype_0 = const()[name = tensor("op_2678_cast_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor output = cast(dtype = var_2678_cast_to_fp32_dtype_0, x = var_2678_cast); + } -> (output); +} \ No newline at end of file diff --git a/ggml-medium-encoder.mlmodelc/weights/weight.bin b/ggml-medium-encoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..fb9924fc07bf1eae79440b58362feaa6a60639c7 --- /dev/null +++ b/ggml-medium-encoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6a188b0e4e3109f28f38f1f47ea2497ffe623923419df8e1ae12cb5f809a1815 +size 614507008 diff --git a/ggml-medium.en-encoder.mlmodelc.zip b/ggml-medium.en-encoder.mlmodelc.zip index b9ab3229626a5850377c38b5a66d5984c53e9425..230301f4917019de25b65fc6ed800989377c41f3 100644 --- a/ggml-medium.en-encoder.mlmodelc.zip +++ b/ggml-medium.en-encoder.mlmodelc.zip @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:cdc44fee3c62b5743913e3147ed75f4e8ecfb52dd7a0f0f7387094b406ff0ee6 +oid sha256:57cf30e4542de7ce7b798573d99e3198ebc435c2fd071672f9fd5c71d2984b99 size 566993085 diff --git a/ggml-medium.en-encoder.mlmodelc/analytics/coremldata.bin b/ggml-medium.en-encoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..cfebf363e7217ce13431621718bc293f474504b9 --- /dev/null +++ b/ggml-medium.en-encoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:adbe456375e7eb3407732a426ecb65bbda86860e4aa801f3a696b70b8a533cdd +size 207 diff --git a/ggml-medium.en-encoder.mlmodelc/coremldata.bin b/ggml-medium.en-encoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..c732ed7b8caeac3c15acda1c2836f3e1f525a5a8 --- /dev/null +++ b/ggml-medium.en-encoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:05fe28591b40616fa0c34ad7b853133623f5300923ec812acb11459c411acf3b +size 149 diff --git a/ggml-medium.en-encoder.mlmodelc/metadata.json b/ggml-medium.en-encoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..a83f08a4fb2b982fbeeece957015e7401dfa6f10 --- /dev/null +++ b/ggml-medium.en-encoder.mlmodelc/metadata.json @@ -0,0 +1,64 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float32", + "formattedType" : "MultiArray (Float32)", + "shortDescription" : "", + "shape" : "[]", + "name" : "output", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 6, + "mlProgramOperationTypeHistogram" : { + "Linear" : 144, + "Matmul" : 48, + "Cast" : 2, + "Conv" : 2, + "Softmax" : 24, + "Add" : 49, + "LayerNorm" : 49, + "Mul" : 48, + "Transpose" : 97, + "Gelu" : 26, + "Reshape" : 96 + }, + "computePrecision" : "Mixed (Float16, Float32, Int32)", + "isUpdatable" : "0", + "availability" : { + "macOS" : "12.0", + "tvOS" : "15.0", + "watchOS" : "8.0", + "iOS" : "15.0", + "macCatalyst" : "15.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float32", + "formattedType" : "MultiArray (Float32 1 × 80 × 3000)", + "shortDescription" : "", + "shape" : "[1, 80, 3000]", + "name" : "logmel_data", + "type" : "MultiArray" + } + ], + "generatedClassName" : "coreml_encoder_medium_en", + "method" : "predict" + } +] \ No newline at end of file diff --git a/ggml-medium.en-encoder.mlmodelc/model.mil b/ggml-medium.en-encoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..4771e10615202abab5bfc7b155ba0872539682da --- /dev/null +++ b/ggml-medium.en-encoder.mlmodelc/model.mil @@ -0,0 +1,1455 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "4.28.4"}, {"coremlc-version", "1436.100.10"}})] +{ + func main(tensor logmel_data) { + tensor var_56 = const()[name = tensor("op_56"), val = tensor(1)]; + tensor var_64 = const()[name = tensor("op_64"), val = tensor([1])]; + tensor var_66 = const()[name = tensor("op_66"), val = tensor([1])]; + tensor var_68_pad_type_0 = const()[name = tensor("op_68_pad_type_0"), val = tensor("custom")]; + tensor var_68_pad_0 = const()[name = tensor("op_68_pad_0"), val = tensor([1, 1])]; + tensor logmel_data_to_fp16_dtype_0 = const()[name = tensor("logmel_data_to_fp16_dtype_0"), val = tensor("fp16")]; + tensor weight_3_to_fp16 = const()[name = tensor("weight_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor bias_3_to_fp16 = const()[name = tensor("bias_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(491648)))]; + tensor cast_727 = cast(dtype = logmel_data_to_fp16_dtype_0, x = logmel_data); + tensor var_68_cast = conv(bias = bias_3_to_fp16, dilations = var_66, groups = var_56, pad = var_68_pad_0, pad_type = var_68_pad_type_0, strides = var_64, weight = weight_3_to_fp16, x = cast_727); + tensor input_1_mode_0 = const()[name = tensor("input_1_mode_0"), val = tensor("EXACT")]; + tensor input_1_cast = gelu(mode = input_1_mode_0, x = var_68_cast); + tensor var_72 = const()[name = tensor("op_72"), val = tensor(1)]; + tensor var_81 = const()[name = tensor("op_81"), val = tensor([2])]; + tensor var_83 = const()[name = tensor("op_83"), val = tensor([1])]; + tensor var_85_pad_type_0 = const()[name = tensor("op_85_pad_type_0"), val = tensor("custom")]; + tensor var_85_pad_0 = const()[name = tensor("op_85_pad_0"), val = tensor([1, 1])]; + tensor weight_7_to_fp16 = const()[name = tensor("weight_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(493760)))]; + tensor bias_7_to_fp16 = const()[name = tensor("bias_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6785280)))]; + tensor var_85_cast = conv(bias = bias_7_to_fp16, dilations = var_83, groups = var_72, pad = var_85_pad_0, pad_type = var_85_pad_type_0, strides = var_81, weight = weight_7_to_fp16, x = input_1_cast); + tensor x_3_mode_0 = const()[name = tensor("x_3_mode_0"), val = tensor("EXACT")]; + tensor x_3_cast = gelu(mode = x_3_mode_0, x = var_85_cast); + tensor var_90 = const()[name = tensor("op_90"), val = tensor([0, 2, 1])]; + tensor positional_embedding_to_fp16 = const()[name = tensor("positional_embedding_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6787392)))]; + tensor transpose_192 = transpose(perm = var_90, x = x_3_cast); + tensor var_93_cast = add(x = transpose_192, y = positional_embedding_to_fp16); + tensor var_106 = const()[name = tensor("op_106"), val = tensor(-1)]; + tensor var_123_axes_0 = const()[name = tensor("op_123_axes_0"), val = tensor([-1])]; + tensor blocks_0_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_0_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9859456)))]; + tensor blocks_0_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_0_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9861568)))]; + tensor var_112_to_fp16 = const()[name = tensor("op_112_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_123_cast = layer_norm(axes = var_123_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_112_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_93_cast); + tensor var_134_to_fp16 = const()[name = tensor("op_134_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9863680)))]; + tensor var_135_to_fp16 = const()[name = tensor("op_135_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11960896)))]; + tensor q_1_cast = linear(bias = var_135_to_fp16, weight = var_134_to_fp16, x = var_123_cast); + tensor var_138_to_fp16 = const()[name = tensor("op_138_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11963008)))]; + tensor k_1_bias_0_to_fp16 = const()[name = tensor("k_1_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14060224)))]; + tensor k_1_cast = linear(bias = k_1_bias_0_to_fp16, weight = var_138_to_fp16, x = var_123_cast); + tensor var_142_to_fp16 = const()[name = tensor("op_142_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14062336)))]; + tensor var_143_to_fp16 = const()[name = tensor("op_143_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16159552)))]; + tensor v_1_cast = linear(bias = var_143_to_fp16, weight = var_142_to_fp16, x = var_123_cast); + tensor var_151 = const()[name = tensor("op_151"), val = tensor([1, 1500, 16, -1])]; + tensor var_152_cast = reshape(shape = var_151, x = q_1_cast); + tensor const_168_to_fp16 = const()[name = tensor("const_168_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_3_cast = mul(x = var_152_cast, y = const_168_to_fp16); + tensor var_158 = const()[name = tensor("op_158"), val = tensor([1, 1500, 16, -1])]; + tensor var_159_cast = reshape(shape = var_158, x = k_1_cast); + tensor const_169_to_fp16 = const()[name = tensor("const_169_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_3_cast = mul(x = var_159_cast, y = const_169_to_fp16); + tensor var_165 = const()[name = tensor("op_165"), val = tensor([1, 1500, 16, -1])]; + tensor var_166_cast = reshape(shape = var_165, x = v_1_cast); + tensor var_167 = const()[name = tensor("op_167"), val = tensor([0, 2, 1, 3])]; + tensor qk_1_transpose_x_0 = const()[name = tensor("qk_1_transpose_x_0"), val = tensor(false)]; + tensor qk_1_transpose_y_0 = const()[name = tensor("qk_1_transpose_y_0"), val = tensor(false)]; + tensor transpose_48_perm_0 = const()[name = tensor("transpose_48_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_49_perm_0 = const()[name = tensor("transpose_49_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_189 = transpose(perm = transpose_49_perm_0, x = k_3_cast); + tensor transpose_190 = transpose(perm = transpose_48_perm_0, x = q_3_cast); + tensor qk_1_cast = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_190, y = transpose_189); + tensor var_171_cast = softmax(axis = var_106, x = qk_1_cast); + tensor var_173_transpose_x_0 = const()[name = tensor("op_173_transpose_x_0"), val = tensor(false)]; + tensor var_173_transpose_y_0 = const()[name = tensor("op_173_transpose_y_0"), val = tensor(false)]; + tensor transpose_191 = transpose(perm = var_167, x = var_166_cast); + tensor var_173_cast = matmul(transpose_x = var_173_transpose_x_0, transpose_y = var_173_transpose_y_0, x = var_171_cast, y = transpose_191); + tensor var_174 = const()[name = tensor("op_174"), val = tensor([0, 2, 1, 3])]; + tensor concat_0 = const()[name = tensor("concat_0"), val = tensor([1, 1500, 1024])]; + tensor transpose_188 = transpose(perm = var_174, x = var_173_cast); + tensor x_11_cast = reshape(shape = concat_0, x = transpose_188); + tensor var_179_to_fp16 = const()[name = tensor("op_179_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16161664)))]; + tensor var_180_to_fp16 = const()[name = tensor("op_180_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18258880)))]; + tensor var_181_cast = linear(bias = var_180_to_fp16, weight = var_179_to_fp16, x = x_11_cast); + tensor x_13_cast = add(x = var_93_cast, y = var_181_cast); + tensor var_187_axes_0 = const()[name = tensor("op_187_axes_0"), val = tensor([-1])]; + tensor blocks_0_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18260992)))]; + tensor blocks_0_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18263104)))]; + tensor var_187_cast = layer_norm(axes = var_187_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_112_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast); + tensor var_196_to_fp16 = const()[name = tensor("op_196_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18265216)))]; + tensor var_197_to_fp16 = const()[name = tensor("op_197_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26653888)))]; + tensor input_9_cast = linear(bias = var_197_to_fp16, weight = var_196_to_fp16, x = var_187_cast); + tensor x_17_mode_0 = const()[name = tensor("x_17_mode_0"), val = tensor("EXACT")]; + tensor x_17_cast = gelu(mode = x_17_mode_0, x = input_9_cast); + tensor var_202_to_fp16 = const()[name = tensor("op_202_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26662144)))]; + tensor var_203_to_fp16 = const()[name = tensor("op_203_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35050816)))]; + tensor var_204_cast = linear(bias = var_203_to_fp16, weight = var_202_to_fp16, x = x_17_cast); + tensor x_19_cast = add(x = x_13_cast, y = var_204_cast); + tensor var_213 = const()[name = tensor("op_213"), val = tensor(-1)]; + tensor var_230_axes_0 = const()[name = tensor("op_230_axes_0"), val = tensor([-1])]; + tensor blocks_1_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_1_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35052928)))]; + tensor blocks_1_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_1_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35055040)))]; + tensor var_219_to_fp16 = const()[name = tensor("op_219_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_230_cast = layer_norm(axes = var_230_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_219_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast); + tensor var_241_to_fp16 = const()[name = tensor("op_241_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35057152)))]; + tensor var_242_to_fp16 = const()[name = tensor("op_242_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37154368)))]; + tensor q_5_cast = linear(bias = var_242_to_fp16, weight = var_241_to_fp16, x = var_230_cast); + tensor var_245_to_fp16 = const()[name = tensor("op_245_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37156480)))]; + tensor k_5_bias_0_to_fp16 = const()[name = tensor("k_5_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39253696)))]; + tensor k_5_cast = linear(bias = k_5_bias_0_to_fp16, weight = var_245_to_fp16, x = var_230_cast); + tensor var_249_to_fp16 = const()[name = tensor("op_249_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39255808)))]; + tensor var_250_to_fp16 = const()[name = tensor("op_250_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(41353024)))]; + tensor v_5_cast = linear(bias = var_250_to_fp16, weight = var_249_to_fp16, x = var_230_cast); + tensor var_258 = const()[name = tensor("op_258"), val = tensor([1, 1500, 16, -1])]; + tensor var_259_cast = reshape(shape = var_258, x = q_5_cast); + tensor const_170_to_fp16 = const()[name = tensor("const_170_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_7_cast = mul(x = var_259_cast, y = const_170_to_fp16); + tensor var_265 = const()[name = tensor("op_265"), val = tensor([1, 1500, 16, -1])]; + tensor var_266_cast = reshape(shape = var_265, x = k_5_cast); + tensor const_171_to_fp16 = const()[name = tensor("const_171_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_7_cast = mul(x = var_266_cast, y = const_171_to_fp16); + tensor var_272 = const()[name = tensor("op_272"), val = tensor([1, 1500, 16, -1])]; + tensor var_273_cast = reshape(shape = var_272, x = v_5_cast); + tensor var_274 = const()[name = tensor("op_274"), val = tensor([0, 2, 1, 3])]; + tensor qk_3_transpose_x_0 = const()[name = tensor("qk_3_transpose_x_0"), val = tensor(false)]; + tensor qk_3_transpose_y_0 = const()[name = tensor("qk_3_transpose_y_0"), val = tensor(false)]; + tensor transpose_50_perm_0 = const()[name = tensor("transpose_50_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_51_perm_0 = const()[name = tensor("transpose_51_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_185 = transpose(perm = transpose_51_perm_0, x = k_7_cast); + tensor transpose_186 = transpose(perm = transpose_50_perm_0, x = q_7_cast); + tensor qk_3_cast = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_186, y = transpose_185); + tensor var_278_cast = softmax(axis = var_213, x = qk_3_cast); + tensor var_280_transpose_x_0 = const()[name = tensor("op_280_transpose_x_0"), val = tensor(false)]; + tensor var_280_transpose_y_0 = const()[name = tensor("op_280_transpose_y_0"), val = tensor(false)]; + tensor transpose_187 = transpose(perm = var_274, x = var_273_cast); + tensor var_280_cast = matmul(transpose_x = var_280_transpose_x_0, transpose_y = var_280_transpose_y_0, x = var_278_cast, y = transpose_187); + tensor var_281 = const()[name = tensor("op_281"), val = tensor([0, 2, 1, 3])]; + tensor concat_1 = const()[name = tensor("concat_1"), val = tensor([1, 1500, 1024])]; + tensor transpose_184 = transpose(perm = var_281, x = var_280_cast); + tensor x_23_cast = reshape(shape = concat_1, x = transpose_184); + tensor var_286_to_fp16 = const()[name = tensor("op_286_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(41355136)))]; + tensor var_287_to_fp16 = const()[name = tensor("op_287_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43452352)))]; + tensor var_288_cast = linear(bias = var_287_to_fp16, weight = var_286_to_fp16, x = x_23_cast); + tensor x_25_cast = add(x = x_19_cast, y = var_288_cast); + tensor var_294_axes_0 = const()[name = tensor("op_294_axes_0"), val = tensor([-1])]; + tensor blocks_1_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43454464)))]; + tensor blocks_1_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43456576)))]; + tensor var_294_cast = layer_norm(axes = var_294_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_219_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast); + tensor var_303_to_fp16 = const()[name = tensor("op_303_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43458688)))]; + tensor var_304_to_fp16 = const()[name = tensor("op_304_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51847360)))]; + tensor input_17_cast = linear(bias = var_304_to_fp16, weight = var_303_to_fp16, x = var_294_cast); + tensor x_29_mode_0 = const()[name = tensor("x_29_mode_0"), val = tensor("EXACT")]; + tensor x_29_cast = gelu(mode = x_29_mode_0, x = input_17_cast); + tensor var_309_to_fp16 = const()[name = tensor("op_309_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51855616)))]; + tensor var_310_to_fp16 = const()[name = tensor("op_310_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60244288)))]; + tensor var_311_cast = linear(bias = var_310_to_fp16, weight = var_309_to_fp16, x = x_29_cast); + tensor x_31_cast = add(x = x_25_cast, y = var_311_cast); + tensor var_320 = const()[name = tensor("op_320"), val = tensor(-1)]; + tensor var_337_axes_0 = const()[name = tensor("op_337_axes_0"), val = tensor([-1])]; + tensor blocks_2_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_2_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60246400)))]; + tensor blocks_2_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_2_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60248512)))]; + tensor var_326_to_fp16 = const()[name = tensor("op_326_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_337_cast = layer_norm(axes = var_337_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_326_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast); + tensor var_348_to_fp16 = const()[name = tensor("op_348_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60250624)))]; + tensor var_349_to_fp16 = const()[name = tensor("op_349_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62347840)))]; + tensor q_9_cast = linear(bias = var_349_to_fp16, weight = var_348_to_fp16, x = var_337_cast); + tensor var_352_to_fp16 = const()[name = tensor("op_352_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62349952)))]; + tensor k_9_bias_0_to_fp16 = const()[name = tensor("k_9_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64447168)))]; + tensor k_9_cast = linear(bias = k_9_bias_0_to_fp16, weight = var_352_to_fp16, x = var_337_cast); + tensor var_356_to_fp16 = const()[name = tensor("op_356_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64449280)))]; + tensor var_357_to_fp16 = const()[name = tensor("op_357_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66546496)))]; + tensor v_9_cast = linear(bias = var_357_to_fp16, weight = var_356_to_fp16, x = var_337_cast); + tensor var_365 = const()[name = tensor("op_365"), val = tensor([1, 1500, 16, -1])]; + tensor var_366_cast = reshape(shape = var_365, x = q_9_cast); + tensor const_172_to_fp16 = const()[name = tensor("const_172_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_11_cast = mul(x = var_366_cast, y = const_172_to_fp16); + tensor var_372 = const()[name = tensor("op_372"), val = tensor([1, 1500, 16, -1])]; + tensor var_373_cast = reshape(shape = var_372, x = k_9_cast); + tensor const_173_to_fp16 = const()[name = tensor("const_173_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_11_cast = mul(x = var_373_cast, y = const_173_to_fp16); + tensor var_379 = const()[name = tensor("op_379"), val = tensor([1, 1500, 16, -1])]; + tensor var_380_cast = reshape(shape = var_379, x = v_9_cast); + tensor var_381 = const()[name = tensor("op_381"), val = tensor([0, 2, 1, 3])]; + tensor qk_5_transpose_x_0 = const()[name = tensor("qk_5_transpose_x_0"), val = tensor(false)]; + tensor qk_5_transpose_y_0 = const()[name = tensor("qk_5_transpose_y_0"), val = tensor(false)]; + tensor transpose_52_perm_0 = const()[name = tensor("transpose_52_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_53_perm_0 = const()[name = tensor("transpose_53_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_181 = transpose(perm = transpose_53_perm_0, x = k_11_cast); + tensor transpose_182 = transpose(perm = transpose_52_perm_0, x = q_11_cast); + tensor qk_5_cast = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_182, y = transpose_181); + tensor var_385_cast = softmax(axis = var_320, x = qk_5_cast); + tensor var_387_transpose_x_0 = const()[name = tensor("op_387_transpose_x_0"), val = tensor(false)]; + tensor var_387_transpose_y_0 = const()[name = tensor("op_387_transpose_y_0"), val = tensor(false)]; + tensor transpose_183 = transpose(perm = var_381, x = var_380_cast); + tensor var_387_cast = matmul(transpose_x = var_387_transpose_x_0, transpose_y = var_387_transpose_y_0, x = var_385_cast, y = transpose_183); + tensor var_388 = const()[name = tensor("op_388"), val = tensor([0, 2, 1, 3])]; + tensor concat_2 = const()[name = tensor("concat_2"), val = tensor([1, 1500, 1024])]; + tensor transpose_180 = transpose(perm = var_388, x = var_387_cast); + tensor x_35_cast = reshape(shape = concat_2, x = transpose_180); + tensor var_393_to_fp16 = const()[name = tensor("op_393_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66548608)))]; + tensor var_394_to_fp16 = const()[name = tensor("op_394_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68645824)))]; + tensor var_395_cast = linear(bias = var_394_to_fp16, weight = var_393_to_fp16, x = x_35_cast); + tensor x_37_cast = add(x = x_31_cast, y = var_395_cast); + tensor var_401_axes_0 = const()[name = tensor("op_401_axes_0"), val = tensor([-1])]; + tensor blocks_2_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68647936)))]; + tensor blocks_2_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68650048)))]; + tensor var_401_cast = layer_norm(axes = var_401_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_326_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast); + tensor var_410_to_fp16 = const()[name = tensor("op_410_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68652160)))]; + tensor var_411_to_fp16 = const()[name = tensor("op_411_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77040832)))]; + tensor input_25_cast = linear(bias = var_411_to_fp16, weight = var_410_to_fp16, x = var_401_cast); + tensor x_41_mode_0 = const()[name = tensor("x_41_mode_0"), val = tensor("EXACT")]; + tensor x_41_cast = gelu(mode = x_41_mode_0, x = input_25_cast); + tensor var_416_to_fp16 = const()[name = tensor("op_416_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77049088)))]; + tensor var_417_to_fp16 = const()[name = tensor("op_417_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85437760)))]; + tensor var_418_cast = linear(bias = var_417_to_fp16, weight = var_416_to_fp16, x = x_41_cast); + tensor x_43_cast = add(x = x_37_cast, y = var_418_cast); + tensor var_427 = const()[name = tensor("op_427"), val = tensor(-1)]; + tensor var_444_axes_0 = const()[name = tensor("op_444_axes_0"), val = tensor([-1])]; + tensor blocks_3_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_3_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85439872)))]; + tensor blocks_3_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_3_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85441984)))]; + tensor var_433_to_fp16 = const()[name = tensor("op_433_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_444_cast = layer_norm(axes = var_444_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_433_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast); + tensor var_455_to_fp16 = const()[name = tensor("op_455_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85444096)))]; + tensor var_456_to_fp16 = const()[name = tensor("op_456_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87541312)))]; + tensor q_13_cast = linear(bias = var_456_to_fp16, weight = var_455_to_fp16, x = var_444_cast); + tensor var_459_to_fp16 = const()[name = tensor("op_459_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87543424)))]; + tensor k_13_bias_0_to_fp16 = const()[name = tensor("k_13_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89640640)))]; + tensor k_13_cast = linear(bias = k_13_bias_0_to_fp16, weight = var_459_to_fp16, x = var_444_cast); + tensor var_463_to_fp16 = const()[name = tensor("op_463_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89642752)))]; + tensor var_464_to_fp16 = const()[name = tensor("op_464_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91739968)))]; + tensor v_13_cast = linear(bias = var_464_to_fp16, weight = var_463_to_fp16, x = var_444_cast); + tensor var_472 = const()[name = tensor("op_472"), val = tensor([1, 1500, 16, -1])]; + tensor var_473_cast = reshape(shape = var_472, x = q_13_cast); + tensor const_174_to_fp16 = const()[name = tensor("const_174_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_15_cast = mul(x = var_473_cast, y = const_174_to_fp16); + tensor var_479 = const()[name = tensor("op_479"), val = tensor([1, 1500, 16, -1])]; + tensor var_480_cast = reshape(shape = var_479, x = k_13_cast); + tensor const_175_to_fp16 = const()[name = tensor("const_175_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_15_cast = mul(x = var_480_cast, y = const_175_to_fp16); + tensor var_486 = const()[name = tensor("op_486"), val = tensor([1, 1500, 16, -1])]; + tensor var_487_cast = reshape(shape = var_486, x = v_13_cast); + tensor var_488 = const()[name = tensor("op_488"), val = tensor([0, 2, 1, 3])]; + tensor qk_7_transpose_x_0 = const()[name = tensor("qk_7_transpose_x_0"), val = tensor(false)]; + tensor qk_7_transpose_y_0 = const()[name = tensor("qk_7_transpose_y_0"), val = tensor(false)]; + tensor transpose_54_perm_0 = const()[name = tensor("transpose_54_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_55_perm_0 = const()[name = tensor("transpose_55_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_177 = transpose(perm = transpose_55_perm_0, x = k_15_cast); + tensor transpose_178 = transpose(perm = transpose_54_perm_0, x = q_15_cast); + tensor qk_7_cast = matmul(transpose_x = qk_7_transpose_x_0, transpose_y = qk_7_transpose_y_0, x = transpose_178, y = transpose_177); + tensor var_492_cast = softmax(axis = var_427, x = qk_7_cast); + tensor var_494_transpose_x_0 = const()[name = tensor("op_494_transpose_x_0"), val = tensor(false)]; + tensor var_494_transpose_y_0 = const()[name = tensor("op_494_transpose_y_0"), val = tensor(false)]; + tensor transpose_179 = transpose(perm = var_488, x = var_487_cast); + tensor var_494_cast = matmul(transpose_x = var_494_transpose_x_0, transpose_y = var_494_transpose_y_0, x = var_492_cast, y = transpose_179); + tensor var_495 = const()[name = tensor("op_495"), val = tensor([0, 2, 1, 3])]; + tensor concat_3 = const()[name = tensor("concat_3"), val = tensor([1, 1500, 1024])]; + tensor transpose_176 = transpose(perm = var_495, x = var_494_cast); + tensor x_47_cast = reshape(shape = concat_3, x = transpose_176); + tensor var_500_to_fp16 = const()[name = tensor("op_500_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91742080)))]; + tensor var_501_to_fp16 = const()[name = tensor("op_501_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93839296)))]; + tensor var_502_cast = linear(bias = var_501_to_fp16, weight = var_500_to_fp16, x = x_47_cast); + tensor x_49_cast = add(x = x_43_cast, y = var_502_cast); + tensor var_508_axes_0 = const()[name = tensor("op_508_axes_0"), val = tensor([-1])]; + tensor blocks_3_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93841408)))]; + tensor blocks_3_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93843520)))]; + tensor var_508_cast = layer_norm(axes = var_508_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_433_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast); + tensor var_517_to_fp16 = const()[name = tensor("op_517_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93845632)))]; + tensor var_518_to_fp16 = const()[name = tensor("op_518_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102234304)))]; + tensor input_33_cast = linear(bias = var_518_to_fp16, weight = var_517_to_fp16, x = var_508_cast); + tensor x_53_mode_0 = const()[name = tensor("x_53_mode_0"), val = tensor("EXACT")]; + tensor x_53_cast = gelu(mode = x_53_mode_0, x = input_33_cast); + tensor var_523_to_fp16 = const()[name = tensor("op_523_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(102242560)))]; + tensor var_524_to_fp16 = const()[name = tensor("op_524_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110631232)))]; + tensor var_525_cast = linear(bias = var_524_to_fp16, weight = var_523_to_fp16, x = x_53_cast); + tensor x_55_cast = add(x = x_49_cast, y = var_525_cast); + tensor var_534 = const()[name = tensor("op_534"), val = tensor(-1)]; + tensor var_551_axes_0 = const()[name = tensor("op_551_axes_0"), val = tensor([-1])]; + tensor blocks_4_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_4_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110633344)))]; + tensor blocks_4_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_4_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110635456)))]; + tensor var_540_to_fp16 = const()[name = tensor("op_540_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_551_cast = layer_norm(axes = var_551_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_540_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast); + tensor var_562_to_fp16 = const()[name = tensor("op_562_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110637568)))]; + tensor var_563_to_fp16 = const()[name = tensor("op_563_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112734784)))]; + tensor q_17_cast = linear(bias = var_563_to_fp16, weight = var_562_to_fp16, x = var_551_cast); + tensor var_566_to_fp16 = const()[name = tensor("op_566_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(112736896)))]; + tensor k_17_bias_0_to_fp16 = const()[name = tensor("k_17_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114834112)))]; + tensor k_17_cast = linear(bias = k_17_bias_0_to_fp16, weight = var_566_to_fp16, x = var_551_cast); + tensor var_570_to_fp16 = const()[name = tensor("op_570_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114836224)))]; + tensor var_571_to_fp16 = const()[name = tensor("op_571_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116933440)))]; + tensor v_17_cast = linear(bias = var_571_to_fp16, weight = var_570_to_fp16, x = var_551_cast); + tensor var_579 = const()[name = tensor("op_579"), val = tensor([1, 1500, 16, -1])]; + tensor var_580_cast = reshape(shape = var_579, x = q_17_cast); + tensor const_176_to_fp16 = const()[name = tensor("const_176_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_19_cast = mul(x = var_580_cast, y = const_176_to_fp16); + tensor var_586 = const()[name = tensor("op_586"), val = tensor([1, 1500, 16, -1])]; + tensor var_587_cast = reshape(shape = var_586, x = k_17_cast); + tensor const_177_to_fp16 = const()[name = tensor("const_177_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_19_cast = mul(x = var_587_cast, y = const_177_to_fp16); + tensor var_593 = const()[name = tensor("op_593"), val = tensor([1, 1500, 16, -1])]; + tensor var_594_cast = reshape(shape = var_593, x = v_17_cast); + tensor var_595 = const()[name = tensor("op_595"), val = tensor([0, 2, 1, 3])]; + tensor qk_9_transpose_x_0 = const()[name = tensor("qk_9_transpose_x_0"), val = tensor(false)]; + tensor qk_9_transpose_y_0 = const()[name = tensor("qk_9_transpose_y_0"), val = tensor(false)]; + tensor transpose_56_perm_0 = const()[name = tensor("transpose_56_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_57_perm_0 = const()[name = tensor("transpose_57_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_173 = transpose(perm = transpose_57_perm_0, x = k_19_cast); + tensor transpose_174 = transpose(perm = transpose_56_perm_0, x = q_19_cast); + tensor qk_9_cast = matmul(transpose_x = qk_9_transpose_x_0, transpose_y = qk_9_transpose_y_0, x = transpose_174, y = transpose_173); + tensor var_599_cast = softmax(axis = var_534, x = qk_9_cast); + tensor var_601_transpose_x_0 = const()[name = tensor("op_601_transpose_x_0"), val = tensor(false)]; + tensor var_601_transpose_y_0 = const()[name = tensor("op_601_transpose_y_0"), val = tensor(false)]; + tensor transpose_175 = transpose(perm = var_595, x = var_594_cast); + tensor var_601_cast = matmul(transpose_x = var_601_transpose_x_0, transpose_y = var_601_transpose_y_0, x = var_599_cast, y = transpose_175); + tensor var_602 = const()[name = tensor("op_602"), val = tensor([0, 2, 1, 3])]; + tensor concat_4 = const()[name = tensor("concat_4"), val = tensor([1, 1500, 1024])]; + tensor transpose_172 = transpose(perm = var_602, x = var_601_cast); + tensor x_59_cast = reshape(shape = concat_4, x = transpose_172); + tensor var_607_to_fp16 = const()[name = tensor("op_607_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(116935552)))]; + tensor var_608_to_fp16 = const()[name = tensor("op_608_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119032768)))]; + tensor var_609_cast = linear(bias = var_608_to_fp16, weight = var_607_to_fp16, x = x_59_cast); + tensor x_61_cast = add(x = x_55_cast, y = var_609_cast); + tensor var_615_axes_0 = const()[name = tensor("op_615_axes_0"), val = tensor([-1])]; + tensor blocks_4_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119034880)))]; + tensor blocks_4_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119036992)))]; + tensor var_615_cast = layer_norm(axes = var_615_axes_0, beta = blocks_4_mlp_ln_bias_to_fp16, epsilon = var_540_to_fp16, gamma = blocks_4_mlp_ln_weight_to_fp16, x = x_61_cast); + tensor var_624_to_fp16 = const()[name = tensor("op_624_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119039104)))]; + tensor var_625_to_fp16 = const()[name = tensor("op_625_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127427776)))]; + tensor input_41_cast = linear(bias = var_625_to_fp16, weight = var_624_to_fp16, x = var_615_cast); + tensor x_65_mode_0 = const()[name = tensor("x_65_mode_0"), val = tensor("EXACT")]; + tensor x_65_cast = gelu(mode = x_65_mode_0, x = input_41_cast); + tensor var_630_to_fp16 = const()[name = tensor("op_630_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(127436032)))]; + tensor var_631_to_fp16 = const()[name = tensor("op_631_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135824704)))]; + tensor var_632_cast = linear(bias = var_631_to_fp16, weight = var_630_to_fp16, x = x_65_cast); + tensor x_67_cast = add(x = x_61_cast, y = var_632_cast); + tensor var_641 = const()[name = tensor("op_641"), val = tensor(-1)]; + tensor var_658_axes_0 = const()[name = tensor("op_658_axes_0"), val = tensor([-1])]; + tensor blocks_5_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_5_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135826816)))]; + tensor blocks_5_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_5_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135828928)))]; + tensor var_647_to_fp16 = const()[name = tensor("op_647_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_658_cast = layer_norm(axes = var_658_axes_0, beta = blocks_5_attn_ln_bias_to_fp16, epsilon = var_647_to_fp16, gamma = blocks_5_attn_ln_weight_to_fp16, x = x_67_cast); + tensor var_669_to_fp16 = const()[name = tensor("op_669_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(135831040)))]; + tensor var_670_to_fp16 = const()[name = tensor("op_670_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137928256)))]; + tensor q_21_cast = linear(bias = var_670_to_fp16, weight = var_669_to_fp16, x = var_658_cast); + tensor var_673_to_fp16 = const()[name = tensor("op_673_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137930368)))]; + tensor k_21_bias_0_to_fp16 = const()[name = tensor("k_21_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(140027584)))]; + tensor k_21_cast = linear(bias = k_21_bias_0_to_fp16, weight = var_673_to_fp16, x = var_658_cast); + tensor var_677_to_fp16 = const()[name = tensor("op_677_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(140029696)))]; + tensor var_678_to_fp16 = const()[name = tensor("op_678_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142126912)))]; + tensor v_21_cast = linear(bias = var_678_to_fp16, weight = var_677_to_fp16, x = var_658_cast); + tensor var_686 = const()[name = tensor("op_686"), val = tensor([1, 1500, 16, -1])]; + tensor var_687_cast = reshape(shape = var_686, x = q_21_cast); + tensor const_178_to_fp16 = const()[name = tensor("const_178_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_23_cast = mul(x = var_687_cast, y = const_178_to_fp16); + tensor var_693 = const()[name = tensor("op_693"), val = tensor([1, 1500, 16, -1])]; + tensor var_694_cast = reshape(shape = var_693, x = k_21_cast); + tensor const_179_to_fp16 = const()[name = tensor("const_179_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_23_cast = mul(x = var_694_cast, y = const_179_to_fp16); + tensor var_700 = const()[name = tensor("op_700"), val = tensor([1, 1500, 16, -1])]; + tensor var_701_cast = reshape(shape = var_700, x = v_21_cast); + tensor var_702 = const()[name = tensor("op_702"), val = tensor([0, 2, 1, 3])]; + tensor qk_11_transpose_x_0 = const()[name = tensor("qk_11_transpose_x_0"), val = tensor(false)]; + tensor qk_11_transpose_y_0 = const()[name = tensor("qk_11_transpose_y_0"), val = tensor(false)]; + tensor transpose_58_perm_0 = const()[name = tensor("transpose_58_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_59_perm_0 = const()[name = tensor("transpose_59_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_169 = transpose(perm = transpose_59_perm_0, x = k_23_cast); + tensor transpose_170 = transpose(perm = transpose_58_perm_0, x = q_23_cast); + tensor qk_11_cast = matmul(transpose_x = qk_11_transpose_x_0, transpose_y = qk_11_transpose_y_0, x = transpose_170, y = transpose_169); + tensor var_706_cast = softmax(axis = var_641, x = qk_11_cast); + tensor var_708_transpose_x_0 = const()[name = tensor("op_708_transpose_x_0"), val = tensor(false)]; + tensor var_708_transpose_y_0 = const()[name = tensor("op_708_transpose_y_0"), val = tensor(false)]; + tensor transpose_171 = transpose(perm = var_702, x = var_701_cast); + tensor var_708_cast = matmul(transpose_x = var_708_transpose_x_0, transpose_y = var_708_transpose_y_0, x = var_706_cast, y = transpose_171); + tensor var_709 = const()[name = tensor("op_709"), val = tensor([0, 2, 1, 3])]; + tensor concat_5 = const()[name = tensor("concat_5"), val = tensor([1, 1500, 1024])]; + tensor transpose_168 = transpose(perm = var_709, x = var_708_cast); + tensor x_71_cast = reshape(shape = concat_5, x = transpose_168); + tensor var_714_to_fp16 = const()[name = tensor("op_714_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(142129024)))]; + tensor var_715_to_fp16 = const()[name = tensor("op_715_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(144226240)))]; + tensor var_716_cast = linear(bias = var_715_to_fp16, weight = var_714_to_fp16, x = x_71_cast); + tensor x_73_cast = add(x = x_67_cast, y = var_716_cast); + tensor var_722_axes_0 = const()[name = tensor("op_722_axes_0"), val = tensor([-1])]; + tensor blocks_5_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(144228352)))]; + tensor blocks_5_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(144230464)))]; + tensor var_722_cast = layer_norm(axes = var_722_axes_0, beta = blocks_5_mlp_ln_bias_to_fp16, epsilon = var_647_to_fp16, gamma = blocks_5_mlp_ln_weight_to_fp16, x = x_73_cast); + tensor var_731_to_fp16 = const()[name = tensor("op_731_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(144232576)))]; + tensor var_732_to_fp16 = const()[name = tensor("op_732_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152621248)))]; + tensor input_49_cast = linear(bias = var_732_to_fp16, weight = var_731_to_fp16, x = var_722_cast); + tensor x_77_mode_0 = const()[name = tensor("x_77_mode_0"), val = tensor("EXACT")]; + tensor x_77_cast = gelu(mode = x_77_mode_0, x = input_49_cast); + tensor var_737_to_fp16 = const()[name = tensor("op_737_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152629504)))]; + tensor var_738_to_fp16 = const()[name = tensor("op_738_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161018176)))]; + tensor var_739_cast = linear(bias = var_738_to_fp16, weight = var_737_to_fp16, x = x_77_cast); + tensor x_79_cast = add(x = x_73_cast, y = var_739_cast); + tensor var_748 = const()[name = tensor("op_748"), val = tensor(-1)]; + tensor var_765_axes_0 = const()[name = tensor("op_765_axes_0"), val = tensor([-1])]; + tensor blocks_6_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_6_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161020288)))]; + tensor blocks_6_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_6_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161022400)))]; + tensor var_754_to_fp16 = const()[name = tensor("op_754_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_765_cast = layer_norm(axes = var_765_axes_0, beta = blocks_6_attn_ln_bias_to_fp16, epsilon = var_754_to_fp16, gamma = blocks_6_attn_ln_weight_to_fp16, x = x_79_cast); + tensor var_776_to_fp16 = const()[name = tensor("op_776_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(161024512)))]; + tensor var_777_to_fp16 = const()[name = tensor("op_777_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163121728)))]; + tensor q_25_cast = linear(bias = var_777_to_fp16, weight = var_776_to_fp16, x = var_765_cast); + tensor var_780_to_fp16 = const()[name = tensor("op_780_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163123840)))]; + tensor k_25_bias_0_to_fp16 = const()[name = tensor("k_25_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165221056)))]; + tensor k_25_cast = linear(bias = k_25_bias_0_to_fp16, weight = var_780_to_fp16, x = var_765_cast); + tensor var_784_to_fp16 = const()[name = tensor("op_784_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165223168)))]; + tensor var_785_to_fp16 = const()[name = tensor("op_785_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(167320384)))]; + tensor v_25_cast = linear(bias = var_785_to_fp16, weight = var_784_to_fp16, x = var_765_cast); + tensor var_793 = const()[name = tensor("op_793"), val = tensor([1, 1500, 16, -1])]; + tensor var_794_cast = reshape(shape = var_793, x = q_25_cast); + tensor const_180_to_fp16 = const()[name = tensor("const_180_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_27_cast = mul(x = var_794_cast, y = const_180_to_fp16); + tensor var_800 = const()[name = tensor("op_800"), val = tensor([1, 1500, 16, -1])]; + tensor var_801_cast = reshape(shape = var_800, x = k_25_cast); + tensor const_181_to_fp16 = const()[name = tensor("const_181_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_27_cast = mul(x = var_801_cast, y = const_181_to_fp16); + tensor var_807 = const()[name = tensor("op_807"), val = tensor([1, 1500, 16, -1])]; + tensor var_808_cast = reshape(shape = var_807, x = v_25_cast); + tensor var_809 = const()[name = tensor("op_809"), val = tensor([0, 2, 1, 3])]; + tensor qk_13_transpose_x_0 = const()[name = tensor("qk_13_transpose_x_0"), val = tensor(false)]; + tensor qk_13_transpose_y_0 = const()[name = tensor("qk_13_transpose_y_0"), val = tensor(false)]; + tensor transpose_60_perm_0 = const()[name = tensor("transpose_60_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_61_perm_0 = const()[name = tensor("transpose_61_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_165 = transpose(perm = transpose_61_perm_0, x = k_27_cast); + tensor transpose_166 = transpose(perm = transpose_60_perm_0, x = q_27_cast); + tensor qk_13_cast = matmul(transpose_x = qk_13_transpose_x_0, transpose_y = qk_13_transpose_y_0, x = transpose_166, y = transpose_165); + tensor var_813_cast = softmax(axis = var_748, x = qk_13_cast); + tensor var_815_transpose_x_0 = const()[name = tensor("op_815_transpose_x_0"), val = tensor(false)]; + tensor var_815_transpose_y_0 = const()[name = tensor("op_815_transpose_y_0"), val = tensor(false)]; + tensor transpose_167 = transpose(perm = var_809, x = var_808_cast); + tensor var_815_cast = matmul(transpose_x = var_815_transpose_x_0, transpose_y = var_815_transpose_y_0, x = var_813_cast, y = transpose_167); + tensor var_816 = const()[name = tensor("op_816"), val = tensor([0, 2, 1, 3])]; + tensor concat_6 = const()[name = tensor("concat_6"), val = tensor([1, 1500, 1024])]; + tensor transpose_164 = transpose(perm = var_816, x = var_815_cast); + tensor x_83_cast = reshape(shape = concat_6, x = transpose_164); + tensor var_821_to_fp16 = const()[name = tensor("op_821_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(167322496)))]; + tensor var_822_to_fp16 = const()[name = tensor("op_822_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(169419712)))]; + tensor var_823_cast = linear(bias = var_822_to_fp16, weight = var_821_to_fp16, x = x_83_cast); + tensor x_85_cast = add(x = x_79_cast, y = var_823_cast); + tensor var_829_axes_0 = const()[name = tensor("op_829_axes_0"), val = tensor([-1])]; + tensor blocks_6_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_6_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(169421824)))]; + tensor blocks_6_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_6_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(169423936)))]; + tensor var_829_cast = layer_norm(axes = var_829_axes_0, beta = blocks_6_mlp_ln_bias_to_fp16, epsilon = var_754_to_fp16, gamma = blocks_6_mlp_ln_weight_to_fp16, x = x_85_cast); + tensor var_838_to_fp16 = const()[name = tensor("op_838_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(169426048)))]; + tensor var_839_to_fp16 = const()[name = tensor("op_839_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(177814720)))]; + tensor input_57_cast = linear(bias = var_839_to_fp16, weight = var_838_to_fp16, x = var_829_cast); + tensor x_89_mode_0 = const()[name = tensor("x_89_mode_0"), val = tensor("EXACT")]; + tensor x_89_cast = gelu(mode = x_89_mode_0, x = input_57_cast); + tensor var_844_to_fp16 = const()[name = tensor("op_844_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(177822976)))]; + tensor var_845_to_fp16 = const()[name = tensor("op_845_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186211648)))]; + tensor var_846_cast = linear(bias = var_845_to_fp16, weight = var_844_to_fp16, x = x_89_cast); + tensor x_91_cast = add(x = x_85_cast, y = var_846_cast); + tensor var_855 = const()[name = tensor("op_855"), val = tensor(-1)]; + tensor var_872_axes_0 = const()[name = tensor("op_872_axes_0"), val = tensor([-1])]; + tensor blocks_7_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_7_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186213760)))]; + tensor blocks_7_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_7_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186215872)))]; + tensor var_861_to_fp16 = const()[name = tensor("op_861_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_872_cast = layer_norm(axes = var_872_axes_0, beta = blocks_7_attn_ln_bias_to_fp16, epsilon = var_861_to_fp16, gamma = blocks_7_attn_ln_weight_to_fp16, x = x_91_cast); + tensor var_883_to_fp16 = const()[name = tensor("op_883_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(186217984)))]; + tensor var_884_to_fp16 = const()[name = tensor("op_884_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(188315200)))]; + tensor q_29_cast = linear(bias = var_884_to_fp16, weight = var_883_to_fp16, x = var_872_cast); + tensor var_887_to_fp16 = const()[name = tensor("op_887_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(188317312)))]; + tensor k_29_bias_0_to_fp16 = const()[name = tensor("k_29_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190414528)))]; + tensor k_29_cast = linear(bias = k_29_bias_0_to_fp16, weight = var_887_to_fp16, x = var_872_cast); + tensor var_891_to_fp16 = const()[name = tensor("op_891_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(190416640)))]; + tensor var_892_to_fp16 = const()[name = tensor("op_892_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(192513856)))]; + tensor v_29_cast = linear(bias = var_892_to_fp16, weight = var_891_to_fp16, x = var_872_cast); + tensor var_900 = const()[name = tensor("op_900"), val = tensor([1, 1500, 16, -1])]; + tensor var_901_cast = reshape(shape = var_900, x = q_29_cast); + tensor const_182_to_fp16 = const()[name = tensor("const_182_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_31_cast = mul(x = var_901_cast, y = const_182_to_fp16); + tensor var_907 = const()[name = tensor("op_907"), val = tensor([1, 1500, 16, -1])]; + tensor var_908_cast = reshape(shape = var_907, x = k_29_cast); + tensor const_183_to_fp16 = const()[name = tensor("const_183_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_31_cast = mul(x = var_908_cast, y = const_183_to_fp16); + tensor var_914 = const()[name = tensor("op_914"), val = tensor([1, 1500, 16, -1])]; + tensor var_915_cast = reshape(shape = var_914, x = v_29_cast); + tensor var_916 = const()[name = tensor("op_916"), val = tensor([0, 2, 1, 3])]; + tensor qk_15_transpose_x_0 = const()[name = tensor("qk_15_transpose_x_0"), val = tensor(false)]; + tensor qk_15_transpose_y_0 = const()[name = tensor("qk_15_transpose_y_0"), val = tensor(false)]; + tensor transpose_62_perm_0 = const()[name = tensor("transpose_62_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_63_perm_0 = const()[name = tensor("transpose_63_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_161 = transpose(perm = transpose_63_perm_0, x = k_31_cast); + tensor transpose_162 = transpose(perm = transpose_62_perm_0, x = q_31_cast); + tensor qk_15_cast = matmul(transpose_x = qk_15_transpose_x_0, transpose_y = qk_15_transpose_y_0, x = transpose_162, y = transpose_161); + tensor var_920_cast = softmax(axis = var_855, x = qk_15_cast); + tensor var_922_transpose_x_0 = const()[name = tensor("op_922_transpose_x_0"), val = tensor(false)]; + tensor var_922_transpose_y_0 = const()[name = tensor("op_922_transpose_y_0"), val = tensor(false)]; + tensor transpose_163 = transpose(perm = var_916, x = var_915_cast); + tensor var_922_cast = matmul(transpose_x = var_922_transpose_x_0, transpose_y = var_922_transpose_y_0, x = var_920_cast, y = transpose_163); + tensor var_923 = const()[name = tensor("op_923"), val = tensor([0, 2, 1, 3])]; + tensor concat_7 = const()[name = tensor("concat_7"), val = tensor([1, 1500, 1024])]; + tensor transpose_160 = transpose(perm = var_923, x = var_922_cast); + tensor x_95_cast = reshape(shape = concat_7, x = transpose_160); + tensor var_928_to_fp16 = const()[name = tensor("op_928_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(192515968)))]; + tensor var_929_to_fp16 = const()[name = tensor("op_929_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194613184)))]; + tensor var_930_cast = linear(bias = var_929_to_fp16, weight = var_928_to_fp16, x = x_95_cast); + tensor x_97_cast = add(x = x_91_cast, y = var_930_cast); + tensor var_936_axes_0 = const()[name = tensor("op_936_axes_0"), val = tensor([-1])]; + tensor blocks_7_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_7_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194615296)))]; + tensor blocks_7_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_7_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194617408)))]; + tensor var_936_cast = layer_norm(axes = var_936_axes_0, beta = blocks_7_mlp_ln_bias_to_fp16, epsilon = var_861_to_fp16, gamma = blocks_7_mlp_ln_weight_to_fp16, x = x_97_cast); + tensor var_945_to_fp16 = const()[name = tensor("op_945_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(194619520)))]; + tensor var_946_to_fp16 = const()[name = tensor("op_946_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(203008192)))]; + tensor input_65_cast = linear(bias = var_946_to_fp16, weight = var_945_to_fp16, x = var_936_cast); + tensor x_101_mode_0 = const()[name = tensor("x_101_mode_0"), val = tensor("EXACT")]; + tensor x_101_cast = gelu(mode = x_101_mode_0, x = input_65_cast); + tensor var_951_to_fp16 = const()[name = tensor("op_951_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(203016448)))]; + tensor var_952_to_fp16 = const()[name = tensor("op_952_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211405120)))]; + tensor var_953_cast = linear(bias = var_952_to_fp16, weight = var_951_to_fp16, x = x_101_cast); + tensor x_103_cast = add(x = x_97_cast, y = var_953_cast); + tensor var_962 = const()[name = tensor("op_962"), val = tensor(-1)]; + tensor var_979_axes_0 = const()[name = tensor("op_979_axes_0"), val = tensor([-1])]; + tensor blocks_8_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_8_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211407232)))]; + tensor blocks_8_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_8_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211409344)))]; + tensor var_968_to_fp16 = const()[name = tensor("op_968_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_979_cast = layer_norm(axes = var_979_axes_0, beta = blocks_8_attn_ln_bias_to_fp16, epsilon = var_968_to_fp16, gamma = blocks_8_attn_ln_weight_to_fp16, x = x_103_cast); + tensor var_990_to_fp16 = const()[name = tensor("op_990_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(211411456)))]; + tensor var_991_to_fp16 = const()[name = tensor("op_991_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(213508672)))]; + tensor q_33_cast = linear(bias = var_991_to_fp16, weight = var_990_to_fp16, x = var_979_cast); + tensor var_994_to_fp16 = const()[name = tensor("op_994_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(213510784)))]; + tensor k_33_bias_0_to_fp16 = const()[name = tensor("k_33_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215608000)))]; + tensor k_33_cast = linear(bias = k_33_bias_0_to_fp16, weight = var_994_to_fp16, x = var_979_cast); + tensor var_998_to_fp16 = const()[name = tensor("op_998_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(215610112)))]; + tensor var_999_to_fp16 = const()[name = tensor("op_999_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217707328)))]; + tensor v_33_cast = linear(bias = var_999_to_fp16, weight = var_998_to_fp16, x = var_979_cast); + tensor var_1007 = const()[name = tensor("op_1007"), val = tensor([1, 1500, 16, -1])]; + tensor var_1008_cast = reshape(shape = var_1007, x = q_33_cast); + tensor const_184_to_fp16 = const()[name = tensor("const_184_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_35_cast = mul(x = var_1008_cast, y = const_184_to_fp16); + tensor var_1014 = const()[name = tensor("op_1014"), val = tensor([1, 1500, 16, -1])]; + tensor var_1015_cast = reshape(shape = var_1014, x = k_33_cast); + tensor const_185_to_fp16 = const()[name = tensor("const_185_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_35_cast = mul(x = var_1015_cast, y = const_185_to_fp16); + tensor var_1021 = const()[name = tensor("op_1021"), val = tensor([1, 1500, 16, -1])]; + tensor var_1022_cast = reshape(shape = var_1021, x = v_33_cast); + tensor var_1023 = const()[name = tensor("op_1023"), val = tensor([0, 2, 1, 3])]; + tensor qk_17_transpose_x_0 = const()[name = tensor("qk_17_transpose_x_0"), val = tensor(false)]; + tensor qk_17_transpose_y_0 = const()[name = tensor("qk_17_transpose_y_0"), val = tensor(false)]; + tensor transpose_64_perm_0 = const()[name = tensor("transpose_64_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_65_perm_0 = const()[name = tensor("transpose_65_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_157 = transpose(perm = transpose_65_perm_0, x = k_35_cast); + tensor transpose_158 = transpose(perm = transpose_64_perm_0, x = q_35_cast); + tensor qk_17_cast = matmul(transpose_x = qk_17_transpose_x_0, transpose_y = qk_17_transpose_y_0, x = transpose_158, y = transpose_157); + tensor var_1027_cast = softmax(axis = var_962, x = qk_17_cast); + tensor var_1029_transpose_x_0 = const()[name = tensor("op_1029_transpose_x_0"), val = tensor(false)]; + tensor var_1029_transpose_y_0 = const()[name = tensor("op_1029_transpose_y_0"), val = tensor(false)]; + tensor transpose_159 = transpose(perm = var_1023, x = var_1022_cast); + tensor var_1029_cast = matmul(transpose_x = var_1029_transpose_x_0, transpose_y = var_1029_transpose_y_0, x = var_1027_cast, y = transpose_159); + tensor var_1030 = const()[name = tensor("op_1030"), val = tensor([0, 2, 1, 3])]; + tensor concat_8 = const()[name = tensor("concat_8"), val = tensor([1, 1500, 1024])]; + tensor transpose_156 = transpose(perm = var_1030, x = var_1029_cast); + tensor x_107_cast = reshape(shape = concat_8, x = transpose_156); + tensor var_1035_to_fp16 = const()[name = tensor("op_1035_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(217709440)))]; + tensor var_1036_to_fp16 = const()[name = tensor("op_1036_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219806656)))]; + tensor var_1037_cast = linear(bias = var_1036_to_fp16, weight = var_1035_to_fp16, x = x_107_cast); + tensor x_109_cast = add(x = x_103_cast, y = var_1037_cast); + tensor var_1043_axes_0 = const()[name = tensor("op_1043_axes_0"), val = tensor([-1])]; + tensor blocks_8_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_8_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219808768)))]; + tensor blocks_8_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_8_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219810880)))]; + tensor var_1043_cast = layer_norm(axes = var_1043_axes_0, beta = blocks_8_mlp_ln_bias_to_fp16, epsilon = var_968_to_fp16, gamma = blocks_8_mlp_ln_weight_to_fp16, x = x_109_cast); + tensor var_1052_to_fp16 = const()[name = tensor("op_1052_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(219812992)))]; + tensor var_1053_to_fp16 = const()[name = tensor("op_1053_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228201664)))]; + tensor input_73_cast = linear(bias = var_1053_to_fp16, weight = var_1052_to_fp16, x = var_1043_cast); + tensor x_113_mode_0 = const()[name = tensor("x_113_mode_0"), val = tensor("EXACT")]; + tensor x_113_cast = gelu(mode = x_113_mode_0, x = input_73_cast); + tensor var_1058_to_fp16 = const()[name = tensor("op_1058_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(228209920)))]; + tensor var_1059_to_fp16 = const()[name = tensor("op_1059_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236598592)))]; + tensor var_1060_cast = linear(bias = var_1059_to_fp16, weight = var_1058_to_fp16, x = x_113_cast); + tensor x_115_cast = add(x = x_109_cast, y = var_1060_cast); + tensor var_1069 = const()[name = tensor("op_1069"), val = tensor(-1)]; + tensor var_1086_axes_0 = const()[name = tensor("op_1086_axes_0"), val = tensor([-1])]; + tensor blocks_9_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_9_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236600704)))]; + tensor blocks_9_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_9_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236602816)))]; + tensor var_1075_to_fp16 = const()[name = tensor("op_1075_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1086_cast = layer_norm(axes = var_1086_axes_0, beta = blocks_9_attn_ln_bias_to_fp16, epsilon = var_1075_to_fp16, gamma = blocks_9_attn_ln_weight_to_fp16, x = x_115_cast); + tensor var_1097_to_fp16 = const()[name = tensor("op_1097_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(236604928)))]; + tensor var_1098_to_fp16 = const()[name = tensor("op_1098_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(238702144)))]; + tensor q_37_cast = linear(bias = var_1098_to_fp16, weight = var_1097_to_fp16, x = var_1086_cast); + tensor var_1101_to_fp16 = const()[name = tensor("op_1101_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(238704256)))]; + tensor k_37_bias_0_to_fp16 = const()[name = tensor("k_37_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240801472)))]; + tensor k_37_cast = linear(bias = k_37_bias_0_to_fp16, weight = var_1101_to_fp16, x = var_1086_cast); + tensor var_1105_to_fp16 = const()[name = tensor("op_1105_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(240803584)))]; + tensor var_1106_to_fp16 = const()[name = tensor("op_1106_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242900800)))]; + tensor v_37_cast = linear(bias = var_1106_to_fp16, weight = var_1105_to_fp16, x = var_1086_cast); + tensor var_1114 = const()[name = tensor("op_1114"), val = tensor([1, 1500, 16, -1])]; + tensor var_1115_cast = reshape(shape = var_1114, x = q_37_cast); + tensor const_186_to_fp16 = const()[name = tensor("const_186_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_39_cast = mul(x = var_1115_cast, y = const_186_to_fp16); + tensor var_1121 = const()[name = tensor("op_1121"), val = tensor([1, 1500, 16, -1])]; + tensor var_1122_cast = reshape(shape = var_1121, x = k_37_cast); + tensor const_187_to_fp16 = const()[name = tensor("const_187_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_39_cast = mul(x = var_1122_cast, y = const_187_to_fp16); + tensor var_1128 = const()[name = tensor("op_1128"), val = tensor([1, 1500, 16, -1])]; + tensor var_1129_cast = reshape(shape = var_1128, x = v_37_cast); + tensor var_1130 = const()[name = tensor("op_1130"), val = tensor([0, 2, 1, 3])]; + tensor qk_19_transpose_x_0 = const()[name = tensor("qk_19_transpose_x_0"), val = tensor(false)]; + tensor qk_19_transpose_y_0 = const()[name = tensor("qk_19_transpose_y_0"), val = tensor(false)]; + tensor transpose_66_perm_0 = const()[name = tensor("transpose_66_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_67_perm_0 = const()[name = tensor("transpose_67_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_153 = transpose(perm = transpose_67_perm_0, x = k_39_cast); + tensor transpose_154 = transpose(perm = transpose_66_perm_0, x = q_39_cast); + tensor qk_19_cast = matmul(transpose_x = qk_19_transpose_x_0, transpose_y = qk_19_transpose_y_0, x = transpose_154, y = transpose_153); + tensor var_1134_cast = softmax(axis = var_1069, x = qk_19_cast); + tensor var_1136_transpose_x_0 = const()[name = tensor("op_1136_transpose_x_0"), val = tensor(false)]; + tensor var_1136_transpose_y_0 = const()[name = tensor("op_1136_transpose_y_0"), val = tensor(false)]; + tensor transpose_155 = transpose(perm = var_1130, x = var_1129_cast); + tensor var_1136_cast = matmul(transpose_x = var_1136_transpose_x_0, transpose_y = var_1136_transpose_y_0, x = var_1134_cast, y = transpose_155); + tensor var_1137 = const()[name = tensor("op_1137"), val = tensor([0, 2, 1, 3])]; + tensor concat_9 = const()[name = tensor("concat_9"), val = tensor([1, 1500, 1024])]; + tensor transpose_152 = transpose(perm = var_1137, x = var_1136_cast); + tensor x_119_cast = reshape(shape = concat_9, x = transpose_152); + tensor var_1142_to_fp16 = const()[name = tensor("op_1142_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(242902912)))]; + tensor var_1143_to_fp16 = const()[name = tensor("op_1143_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(245000128)))]; + tensor var_1144_cast = linear(bias = var_1143_to_fp16, weight = var_1142_to_fp16, x = x_119_cast); + tensor x_121_cast = add(x = x_115_cast, y = var_1144_cast); + tensor var_1150_axes_0 = const()[name = tensor("op_1150_axes_0"), val = tensor([-1])]; + tensor blocks_9_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_9_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(245002240)))]; + tensor blocks_9_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_9_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(245004352)))]; + tensor var_1150_cast = layer_norm(axes = var_1150_axes_0, beta = blocks_9_mlp_ln_bias_to_fp16, epsilon = var_1075_to_fp16, gamma = blocks_9_mlp_ln_weight_to_fp16, x = x_121_cast); + tensor var_1159_to_fp16 = const()[name = tensor("op_1159_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(245006464)))]; + tensor var_1160_to_fp16 = const()[name = tensor("op_1160_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(253395136)))]; + tensor input_81_cast = linear(bias = var_1160_to_fp16, weight = var_1159_to_fp16, x = var_1150_cast); + tensor x_125_mode_0 = const()[name = tensor("x_125_mode_0"), val = tensor("EXACT")]; + tensor x_125_cast = gelu(mode = x_125_mode_0, x = input_81_cast); + tensor var_1165_to_fp16 = const()[name = tensor("op_1165_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(253403392)))]; + tensor var_1166_to_fp16 = const()[name = tensor("op_1166_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261792064)))]; + tensor var_1167_cast = linear(bias = var_1166_to_fp16, weight = var_1165_to_fp16, x = x_125_cast); + tensor x_127_cast = add(x = x_121_cast, y = var_1167_cast); + tensor var_1176 = const()[name = tensor("op_1176"), val = tensor(-1)]; + tensor var_1193_axes_0 = const()[name = tensor("op_1193_axes_0"), val = tensor([-1])]; + tensor blocks_10_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_10_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261794176)))]; + tensor blocks_10_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_10_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261796288)))]; + tensor var_1182_to_fp16 = const()[name = tensor("op_1182_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1193_cast = layer_norm(axes = var_1193_axes_0, beta = blocks_10_attn_ln_bias_to_fp16, epsilon = var_1182_to_fp16, gamma = blocks_10_attn_ln_weight_to_fp16, x = x_127_cast); + tensor var_1204_to_fp16 = const()[name = tensor("op_1204_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(261798400)))]; + tensor var_1205_to_fp16 = const()[name = tensor("op_1205_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263895616)))]; + tensor q_41_cast = linear(bias = var_1205_to_fp16, weight = var_1204_to_fp16, x = var_1193_cast); + tensor var_1208_to_fp16 = const()[name = tensor("op_1208_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(263897728)))]; + tensor k_41_bias_0_to_fp16 = const()[name = tensor("k_41_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265994944)))]; + tensor k_41_cast = linear(bias = k_41_bias_0_to_fp16, weight = var_1208_to_fp16, x = var_1193_cast); + tensor var_1212_to_fp16 = const()[name = tensor("op_1212_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(265997056)))]; + tensor var_1213_to_fp16 = const()[name = tensor("op_1213_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268094272)))]; + tensor v_41_cast = linear(bias = var_1213_to_fp16, weight = var_1212_to_fp16, x = var_1193_cast); + tensor var_1221 = const()[name = tensor("op_1221"), val = tensor([1, 1500, 16, -1])]; + tensor var_1222_cast = reshape(shape = var_1221, x = q_41_cast); + tensor const_188_to_fp16 = const()[name = tensor("const_188_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_43_cast = mul(x = var_1222_cast, y = const_188_to_fp16); + tensor var_1228 = const()[name = tensor("op_1228"), val = tensor([1, 1500, 16, -1])]; + tensor var_1229_cast = reshape(shape = var_1228, x = k_41_cast); + tensor const_189_to_fp16 = const()[name = tensor("const_189_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_43_cast = mul(x = var_1229_cast, y = const_189_to_fp16); + tensor var_1235 = const()[name = tensor("op_1235"), val = tensor([1, 1500, 16, -1])]; + tensor var_1236_cast = reshape(shape = var_1235, x = v_41_cast); + tensor var_1237 = const()[name = tensor("op_1237"), val = tensor([0, 2, 1, 3])]; + tensor qk_21_transpose_x_0 = const()[name = tensor("qk_21_transpose_x_0"), val = tensor(false)]; + tensor qk_21_transpose_y_0 = const()[name = tensor("qk_21_transpose_y_0"), val = tensor(false)]; + tensor transpose_68_perm_0 = const()[name = tensor("transpose_68_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_69_perm_0 = const()[name = tensor("transpose_69_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_149 = transpose(perm = transpose_69_perm_0, x = k_43_cast); + tensor transpose_150 = transpose(perm = transpose_68_perm_0, x = q_43_cast); + tensor qk_21_cast = matmul(transpose_x = qk_21_transpose_x_0, transpose_y = qk_21_transpose_y_0, x = transpose_150, y = transpose_149); + tensor var_1241_cast = softmax(axis = var_1176, x = qk_21_cast); + tensor var_1243_transpose_x_0 = const()[name = tensor("op_1243_transpose_x_0"), val = tensor(false)]; + tensor var_1243_transpose_y_0 = const()[name = tensor("op_1243_transpose_y_0"), val = tensor(false)]; + tensor transpose_151 = transpose(perm = var_1237, x = var_1236_cast); + tensor var_1243_cast = matmul(transpose_x = var_1243_transpose_x_0, transpose_y = var_1243_transpose_y_0, x = var_1241_cast, y = transpose_151); + tensor var_1244 = const()[name = tensor("op_1244"), val = tensor([0, 2, 1, 3])]; + tensor concat_10 = const()[name = tensor("concat_10"), val = tensor([1, 1500, 1024])]; + tensor transpose_148 = transpose(perm = var_1244, x = var_1243_cast); + tensor x_131_cast = reshape(shape = concat_10, x = transpose_148); + tensor var_1249_to_fp16 = const()[name = tensor("op_1249_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(268096384)))]; + tensor var_1250_to_fp16 = const()[name = tensor("op_1250_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(270193600)))]; + tensor var_1251_cast = linear(bias = var_1250_to_fp16, weight = var_1249_to_fp16, x = x_131_cast); + tensor x_133_cast = add(x = x_127_cast, y = var_1251_cast); + tensor var_1257_axes_0 = const()[name = tensor("op_1257_axes_0"), val = tensor([-1])]; + tensor blocks_10_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_10_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(270195712)))]; + tensor blocks_10_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_10_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(270197824)))]; + tensor var_1257_cast = layer_norm(axes = var_1257_axes_0, beta = blocks_10_mlp_ln_bias_to_fp16, epsilon = var_1182_to_fp16, gamma = blocks_10_mlp_ln_weight_to_fp16, x = x_133_cast); + tensor var_1266_to_fp16 = const()[name = tensor("op_1266_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(270199936)))]; + tensor var_1267_to_fp16 = const()[name = tensor("op_1267_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(278588608)))]; + tensor input_89_cast = linear(bias = var_1267_to_fp16, weight = var_1266_to_fp16, x = var_1257_cast); + tensor x_137_mode_0 = const()[name = tensor("x_137_mode_0"), val = tensor("EXACT")]; + tensor x_137_cast = gelu(mode = x_137_mode_0, x = input_89_cast); + tensor var_1272_to_fp16 = const()[name = tensor("op_1272_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(278596864)))]; + tensor var_1273_to_fp16 = const()[name = tensor("op_1273_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286985536)))]; + tensor var_1274_cast = linear(bias = var_1273_to_fp16, weight = var_1272_to_fp16, x = x_137_cast); + tensor x_139_cast = add(x = x_133_cast, y = var_1274_cast); + tensor var_1283 = const()[name = tensor("op_1283"), val = tensor(-1)]; + tensor var_1300_axes_0 = const()[name = tensor("op_1300_axes_0"), val = tensor([-1])]; + tensor blocks_11_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_11_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286987648)))]; + tensor blocks_11_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_11_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286989760)))]; + tensor var_1289_to_fp16 = const()[name = tensor("op_1289_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1300_cast = layer_norm(axes = var_1300_axes_0, beta = blocks_11_attn_ln_bias_to_fp16, epsilon = var_1289_to_fp16, gamma = blocks_11_attn_ln_weight_to_fp16, x = x_139_cast); + tensor var_1311_to_fp16 = const()[name = tensor("op_1311_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(286991872)))]; + tensor var_1312_to_fp16 = const()[name = tensor("op_1312_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289089088)))]; + tensor q_45_cast = linear(bias = var_1312_to_fp16, weight = var_1311_to_fp16, x = var_1300_cast); + tensor var_1315_to_fp16 = const()[name = tensor("op_1315_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(289091200)))]; + tensor k_45_bias_0_to_fp16 = const()[name = tensor("k_45_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291188416)))]; + tensor k_45_cast = linear(bias = k_45_bias_0_to_fp16, weight = var_1315_to_fp16, x = var_1300_cast); + tensor var_1319_to_fp16 = const()[name = tensor("op_1319_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(291190528)))]; + tensor var_1320_to_fp16 = const()[name = tensor("op_1320_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293287744)))]; + tensor v_45_cast = linear(bias = var_1320_to_fp16, weight = var_1319_to_fp16, x = var_1300_cast); + tensor var_1328 = const()[name = tensor("op_1328"), val = tensor([1, 1500, 16, -1])]; + tensor var_1329_cast = reshape(shape = var_1328, x = q_45_cast); + tensor const_190_to_fp16 = const()[name = tensor("const_190_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_47_cast = mul(x = var_1329_cast, y = const_190_to_fp16); + tensor var_1335 = const()[name = tensor("op_1335"), val = tensor([1, 1500, 16, -1])]; + tensor var_1336_cast = reshape(shape = var_1335, x = k_45_cast); + tensor const_191_to_fp16 = const()[name = tensor("const_191_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_47_cast = mul(x = var_1336_cast, y = const_191_to_fp16); + tensor var_1342 = const()[name = tensor("op_1342"), val = tensor([1, 1500, 16, -1])]; + tensor var_1343_cast = reshape(shape = var_1342, x = v_45_cast); + tensor var_1344 = const()[name = tensor("op_1344"), val = tensor([0, 2, 1, 3])]; + tensor qk_23_transpose_x_0 = const()[name = tensor("qk_23_transpose_x_0"), val = tensor(false)]; + tensor qk_23_transpose_y_0 = const()[name = tensor("qk_23_transpose_y_0"), val = tensor(false)]; + tensor transpose_70_perm_0 = const()[name = tensor("transpose_70_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_71_perm_0 = const()[name = tensor("transpose_71_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_145 = transpose(perm = transpose_71_perm_0, x = k_47_cast); + tensor transpose_146 = transpose(perm = transpose_70_perm_0, x = q_47_cast); + tensor qk_23_cast = matmul(transpose_x = qk_23_transpose_x_0, transpose_y = qk_23_transpose_y_0, x = transpose_146, y = transpose_145); + tensor var_1348_cast = softmax(axis = var_1283, x = qk_23_cast); + tensor var_1350_transpose_x_0 = const()[name = tensor("op_1350_transpose_x_0"), val = tensor(false)]; + tensor var_1350_transpose_y_0 = const()[name = tensor("op_1350_transpose_y_0"), val = tensor(false)]; + tensor transpose_147 = transpose(perm = var_1344, x = var_1343_cast); + tensor var_1350_cast = matmul(transpose_x = var_1350_transpose_x_0, transpose_y = var_1350_transpose_y_0, x = var_1348_cast, y = transpose_147); + tensor var_1351 = const()[name = tensor("op_1351"), val = tensor([0, 2, 1, 3])]; + tensor concat_11 = const()[name = tensor("concat_11"), val = tensor([1, 1500, 1024])]; + tensor transpose_144 = transpose(perm = var_1351, x = var_1350_cast); + tensor x_143_cast = reshape(shape = concat_11, x = transpose_144); + tensor var_1356_to_fp16 = const()[name = tensor("op_1356_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(293289856)))]; + tensor var_1357_to_fp16 = const()[name = tensor("op_1357_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(295387072)))]; + tensor var_1358_cast = linear(bias = var_1357_to_fp16, weight = var_1356_to_fp16, x = x_143_cast); + tensor x_145_cast = add(x = x_139_cast, y = var_1358_cast); + tensor var_1364_axes_0 = const()[name = tensor("op_1364_axes_0"), val = tensor([-1])]; + tensor blocks_11_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_11_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(295389184)))]; + tensor blocks_11_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_11_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(295391296)))]; + tensor var_1364_cast = layer_norm(axes = var_1364_axes_0, beta = blocks_11_mlp_ln_bias_to_fp16, epsilon = var_1289_to_fp16, gamma = blocks_11_mlp_ln_weight_to_fp16, x = x_145_cast); + tensor var_1373_to_fp16 = const()[name = tensor("op_1373_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(295393408)))]; + tensor var_1374_to_fp16 = const()[name = tensor("op_1374_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303782080)))]; + tensor input_97_cast = linear(bias = var_1374_to_fp16, weight = var_1373_to_fp16, x = var_1364_cast); + tensor x_149_mode_0 = const()[name = tensor("x_149_mode_0"), val = tensor("EXACT")]; + tensor x_149_cast = gelu(mode = x_149_mode_0, x = input_97_cast); + tensor var_1379_to_fp16 = const()[name = tensor("op_1379_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(303790336)))]; + tensor var_1380_to_fp16 = const()[name = tensor("op_1380_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312179008)))]; + tensor var_1381_cast = linear(bias = var_1380_to_fp16, weight = var_1379_to_fp16, x = x_149_cast); + tensor x_151_cast = add(x = x_145_cast, y = var_1381_cast); + tensor var_1390 = const()[name = tensor("op_1390"), val = tensor(-1)]; + tensor var_1407_axes_0 = const()[name = tensor("op_1407_axes_0"), val = tensor([-1])]; + tensor blocks_12_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_12_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312181120)))]; + tensor blocks_12_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_12_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312183232)))]; + tensor var_1396_to_fp16 = const()[name = tensor("op_1396_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1407_cast = layer_norm(axes = var_1407_axes_0, beta = blocks_12_attn_ln_bias_to_fp16, epsilon = var_1396_to_fp16, gamma = blocks_12_attn_ln_weight_to_fp16, x = x_151_cast); + tensor var_1418_to_fp16 = const()[name = tensor("op_1418_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(312185344)))]; + tensor var_1419_to_fp16 = const()[name = tensor("op_1419_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314282560)))]; + tensor q_49_cast = linear(bias = var_1419_to_fp16, weight = var_1418_to_fp16, x = var_1407_cast); + tensor var_1422_to_fp16 = const()[name = tensor("op_1422_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(314284672)))]; + tensor k_49_bias_0_to_fp16 = const()[name = tensor("k_49_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316381888)))]; + tensor k_49_cast = linear(bias = k_49_bias_0_to_fp16, weight = var_1422_to_fp16, x = var_1407_cast); + tensor var_1426_to_fp16 = const()[name = tensor("op_1426_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(316384000)))]; + tensor var_1427_to_fp16 = const()[name = tensor("op_1427_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(318481216)))]; + tensor v_49_cast = linear(bias = var_1427_to_fp16, weight = var_1426_to_fp16, x = var_1407_cast); + tensor var_1435 = const()[name = tensor("op_1435"), val = tensor([1, 1500, 16, -1])]; + tensor var_1436_cast = reshape(shape = var_1435, x = q_49_cast); + tensor const_192_to_fp16 = const()[name = tensor("const_192_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_51_cast = mul(x = var_1436_cast, y = const_192_to_fp16); + tensor var_1442 = const()[name = tensor("op_1442"), val = tensor([1, 1500, 16, -1])]; + tensor var_1443_cast = reshape(shape = var_1442, x = k_49_cast); + tensor const_193_to_fp16 = const()[name = tensor("const_193_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_51_cast = mul(x = var_1443_cast, y = const_193_to_fp16); + tensor var_1449 = const()[name = tensor("op_1449"), val = tensor([1, 1500, 16, -1])]; + tensor var_1450_cast = reshape(shape = var_1449, x = v_49_cast); + tensor var_1451 = const()[name = tensor("op_1451"), val = tensor([0, 2, 1, 3])]; + tensor qk_25_transpose_x_0 = const()[name = tensor("qk_25_transpose_x_0"), val = tensor(false)]; + tensor qk_25_transpose_y_0 = const()[name = tensor("qk_25_transpose_y_0"), val = tensor(false)]; + tensor transpose_72_perm_0 = const()[name = tensor("transpose_72_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_73_perm_0 = const()[name = tensor("transpose_73_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_141 = transpose(perm = transpose_73_perm_0, x = k_51_cast); + tensor transpose_142 = transpose(perm = transpose_72_perm_0, x = q_51_cast); + tensor qk_25_cast = matmul(transpose_x = qk_25_transpose_x_0, transpose_y = qk_25_transpose_y_0, x = transpose_142, y = transpose_141); + tensor var_1455_cast = softmax(axis = var_1390, x = qk_25_cast); + tensor var_1457_transpose_x_0 = const()[name = tensor("op_1457_transpose_x_0"), val = tensor(false)]; + tensor var_1457_transpose_y_0 = const()[name = tensor("op_1457_transpose_y_0"), val = tensor(false)]; + tensor transpose_143 = transpose(perm = var_1451, x = var_1450_cast); + tensor var_1457_cast = matmul(transpose_x = var_1457_transpose_x_0, transpose_y = var_1457_transpose_y_0, x = var_1455_cast, y = transpose_143); + tensor var_1458 = const()[name = tensor("op_1458"), val = tensor([0, 2, 1, 3])]; + tensor concat_12 = const()[name = tensor("concat_12"), val = tensor([1, 1500, 1024])]; + tensor transpose_140 = transpose(perm = var_1458, x = var_1457_cast); + tensor x_155_cast = reshape(shape = concat_12, x = transpose_140); + tensor var_1463_to_fp16 = const()[name = tensor("op_1463_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(318483328)))]; + tensor var_1464_to_fp16 = const()[name = tensor("op_1464_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(320580544)))]; + tensor var_1465_cast = linear(bias = var_1464_to_fp16, weight = var_1463_to_fp16, x = x_155_cast); + tensor x_157_cast = add(x = x_151_cast, y = var_1465_cast); + tensor var_1471_axes_0 = const()[name = tensor("op_1471_axes_0"), val = tensor([-1])]; + tensor blocks_12_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_12_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(320582656)))]; + tensor blocks_12_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_12_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(320584768)))]; + tensor var_1471_cast = layer_norm(axes = var_1471_axes_0, beta = blocks_12_mlp_ln_bias_to_fp16, epsilon = var_1396_to_fp16, gamma = blocks_12_mlp_ln_weight_to_fp16, x = x_157_cast); + tensor var_1480_to_fp16 = const()[name = tensor("op_1480_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(320586880)))]; + tensor var_1481_to_fp16 = const()[name = tensor("op_1481_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(328975552)))]; + tensor input_105_cast = linear(bias = var_1481_to_fp16, weight = var_1480_to_fp16, x = var_1471_cast); + tensor x_161_mode_0 = const()[name = tensor("x_161_mode_0"), val = tensor("EXACT")]; + tensor x_161_cast = gelu(mode = x_161_mode_0, x = input_105_cast); + tensor var_1486_to_fp16 = const()[name = tensor("op_1486_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(328983808)))]; + tensor var_1487_to_fp16 = const()[name = tensor("op_1487_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337372480)))]; + tensor var_1488_cast = linear(bias = var_1487_to_fp16, weight = var_1486_to_fp16, x = x_161_cast); + tensor x_163_cast = add(x = x_157_cast, y = var_1488_cast); + tensor var_1497 = const()[name = tensor("op_1497"), val = tensor(-1)]; + tensor var_1514_axes_0 = const()[name = tensor("op_1514_axes_0"), val = tensor([-1])]; + tensor blocks_13_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_13_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337374592)))]; + tensor blocks_13_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_13_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337376704)))]; + tensor var_1503_to_fp16 = const()[name = tensor("op_1503_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1514_cast = layer_norm(axes = var_1514_axes_0, beta = blocks_13_attn_ln_bias_to_fp16, epsilon = var_1503_to_fp16, gamma = blocks_13_attn_ln_weight_to_fp16, x = x_163_cast); + tensor var_1525_to_fp16 = const()[name = tensor("op_1525_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(337378816)))]; + tensor var_1526_to_fp16 = const()[name = tensor("op_1526_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339476032)))]; + tensor q_53_cast = linear(bias = var_1526_to_fp16, weight = var_1525_to_fp16, x = var_1514_cast); + tensor var_1529_to_fp16 = const()[name = tensor("op_1529_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(339478144)))]; + tensor k_53_bias_0_to_fp16 = const()[name = tensor("k_53_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(341575360)))]; + tensor k_53_cast = linear(bias = k_53_bias_0_to_fp16, weight = var_1529_to_fp16, x = var_1514_cast); + tensor var_1533_to_fp16 = const()[name = tensor("op_1533_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(341577472)))]; + tensor var_1534_to_fp16 = const()[name = tensor("op_1534_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(343674688)))]; + tensor v_53_cast = linear(bias = var_1534_to_fp16, weight = var_1533_to_fp16, x = var_1514_cast); + tensor var_1542 = const()[name = tensor("op_1542"), val = tensor([1, 1500, 16, -1])]; + tensor var_1543_cast = reshape(shape = var_1542, x = q_53_cast); + tensor const_194_to_fp16 = const()[name = tensor("const_194_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_55_cast = mul(x = var_1543_cast, y = const_194_to_fp16); + tensor var_1549 = const()[name = tensor("op_1549"), val = tensor([1, 1500, 16, -1])]; + tensor var_1550_cast = reshape(shape = var_1549, x = k_53_cast); + tensor const_195_to_fp16 = const()[name = tensor("const_195_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_55_cast = mul(x = var_1550_cast, y = const_195_to_fp16); + tensor var_1556 = const()[name = tensor("op_1556"), val = tensor([1, 1500, 16, -1])]; + tensor var_1557_cast = reshape(shape = var_1556, x = v_53_cast); + tensor var_1558 = const()[name = tensor("op_1558"), val = tensor([0, 2, 1, 3])]; + tensor qk_27_transpose_x_0 = const()[name = tensor("qk_27_transpose_x_0"), val = tensor(false)]; + tensor qk_27_transpose_y_0 = const()[name = tensor("qk_27_transpose_y_0"), val = tensor(false)]; + tensor transpose_74_perm_0 = const()[name = tensor("transpose_74_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_75_perm_0 = const()[name = tensor("transpose_75_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_137 = transpose(perm = transpose_75_perm_0, x = k_55_cast); + tensor transpose_138 = transpose(perm = transpose_74_perm_0, x = q_55_cast); + tensor qk_27_cast = matmul(transpose_x = qk_27_transpose_x_0, transpose_y = qk_27_transpose_y_0, x = transpose_138, y = transpose_137); + tensor var_1562_cast = softmax(axis = var_1497, x = qk_27_cast); + tensor var_1564_transpose_x_0 = const()[name = tensor("op_1564_transpose_x_0"), val = tensor(false)]; + tensor var_1564_transpose_y_0 = const()[name = tensor("op_1564_transpose_y_0"), val = tensor(false)]; + tensor transpose_139 = transpose(perm = var_1558, x = var_1557_cast); + tensor var_1564_cast = matmul(transpose_x = var_1564_transpose_x_0, transpose_y = var_1564_transpose_y_0, x = var_1562_cast, y = transpose_139); + tensor var_1565 = const()[name = tensor("op_1565"), val = tensor([0, 2, 1, 3])]; + tensor concat_13 = const()[name = tensor("concat_13"), val = tensor([1, 1500, 1024])]; + tensor transpose_136 = transpose(perm = var_1565, x = var_1564_cast); + tensor x_167_cast = reshape(shape = concat_13, x = transpose_136); + tensor var_1570_to_fp16 = const()[name = tensor("op_1570_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(343676800)))]; + tensor var_1571_to_fp16 = const()[name = tensor("op_1571_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(345774016)))]; + tensor var_1572_cast = linear(bias = var_1571_to_fp16, weight = var_1570_to_fp16, x = x_167_cast); + tensor x_169_cast = add(x = x_163_cast, y = var_1572_cast); + tensor var_1578_axes_0 = const()[name = tensor("op_1578_axes_0"), val = tensor([-1])]; + tensor blocks_13_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_13_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(345776128)))]; + tensor blocks_13_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_13_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(345778240)))]; + tensor var_1578_cast = layer_norm(axes = var_1578_axes_0, beta = blocks_13_mlp_ln_bias_to_fp16, epsilon = var_1503_to_fp16, gamma = blocks_13_mlp_ln_weight_to_fp16, x = x_169_cast); + tensor var_1587_to_fp16 = const()[name = tensor("op_1587_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(345780352)))]; + tensor var_1588_to_fp16 = const()[name = tensor("op_1588_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(354169024)))]; + tensor input_113_cast = linear(bias = var_1588_to_fp16, weight = var_1587_to_fp16, x = var_1578_cast); + tensor x_173_mode_0 = const()[name = tensor("x_173_mode_0"), val = tensor("EXACT")]; + tensor x_173_cast = gelu(mode = x_173_mode_0, x = input_113_cast); + tensor var_1593_to_fp16 = const()[name = tensor("op_1593_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(354177280)))]; + tensor var_1594_to_fp16 = const()[name = tensor("op_1594_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362565952)))]; + tensor var_1595_cast = linear(bias = var_1594_to_fp16, weight = var_1593_to_fp16, x = x_173_cast); + tensor x_175_cast = add(x = x_169_cast, y = var_1595_cast); + tensor var_1604 = const()[name = tensor("op_1604"), val = tensor(-1)]; + tensor var_1621_axes_0 = const()[name = tensor("op_1621_axes_0"), val = tensor([-1])]; + tensor blocks_14_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_14_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362568064)))]; + tensor blocks_14_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_14_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362570176)))]; + tensor var_1610_to_fp16 = const()[name = tensor("op_1610_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1621_cast = layer_norm(axes = var_1621_axes_0, beta = blocks_14_attn_ln_bias_to_fp16, epsilon = var_1610_to_fp16, gamma = blocks_14_attn_ln_weight_to_fp16, x = x_175_cast); + tensor var_1632_to_fp16 = const()[name = tensor("op_1632_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(362572288)))]; + tensor var_1633_to_fp16 = const()[name = tensor("op_1633_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364669504)))]; + tensor q_57_cast = linear(bias = var_1633_to_fp16, weight = var_1632_to_fp16, x = var_1621_cast); + tensor var_1636_to_fp16 = const()[name = tensor("op_1636_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(364671616)))]; + tensor k_57_bias_0_to_fp16 = const()[name = tensor("k_57_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(366768832)))]; + tensor k_57_cast = linear(bias = k_57_bias_0_to_fp16, weight = var_1636_to_fp16, x = var_1621_cast); + tensor var_1640_to_fp16 = const()[name = tensor("op_1640_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(366770944)))]; + tensor var_1641_to_fp16 = const()[name = tensor("op_1641_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368868160)))]; + tensor v_57_cast = linear(bias = var_1641_to_fp16, weight = var_1640_to_fp16, x = var_1621_cast); + tensor var_1649 = const()[name = tensor("op_1649"), val = tensor([1, 1500, 16, -1])]; + tensor var_1650_cast = reshape(shape = var_1649, x = q_57_cast); + tensor const_196_to_fp16 = const()[name = tensor("const_196_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_59_cast = mul(x = var_1650_cast, y = const_196_to_fp16); + tensor var_1656 = const()[name = tensor("op_1656"), val = tensor([1, 1500, 16, -1])]; + tensor var_1657_cast = reshape(shape = var_1656, x = k_57_cast); + tensor const_197_to_fp16 = const()[name = tensor("const_197_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_59_cast = mul(x = var_1657_cast, y = const_197_to_fp16); + tensor var_1663 = const()[name = tensor("op_1663"), val = tensor([1, 1500, 16, -1])]; + tensor var_1664_cast = reshape(shape = var_1663, x = v_57_cast); + tensor var_1665 = const()[name = tensor("op_1665"), val = tensor([0, 2, 1, 3])]; + tensor qk_29_transpose_x_0 = const()[name = tensor("qk_29_transpose_x_0"), val = tensor(false)]; + tensor qk_29_transpose_y_0 = const()[name = tensor("qk_29_transpose_y_0"), val = tensor(false)]; + tensor transpose_76_perm_0 = const()[name = tensor("transpose_76_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_77_perm_0 = const()[name = tensor("transpose_77_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_133 = transpose(perm = transpose_77_perm_0, x = k_59_cast); + tensor transpose_134 = transpose(perm = transpose_76_perm_0, x = q_59_cast); + tensor qk_29_cast = matmul(transpose_x = qk_29_transpose_x_0, transpose_y = qk_29_transpose_y_0, x = transpose_134, y = transpose_133); + tensor var_1669_cast = softmax(axis = var_1604, x = qk_29_cast); + tensor var_1671_transpose_x_0 = const()[name = tensor("op_1671_transpose_x_0"), val = tensor(false)]; + tensor var_1671_transpose_y_0 = const()[name = tensor("op_1671_transpose_y_0"), val = tensor(false)]; + tensor transpose_135 = transpose(perm = var_1665, x = var_1664_cast); + tensor var_1671_cast = matmul(transpose_x = var_1671_transpose_x_0, transpose_y = var_1671_transpose_y_0, x = var_1669_cast, y = transpose_135); + tensor var_1672 = const()[name = tensor("op_1672"), val = tensor([0, 2, 1, 3])]; + tensor concat_14 = const()[name = tensor("concat_14"), val = tensor([1, 1500, 1024])]; + tensor transpose_132 = transpose(perm = var_1672, x = var_1671_cast); + tensor x_179_cast = reshape(shape = concat_14, x = transpose_132); + tensor var_1677_to_fp16 = const()[name = tensor("op_1677_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368870272)))]; + tensor var_1678_to_fp16 = const()[name = tensor("op_1678_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(370967488)))]; + tensor var_1679_cast = linear(bias = var_1678_to_fp16, weight = var_1677_to_fp16, x = x_179_cast); + tensor x_181_cast = add(x = x_175_cast, y = var_1679_cast); + tensor var_1685_axes_0 = const()[name = tensor("op_1685_axes_0"), val = tensor([-1])]; + tensor blocks_14_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_14_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(370969600)))]; + tensor blocks_14_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_14_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(370971712)))]; + tensor var_1685_cast = layer_norm(axes = var_1685_axes_0, beta = blocks_14_mlp_ln_bias_to_fp16, epsilon = var_1610_to_fp16, gamma = blocks_14_mlp_ln_weight_to_fp16, x = x_181_cast); + tensor var_1694_to_fp16 = const()[name = tensor("op_1694_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(370973824)))]; + tensor var_1695_to_fp16 = const()[name = tensor("op_1695_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(379362496)))]; + tensor input_121_cast = linear(bias = var_1695_to_fp16, weight = var_1694_to_fp16, x = var_1685_cast); + tensor x_185_mode_0 = const()[name = tensor("x_185_mode_0"), val = tensor("EXACT")]; + tensor x_185_cast = gelu(mode = x_185_mode_0, x = input_121_cast); + tensor var_1700_to_fp16 = const()[name = tensor("op_1700_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(379370752)))]; + tensor var_1701_to_fp16 = const()[name = tensor("op_1701_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387759424)))]; + tensor var_1702_cast = linear(bias = var_1701_to_fp16, weight = var_1700_to_fp16, x = x_185_cast); + tensor x_187_cast = add(x = x_181_cast, y = var_1702_cast); + tensor var_1711 = const()[name = tensor("op_1711"), val = tensor(-1)]; + tensor var_1728_axes_0 = const()[name = tensor("op_1728_axes_0"), val = tensor([-1])]; + tensor blocks_15_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_15_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387761536)))]; + tensor blocks_15_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_15_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387763648)))]; + tensor var_1717_to_fp16 = const()[name = tensor("op_1717_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1728_cast = layer_norm(axes = var_1728_axes_0, beta = blocks_15_attn_ln_bias_to_fp16, epsilon = var_1717_to_fp16, gamma = blocks_15_attn_ln_weight_to_fp16, x = x_187_cast); + tensor var_1739_to_fp16 = const()[name = tensor("op_1739_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(387765760)))]; + tensor var_1740_to_fp16 = const()[name = tensor("op_1740_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(389862976)))]; + tensor q_61_cast = linear(bias = var_1740_to_fp16, weight = var_1739_to_fp16, x = var_1728_cast); + tensor var_1743_to_fp16 = const()[name = tensor("op_1743_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(389865088)))]; + tensor k_61_bias_0_to_fp16 = const()[name = tensor("k_61_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(391962304)))]; + tensor k_61_cast = linear(bias = k_61_bias_0_to_fp16, weight = var_1743_to_fp16, x = var_1728_cast); + tensor var_1747_to_fp16 = const()[name = tensor("op_1747_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(391964416)))]; + tensor var_1748_to_fp16 = const()[name = tensor("op_1748_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(394061632)))]; + tensor v_61_cast = linear(bias = var_1748_to_fp16, weight = var_1747_to_fp16, x = var_1728_cast); + tensor var_1756 = const()[name = tensor("op_1756"), val = tensor([1, 1500, 16, -1])]; + tensor var_1757_cast = reshape(shape = var_1756, x = q_61_cast); + tensor const_198_to_fp16 = const()[name = tensor("const_198_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_63_cast = mul(x = var_1757_cast, y = const_198_to_fp16); + tensor var_1763 = const()[name = tensor("op_1763"), val = tensor([1, 1500, 16, -1])]; + tensor var_1764_cast = reshape(shape = var_1763, x = k_61_cast); + tensor const_199_to_fp16 = const()[name = tensor("const_199_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_63_cast = mul(x = var_1764_cast, y = const_199_to_fp16); + tensor var_1770 = const()[name = tensor("op_1770"), val = tensor([1, 1500, 16, -1])]; + tensor var_1771_cast = reshape(shape = var_1770, x = v_61_cast); + tensor var_1772 = const()[name = tensor("op_1772"), val = tensor([0, 2, 1, 3])]; + tensor qk_31_transpose_x_0 = const()[name = tensor("qk_31_transpose_x_0"), val = tensor(false)]; + tensor qk_31_transpose_y_0 = const()[name = tensor("qk_31_transpose_y_0"), val = tensor(false)]; + tensor transpose_78_perm_0 = const()[name = tensor("transpose_78_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_79_perm_0 = const()[name = tensor("transpose_79_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_129 = transpose(perm = transpose_79_perm_0, x = k_63_cast); + tensor transpose_130 = transpose(perm = transpose_78_perm_0, x = q_63_cast); + tensor qk_31_cast = matmul(transpose_x = qk_31_transpose_x_0, transpose_y = qk_31_transpose_y_0, x = transpose_130, y = transpose_129); + tensor var_1776_cast = softmax(axis = var_1711, x = qk_31_cast); + tensor var_1778_transpose_x_0 = const()[name = tensor("op_1778_transpose_x_0"), val = tensor(false)]; + tensor var_1778_transpose_y_0 = const()[name = tensor("op_1778_transpose_y_0"), val = tensor(false)]; + tensor transpose_131 = transpose(perm = var_1772, x = var_1771_cast); + tensor var_1778_cast = matmul(transpose_x = var_1778_transpose_x_0, transpose_y = var_1778_transpose_y_0, x = var_1776_cast, y = transpose_131); + tensor var_1779 = const()[name = tensor("op_1779"), val = tensor([0, 2, 1, 3])]; + tensor concat_15 = const()[name = tensor("concat_15"), val = tensor([1, 1500, 1024])]; + tensor transpose_128 = transpose(perm = var_1779, x = var_1778_cast); + tensor x_191_cast = reshape(shape = concat_15, x = transpose_128); + tensor var_1784_to_fp16 = const()[name = tensor("op_1784_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(394063744)))]; + tensor var_1785_to_fp16 = const()[name = tensor("op_1785_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(396160960)))]; + tensor var_1786_cast = linear(bias = var_1785_to_fp16, weight = var_1784_to_fp16, x = x_191_cast); + tensor x_193_cast = add(x = x_187_cast, y = var_1786_cast); + tensor var_1792_axes_0 = const()[name = tensor("op_1792_axes_0"), val = tensor([-1])]; + tensor blocks_15_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_15_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(396163072)))]; + tensor blocks_15_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_15_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(396165184)))]; + tensor var_1792_cast = layer_norm(axes = var_1792_axes_0, beta = blocks_15_mlp_ln_bias_to_fp16, epsilon = var_1717_to_fp16, gamma = blocks_15_mlp_ln_weight_to_fp16, x = x_193_cast); + tensor var_1801_to_fp16 = const()[name = tensor("op_1801_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(396167296)))]; + tensor var_1802_to_fp16 = const()[name = tensor("op_1802_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(404555968)))]; + tensor input_129_cast = linear(bias = var_1802_to_fp16, weight = var_1801_to_fp16, x = var_1792_cast); + tensor x_197_mode_0 = const()[name = tensor("x_197_mode_0"), val = tensor("EXACT")]; + tensor x_197_cast = gelu(mode = x_197_mode_0, x = input_129_cast); + tensor var_1807_to_fp16 = const()[name = tensor("op_1807_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(404564224)))]; + tensor var_1808_to_fp16 = const()[name = tensor("op_1808_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(412952896)))]; + tensor var_1809_cast = linear(bias = var_1808_to_fp16, weight = var_1807_to_fp16, x = x_197_cast); + tensor x_199_cast = add(x = x_193_cast, y = var_1809_cast); + tensor var_1818 = const()[name = tensor("op_1818"), val = tensor(-1)]; + tensor var_1835_axes_0 = const()[name = tensor("op_1835_axes_0"), val = tensor([-1])]; + tensor blocks_16_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_16_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(412955008)))]; + tensor blocks_16_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_16_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(412957120)))]; + tensor var_1824_to_fp16 = const()[name = tensor("op_1824_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1835_cast = layer_norm(axes = var_1835_axes_0, beta = blocks_16_attn_ln_bias_to_fp16, epsilon = var_1824_to_fp16, gamma = blocks_16_attn_ln_weight_to_fp16, x = x_199_cast); + tensor var_1846_to_fp16 = const()[name = tensor("op_1846_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(412959232)))]; + tensor var_1847_to_fp16 = const()[name = tensor("op_1847_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(415056448)))]; + tensor q_65_cast = linear(bias = var_1847_to_fp16, weight = var_1846_to_fp16, x = var_1835_cast); + tensor var_1850_to_fp16 = const()[name = tensor("op_1850_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(415058560)))]; + tensor k_65_bias_0_to_fp16 = const()[name = tensor("k_65_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(417155776)))]; + tensor k_65_cast = linear(bias = k_65_bias_0_to_fp16, weight = var_1850_to_fp16, x = var_1835_cast); + tensor var_1854_to_fp16 = const()[name = tensor("op_1854_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(417157888)))]; + tensor var_1855_to_fp16 = const()[name = tensor("op_1855_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(419255104)))]; + tensor v_65_cast = linear(bias = var_1855_to_fp16, weight = var_1854_to_fp16, x = var_1835_cast); + tensor var_1863 = const()[name = tensor("op_1863"), val = tensor([1, 1500, 16, -1])]; + tensor var_1864_cast = reshape(shape = var_1863, x = q_65_cast); + tensor const_200_to_fp16 = const()[name = tensor("const_200_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_67_cast = mul(x = var_1864_cast, y = const_200_to_fp16); + tensor var_1870 = const()[name = tensor("op_1870"), val = tensor([1, 1500, 16, -1])]; + tensor var_1871_cast = reshape(shape = var_1870, x = k_65_cast); + tensor const_201_to_fp16 = const()[name = tensor("const_201_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_67_cast = mul(x = var_1871_cast, y = const_201_to_fp16); + tensor var_1877 = const()[name = tensor("op_1877"), val = tensor([1, 1500, 16, -1])]; + tensor var_1878_cast = reshape(shape = var_1877, x = v_65_cast); + tensor var_1879 = const()[name = tensor("op_1879"), val = tensor([0, 2, 1, 3])]; + tensor qk_33_transpose_x_0 = const()[name = tensor("qk_33_transpose_x_0"), val = tensor(false)]; + tensor qk_33_transpose_y_0 = const()[name = tensor("qk_33_transpose_y_0"), val = tensor(false)]; + tensor transpose_80_perm_0 = const()[name = tensor("transpose_80_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_81_perm_0 = const()[name = tensor("transpose_81_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_125 = transpose(perm = transpose_81_perm_0, x = k_67_cast); + tensor transpose_126 = transpose(perm = transpose_80_perm_0, x = q_67_cast); + tensor qk_33_cast = matmul(transpose_x = qk_33_transpose_x_0, transpose_y = qk_33_transpose_y_0, x = transpose_126, y = transpose_125); + tensor var_1883_cast = softmax(axis = var_1818, x = qk_33_cast); + tensor var_1885_transpose_x_0 = const()[name = tensor("op_1885_transpose_x_0"), val = tensor(false)]; + tensor var_1885_transpose_y_0 = const()[name = tensor("op_1885_transpose_y_0"), val = tensor(false)]; + tensor transpose_127 = transpose(perm = var_1879, x = var_1878_cast); + tensor var_1885_cast = matmul(transpose_x = var_1885_transpose_x_0, transpose_y = var_1885_transpose_y_0, x = var_1883_cast, y = transpose_127); + tensor var_1886 = const()[name = tensor("op_1886"), val = tensor([0, 2, 1, 3])]; + tensor concat_16 = const()[name = tensor("concat_16"), val = tensor([1, 1500, 1024])]; + tensor transpose_124 = transpose(perm = var_1886, x = var_1885_cast); + tensor x_203_cast = reshape(shape = concat_16, x = transpose_124); + tensor var_1891_to_fp16 = const()[name = tensor("op_1891_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(419257216)))]; + tensor var_1892_to_fp16 = const()[name = tensor("op_1892_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(421354432)))]; + tensor var_1893_cast = linear(bias = var_1892_to_fp16, weight = var_1891_to_fp16, x = x_203_cast); + tensor x_205_cast = add(x = x_199_cast, y = var_1893_cast); + tensor var_1899_axes_0 = const()[name = tensor("op_1899_axes_0"), val = tensor([-1])]; + tensor blocks_16_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_16_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(421356544)))]; + tensor blocks_16_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_16_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(421358656)))]; + tensor var_1899_cast = layer_norm(axes = var_1899_axes_0, beta = blocks_16_mlp_ln_bias_to_fp16, epsilon = var_1824_to_fp16, gamma = blocks_16_mlp_ln_weight_to_fp16, x = x_205_cast); + tensor var_1908_to_fp16 = const()[name = tensor("op_1908_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(421360768)))]; + tensor var_1909_to_fp16 = const()[name = tensor("op_1909_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(429749440)))]; + tensor input_137_cast = linear(bias = var_1909_to_fp16, weight = var_1908_to_fp16, x = var_1899_cast); + tensor x_209_mode_0 = const()[name = tensor("x_209_mode_0"), val = tensor("EXACT")]; + tensor x_209_cast = gelu(mode = x_209_mode_0, x = input_137_cast); + tensor var_1914_to_fp16 = const()[name = tensor("op_1914_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(429757696)))]; + tensor var_1915_to_fp16 = const()[name = tensor("op_1915_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(438146368)))]; + tensor var_1916_cast = linear(bias = var_1915_to_fp16, weight = var_1914_to_fp16, x = x_209_cast); + tensor x_211_cast = add(x = x_205_cast, y = var_1916_cast); + tensor var_1925 = const()[name = tensor("op_1925"), val = tensor(-1)]; + tensor var_1942_axes_0 = const()[name = tensor("op_1942_axes_0"), val = tensor([-1])]; + tensor blocks_17_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_17_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(438148480)))]; + tensor blocks_17_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_17_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(438150592)))]; + tensor var_1931_to_fp16 = const()[name = tensor("op_1931_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1942_cast = layer_norm(axes = var_1942_axes_0, beta = blocks_17_attn_ln_bias_to_fp16, epsilon = var_1931_to_fp16, gamma = blocks_17_attn_ln_weight_to_fp16, x = x_211_cast); + tensor var_1953_to_fp16 = const()[name = tensor("op_1953_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(438152704)))]; + tensor var_1954_to_fp16 = const()[name = tensor("op_1954_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(440249920)))]; + tensor q_69_cast = linear(bias = var_1954_to_fp16, weight = var_1953_to_fp16, x = var_1942_cast); + tensor var_1957_to_fp16 = const()[name = tensor("op_1957_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(440252032)))]; + tensor k_69_bias_0_to_fp16 = const()[name = tensor("k_69_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(442349248)))]; + tensor k_69_cast = linear(bias = k_69_bias_0_to_fp16, weight = var_1957_to_fp16, x = var_1942_cast); + tensor var_1961_to_fp16 = const()[name = tensor("op_1961_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(442351360)))]; + tensor var_1962_to_fp16 = const()[name = tensor("op_1962_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(444448576)))]; + tensor v_69_cast = linear(bias = var_1962_to_fp16, weight = var_1961_to_fp16, x = var_1942_cast); + tensor var_1970 = const()[name = tensor("op_1970"), val = tensor([1, 1500, 16, -1])]; + tensor var_1971_cast = reshape(shape = var_1970, x = q_69_cast); + tensor const_202_to_fp16 = const()[name = tensor("const_202_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_71_cast = mul(x = var_1971_cast, y = const_202_to_fp16); + tensor var_1977 = const()[name = tensor("op_1977"), val = tensor([1, 1500, 16, -1])]; + tensor var_1978_cast = reshape(shape = var_1977, x = k_69_cast); + tensor const_203_to_fp16 = const()[name = tensor("const_203_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_71_cast = mul(x = var_1978_cast, y = const_203_to_fp16); + tensor var_1984 = const()[name = tensor("op_1984"), val = tensor([1, 1500, 16, -1])]; + tensor var_1985_cast = reshape(shape = var_1984, x = v_69_cast); + tensor var_1986 = const()[name = tensor("op_1986"), val = tensor([0, 2, 1, 3])]; + tensor qk_35_transpose_x_0 = const()[name = tensor("qk_35_transpose_x_0"), val = tensor(false)]; + tensor qk_35_transpose_y_0 = const()[name = tensor("qk_35_transpose_y_0"), val = tensor(false)]; + tensor transpose_82_perm_0 = const()[name = tensor("transpose_82_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_83_perm_0 = const()[name = tensor("transpose_83_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_121 = transpose(perm = transpose_83_perm_0, x = k_71_cast); + tensor transpose_122 = transpose(perm = transpose_82_perm_0, x = q_71_cast); + tensor qk_35_cast = matmul(transpose_x = qk_35_transpose_x_0, transpose_y = qk_35_transpose_y_0, x = transpose_122, y = transpose_121); + tensor var_1990_cast = softmax(axis = var_1925, x = qk_35_cast); + tensor var_1992_transpose_x_0 = const()[name = tensor("op_1992_transpose_x_0"), val = tensor(false)]; + tensor var_1992_transpose_y_0 = const()[name = tensor("op_1992_transpose_y_0"), val = tensor(false)]; + tensor transpose_123 = transpose(perm = var_1986, x = var_1985_cast); + tensor var_1992_cast = matmul(transpose_x = var_1992_transpose_x_0, transpose_y = var_1992_transpose_y_0, x = var_1990_cast, y = transpose_123); + tensor var_1993 = const()[name = tensor("op_1993"), val = tensor([0, 2, 1, 3])]; + tensor concat_17 = const()[name = tensor("concat_17"), val = tensor([1, 1500, 1024])]; + tensor transpose_120 = transpose(perm = var_1993, x = var_1992_cast); + tensor x_215_cast = reshape(shape = concat_17, x = transpose_120); + tensor var_1998_to_fp16 = const()[name = tensor("op_1998_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(444450688)))]; + tensor var_1999_to_fp16 = const()[name = tensor("op_1999_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(446547904)))]; + tensor var_2000_cast = linear(bias = var_1999_to_fp16, weight = var_1998_to_fp16, x = x_215_cast); + tensor x_217_cast = add(x = x_211_cast, y = var_2000_cast); + tensor var_2006_axes_0 = const()[name = tensor("op_2006_axes_0"), val = tensor([-1])]; + tensor blocks_17_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_17_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(446550016)))]; + tensor blocks_17_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_17_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(446552128)))]; + tensor var_2006_cast = layer_norm(axes = var_2006_axes_0, beta = blocks_17_mlp_ln_bias_to_fp16, epsilon = var_1931_to_fp16, gamma = blocks_17_mlp_ln_weight_to_fp16, x = x_217_cast); + tensor var_2015_to_fp16 = const()[name = tensor("op_2015_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(446554240)))]; + tensor var_2016_to_fp16 = const()[name = tensor("op_2016_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(454942912)))]; + tensor input_145_cast = linear(bias = var_2016_to_fp16, weight = var_2015_to_fp16, x = var_2006_cast); + tensor x_221_mode_0 = const()[name = tensor("x_221_mode_0"), val = tensor("EXACT")]; + tensor x_221_cast = gelu(mode = x_221_mode_0, x = input_145_cast); + tensor var_2021_to_fp16 = const()[name = tensor("op_2021_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(454951168)))]; + tensor var_2022_to_fp16 = const()[name = tensor("op_2022_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(463339840)))]; + tensor var_2023_cast = linear(bias = var_2022_to_fp16, weight = var_2021_to_fp16, x = x_221_cast); + tensor x_223_cast = add(x = x_217_cast, y = var_2023_cast); + tensor var_2032 = const()[name = tensor("op_2032"), val = tensor(-1)]; + tensor var_2049_axes_0 = const()[name = tensor("op_2049_axes_0"), val = tensor([-1])]; + tensor blocks_18_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_18_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(463341952)))]; + tensor blocks_18_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_18_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(463344064)))]; + tensor var_2038_to_fp16 = const()[name = tensor("op_2038_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2049_cast = layer_norm(axes = var_2049_axes_0, beta = blocks_18_attn_ln_bias_to_fp16, epsilon = var_2038_to_fp16, gamma = blocks_18_attn_ln_weight_to_fp16, x = x_223_cast); + tensor var_2060_to_fp16 = const()[name = tensor("op_2060_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(463346176)))]; + tensor var_2061_to_fp16 = const()[name = tensor("op_2061_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(465443392)))]; + tensor q_73_cast = linear(bias = var_2061_to_fp16, weight = var_2060_to_fp16, x = var_2049_cast); + tensor var_2064_to_fp16 = const()[name = tensor("op_2064_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(465445504)))]; + tensor k_73_bias_0_to_fp16 = const()[name = tensor("k_73_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(467542720)))]; + tensor k_73_cast = linear(bias = k_73_bias_0_to_fp16, weight = var_2064_to_fp16, x = var_2049_cast); + tensor var_2068_to_fp16 = const()[name = tensor("op_2068_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(467544832)))]; + tensor var_2069_to_fp16 = const()[name = tensor("op_2069_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(469642048)))]; + tensor v_73_cast = linear(bias = var_2069_to_fp16, weight = var_2068_to_fp16, x = var_2049_cast); + tensor var_2077 = const()[name = tensor("op_2077"), val = tensor([1, 1500, 16, -1])]; + tensor var_2078_cast = reshape(shape = var_2077, x = q_73_cast); + tensor const_204_to_fp16 = const()[name = tensor("const_204_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_75_cast = mul(x = var_2078_cast, y = const_204_to_fp16); + tensor var_2084 = const()[name = tensor("op_2084"), val = tensor([1, 1500, 16, -1])]; + tensor var_2085_cast = reshape(shape = var_2084, x = k_73_cast); + tensor const_205_to_fp16 = const()[name = tensor("const_205_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_75_cast = mul(x = var_2085_cast, y = const_205_to_fp16); + tensor var_2091 = const()[name = tensor("op_2091"), val = tensor([1, 1500, 16, -1])]; + tensor var_2092_cast = reshape(shape = var_2091, x = v_73_cast); + tensor var_2093 = const()[name = tensor("op_2093"), val = tensor([0, 2, 1, 3])]; + tensor qk_37_transpose_x_0 = const()[name = tensor("qk_37_transpose_x_0"), val = tensor(false)]; + tensor qk_37_transpose_y_0 = const()[name = tensor("qk_37_transpose_y_0"), val = tensor(false)]; + tensor transpose_84_perm_0 = const()[name = tensor("transpose_84_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_85_perm_0 = const()[name = tensor("transpose_85_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_117 = transpose(perm = transpose_85_perm_0, x = k_75_cast); + tensor transpose_118 = transpose(perm = transpose_84_perm_0, x = q_75_cast); + tensor qk_37_cast = matmul(transpose_x = qk_37_transpose_x_0, transpose_y = qk_37_transpose_y_0, x = transpose_118, y = transpose_117); + tensor var_2097_cast = softmax(axis = var_2032, x = qk_37_cast); + tensor var_2099_transpose_x_0 = const()[name = tensor("op_2099_transpose_x_0"), val = tensor(false)]; + tensor var_2099_transpose_y_0 = const()[name = tensor("op_2099_transpose_y_0"), val = tensor(false)]; + tensor transpose_119 = transpose(perm = var_2093, x = var_2092_cast); + tensor var_2099_cast = matmul(transpose_x = var_2099_transpose_x_0, transpose_y = var_2099_transpose_y_0, x = var_2097_cast, y = transpose_119); + tensor var_2100 = const()[name = tensor("op_2100"), val = tensor([0, 2, 1, 3])]; + tensor concat_18 = const()[name = tensor("concat_18"), val = tensor([1, 1500, 1024])]; + tensor transpose_116 = transpose(perm = var_2100, x = var_2099_cast); + tensor x_227_cast = reshape(shape = concat_18, x = transpose_116); + tensor var_2105_to_fp16 = const()[name = tensor("op_2105_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(469644160)))]; + tensor var_2106_to_fp16 = const()[name = tensor("op_2106_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(471741376)))]; + tensor var_2107_cast = linear(bias = var_2106_to_fp16, weight = var_2105_to_fp16, x = x_227_cast); + tensor x_229_cast = add(x = x_223_cast, y = var_2107_cast); + tensor var_2113_axes_0 = const()[name = tensor("op_2113_axes_0"), val = tensor([-1])]; + tensor blocks_18_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_18_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(471743488)))]; + tensor blocks_18_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_18_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(471745600)))]; + tensor var_2113_cast = layer_norm(axes = var_2113_axes_0, beta = blocks_18_mlp_ln_bias_to_fp16, epsilon = var_2038_to_fp16, gamma = blocks_18_mlp_ln_weight_to_fp16, x = x_229_cast); + tensor var_2122_to_fp16 = const()[name = tensor("op_2122_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(471747712)))]; + tensor var_2123_to_fp16 = const()[name = tensor("op_2123_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(480136384)))]; + tensor input_153_cast = linear(bias = var_2123_to_fp16, weight = var_2122_to_fp16, x = var_2113_cast); + tensor x_233_mode_0 = const()[name = tensor("x_233_mode_0"), val = tensor("EXACT")]; + tensor x_233_cast = gelu(mode = x_233_mode_0, x = input_153_cast); + tensor var_2128_to_fp16 = const()[name = tensor("op_2128_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(480144640)))]; + tensor var_2129_to_fp16 = const()[name = tensor("op_2129_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(488533312)))]; + tensor var_2130_cast = linear(bias = var_2129_to_fp16, weight = var_2128_to_fp16, x = x_233_cast); + tensor x_235_cast = add(x = x_229_cast, y = var_2130_cast); + tensor var_2139 = const()[name = tensor("op_2139"), val = tensor(-1)]; + tensor var_2156_axes_0 = const()[name = tensor("op_2156_axes_0"), val = tensor([-1])]; + tensor blocks_19_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_19_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(488535424)))]; + tensor blocks_19_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_19_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(488537536)))]; + tensor var_2145_to_fp16 = const()[name = tensor("op_2145_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2156_cast = layer_norm(axes = var_2156_axes_0, beta = blocks_19_attn_ln_bias_to_fp16, epsilon = var_2145_to_fp16, gamma = blocks_19_attn_ln_weight_to_fp16, x = x_235_cast); + tensor var_2167_to_fp16 = const()[name = tensor("op_2167_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(488539648)))]; + tensor var_2168_to_fp16 = const()[name = tensor("op_2168_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(490636864)))]; + tensor q_77_cast = linear(bias = var_2168_to_fp16, weight = var_2167_to_fp16, x = var_2156_cast); + tensor var_2171_to_fp16 = const()[name = tensor("op_2171_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(490638976)))]; + tensor k_77_bias_0_to_fp16 = const()[name = tensor("k_77_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(492736192)))]; + tensor k_77_cast = linear(bias = k_77_bias_0_to_fp16, weight = var_2171_to_fp16, x = var_2156_cast); + tensor var_2175_to_fp16 = const()[name = tensor("op_2175_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(492738304)))]; + tensor var_2176_to_fp16 = const()[name = tensor("op_2176_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(494835520)))]; + tensor v_77_cast = linear(bias = var_2176_to_fp16, weight = var_2175_to_fp16, x = var_2156_cast); + tensor var_2184 = const()[name = tensor("op_2184"), val = tensor([1, 1500, 16, -1])]; + tensor var_2185_cast = reshape(shape = var_2184, x = q_77_cast); + tensor const_206_to_fp16 = const()[name = tensor("const_206_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_79_cast = mul(x = var_2185_cast, y = const_206_to_fp16); + tensor var_2191 = const()[name = tensor("op_2191"), val = tensor([1, 1500, 16, -1])]; + tensor var_2192_cast = reshape(shape = var_2191, x = k_77_cast); + tensor const_207_to_fp16 = const()[name = tensor("const_207_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_79_cast = mul(x = var_2192_cast, y = const_207_to_fp16); + tensor var_2198 = const()[name = tensor("op_2198"), val = tensor([1, 1500, 16, -1])]; + tensor var_2199_cast = reshape(shape = var_2198, x = v_77_cast); + tensor var_2200 = const()[name = tensor("op_2200"), val = tensor([0, 2, 1, 3])]; + tensor qk_39_transpose_x_0 = const()[name = tensor("qk_39_transpose_x_0"), val = tensor(false)]; + tensor qk_39_transpose_y_0 = const()[name = tensor("qk_39_transpose_y_0"), val = tensor(false)]; + tensor transpose_86_perm_0 = const()[name = tensor("transpose_86_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_87_perm_0 = const()[name = tensor("transpose_87_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_113 = transpose(perm = transpose_87_perm_0, x = k_79_cast); + tensor transpose_114 = transpose(perm = transpose_86_perm_0, x = q_79_cast); + tensor qk_39_cast = matmul(transpose_x = qk_39_transpose_x_0, transpose_y = qk_39_transpose_y_0, x = transpose_114, y = transpose_113); + tensor var_2204_cast = softmax(axis = var_2139, x = qk_39_cast); + tensor var_2206_transpose_x_0 = const()[name = tensor("op_2206_transpose_x_0"), val = tensor(false)]; + tensor var_2206_transpose_y_0 = const()[name = tensor("op_2206_transpose_y_0"), val = tensor(false)]; + tensor transpose_115 = transpose(perm = var_2200, x = var_2199_cast); + tensor var_2206_cast = matmul(transpose_x = var_2206_transpose_x_0, transpose_y = var_2206_transpose_y_0, x = var_2204_cast, y = transpose_115); + tensor var_2207 = const()[name = tensor("op_2207"), val = tensor([0, 2, 1, 3])]; + tensor concat_19 = const()[name = tensor("concat_19"), val = tensor([1, 1500, 1024])]; + tensor transpose_112 = transpose(perm = var_2207, x = var_2206_cast); + tensor x_239_cast = reshape(shape = concat_19, x = transpose_112); + tensor var_2212_to_fp16 = const()[name = tensor("op_2212_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(494837632)))]; + tensor var_2213_to_fp16 = const()[name = tensor("op_2213_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496934848)))]; + tensor var_2214_cast = linear(bias = var_2213_to_fp16, weight = var_2212_to_fp16, x = x_239_cast); + tensor x_241_cast = add(x = x_235_cast, y = var_2214_cast); + tensor var_2220_axes_0 = const()[name = tensor("op_2220_axes_0"), val = tensor([-1])]; + tensor blocks_19_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_19_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496936960)))]; + tensor blocks_19_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_19_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496939072)))]; + tensor var_2220_cast = layer_norm(axes = var_2220_axes_0, beta = blocks_19_mlp_ln_bias_to_fp16, epsilon = var_2145_to_fp16, gamma = blocks_19_mlp_ln_weight_to_fp16, x = x_241_cast); + tensor var_2229_to_fp16 = const()[name = tensor("op_2229_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(496941184)))]; + tensor var_2230_to_fp16 = const()[name = tensor("op_2230_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(505329856)))]; + tensor input_161_cast = linear(bias = var_2230_to_fp16, weight = var_2229_to_fp16, x = var_2220_cast); + tensor x_245_mode_0 = const()[name = tensor("x_245_mode_0"), val = tensor("EXACT")]; + tensor x_245_cast = gelu(mode = x_245_mode_0, x = input_161_cast); + tensor var_2235_to_fp16 = const()[name = tensor("op_2235_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(505338112)))]; + tensor var_2236_to_fp16 = const()[name = tensor("op_2236_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(513726784)))]; + tensor var_2237_cast = linear(bias = var_2236_to_fp16, weight = var_2235_to_fp16, x = x_245_cast); + tensor x_247_cast = add(x = x_241_cast, y = var_2237_cast); + tensor var_2246 = const()[name = tensor("op_2246"), val = tensor(-1)]; + tensor var_2263_axes_0 = const()[name = tensor("op_2263_axes_0"), val = tensor([-1])]; + tensor blocks_20_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_20_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(513728896)))]; + tensor blocks_20_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_20_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(513731008)))]; + tensor var_2252_to_fp16 = const()[name = tensor("op_2252_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2263_cast = layer_norm(axes = var_2263_axes_0, beta = blocks_20_attn_ln_bias_to_fp16, epsilon = var_2252_to_fp16, gamma = blocks_20_attn_ln_weight_to_fp16, x = x_247_cast); + tensor var_2274_to_fp16 = const()[name = tensor("op_2274_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(513733120)))]; + tensor var_2275_to_fp16 = const()[name = tensor("op_2275_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(515830336)))]; + tensor q_81_cast = linear(bias = var_2275_to_fp16, weight = var_2274_to_fp16, x = var_2263_cast); + tensor var_2278_to_fp16 = const()[name = tensor("op_2278_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(515832448)))]; + tensor k_81_bias_0_to_fp16 = const()[name = tensor("k_81_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(517929664)))]; + tensor k_81_cast = linear(bias = k_81_bias_0_to_fp16, weight = var_2278_to_fp16, x = var_2263_cast); + tensor var_2282_to_fp16 = const()[name = tensor("op_2282_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(517931776)))]; + tensor var_2283_to_fp16 = const()[name = tensor("op_2283_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(520028992)))]; + tensor v_81_cast = linear(bias = var_2283_to_fp16, weight = var_2282_to_fp16, x = var_2263_cast); + tensor var_2291 = const()[name = tensor("op_2291"), val = tensor([1, 1500, 16, -1])]; + tensor var_2292_cast = reshape(shape = var_2291, x = q_81_cast); + tensor const_208_to_fp16 = const()[name = tensor("const_208_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_83_cast = mul(x = var_2292_cast, y = const_208_to_fp16); + tensor var_2298 = const()[name = tensor("op_2298"), val = tensor([1, 1500, 16, -1])]; + tensor var_2299_cast = reshape(shape = var_2298, x = k_81_cast); + tensor const_209_to_fp16 = const()[name = tensor("const_209_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_83_cast = mul(x = var_2299_cast, y = const_209_to_fp16); + tensor var_2305 = const()[name = tensor("op_2305"), val = tensor([1, 1500, 16, -1])]; + tensor var_2306_cast = reshape(shape = var_2305, x = v_81_cast); + tensor var_2307 = const()[name = tensor("op_2307"), val = tensor([0, 2, 1, 3])]; + tensor qk_41_transpose_x_0 = const()[name = tensor("qk_41_transpose_x_0"), val = tensor(false)]; + tensor qk_41_transpose_y_0 = const()[name = tensor("qk_41_transpose_y_0"), val = tensor(false)]; + tensor transpose_88_perm_0 = const()[name = tensor("transpose_88_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_89_perm_0 = const()[name = tensor("transpose_89_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_109 = transpose(perm = transpose_89_perm_0, x = k_83_cast); + tensor transpose_110 = transpose(perm = transpose_88_perm_0, x = q_83_cast); + tensor qk_41_cast = matmul(transpose_x = qk_41_transpose_x_0, transpose_y = qk_41_transpose_y_0, x = transpose_110, y = transpose_109); + tensor var_2311_cast = softmax(axis = var_2246, x = qk_41_cast); + tensor var_2313_transpose_x_0 = const()[name = tensor("op_2313_transpose_x_0"), val = tensor(false)]; + tensor var_2313_transpose_y_0 = const()[name = tensor("op_2313_transpose_y_0"), val = tensor(false)]; + tensor transpose_111 = transpose(perm = var_2307, x = var_2306_cast); + tensor var_2313_cast = matmul(transpose_x = var_2313_transpose_x_0, transpose_y = var_2313_transpose_y_0, x = var_2311_cast, y = transpose_111); + tensor var_2314 = const()[name = tensor("op_2314"), val = tensor([0, 2, 1, 3])]; + tensor concat_20 = const()[name = tensor("concat_20"), val = tensor([1, 1500, 1024])]; + tensor transpose_108 = transpose(perm = var_2314, x = var_2313_cast); + tensor x_251_cast = reshape(shape = concat_20, x = transpose_108); + tensor var_2319_to_fp16 = const()[name = tensor("op_2319_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(520031104)))]; + tensor var_2320_to_fp16 = const()[name = tensor("op_2320_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(522128320)))]; + tensor var_2321_cast = linear(bias = var_2320_to_fp16, weight = var_2319_to_fp16, x = x_251_cast); + tensor x_253_cast = add(x = x_247_cast, y = var_2321_cast); + tensor var_2327_axes_0 = const()[name = tensor("op_2327_axes_0"), val = tensor([-1])]; + tensor blocks_20_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_20_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(522130432)))]; + tensor blocks_20_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_20_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(522132544)))]; + tensor var_2327_cast = layer_norm(axes = var_2327_axes_0, beta = blocks_20_mlp_ln_bias_to_fp16, epsilon = var_2252_to_fp16, gamma = blocks_20_mlp_ln_weight_to_fp16, x = x_253_cast); + tensor var_2336_to_fp16 = const()[name = tensor("op_2336_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(522134656)))]; + tensor var_2337_to_fp16 = const()[name = tensor("op_2337_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(530523328)))]; + tensor input_169_cast = linear(bias = var_2337_to_fp16, weight = var_2336_to_fp16, x = var_2327_cast); + tensor x_257_mode_0 = const()[name = tensor("x_257_mode_0"), val = tensor("EXACT")]; + tensor x_257_cast = gelu(mode = x_257_mode_0, x = input_169_cast); + tensor var_2342_to_fp16 = const()[name = tensor("op_2342_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(530531584)))]; + tensor var_2343_to_fp16 = const()[name = tensor("op_2343_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(538920256)))]; + tensor var_2344_cast = linear(bias = var_2343_to_fp16, weight = var_2342_to_fp16, x = x_257_cast); + tensor x_259_cast = add(x = x_253_cast, y = var_2344_cast); + tensor var_2353 = const()[name = tensor("op_2353"), val = tensor(-1)]; + tensor var_2370_axes_0 = const()[name = tensor("op_2370_axes_0"), val = tensor([-1])]; + tensor blocks_21_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_21_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(538922368)))]; + tensor blocks_21_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_21_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(538924480)))]; + tensor var_2359_to_fp16 = const()[name = tensor("op_2359_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2370_cast = layer_norm(axes = var_2370_axes_0, beta = blocks_21_attn_ln_bias_to_fp16, epsilon = var_2359_to_fp16, gamma = blocks_21_attn_ln_weight_to_fp16, x = x_259_cast); + tensor var_2381_to_fp16 = const()[name = tensor("op_2381_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(538926592)))]; + tensor var_2382_to_fp16 = const()[name = tensor("op_2382_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(541023808)))]; + tensor q_85_cast = linear(bias = var_2382_to_fp16, weight = var_2381_to_fp16, x = var_2370_cast); + tensor var_2385_to_fp16 = const()[name = tensor("op_2385_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(541025920)))]; + tensor k_85_bias_0_to_fp16 = const()[name = tensor("k_85_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(543123136)))]; + tensor k_85_cast = linear(bias = k_85_bias_0_to_fp16, weight = var_2385_to_fp16, x = var_2370_cast); + tensor var_2389_to_fp16 = const()[name = tensor("op_2389_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(543125248)))]; + tensor var_2390_to_fp16 = const()[name = tensor("op_2390_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(545222464)))]; + tensor v_85_cast = linear(bias = var_2390_to_fp16, weight = var_2389_to_fp16, x = var_2370_cast); + tensor var_2398 = const()[name = tensor("op_2398"), val = tensor([1, 1500, 16, -1])]; + tensor var_2399_cast = reshape(shape = var_2398, x = q_85_cast); + tensor const_210_to_fp16 = const()[name = tensor("const_210_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_87_cast = mul(x = var_2399_cast, y = const_210_to_fp16); + tensor var_2405 = const()[name = tensor("op_2405"), val = tensor([1, 1500, 16, -1])]; + tensor var_2406_cast = reshape(shape = var_2405, x = k_85_cast); + tensor const_211_to_fp16 = const()[name = tensor("const_211_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_87_cast = mul(x = var_2406_cast, y = const_211_to_fp16); + tensor var_2412 = const()[name = tensor("op_2412"), val = tensor([1, 1500, 16, -1])]; + tensor var_2413_cast = reshape(shape = var_2412, x = v_85_cast); + tensor var_2414 = const()[name = tensor("op_2414"), val = tensor([0, 2, 1, 3])]; + tensor qk_43_transpose_x_0 = const()[name = tensor("qk_43_transpose_x_0"), val = tensor(false)]; + tensor qk_43_transpose_y_0 = const()[name = tensor("qk_43_transpose_y_0"), val = tensor(false)]; + tensor transpose_90_perm_0 = const()[name = tensor("transpose_90_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_91_perm_0 = const()[name = tensor("transpose_91_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_105 = transpose(perm = transpose_91_perm_0, x = k_87_cast); + tensor transpose_106 = transpose(perm = transpose_90_perm_0, x = q_87_cast); + tensor qk_43_cast = matmul(transpose_x = qk_43_transpose_x_0, transpose_y = qk_43_transpose_y_0, x = transpose_106, y = transpose_105); + tensor var_2418_cast = softmax(axis = var_2353, x = qk_43_cast); + tensor var_2420_transpose_x_0 = const()[name = tensor("op_2420_transpose_x_0"), val = tensor(false)]; + tensor var_2420_transpose_y_0 = const()[name = tensor("op_2420_transpose_y_0"), val = tensor(false)]; + tensor transpose_107 = transpose(perm = var_2414, x = var_2413_cast); + tensor var_2420_cast = matmul(transpose_x = var_2420_transpose_x_0, transpose_y = var_2420_transpose_y_0, x = var_2418_cast, y = transpose_107); + tensor var_2421 = const()[name = tensor("op_2421"), val = tensor([0, 2, 1, 3])]; + tensor concat_21 = const()[name = tensor("concat_21"), val = tensor([1, 1500, 1024])]; + tensor transpose_104 = transpose(perm = var_2421, x = var_2420_cast); + tensor x_263_cast = reshape(shape = concat_21, x = transpose_104); + tensor var_2426_to_fp16 = const()[name = tensor("op_2426_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(545224576)))]; + tensor var_2427_to_fp16 = const()[name = tensor("op_2427_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(547321792)))]; + tensor var_2428_cast = linear(bias = var_2427_to_fp16, weight = var_2426_to_fp16, x = x_263_cast); + tensor x_265_cast = add(x = x_259_cast, y = var_2428_cast); + tensor var_2434_axes_0 = const()[name = tensor("op_2434_axes_0"), val = tensor([-1])]; + tensor blocks_21_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_21_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(547323904)))]; + tensor blocks_21_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_21_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(547326016)))]; + tensor var_2434_cast = layer_norm(axes = var_2434_axes_0, beta = blocks_21_mlp_ln_bias_to_fp16, epsilon = var_2359_to_fp16, gamma = blocks_21_mlp_ln_weight_to_fp16, x = x_265_cast); + tensor var_2443_to_fp16 = const()[name = tensor("op_2443_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(547328128)))]; + tensor var_2444_to_fp16 = const()[name = tensor("op_2444_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(555716800)))]; + tensor input_177_cast = linear(bias = var_2444_to_fp16, weight = var_2443_to_fp16, x = var_2434_cast); + tensor x_269_mode_0 = const()[name = tensor("x_269_mode_0"), val = tensor("EXACT")]; + tensor x_269_cast = gelu(mode = x_269_mode_0, x = input_177_cast); + tensor var_2449_to_fp16 = const()[name = tensor("op_2449_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(555725056)))]; + tensor var_2450_to_fp16 = const()[name = tensor("op_2450_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(564113728)))]; + tensor var_2451_cast = linear(bias = var_2450_to_fp16, weight = var_2449_to_fp16, x = x_269_cast); + tensor x_271_cast = add(x = x_265_cast, y = var_2451_cast); + tensor var_2460 = const()[name = tensor("op_2460"), val = tensor(-1)]; + tensor var_2477_axes_0 = const()[name = tensor("op_2477_axes_0"), val = tensor([-1])]; + tensor blocks_22_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_22_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(564115840)))]; + tensor blocks_22_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_22_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(564117952)))]; + tensor var_2466_to_fp16 = const()[name = tensor("op_2466_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2477_cast = layer_norm(axes = var_2477_axes_0, beta = blocks_22_attn_ln_bias_to_fp16, epsilon = var_2466_to_fp16, gamma = blocks_22_attn_ln_weight_to_fp16, x = x_271_cast); + tensor var_2488_to_fp16 = const()[name = tensor("op_2488_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(564120064)))]; + tensor var_2489_to_fp16 = const()[name = tensor("op_2489_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(566217280)))]; + tensor q_89_cast = linear(bias = var_2489_to_fp16, weight = var_2488_to_fp16, x = var_2477_cast); + tensor var_2492_to_fp16 = const()[name = tensor("op_2492_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(566219392)))]; + tensor k_89_bias_0_to_fp16 = const()[name = tensor("k_89_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568316608)))]; + tensor k_89_cast = linear(bias = k_89_bias_0_to_fp16, weight = var_2492_to_fp16, x = var_2477_cast); + tensor var_2496_to_fp16 = const()[name = tensor("op_2496_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(568318720)))]; + tensor var_2497_to_fp16 = const()[name = tensor("op_2497_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(570415936)))]; + tensor v_89_cast = linear(bias = var_2497_to_fp16, weight = var_2496_to_fp16, x = var_2477_cast); + tensor var_2505 = const()[name = tensor("op_2505"), val = tensor([1, 1500, 16, -1])]; + tensor var_2506_cast = reshape(shape = var_2505, x = q_89_cast); + tensor const_212_to_fp16 = const()[name = tensor("const_212_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_91_cast = mul(x = var_2506_cast, y = const_212_to_fp16); + tensor var_2512 = const()[name = tensor("op_2512"), val = tensor([1, 1500, 16, -1])]; + tensor var_2513_cast = reshape(shape = var_2512, x = k_89_cast); + tensor const_213_to_fp16 = const()[name = tensor("const_213_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_91_cast = mul(x = var_2513_cast, y = const_213_to_fp16); + tensor var_2519 = const()[name = tensor("op_2519"), val = tensor([1, 1500, 16, -1])]; + tensor var_2520_cast = reshape(shape = var_2519, x = v_89_cast); + tensor var_2521 = const()[name = tensor("op_2521"), val = tensor([0, 2, 1, 3])]; + tensor qk_45_transpose_x_0 = const()[name = tensor("qk_45_transpose_x_0"), val = tensor(false)]; + tensor qk_45_transpose_y_0 = const()[name = tensor("qk_45_transpose_y_0"), val = tensor(false)]; + tensor transpose_92_perm_0 = const()[name = tensor("transpose_92_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_93_perm_0 = const()[name = tensor("transpose_93_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_101 = transpose(perm = transpose_93_perm_0, x = k_91_cast); + tensor transpose_102 = transpose(perm = transpose_92_perm_0, x = q_91_cast); + tensor qk_45_cast = matmul(transpose_x = qk_45_transpose_x_0, transpose_y = qk_45_transpose_y_0, x = transpose_102, y = transpose_101); + tensor var_2525_cast = softmax(axis = var_2460, x = qk_45_cast); + tensor var_2527_transpose_x_0 = const()[name = tensor("op_2527_transpose_x_0"), val = tensor(false)]; + tensor var_2527_transpose_y_0 = const()[name = tensor("op_2527_transpose_y_0"), val = tensor(false)]; + tensor transpose_103 = transpose(perm = var_2521, x = var_2520_cast); + tensor var_2527_cast = matmul(transpose_x = var_2527_transpose_x_0, transpose_y = var_2527_transpose_y_0, x = var_2525_cast, y = transpose_103); + tensor var_2528 = const()[name = tensor("op_2528"), val = tensor([0, 2, 1, 3])]; + tensor concat_22 = const()[name = tensor("concat_22"), val = tensor([1, 1500, 1024])]; + tensor transpose_100 = transpose(perm = var_2528, x = var_2527_cast); + tensor x_275_cast = reshape(shape = concat_22, x = transpose_100); + tensor var_2533_to_fp16 = const()[name = tensor("op_2533_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(570418048)))]; + tensor var_2534_to_fp16 = const()[name = tensor("op_2534_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(572515264)))]; + tensor var_2535_cast = linear(bias = var_2534_to_fp16, weight = var_2533_to_fp16, x = x_275_cast); + tensor x_277_cast = add(x = x_271_cast, y = var_2535_cast); + tensor var_2541_axes_0 = const()[name = tensor("op_2541_axes_0"), val = tensor([-1])]; + tensor blocks_22_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_22_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(572517376)))]; + tensor blocks_22_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_22_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(572519488)))]; + tensor var_2541_cast = layer_norm(axes = var_2541_axes_0, beta = blocks_22_mlp_ln_bias_to_fp16, epsilon = var_2466_to_fp16, gamma = blocks_22_mlp_ln_weight_to_fp16, x = x_277_cast); + tensor var_2550_to_fp16 = const()[name = tensor("op_2550_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(572521600)))]; + tensor var_2551_to_fp16 = const()[name = tensor("op_2551_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(580910272)))]; + tensor input_185_cast = linear(bias = var_2551_to_fp16, weight = var_2550_to_fp16, x = var_2541_cast); + tensor x_281_mode_0 = const()[name = tensor("x_281_mode_0"), val = tensor("EXACT")]; + tensor x_281_cast = gelu(mode = x_281_mode_0, x = input_185_cast); + tensor var_2556_to_fp16 = const()[name = tensor("op_2556_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(580918528)))]; + tensor var_2557_to_fp16 = const()[name = tensor("op_2557_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(589307200)))]; + tensor var_2558_cast = linear(bias = var_2557_to_fp16, weight = var_2556_to_fp16, x = x_281_cast); + tensor x_283_cast = add(x = x_277_cast, y = var_2558_cast); + tensor var_2567 = const()[name = tensor("op_2567"), val = tensor(-1)]; + tensor var_2584_axes_0 = const()[name = tensor("op_2584_axes_0"), val = tensor([-1])]; + tensor blocks_23_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_23_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(589309312)))]; + tensor blocks_23_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_23_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(589311424)))]; + tensor var_2573_to_fp16 = const()[name = tensor("op_2573_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2584_cast = layer_norm(axes = var_2584_axes_0, beta = blocks_23_attn_ln_bias_to_fp16, epsilon = var_2573_to_fp16, gamma = blocks_23_attn_ln_weight_to_fp16, x = x_283_cast); + tensor var_2595_to_fp16 = const()[name = tensor("op_2595_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(589313536)))]; + tensor var_2596_to_fp16 = const()[name = tensor("op_2596_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(591410752)))]; + tensor q_93_cast = linear(bias = var_2596_to_fp16, weight = var_2595_to_fp16, x = var_2584_cast); + tensor var_2599_to_fp16 = const()[name = tensor("op_2599_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(591412864)))]; + tensor k_93_bias_0_to_fp16 = const()[name = tensor("k_93_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(593510080)))]; + tensor k_93_cast = linear(bias = k_93_bias_0_to_fp16, weight = var_2599_to_fp16, x = var_2584_cast); + tensor var_2603_to_fp16 = const()[name = tensor("op_2603_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(593512192)))]; + tensor var_2604_to_fp16 = const()[name = tensor("op_2604_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(595609408)))]; + tensor v_93_cast = linear(bias = var_2604_to_fp16, weight = var_2603_to_fp16, x = var_2584_cast); + tensor var_2612 = const()[name = tensor("op_2612"), val = tensor([1, 1500, 16, -1])]; + tensor var_2613_cast = reshape(shape = var_2612, x = q_93_cast); + tensor const_214_to_fp16 = const()[name = tensor("const_214_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_cast = mul(x = var_2613_cast, y = const_214_to_fp16); + tensor var_2619 = const()[name = tensor("op_2619"), val = tensor([1, 1500, 16, -1])]; + tensor var_2620_cast = reshape(shape = var_2619, x = k_93_cast); + tensor const_215_to_fp16 = const()[name = tensor("const_215_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_cast = mul(x = var_2620_cast, y = const_215_to_fp16); + tensor var_2626 = const()[name = tensor("op_2626"), val = tensor([1, 1500, 16, -1])]; + tensor var_2627_cast = reshape(shape = var_2626, x = v_93_cast); + tensor var_2628 = const()[name = tensor("op_2628"), val = tensor([0, 2, 1, 3])]; + tensor qk_transpose_x_0 = const()[name = tensor("qk_transpose_x_0"), val = tensor(false)]; + tensor qk_transpose_y_0 = const()[name = tensor("qk_transpose_y_0"), val = tensor(false)]; + tensor transpose_94_perm_0 = const()[name = tensor("transpose_94_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_95_perm_0 = const()[name = tensor("transpose_95_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_97 = transpose(perm = transpose_95_perm_0, x = k_cast); + tensor transpose_98 = transpose(perm = transpose_94_perm_0, x = q_cast); + tensor qk_cast = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_98, y = transpose_97); + tensor var_2632_cast = softmax(axis = var_2567, x = qk_cast); + tensor var_2634_transpose_x_0 = const()[name = tensor("op_2634_transpose_x_0"), val = tensor(false)]; + tensor var_2634_transpose_y_0 = const()[name = tensor("op_2634_transpose_y_0"), val = tensor(false)]; + tensor transpose_99 = transpose(perm = var_2628, x = var_2627_cast); + tensor var_2634_cast = matmul(transpose_x = var_2634_transpose_x_0, transpose_y = var_2634_transpose_y_0, x = var_2632_cast, y = transpose_99); + tensor var_2635 = const()[name = tensor("op_2635"), val = tensor([0, 2, 1, 3])]; + tensor concat_23 = const()[name = tensor("concat_23"), val = tensor([1, 1500, 1024])]; + tensor transpose_96 = transpose(perm = var_2635, x = var_2634_cast); + tensor x_287_cast = reshape(shape = concat_23, x = transpose_96); + tensor var_2640_to_fp16 = const()[name = tensor("op_2640_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(595611520)))]; + tensor var_2641_to_fp16 = const()[name = tensor("op_2641_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(597708736)))]; + tensor var_2642_cast = linear(bias = var_2641_to_fp16, weight = var_2640_to_fp16, x = x_287_cast); + tensor x_289_cast = add(x = x_283_cast, y = var_2642_cast); + tensor var_2648_axes_0 = const()[name = tensor("op_2648_axes_0"), val = tensor([-1])]; + tensor blocks_23_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_23_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(597710848)))]; + tensor blocks_23_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_23_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(597712960)))]; + tensor var_2648_cast = layer_norm(axes = var_2648_axes_0, beta = blocks_23_mlp_ln_bias_to_fp16, epsilon = var_2573_to_fp16, gamma = blocks_23_mlp_ln_weight_to_fp16, x = x_289_cast); + tensor var_2657_to_fp16 = const()[name = tensor("op_2657_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(597715072)))]; + tensor var_2658_to_fp16 = const()[name = tensor("op_2658_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(606103744)))]; + tensor input_193_cast = linear(bias = var_2658_to_fp16, weight = var_2657_to_fp16, x = var_2648_cast); + tensor x_293_mode_0 = const()[name = tensor("x_293_mode_0"), val = tensor("EXACT")]; + tensor x_293_cast = gelu(mode = x_293_mode_0, x = input_193_cast); + tensor var_2663_to_fp16 = const()[name = tensor("op_2663_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(606112000)))]; + tensor var_2664_to_fp16 = const()[name = tensor("op_2664_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614500672)))]; + tensor var_2665_cast = linear(bias = var_2664_to_fp16, weight = var_2663_to_fp16, x = x_293_cast); + tensor x_cast = add(x = x_289_cast, y = var_2665_cast); + tensor var_2678_axes_0 = const()[name = tensor("op_2678_axes_0"), val = tensor([-1])]; + tensor ln_post_weight_to_fp16 = const()[name = tensor("ln_post_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614502784)))]; + tensor ln_post_bias_to_fp16 = const()[name = tensor("ln_post_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(614504896)))]; + tensor var_2669_to_fp16 = const()[name = tensor("op_2669_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_2678_cast = layer_norm(axes = var_2678_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_2669_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast); + tensor var_2678_cast_to_fp32_dtype_0 = const()[name = tensor("op_2678_cast_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor output = cast(dtype = var_2678_cast_to_fp32_dtype_0, x = var_2678_cast); + } -> (output); +} \ No newline at end of file diff --git a/ggml-medium.en-encoder.mlmodelc/weights/weight.bin b/ggml-medium.en-encoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..ef453169cca80cb0fb9fb7b6f6939e34b6acf521 --- /dev/null +++ b/ggml-medium.en-encoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:63a6c6a2c6cb26aaca930ccc03fe96e2e5820d1ae2802010900eff36b226674d +size 614507008 diff --git a/ggml-small-encoder.mlmodelc.zip b/ggml-small-encoder.mlmodelc.zip index f5a0d8d0634acd13b9e19b9587e0a3473af5be02..57825c9265a7889fb11286cd22f82a7823832a16 100644 --- a/ggml-small-encoder.mlmodelc.zip +++ b/ggml-small-encoder.mlmodelc.zip @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:de43fb9fed471e95c19e60ae67575c2bf09e8fb607016da171b06ddad313988b +oid sha256:c8b65b12409d1fd32029d42992903f4e5aef6679da3a4835bcb85738a19e1057 size 163083239 diff --git a/ggml-small-encoder.mlmodelc/analytics/coremldata.bin b/ggml-small-encoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..9af8eb5beab7ec5764639dbc0fead5c8eb711493 --- /dev/null +++ b/ggml-small-encoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:18ad2072ae82872c2ba8a187071e1e7d6c1105253685e7aa95138adcf07874e0 +size 207 diff --git a/ggml-small-encoder.mlmodelc/coremldata.bin b/ggml-small-encoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..c732ed7b8caeac3c15acda1c2836f3e1f525a5a8 --- /dev/null +++ b/ggml-small-encoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:05fe28591b40616fa0c34ad7b853133623f5300923ec812acb11459c411acf3b +size 149 diff --git a/ggml-small-encoder.mlmodelc/metadata.json b/ggml-small-encoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..9cd2f12c33b3b1a5fb4748833753f83b633e1628 --- /dev/null +++ b/ggml-small-encoder.mlmodelc/metadata.json @@ -0,0 +1,64 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float32", + "formattedType" : "MultiArray (Float32)", + "shortDescription" : "", + "shape" : "[]", + "name" : "output", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 6, + "mlProgramOperationTypeHistogram" : { + "Linear" : 72, + "Matmul" : 24, + "Cast" : 2, + "Conv" : 2, + "Softmax" : 12, + "Add" : 25, + "LayerNorm" : 25, + "Mul" : 24, + "Transpose" : 49, + "Gelu" : 14, + "Reshape" : 48 + }, + "computePrecision" : "Mixed (Float16, Float32, Int32)", + "isUpdatable" : "0", + "availability" : { + "macOS" : "12.0", + "tvOS" : "15.0", + "watchOS" : "8.0", + "iOS" : "15.0", + "macCatalyst" : "15.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float32", + "formattedType" : "MultiArray (Float32 1 × 80 × 3000)", + "shortDescription" : "", + "shape" : "[1, 80, 3000]", + "name" : "logmel_data", + "type" : "MultiArray" + } + ], + "generatedClassName" : "coreml_encoder_small", + "method" : "predict" + } +] \ No newline at end of file diff --git a/ggml-small-encoder.mlmodelc/model.mil b/ggml-small-encoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..8149f70d09bf9f27147da1dfaed40fe475e486c2 --- /dev/null +++ b/ggml-small-encoder.mlmodelc/model.mil @@ -0,0 +1,747 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "4.28.4"}, {"coremlc-version", "1436.100.10"}})] +{ + func main(tensor logmel_data) { + tensor var_32 = const()[name = tensor("op_32"), val = tensor(1)]; + tensor var_40 = const()[name = tensor("op_40"), val = tensor([1])]; + tensor var_42 = const()[name = tensor("op_42"), val = tensor([1])]; + tensor var_44_pad_type_0 = const()[name = tensor("op_44_pad_type_0"), val = tensor("custom")]; + tensor var_44_pad_0 = const()[name = tensor("op_44_pad_0"), val = tensor([1, 1])]; + tensor logmel_data_to_fp16_dtype_0 = const()[name = tensor("logmel_data_to_fp16_dtype_0"), val = tensor("fp16")]; + tensor weight_3_to_fp16 = const()[name = tensor("weight_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor bias_3_to_fp16 = const()[name = tensor("bias_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368768)))]; + tensor cast_367 = cast(dtype = logmel_data_to_fp16_dtype_0, x = logmel_data); + tensor var_44_cast = conv(bias = bias_3_to_fp16, dilations = var_42, groups = var_32, pad = var_44_pad_0, pad_type = var_44_pad_type_0, strides = var_40, weight = weight_3_to_fp16, x = cast_367); + tensor input_1_mode_0 = const()[name = tensor("input_1_mode_0"), val = tensor("EXACT")]; + tensor input_1_cast = gelu(mode = input_1_mode_0, x = var_44_cast); + tensor var_48 = const()[name = tensor("op_48"), val = tensor(1)]; + tensor var_57 = const()[name = tensor("op_57"), val = tensor([2])]; + tensor var_59 = const()[name = tensor("op_59"), val = tensor([1])]; + tensor var_61_pad_type_0 = const()[name = tensor("op_61_pad_type_0"), val = tensor("custom")]; + tensor var_61_pad_0 = const()[name = tensor("op_61_pad_0"), val = tensor([1, 1])]; + tensor weight_7_to_fp16 = const()[name = tensor("weight_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(370368)))]; + tensor bias_7_to_fp16 = const()[name = tensor("bias_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3909376)))]; + tensor var_61_cast = conv(bias = bias_7_to_fp16, dilations = var_59, groups = var_48, pad = var_61_pad_0, pad_type = var_61_pad_type_0, strides = var_57, weight = weight_7_to_fp16, x = input_1_cast); + tensor x_3_mode_0 = const()[name = tensor("x_3_mode_0"), val = tensor("EXACT")]; + tensor x_3_cast = gelu(mode = x_3_mode_0, x = var_61_cast); + tensor var_66 = const()[name = tensor("op_66"), val = tensor([0, 2, 1])]; + tensor positional_embedding_to_fp16 = const()[name = tensor("positional_embedding_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3910976)))]; + tensor transpose_96 = transpose(perm = var_66, x = x_3_cast); + tensor var_69_cast = add(x = transpose_96, y = positional_embedding_to_fp16); + tensor var_82 = const()[name = tensor("op_82"), val = tensor(-1)]; + tensor var_99_axes_0 = const()[name = tensor("op_99_axes_0"), val = tensor([-1])]; + tensor blocks_0_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_0_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6215040)))]; + tensor blocks_0_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_0_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6216640)))]; + tensor var_88_to_fp16 = const()[name = tensor("op_88_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_99_cast = layer_norm(axes = var_99_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_88_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_69_cast); + tensor var_110_to_fp16 = const()[name = tensor("op_110_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6218240)))]; + tensor var_111_to_fp16 = const()[name = tensor("op_111_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7397952)))]; + tensor q_1_cast = linear(bias = var_111_to_fp16, weight = var_110_to_fp16, x = var_99_cast); + tensor var_114_to_fp16 = const()[name = tensor("op_114_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7399552)))]; + tensor k_1_bias_0_to_fp16 = const()[name = tensor("k_1_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8579264)))]; + tensor k_1_cast = linear(bias = k_1_bias_0_to_fp16, weight = var_114_to_fp16, x = var_99_cast); + tensor var_118_to_fp16 = const()[name = tensor("op_118_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8580864)))]; + tensor var_119_to_fp16 = const()[name = tensor("op_119_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9760576)))]; + tensor v_1_cast = linear(bias = var_119_to_fp16, weight = var_118_to_fp16, x = var_99_cast); + tensor var_127 = const()[name = tensor("op_127"), val = tensor([1, 1500, 12, -1])]; + tensor var_128_cast = reshape(shape = var_127, x = q_1_cast); + tensor const_84_to_fp16 = const()[name = tensor("const_84_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_3_cast = mul(x = var_128_cast, y = const_84_to_fp16); + tensor var_134 = const()[name = tensor("op_134"), val = tensor([1, 1500, 12, -1])]; + tensor var_135_cast = reshape(shape = var_134, x = k_1_cast); + tensor const_85_to_fp16 = const()[name = tensor("const_85_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_3_cast = mul(x = var_135_cast, y = const_85_to_fp16); + tensor var_141 = const()[name = tensor("op_141"), val = tensor([1, 1500, 12, -1])]; + tensor var_142_cast = reshape(shape = var_141, x = v_1_cast); + tensor var_143 = const()[name = tensor("op_143"), val = tensor([0, 2, 1, 3])]; + tensor qk_1_transpose_x_0 = const()[name = tensor("qk_1_transpose_x_0"), val = tensor(false)]; + tensor qk_1_transpose_y_0 = const()[name = tensor("qk_1_transpose_y_0"), val = tensor(false)]; + tensor transpose_24_perm_0 = const()[name = tensor("transpose_24_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_25_perm_0 = const()[name = tensor("transpose_25_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_93 = transpose(perm = transpose_25_perm_0, x = k_3_cast); + tensor transpose_94 = transpose(perm = transpose_24_perm_0, x = q_3_cast); + tensor qk_1_cast = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_94, y = transpose_93); + tensor var_147_cast = softmax(axis = var_82, x = qk_1_cast); + tensor var_149_transpose_x_0 = const()[name = tensor("op_149_transpose_x_0"), val = tensor(false)]; + tensor var_149_transpose_y_0 = const()[name = tensor("op_149_transpose_y_0"), val = tensor(false)]; + tensor transpose_95 = transpose(perm = var_143, x = var_142_cast); + tensor var_149_cast = matmul(transpose_x = var_149_transpose_x_0, transpose_y = var_149_transpose_y_0, x = var_147_cast, y = transpose_95); + tensor var_150 = const()[name = tensor("op_150"), val = tensor([0, 2, 1, 3])]; + tensor concat_0 = const()[name = tensor("concat_0"), val = tensor([1, 1500, 768])]; + tensor transpose_92 = transpose(perm = var_150, x = var_149_cast); + tensor x_11_cast = reshape(shape = concat_0, x = transpose_92); + tensor var_155_to_fp16 = const()[name = tensor("op_155_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9762176)))]; + tensor var_156_to_fp16 = const()[name = tensor("op_156_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10941888)))]; + tensor var_157_cast = linear(bias = var_156_to_fp16, weight = var_155_to_fp16, x = x_11_cast); + tensor x_13_cast = add(x = var_69_cast, y = var_157_cast); + tensor var_163_axes_0 = const()[name = tensor("op_163_axes_0"), val = tensor([-1])]; + tensor blocks_0_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10943488)))]; + tensor blocks_0_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10945088)))]; + tensor var_163_cast = layer_norm(axes = var_163_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_88_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast); + tensor var_172_to_fp16 = const()[name = tensor("op_172_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10946688)))]; + tensor var_173_to_fp16 = const()[name = tensor("op_173_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15665344)))]; + tensor input_9_cast = linear(bias = var_173_to_fp16, weight = var_172_to_fp16, x = var_163_cast); + tensor x_17_mode_0 = const()[name = tensor("x_17_mode_0"), val = tensor("EXACT")]; + tensor x_17_cast = gelu(mode = x_17_mode_0, x = input_9_cast); + tensor var_178_to_fp16 = const()[name = tensor("op_178_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15671552)))]; + tensor var_179_to_fp16 = const()[name = tensor("op_179_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20390208)))]; + tensor var_180_cast = linear(bias = var_179_to_fp16, weight = var_178_to_fp16, x = x_17_cast); + tensor x_19_cast = add(x = x_13_cast, y = var_180_cast); + tensor var_189 = const()[name = tensor("op_189"), val = tensor(-1)]; + tensor var_206_axes_0 = const()[name = tensor("op_206_axes_0"), val = tensor([-1])]; + tensor blocks_1_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_1_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20391808)))]; + tensor blocks_1_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_1_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20393408)))]; + tensor var_195_to_fp16 = const()[name = tensor("op_195_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_206_cast = layer_norm(axes = var_206_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_195_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast); + tensor var_217_to_fp16 = const()[name = tensor("op_217_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20395008)))]; + tensor var_218_to_fp16 = const()[name = tensor("op_218_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21574720)))]; + tensor q_5_cast = linear(bias = var_218_to_fp16, weight = var_217_to_fp16, x = var_206_cast); + tensor var_221_to_fp16 = const()[name = tensor("op_221_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21576320)))]; + tensor k_5_bias_0_to_fp16 = const()[name = tensor("k_5_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22756032)))]; + tensor k_5_cast = linear(bias = k_5_bias_0_to_fp16, weight = var_221_to_fp16, x = var_206_cast); + tensor var_225_to_fp16 = const()[name = tensor("op_225_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22757632)))]; + tensor var_226_to_fp16 = const()[name = tensor("op_226_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23937344)))]; + tensor v_5_cast = linear(bias = var_226_to_fp16, weight = var_225_to_fp16, x = var_206_cast); + tensor var_234 = const()[name = tensor("op_234"), val = tensor([1, 1500, 12, -1])]; + tensor var_235_cast = reshape(shape = var_234, x = q_5_cast); + tensor const_86_to_fp16 = const()[name = tensor("const_86_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_7_cast = mul(x = var_235_cast, y = const_86_to_fp16); + tensor var_241 = const()[name = tensor("op_241"), val = tensor([1, 1500, 12, -1])]; + tensor var_242_cast = reshape(shape = var_241, x = k_5_cast); + tensor const_87_to_fp16 = const()[name = tensor("const_87_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_7_cast = mul(x = var_242_cast, y = const_87_to_fp16); + tensor var_248 = const()[name = tensor("op_248"), val = tensor([1, 1500, 12, -1])]; + tensor var_249_cast = reshape(shape = var_248, x = v_5_cast); + tensor var_250 = const()[name = tensor("op_250"), val = tensor([0, 2, 1, 3])]; + tensor qk_3_transpose_x_0 = const()[name = tensor("qk_3_transpose_x_0"), val = tensor(false)]; + tensor qk_3_transpose_y_0 = const()[name = tensor("qk_3_transpose_y_0"), val = tensor(false)]; + tensor transpose_26_perm_0 = const()[name = tensor("transpose_26_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_27_perm_0 = const()[name = tensor("transpose_27_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_89 = transpose(perm = transpose_27_perm_0, x = k_7_cast); + tensor transpose_90 = transpose(perm = transpose_26_perm_0, x = q_7_cast); + tensor qk_3_cast = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_90, y = transpose_89); + tensor var_254_cast = softmax(axis = var_189, x = qk_3_cast); + tensor var_256_transpose_x_0 = const()[name = tensor("op_256_transpose_x_0"), val = tensor(false)]; + tensor var_256_transpose_y_0 = const()[name = tensor("op_256_transpose_y_0"), val = tensor(false)]; + tensor transpose_91 = transpose(perm = var_250, x = var_249_cast); + tensor var_256_cast = matmul(transpose_x = var_256_transpose_x_0, transpose_y = var_256_transpose_y_0, x = var_254_cast, y = transpose_91); + tensor var_257 = const()[name = tensor("op_257"), val = tensor([0, 2, 1, 3])]; + tensor concat_1 = const()[name = tensor("concat_1"), val = tensor([1, 1500, 768])]; + tensor transpose_88 = transpose(perm = var_257, x = var_256_cast); + tensor x_23_cast = reshape(shape = concat_1, x = transpose_88); + tensor var_262_to_fp16 = const()[name = tensor("op_262_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23938944)))]; + tensor var_263_to_fp16 = const()[name = tensor("op_263_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25118656)))]; + tensor var_264_cast = linear(bias = var_263_to_fp16, weight = var_262_to_fp16, x = x_23_cast); + tensor x_25_cast = add(x = x_19_cast, y = var_264_cast); + tensor var_270_axes_0 = const()[name = tensor("op_270_axes_0"), val = tensor([-1])]; + tensor blocks_1_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25120256)))]; + tensor blocks_1_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25121856)))]; + tensor var_270_cast = layer_norm(axes = var_270_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_195_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast); + tensor var_279_to_fp16 = const()[name = tensor("op_279_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25123456)))]; + tensor var_280_to_fp16 = const()[name = tensor("op_280_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29842112)))]; + tensor input_17_cast = linear(bias = var_280_to_fp16, weight = var_279_to_fp16, x = var_270_cast); + tensor x_29_mode_0 = const()[name = tensor("x_29_mode_0"), val = tensor("EXACT")]; + tensor x_29_cast = gelu(mode = x_29_mode_0, x = input_17_cast); + tensor var_285_to_fp16 = const()[name = tensor("op_285_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29848320)))]; + tensor var_286_to_fp16 = const()[name = tensor("op_286_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34566976)))]; + tensor var_287_cast = linear(bias = var_286_to_fp16, weight = var_285_to_fp16, x = x_29_cast); + tensor x_31_cast = add(x = x_25_cast, y = var_287_cast); + tensor var_296 = const()[name = tensor("op_296"), val = tensor(-1)]; + tensor var_313_axes_0 = const()[name = tensor("op_313_axes_0"), val = tensor([-1])]; + tensor blocks_2_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_2_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34568576)))]; + tensor blocks_2_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_2_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34570176)))]; + tensor var_302_to_fp16 = const()[name = tensor("op_302_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_313_cast = layer_norm(axes = var_313_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_302_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast); + tensor var_324_to_fp16 = const()[name = tensor("op_324_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34571776)))]; + tensor var_325_to_fp16 = const()[name = tensor("op_325_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35751488)))]; + tensor q_9_cast = linear(bias = var_325_to_fp16, weight = var_324_to_fp16, x = var_313_cast); + tensor var_328_to_fp16 = const()[name = tensor("op_328_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35753088)))]; + tensor k_9_bias_0_to_fp16 = const()[name = tensor("k_9_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36932800)))]; + tensor k_9_cast = linear(bias = k_9_bias_0_to_fp16, weight = var_328_to_fp16, x = var_313_cast); + tensor var_332_to_fp16 = const()[name = tensor("op_332_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36934400)))]; + tensor var_333_to_fp16 = const()[name = tensor("op_333_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38114112)))]; + tensor v_9_cast = linear(bias = var_333_to_fp16, weight = var_332_to_fp16, x = var_313_cast); + tensor var_341 = const()[name = tensor("op_341"), val = tensor([1, 1500, 12, -1])]; + tensor var_342_cast = reshape(shape = var_341, x = q_9_cast); + tensor const_88_to_fp16 = const()[name = tensor("const_88_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_11_cast = mul(x = var_342_cast, y = const_88_to_fp16); + tensor var_348 = const()[name = tensor("op_348"), val = tensor([1, 1500, 12, -1])]; + tensor var_349_cast = reshape(shape = var_348, x = k_9_cast); + tensor const_89_to_fp16 = const()[name = tensor("const_89_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_11_cast = mul(x = var_349_cast, y = const_89_to_fp16); + tensor var_355 = const()[name = tensor("op_355"), val = tensor([1, 1500, 12, -1])]; + tensor var_356_cast = reshape(shape = var_355, x = v_9_cast); + tensor var_357 = const()[name = tensor("op_357"), val = tensor([0, 2, 1, 3])]; + tensor qk_5_transpose_x_0 = const()[name = tensor("qk_5_transpose_x_0"), val = tensor(false)]; + tensor qk_5_transpose_y_0 = const()[name = tensor("qk_5_transpose_y_0"), val = tensor(false)]; + tensor transpose_28_perm_0 = const()[name = tensor("transpose_28_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_29_perm_0 = const()[name = tensor("transpose_29_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_85 = transpose(perm = transpose_29_perm_0, x = k_11_cast); + tensor transpose_86 = transpose(perm = transpose_28_perm_0, x = q_11_cast); + tensor qk_5_cast = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_86, y = transpose_85); + tensor var_361_cast = softmax(axis = var_296, x = qk_5_cast); + tensor var_363_transpose_x_0 = const()[name = tensor("op_363_transpose_x_0"), val = tensor(false)]; + tensor var_363_transpose_y_0 = const()[name = tensor("op_363_transpose_y_0"), val = tensor(false)]; + tensor transpose_87 = transpose(perm = var_357, x = var_356_cast); + tensor var_363_cast = matmul(transpose_x = var_363_transpose_x_0, transpose_y = var_363_transpose_y_0, x = var_361_cast, y = transpose_87); + tensor var_364 = const()[name = tensor("op_364"), val = tensor([0, 2, 1, 3])]; + tensor concat_2 = const()[name = tensor("concat_2"), val = tensor([1, 1500, 768])]; + tensor transpose_84 = transpose(perm = var_364, x = var_363_cast); + tensor x_35_cast = reshape(shape = concat_2, x = transpose_84); + tensor var_369_to_fp16 = const()[name = tensor("op_369_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38115712)))]; + tensor var_370_to_fp16 = const()[name = tensor("op_370_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39295424)))]; + tensor var_371_cast = linear(bias = var_370_to_fp16, weight = var_369_to_fp16, x = x_35_cast); + tensor x_37_cast = add(x = x_31_cast, y = var_371_cast); + tensor var_377_axes_0 = const()[name = tensor("op_377_axes_0"), val = tensor([-1])]; + tensor blocks_2_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39297024)))]; + tensor blocks_2_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39298624)))]; + tensor var_377_cast = layer_norm(axes = var_377_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_302_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast); + tensor var_386_to_fp16 = const()[name = tensor("op_386_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39300224)))]; + tensor var_387_to_fp16 = const()[name = tensor("op_387_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44018880)))]; + tensor input_25_cast = linear(bias = var_387_to_fp16, weight = var_386_to_fp16, x = var_377_cast); + tensor x_41_mode_0 = const()[name = tensor("x_41_mode_0"), val = tensor("EXACT")]; + tensor x_41_cast = gelu(mode = x_41_mode_0, x = input_25_cast); + tensor var_392_to_fp16 = const()[name = tensor("op_392_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44025088)))]; + tensor var_393_to_fp16 = const()[name = tensor("op_393_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48743744)))]; + tensor var_394_cast = linear(bias = var_393_to_fp16, weight = var_392_to_fp16, x = x_41_cast); + tensor x_43_cast = add(x = x_37_cast, y = var_394_cast); + tensor var_403 = const()[name = tensor("op_403"), val = tensor(-1)]; + tensor var_420_axes_0 = const()[name = tensor("op_420_axes_0"), val = tensor([-1])]; + tensor blocks_3_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_3_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48745344)))]; + tensor blocks_3_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_3_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48746944)))]; + tensor var_409_to_fp16 = const()[name = tensor("op_409_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_420_cast = layer_norm(axes = var_420_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_409_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast); + tensor var_431_to_fp16 = const()[name = tensor("op_431_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48748544)))]; + tensor var_432_to_fp16 = const()[name = tensor("op_432_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49928256)))]; + tensor q_13_cast = linear(bias = var_432_to_fp16, weight = var_431_to_fp16, x = var_420_cast); + tensor var_435_to_fp16 = const()[name = tensor("op_435_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49929856)))]; + tensor k_13_bias_0_to_fp16 = const()[name = tensor("k_13_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51109568)))]; + tensor k_13_cast = linear(bias = k_13_bias_0_to_fp16, weight = var_435_to_fp16, x = var_420_cast); + tensor var_439_to_fp16 = const()[name = tensor("op_439_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51111168)))]; + tensor var_440_to_fp16 = const()[name = tensor("op_440_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(52290880)))]; + tensor v_13_cast = linear(bias = var_440_to_fp16, weight = var_439_to_fp16, x = var_420_cast); + tensor var_448 = const()[name = tensor("op_448"), val = tensor([1, 1500, 12, -1])]; + tensor var_449_cast = reshape(shape = var_448, x = q_13_cast); + tensor const_90_to_fp16 = const()[name = tensor("const_90_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_15_cast = mul(x = var_449_cast, y = const_90_to_fp16); + tensor var_455 = const()[name = tensor("op_455"), val = tensor([1, 1500, 12, -1])]; + tensor var_456_cast = reshape(shape = var_455, x = k_13_cast); + tensor const_91_to_fp16 = const()[name = tensor("const_91_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_15_cast = mul(x = var_456_cast, y = const_91_to_fp16); + tensor var_462 = const()[name = tensor("op_462"), val = tensor([1, 1500, 12, -1])]; + tensor var_463_cast = reshape(shape = var_462, x = v_13_cast); + tensor var_464 = const()[name = tensor("op_464"), val = tensor([0, 2, 1, 3])]; + tensor qk_7_transpose_x_0 = const()[name = tensor("qk_7_transpose_x_0"), val = tensor(false)]; + tensor qk_7_transpose_y_0 = const()[name = tensor("qk_7_transpose_y_0"), val = tensor(false)]; + tensor transpose_30_perm_0 = const()[name = tensor("transpose_30_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_31_perm_0 = const()[name = tensor("transpose_31_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_81 = transpose(perm = transpose_31_perm_0, x = k_15_cast); + tensor transpose_82 = transpose(perm = transpose_30_perm_0, x = q_15_cast); + tensor qk_7_cast = matmul(transpose_x = qk_7_transpose_x_0, transpose_y = qk_7_transpose_y_0, x = transpose_82, y = transpose_81); + tensor var_468_cast = softmax(axis = var_403, x = qk_7_cast); + tensor var_470_transpose_x_0 = const()[name = tensor("op_470_transpose_x_0"), val = tensor(false)]; + tensor var_470_transpose_y_0 = const()[name = tensor("op_470_transpose_y_0"), val = tensor(false)]; + tensor transpose_83 = transpose(perm = var_464, x = var_463_cast); + tensor var_470_cast = matmul(transpose_x = var_470_transpose_x_0, transpose_y = var_470_transpose_y_0, x = var_468_cast, y = transpose_83); + tensor var_471 = const()[name = tensor("op_471"), val = tensor([0, 2, 1, 3])]; + tensor concat_3 = const()[name = tensor("concat_3"), val = tensor([1, 1500, 768])]; + tensor transpose_80 = transpose(perm = var_471, x = var_470_cast); + tensor x_47_cast = reshape(shape = concat_3, x = transpose_80); + tensor var_476_to_fp16 = const()[name = tensor("op_476_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(52292480)))]; + tensor var_477_to_fp16 = const()[name = tensor("op_477_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53472192)))]; + tensor var_478_cast = linear(bias = var_477_to_fp16, weight = var_476_to_fp16, x = x_47_cast); + tensor x_49_cast = add(x = x_43_cast, y = var_478_cast); + tensor var_484_axes_0 = const()[name = tensor("op_484_axes_0"), val = tensor([-1])]; + tensor blocks_3_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53473792)))]; + tensor blocks_3_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53475392)))]; + tensor var_484_cast = layer_norm(axes = var_484_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_409_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast); + tensor var_493_to_fp16 = const()[name = tensor("op_493_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53476992)))]; + tensor var_494_to_fp16 = const()[name = tensor("op_494_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(58195648)))]; + tensor input_33_cast = linear(bias = var_494_to_fp16, weight = var_493_to_fp16, x = var_484_cast); + tensor x_53_mode_0 = const()[name = tensor("x_53_mode_0"), val = tensor("EXACT")]; + tensor x_53_cast = gelu(mode = x_53_mode_0, x = input_33_cast); + tensor var_499_to_fp16 = const()[name = tensor("op_499_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(58201856)))]; + tensor var_500_to_fp16 = const()[name = tensor("op_500_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62920512)))]; + tensor var_501_cast = linear(bias = var_500_to_fp16, weight = var_499_to_fp16, x = x_53_cast); + tensor x_55_cast = add(x = x_49_cast, y = var_501_cast); + tensor var_510 = const()[name = tensor("op_510"), val = tensor(-1)]; + tensor var_527_axes_0 = const()[name = tensor("op_527_axes_0"), val = tensor([-1])]; + tensor blocks_4_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_4_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62922112)))]; + tensor blocks_4_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_4_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62923712)))]; + tensor var_516_to_fp16 = const()[name = tensor("op_516_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_527_cast = layer_norm(axes = var_527_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_516_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast); + tensor var_538_to_fp16 = const()[name = tensor("op_538_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62925312)))]; + tensor var_539_to_fp16 = const()[name = tensor("op_539_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64105024)))]; + tensor q_17_cast = linear(bias = var_539_to_fp16, weight = var_538_to_fp16, x = var_527_cast); + tensor var_542_to_fp16 = const()[name = tensor("op_542_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64106624)))]; + tensor k_17_bias_0_to_fp16 = const()[name = tensor("k_17_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65286336)))]; + tensor k_17_cast = linear(bias = k_17_bias_0_to_fp16, weight = var_542_to_fp16, x = var_527_cast); + tensor var_546_to_fp16 = const()[name = tensor("op_546_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65287936)))]; + tensor var_547_to_fp16 = const()[name = tensor("op_547_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66467648)))]; + tensor v_17_cast = linear(bias = var_547_to_fp16, weight = var_546_to_fp16, x = var_527_cast); + tensor var_555 = const()[name = tensor("op_555"), val = tensor([1, 1500, 12, -1])]; + tensor var_556_cast = reshape(shape = var_555, x = q_17_cast); + tensor const_92_to_fp16 = const()[name = tensor("const_92_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_19_cast = mul(x = var_556_cast, y = const_92_to_fp16); + tensor var_562 = const()[name = tensor("op_562"), val = tensor([1, 1500, 12, -1])]; + tensor var_563_cast = reshape(shape = var_562, x = k_17_cast); + tensor const_93_to_fp16 = const()[name = tensor("const_93_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_19_cast = mul(x = var_563_cast, y = const_93_to_fp16); + tensor var_569 = const()[name = tensor("op_569"), val = tensor([1, 1500, 12, -1])]; + tensor var_570_cast = reshape(shape = var_569, x = v_17_cast); + tensor var_571 = const()[name = tensor("op_571"), val = tensor([0, 2, 1, 3])]; + tensor qk_9_transpose_x_0 = const()[name = tensor("qk_9_transpose_x_0"), val = tensor(false)]; + tensor qk_9_transpose_y_0 = const()[name = tensor("qk_9_transpose_y_0"), val = tensor(false)]; + tensor transpose_32_perm_0 = const()[name = tensor("transpose_32_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_33_perm_0 = const()[name = tensor("transpose_33_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_77 = transpose(perm = transpose_33_perm_0, x = k_19_cast); + tensor transpose_78 = transpose(perm = transpose_32_perm_0, x = q_19_cast); + tensor qk_9_cast = matmul(transpose_x = qk_9_transpose_x_0, transpose_y = qk_9_transpose_y_0, x = transpose_78, y = transpose_77); + tensor var_575_cast = softmax(axis = var_510, x = qk_9_cast); + tensor var_577_transpose_x_0 = const()[name = tensor("op_577_transpose_x_0"), val = tensor(false)]; + tensor var_577_transpose_y_0 = const()[name = tensor("op_577_transpose_y_0"), val = tensor(false)]; + tensor transpose_79 = transpose(perm = var_571, x = var_570_cast); + tensor var_577_cast = matmul(transpose_x = var_577_transpose_x_0, transpose_y = var_577_transpose_y_0, x = var_575_cast, y = transpose_79); + tensor var_578 = const()[name = tensor("op_578"), val = tensor([0, 2, 1, 3])]; + tensor concat_4 = const()[name = tensor("concat_4"), val = tensor([1, 1500, 768])]; + tensor transpose_76 = transpose(perm = var_578, x = var_577_cast); + tensor x_59_cast = reshape(shape = concat_4, x = transpose_76); + tensor var_583_to_fp16 = const()[name = tensor("op_583_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66469248)))]; + tensor var_584_to_fp16 = const()[name = tensor("op_584_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67648960)))]; + tensor var_585_cast = linear(bias = var_584_to_fp16, weight = var_583_to_fp16, x = x_59_cast); + tensor x_61_cast = add(x = x_55_cast, y = var_585_cast); + tensor var_591_axes_0 = const()[name = tensor("op_591_axes_0"), val = tensor([-1])]; + tensor blocks_4_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67650560)))]; + tensor blocks_4_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67652160)))]; + tensor var_591_cast = layer_norm(axes = var_591_axes_0, beta = blocks_4_mlp_ln_bias_to_fp16, epsilon = var_516_to_fp16, gamma = blocks_4_mlp_ln_weight_to_fp16, x = x_61_cast); + tensor var_600_to_fp16 = const()[name = tensor("op_600_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67653760)))]; + tensor var_601_to_fp16 = const()[name = tensor("op_601_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(72372416)))]; + tensor input_41_cast = linear(bias = var_601_to_fp16, weight = var_600_to_fp16, x = var_591_cast); + tensor x_65_mode_0 = const()[name = tensor("x_65_mode_0"), val = tensor("EXACT")]; + tensor x_65_cast = gelu(mode = x_65_mode_0, x = input_41_cast); + tensor var_606_to_fp16 = const()[name = tensor("op_606_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(72378624)))]; + tensor var_607_to_fp16 = const()[name = tensor("op_607_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77097280)))]; + tensor var_608_cast = linear(bias = var_607_to_fp16, weight = var_606_to_fp16, x = x_65_cast); + tensor x_67_cast = add(x = x_61_cast, y = var_608_cast); + tensor var_617 = const()[name = tensor("op_617"), val = tensor(-1)]; + tensor var_634_axes_0 = const()[name = tensor("op_634_axes_0"), val = tensor([-1])]; + tensor blocks_5_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_5_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77098880)))]; + tensor blocks_5_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_5_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77100480)))]; + tensor var_623_to_fp16 = const()[name = tensor("op_623_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_634_cast = layer_norm(axes = var_634_axes_0, beta = blocks_5_attn_ln_bias_to_fp16, epsilon = var_623_to_fp16, gamma = blocks_5_attn_ln_weight_to_fp16, x = x_67_cast); + tensor var_645_to_fp16 = const()[name = tensor("op_645_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77102080)))]; + tensor var_646_to_fp16 = const()[name = tensor("op_646_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78281792)))]; + tensor q_21_cast = linear(bias = var_646_to_fp16, weight = var_645_to_fp16, x = var_634_cast); + tensor var_649_to_fp16 = const()[name = tensor("op_649_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78283392)))]; + tensor k_21_bias_0_to_fp16 = const()[name = tensor("k_21_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79463104)))]; + tensor k_21_cast = linear(bias = k_21_bias_0_to_fp16, weight = var_649_to_fp16, x = var_634_cast); + tensor var_653_to_fp16 = const()[name = tensor("op_653_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79464704)))]; + tensor var_654_to_fp16 = const()[name = tensor("op_654_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80644416)))]; + tensor v_21_cast = linear(bias = var_654_to_fp16, weight = var_653_to_fp16, x = var_634_cast); + tensor var_662 = const()[name = tensor("op_662"), val = tensor([1, 1500, 12, -1])]; + tensor var_663_cast = reshape(shape = var_662, x = q_21_cast); + tensor const_94_to_fp16 = const()[name = tensor("const_94_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_23_cast = mul(x = var_663_cast, y = const_94_to_fp16); + tensor var_669 = const()[name = tensor("op_669"), val = tensor([1, 1500, 12, -1])]; + tensor var_670_cast = reshape(shape = var_669, x = k_21_cast); + tensor const_95_to_fp16 = const()[name = tensor("const_95_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_23_cast = mul(x = var_670_cast, y = const_95_to_fp16); + tensor var_676 = const()[name = tensor("op_676"), val = tensor([1, 1500, 12, -1])]; + tensor var_677_cast = reshape(shape = var_676, x = v_21_cast); + tensor var_678 = const()[name = tensor("op_678"), val = tensor([0, 2, 1, 3])]; + tensor qk_11_transpose_x_0 = const()[name = tensor("qk_11_transpose_x_0"), val = tensor(false)]; + tensor qk_11_transpose_y_0 = const()[name = tensor("qk_11_transpose_y_0"), val = tensor(false)]; + tensor transpose_34_perm_0 = const()[name = tensor("transpose_34_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_35_perm_0 = const()[name = tensor("transpose_35_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_73 = transpose(perm = transpose_35_perm_0, x = k_23_cast); + tensor transpose_74 = transpose(perm = transpose_34_perm_0, x = q_23_cast); + tensor qk_11_cast = matmul(transpose_x = qk_11_transpose_x_0, transpose_y = qk_11_transpose_y_0, x = transpose_74, y = transpose_73); + tensor var_682_cast = softmax(axis = var_617, x = qk_11_cast); + tensor var_684_transpose_x_0 = const()[name = tensor("op_684_transpose_x_0"), val = tensor(false)]; + tensor var_684_transpose_y_0 = const()[name = tensor("op_684_transpose_y_0"), val = tensor(false)]; + tensor transpose_75 = transpose(perm = var_678, x = var_677_cast); + tensor var_684_cast = matmul(transpose_x = var_684_transpose_x_0, transpose_y = var_684_transpose_y_0, x = var_682_cast, y = transpose_75); + tensor var_685 = const()[name = tensor("op_685"), val = tensor([0, 2, 1, 3])]; + tensor concat_5 = const()[name = tensor("concat_5"), val = tensor([1, 1500, 768])]; + tensor transpose_72 = transpose(perm = var_685, x = var_684_cast); + tensor x_71_cast = reshape(shape = concat_5, x = transpose_72); + tensor var_690_to_fp16 = const()[name = tensor("op_690_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80646016)))]; + tensor var_691_to_fp16 = const()[name = tensor("op_691_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81825728)))]; + tensor var_692_cast = linear(bias = var_691_to_fp16, weight = var_690_to_fp16, x = x_71_cast); + tensor x_73_cast = add(x = x_67_cast, y = var_692_cast); + tensor var_698_axes_0 = const()[name = tensor("op_698_axes_0"), val = tensor([-1])]; + tensor blocks_5_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81827328)))]; + tensor blocks_5_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81828928)))]; + tensor var_698_cast = layer_norm(axes = var_698_axes_0, beta = blocks_5_mlp_ln_bias_to_fp16, epsilon = var_623_to_fp16, gamma = blocks_5_mlp_ln_weight_to_fp16, x = x_73_cast); + tensor var_707_to_fp16 = const()[name = tensor("op_707_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81830528)))]; + tensor var_708_to_fp16 = const()[name = tensor("op_708_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86549184)))]; + tensor input_49_cast = linear(bias = var_708_to_fp16, weight = var_707_to_fp16, x = var_698_cast); + tensor x_77_mode_0 = const()[name = tensor("x_77_mode_0"), val = tensor("EXACT")]; + tensor x_77_cast = gelu(mode = x_77_mode_0, x = input_49_cast); + tensor var_713_to_fp16 = const()[name = tensor("op_713_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86555392)))]; + tensor var_714_to_fp16 = const()[name = tensor("op_714_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91274048)))]; + tensor var_715_cast = linear(bias = var_714_to_fp16, weight = var_713_to_fp16, x = x_77_cast); + tensor x_79_cast = add(x = x_73_cast, y = var_715_cast); + tensor var_724 = const()[name = tensor("op_724"), val = tensor(-1)]; + tensor var_741_axes_0 = const()[name = tensor("op_741_axes_0"), val = tensor([-1])]; + tensor blocks_6_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_6_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91275648)))]; + tensor blocks_6_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_6_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91277248)))]; + tensor var_730_to_fp16 = const()[name = tensor("op_730_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_741_cast = layer_norm(axes = var_741_axes_0, beta = blocks_6_attn_ln_bias_to_fp16, epsilon = var_730_to_fp16, gamma = blocks_6_attn_ln_weight_to_fp16, x = x_79_cast); + tensor var_752_to_fp16 = const()[name = tensor("op_752_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91278848)))]; + tensor var_753_to_fp16 = const()[name = tensor("op_753_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92458560)))]; + tensor q_25_cast = linear(bias = var_753_to_fp16, weight = var_752_to_fp16, x = var_741_cast); + tensor var_756_to_fp16 = const()[name = tensor("op_756_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92460160)))]; + tensor k_25_bias_0_to_fp16 = const()[name = tensor("k_25_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93639872)))]; + tensor k_25_cast = linear(bias = k_25_bias_0_to_fp16, weight = var_756_to_fp16, x = var_741_cast); + tensor var_760_to_fp16 = const()[name = tensor("op_760_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93641472)))]; + tensor var_761_to_fp16 = const()[name = tensor("op_761_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(94821184)))]; + tensor v_25_cast = linear(bias = var_761_to_fp16, weight = var_760_to_fp16, x = var_741_cast); + tensor var_769 = const()[name = tensor("op_769"), val = tensor([1, 1500, 12, -1])]; + tensor var_770_cast = reshape(shape = var_769, x = q_25_cast); + tensor const_96_to_fp16 = const()[name = tensor("const_96_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_27_cast = mul(x = var_770_cast, y = const_96_to_fp16); + tensor var_776 = const()[name = tensor("op_776"), val = tensor([1, 1500, 12, -1])]; + tensor var_777_cast = reshape(shape = var_776, x = k_25_cast); + tensor const_97_to_fp16 = const()[name = tensor("const_97_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_27_cast = mul(x = var_777_cast, y = const_97_to_fp16); + tensor var_783 = const()[name = tensor("op_783"), val = tensor([1, 1500, 12, -1])]; + tensor var_784_cast = reshape(shape = var_783, x = v_25_cast); + tensor var_785 = const()[name = tensor("op_785"), val = tensor([0, 2, 1, 3])]; + tensor qk_13_transpose_x_0 = const()[name = tensor("qk_13_transpose_x_0"), val = tensor(false)]; + tensor qk_13_transpose_y_0 = const()[name = tensor("qk_13_transpose_y_0"), val = tensor(false)]; + tensor transpose_36_perm_0 = const()[name = tensor("transpose_36_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_37_perm_0 = const()[name = tensor("transpose_37_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_69 = transpose(perm = transpose_37_perm_0, x = k_27_cast); + tensor transpose_70 = transpose(perm = transpose_36_perm_0, x = q_27_cast); + tensor qk_13_cast = matmul(transpose_x = qk_13_transpose_x_0, transpose_y = qk_13_transpose_y_0, x = transpose_70, y = transpose_69); + tensor var_789_cast = softmax(axis = var_724, x = qk_13_cast); + tensor var_791_transpose_x_0 = const()[name = tensor("op_791_transpose_x_0"), val = tensor(false)]; + tensor var_791_transpose_y_0 = const()[name = tensor("op_791_transpose_y_0"), val = tensor(false)]; + tensor transpose_71 = transpose(perm = var_785, x = var_784_cast); + tensor var_791_cast = matmul(transpose_x = var_791_transpose_x_0, transpose_y = var_791_transpose_y_0, x = var_789_cast, y = transpose_71); + tensor var_792 = const()[name = tensor("op_792"), val = tensor([0, 2, 1, 3])]; + tensor concat_6 = const()[name = tensor("concat_6"), val = tensor([1, 1500, 768])]; + tensor transpose_68 = transpose(perm = var_792, x = var_791_cast); + tensor x_83_cast = reshape(shape = concat_6, x = transpose_68); + tensor var_797_to_fp16 = const()[name = tensor("op_797_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(94822784)))]; + tensor var_798_to_fp16 = const()[name = tensor("op_798_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96002496)))]; + tensor var_799_cast = linear(bias = var_798_to_fp16, weight = var_797_to_fp16, x = x_83_cast); + tensor x_85_cast = add(x = x_79_cast, y = var_799_cast); + tensor var_805_axes_0 = const()[name = tensor("op_805_axes_0"), val = tensor([-1])]; + tensor blocks_6_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_6_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96004096)))]; + tensor blocks_6_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_6_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96005696)))]; + tensor var_805_cast = layer_norm(axes = var_805_axes_0, beta = blocks_6_mlp_ln_bias_to_fp16, epsilon = var_730_to_fp16, gamma = blocks_6_mlp_ln_weight_to_fp16, x = x_85_cast); + tensor var_814_to_fp16 = const()[name = tensor("op_814_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96007296)))]; + tensor var_815_to_fp16 = const()[name = tensor("op_815_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100725952)))]; + tensor input_57_cast = linear(bias = var_815_to_fp16, weight = var_814_to_fp16, x = var_805_cast); + tensor x_89_mode_0 = const()[name = tensor("x_89_mode_0"), val = tensor("EXACT")]; + tensor x_89_cast = gelu(mode = x_89_mode_0, x = input_57_cast); + tensor var_820_to_fp16 = const()[name = tensor("op_820_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100732160)))]; + tensor var_821_to_fp16 = const()[name = tensor("op_821_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105450816)))]; + tensor var_822_cast = linear(bias = var_821_to_fp16, weight = var_820_to_fp16, x = x_89_cast); + tensor x_91_cast = add(x = x_85_cast, y = var_822_cast); + tensor var_831 = const()[name = tensor("op_831"), val = tensor(-1)]; + tensor var_848_axes_0 = const()[name = tensor("op_848_axes_0"), val = tensor([-1])]; + tensor blocks_7_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_7_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105452416)))]; + tensor blocks_7_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_7_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105454016)))]; + tensor var_837_to_fp16 = const()[name = tensor("op_837_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_848_cast = layer_norm(axes = var_848_axes_0, beta = blocks_7_attn_ln_bias_to_fp16, epsilon = var_837_to_fp16, gamma = blocks_7_attn_ln_weight_to_fp16, x = x_91_cast); + tensor var_859_to_fp16 = const()[name = tensor("op_859_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105455616)))]; + tensor var_860_to_fp16 = const()[name = tensor("op_860_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106635328)))]; + tensor q_29_cast = linear(bias = var_860_to_fp16, weight = var_859_to_fp16, x = var_848_cast); + tensor var_863_to_fp16 = const()[name = tensor("op_863_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106636928)))]; + tensor k_29_bias_0_to_fp16 = const()[name = tensor("k_29_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107816640)))]; + tensor k_29_cast = linear(bias = k_29_bias_0_to_fp16, weight = var_863_to_fp16, x = var_848_cast); + tensor var_867_to_fp16 = const()[name = tensor("op_867_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107818240)))]; + tensor var_868_to_fp16 = const()[name = tensor("op_868_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108997952)))]; + tensor v_29_cast = linear(bias = var_868_to_fp16, weight = var_867_to_fp16, x = var_848_cast); + tensor var_876 = const()[name = tensor("op_876"), val = tensor([1, 1500, 12, -1])]; + tensor var_877_cast = reshape(shape = var_876, x = q_29_cast); + tensor const_98_to_fp16 = const()[name = tensor("const_98_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_31_cast = mul(x = var_877_cast, y = const_98_to_fp16); + tensor var_883 = const()[name = tensor("op_883"), val = tensor([1, 1500, 12, -1])]; + tensor var_884_cast = reshape(shape = var_883, x = k_29_cast); + tensor const_99_to_fp16 = const()[name = tensor("const_99_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_31_cast = mul(x = var_884_cast, y = const_99_to_fp16); + tensor var_890 = const()[name = tensor("op_890"), val = tensor([1, 1500, 12, -1])]; + tensor var_891_cast = reshape(shape = var_890, x = v_29_cast); + tensor var_892 = const()[name = tensor("op_892"), val = tensor([0, 2, 1, 3])]; + tensor qk_15_transpose_x_0 = const()[name = tensor("qk_15_transpose_x_0"), val = tensor(false)]; + tensor qk_15_transpose_y_0 = const()[name = tensor("qk_15_transpose_y_0"), val = tensor(false)]; + tensor transpose_38_perm_0 = const()[name = tensor("transpose_38_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_39_perm_0 = const()[name = tensor("transpose_39_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_65 = transpose(perm = transpose_39_perm_0, x = k_31_cast); + tensor transpose_66 = transpose(perm = transpose_38_perm_0, x = q_31_cast); + tensor qk_15_cast = matmul(transpose_x = qk_15_transpose_x_0, transpose_y = qk_15_transpose_y_0, x = transpose_66, y = transpose_65); + tensor var_896_cast = softmax(axis = var_831, x = qk_15_cast); + tensor var_898_transpose_x_0 = const()[name = tensor("op_898_transpose_x_0"), val = tensor(false)]; + tensor var_898_transpose_y_0 = const()[name = tensor("op_898_transpose_y_0"), val = tensor(false)]; + tensor transpose_67 = transpose(perm = var_892, x = var_891_cast); + tensor var_898_cast = matmul(transpose_x = var_898_transpose_x_0, transpose_y = var_898_transpose_y_0, x = var_896_cast, y = transpose_67); + tensor var_899 = const()[name = tensor("op_899"), val = tensor([0, 2, 1, 3])]; + tensor concat_7 = const()[name = tensor("concat_7"), val = tensor([1, 1500, 768])]; + tensor transpose_64 = transpose(perm = var_899, x = var_898_cast); + tensor x_95_cast = reshape(shape = concat_7, x = transpose_64); + tensor var_904_to_fp16 = const()[name = tensor("op_904_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108999552)))]; + tensor var_905_to_fp16 = const()[name = tensor("op_905_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110179264)))]; + tensor var_906_cast = linear(bias = var_905_to_fp16, weight = var_904_to_fp16, x = x_95_cast); + tensor x_97_cast = add(x = x_91_cast, y = var_906_cast); + tensor var_912_axes_0 = const()[name = tensor("op_912_axes_0"), val = tensor([-1])]; + tensor blocks_7_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_7_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110180864)))]; + tensor blocks_7_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_7_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110182464)))]; + tensor var_912_cast = layer_norm(axes = var_912_axes_0, beta = blocks_7_mlp_ln_bias_to_fp16, epsilon = var_837_to_fp16, gamma = blocks_7_mlp_ln_weight_to_fp16, x = x_97_cast); + tensor var_921_to_fp16 = const()[name = tensor("op_921_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110184064)))]; + tensor var_922_to_fp16 = const()[name = tensor("op_922_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114902720)))]; + tensor input_65_cast = linear(bias = var_922_to_fp16, weight = var_921_to_fp16, x = var_912_cast); + tensor x_101_mode_0 = const()[name = tensor("x_101_mode_0"), val = tensor("EXACT")]; + tensor x_101_cast = gelu(mode = x_101_mode_0, x = input_65_cast); + tensor var_927_to_fp16 = const()[name = tensor("op_927_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114908928)))]; + tensor var_928_to_fp16 = const()[name = tensor("op_928_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119627584)))]; + tensor var_929_cast = linear(bias = var_928_to_fp16, weight = var_927_to_fp16, x = x_101_cast); + tensor x_103_cast = add(x = x_97_cast, y = var_929_cast); + tensor var_938 = const()[name = tensor("op_938"), val = tensor(-1)]; + tensor var_955_axes_0 = const()[name = tensor("op_955_axes_0"), val = tensor([-1])]; + tensor blocks_8_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_8_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119629184)))]; + tensor blocks_8_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_8_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119630784)))]; + tensor var_944_to_fp16 = const()[name = tensor("op_944_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_955_cast = layer_norm(axes = var_955_axes_0, beta = blocks_8_attn_ln_bias_to_fp16, epsilon = var_944_to_fp16, gamma = blocks_8_attn_ln_weight_to_fp16, x = x_103_cast); + tensor var_966_to_fp16 = const()[name = tensor("op_966_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119632384)))]; + tensor var_967_to_fp16 = const()[name = tensor("op_967_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120812096)))]; + tensor q_33_cast = linear(bias = var_967_to_fp16, weight = var_966_to_fp16, x = var_955_cast); + tensor var_970_to_fp16 = const()[name = tensor("op_970_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120813696)))]; + tensor k_33_bias_0_to_fp16 = const()[name = tensor("k_33_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121993408)))]; + tensor k_33_cast = linear(bias = k_33_bias_0_to_fp16, weight = var_970_to_fp16, x = var_955_cast); + tensor var_974_to_fp16 = const()[name = tensor("op_974_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121995008)))]; + tensor var_975_to_fp16 = const()[name = tensor("op_975_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123174720)))]; + tensor v_33_cast = linear(bias = var_975_to_fp16, weight = var_974_to_fp16, x = var_955_cast); + tensor var_983 = const()[name = tensor("op_983"), val = tensor([1, 1500, 12, -1])]; + tensor var_984_cast = reshape(shape = var_983, x = q_33_cast); + tensor const_100_to_fp16 = const()[name = tensor("const_100_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_35_cast = mul(x = var_984_cast, y = const_100_to_fp16); + tensor var_990 = const()[name = tensor("op_990"), val = tensor([1, 1500, 12, -1])]; + tensor var_991_cast = reshape(shape = var_990, x = k_33_cast); + tensor const_101_to_fp16 = const()[name = tensor("const_101_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_35_cast = mul(x = var_991_cast, y = const_101_to_fp16); + tensor var_997 = const()[name = tensor("op_997"), val = tensor([1, 1500, 12, -1])]; + tensor var_998_cast = reshape(shape = var_997, x = v_33_cast); + tensor var_999 = const()[name = tensor("op_999"), val = tensor([0, 2, 1, 3])]; + tensor qk_17_transpose_x_0 = const()[name = tensor("qk_17_transpose_x_0"), val = tensor(false)]; + tensor qk_17_transpose_y_0 = const()[name = tensor("qk_17_transpose_y_0"), val = tensor(false)]; + tensor transpose_40_perm_0 = const()[name = tensor("transpose_40_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_41_perm_0 = const()[name = tensor("transpose_41_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_61 = transpose(perm = transpose_41_perm_0, x = k_35_cast); + tensor transpose_62 = transpose(perm = transpose_40_perm_0, x = q_35_cast); + tensor qk_17_cast = matmul(transpose_x = qk_17_transpose_x_0, transpose_y = qk_17_transpose_y_0, x = transpose_62, y = transpose_61); + tensor var_1003_cast = softmax(axis = var_938, x = qk_17_cast); + tensor var_1005_transpose_x_0 = const()[name = tensor("op_1005_transpose_x_0"), val = tensor(false)]; + tensor var_1005_transpose_y_0 = const()[name = tensor("op_1005_transpose_y_0"), val = tensor(false)]; + tensor transpose_63 = transpose(perm = var_999, x = var_998_cast); + tensor var_1005_cast = matmul(transpose_x = var_1005_transpose_x_0, transpose_y = var_1005_transpose_y_0, x = var_1003_cast, y = transpose_63); + tensor var_1006 = const()[name = tensor("op_1006"), val = tensor([0, 2, 1, 3])]; + tensor concat_8 = const()[name = tensor("concat_8"), val = tensor([1, 1500, 768])]; + tensor transpose_60 = transpose(perm = var_1006, x = var_1005_cast); + tensor x_107_cast = reshape(shape = concat_8, x = transpose_60); + tensor var_1011_to_fp16 = const()[name = tensor("op_1011_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123176320)))]; + tensor var_1012_to_fp16 = const()[name = tensor("op_1012_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124356032)))]; + tensor var_1013_cast = linear(bias = var_1012_to_fp16, weight = var_1011_to_fp16, x = x_107_cast); + tensor x_109_cast = add(x = x_103_cast, y = var_1013_cast); + tensor var_1019_axes_0 = const()[name = tensor("op_1019_axes_0"), val = tensor([-1])]; + tensor blocks_8_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_8_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124357632)))]; + tensor blocks_8_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_8_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124359232)))]; + tensor var_1019_cast = layer_norm(axes = var_1019_axes_0, beta = blocks_8_mlp_ln_bias_to_fp16, epsilon = var_944_to_fp16, gamma = blocks_8_mlp_ln_weight_to_fp16, x = x_109_cast); + tensor var_1028_to_fp16 = const()[name = tensor("op_1028_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124360832)))]; + tensor var_1029_to_fp16 = const()[name = tensor("op_1029_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129079488)))]; + tensor input_73_cast = linear(bias = var_1029_to_fp16, weight = var_1028_to_fp16, x = var_1019_cast); + tensor x_113_mode_0 = const()[name = tensor("x_113_mode_0"), val = tensor("EXACT")]; + tensor x_113_cast = gelu(mode = x_113_mode_0, x = input_73_cast); + tensor var_1034_to_fp16 = const()[name = tensor("op_1034_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129085696)))]; + tensor var_1035_to_fp16 = const()[name = tensor("op_1035_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133804352)))]; + tensor var_1036_cast = linear(bias = var_1035_to_fp16, weight = var_1034_to_fp16, x = x_113_cast); + tensor x_115_cast = add(x = x_109_cast, y = var_1036_cast); + tensor var_1045 = const()[name = tensor("op_1045"), val = tensor(-1)]; + tensor var_1062_axes_0 = const()[name = tensor("op_1062_axes_0"), val = tensor([-1])]; + tensor blocks_9_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_9_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133805952)))]; + tensor blocks_9_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_9_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133807552)))]; + tensor var_1051_to_fp16 = const()[name = tensor("op_1051_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1062_cast = layer_norm(axes = var_1062_axes_0, beta = blocks_9_attn_ln_bias_to_fp16, epsilon = var_1051_to_fp16, gamma = blocks_9_attn_ln_weight_to_fp16, x = x_115_cast); + tensor var_1073_to_fp16 = const()[name = tensor("op_1073_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133809152)))]; + tensor var_1074_to_fp16 = const()[name = tensor("op_1074_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134988864)))]; + tensor q_37_cast = linear(bias = var_1074_to_fp16, weight = var_1073_to_fp16, x = var_1062_cast); + tensor var_1077_to_fp16 = const()[name = tensor("op_1077_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134990464)))]; + tensor k_37_bias_0_to_fp16 = const()[name = tensor("k_37_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136170176)))]; + tensor k_37_cast = linear(bias = k_37_bias_0_to_fp16, weight = var_1077_to_fp16, x = var_1062_cast); + tensor var_1081_to_fp16 = const()[name = tensor("op_1081_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136171776)))]; + tensor var_1082_to_fp16 = const()[name = tensor("op_1082_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137351488)))]; + tensor v_37_cast = linear(bias = var_1082_to_fp16, weight = var_1081_to_fp16, x = var_1062_cast); + tensor var_1090 = const()[name = tensor("op_1090"), val = tensor([1, 1500, 12, -1])]; + tensor var_1091_cast = reshape(shape = var_1090, x = q_37_cast); + tensor const_102_to_fp16 = const()[name = tensor("const_102_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_39_cast = mul(x = var_1091_cast, y = const_102_to_fp16); + tensor var_1097 = const()[name = tensor("op_1097"), val = tensor([1, 1500, 12, -1])]; + tensor var_1098_cast = reshape(shape = var_1097, x = k_37_cast); + tensor const_103_to_fp16 = const()[name = tensor("const_103_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_39_cast = mul(x = var_1098_cast, y = const_103_to_fp16); + tensor var_1104 = const()[name = tensor("op_1104"), val = tensor([1, 1500, 12, -1])]; + tensor var_1105_cast = reshape(shape = var_1104, x = v_37_cast); + tensor var_1106 = const()[name = tensor("op_1106"), val = tensor([0, 2, 1, 3])]; + tensor qk_19_transpose_x_0 = const()[name = tensor("qk_19_transpose_x_0"), val = tensor(false)]; + tensor qk_19_transpose_y_0 = const()[name = tensor("qk_19_transpose_y_0"), val = tensor(false)]; + tensor transpose_42_perm_0 = const()[name = tensor("transpose_42_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_43_perm_0 = const()[name = tensor("transpose_43_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_57 = transpose(perm = transpose_43_perm_0, x = k_39_cast); + tensor transpose_58 = transpose(perm = transpose_42_perm_0, x = q_39_cast); + tensor qk_19_cast = matmul(transpose_x = qk_19_transpose_x_0, transpose_y = qk_19_transpose_y_0, x = transpose_58, y = transpose_57); + tensor var_1110_cast = softmax(axis = var_1045, x = qk_19_cast); + tensor var_1112_transpose_x_0 = const()[name = tensor("op_1112_transpose_x_0"), val = tensor(false)]; + tensor var_1112_transpose_y_0 = const()[name = tensor("op_1112_transpose_y_0"), val = tensor(false)]; + tensor transpose_59 = transpose(perm = var_1106, x = var_1105_cast); + tensor var_1112_cast = matmul(transpose_x = var_1112_transpose_x_0, transpose_y = var_1112_transpose_y_0, x = var_1110_cast, y = transpose_59); + tensor var_1113 = const()[name = tensor("op_1113"), val = tensor([0, 2, 1, 3])]; + tensor concat_9 = const()[name = tensor("concat_9"), val = tensor([1, 1500, 768])]; + tensor transpose_56 = transpose(perm = var_1113, x = var_1112_cast); + tensor x_119_cast = reshape(shape = concat_9, x = transpose_56); + tensor var_1118_to_fp16 = const()[name = tensor("op_1118_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137353088)))]; + tensor var_1119_to_fp16 = const()[name = tensor("op_1119_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138532800)))]; + tensor var_1120_cast = linear(bias = var_1119_to_fp16, weight = var_1118_to_fp16, x = x_119_cast); + tensor x_121_cast = add(x = x_115_cast, y = var_1120_cast); + tensor var_1126_axes_0 = const()[name = tensor("op_1126_axes_0"), val = tensor([-1])]; + tensor blocks_9_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_9_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138534400)))]; + tensor blocks_9_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_9_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138536000)))]; + tensor var_1126_cast = layer_norm(axes = var_1126_axes_0, beta = blocks_9_mlp_ln_bias_to_fp16, epsilon = var_1051_to_fp16, gamma = blocks_9_mlp_ln_weight_to_fp16, x = x_121_cast); + tensor var_1135_to_fp16 = const()[name = tensor("op_1135_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138537600)))]; + tensor var_1136_to_fp16 = const()[name = tensor("op_1136_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143256256)))]; + tensor input_81_cast = linear(bias = var_1136_to_fp16, weight = var_1135_to_fp16, x = var_1126_cast); + tensor x_125_mode_0 = const()[name = tensor("x_125_mode_0"), val = tensor("EXACT")]; + tensor x_125_cast = gelu(mode = x_125_mode_0, x = input_81_cast); + tensor var_1141_to_fp16 = const()[name = tensor("op_1141_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143262464)))]; + tensor var_1142_to_fp16 = const()[name = tensor("op_1142_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147981120)))]; + tensor var_1143_cast = linear(bias = var_1142_to_fp16, weight = var_1141_to_fp16, x = x_125_cast); + tensor x_127_cast = add(x = x_121_cast, y = var_1143_cast); + tensor var_1152 = const()[name = tensor("op_1152"), val = tensor(-1)]; + tensor var_1169_axes_0 = const()[name = tensor("op_1169_axes_0"), val = tensor([-1])]; + tensor blocks_10_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_10_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147982720)))]; + tensor blocks_10_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_10_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147984320)))]; + tensor var_1158_to_fp16 = const()[name = tensor("op_1158_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1169_cast = layer_norm(axes = var_1169_axes_0, beta = blocks_10_attn_ln_bias_to_fp16, epsilon = var_1158_to_fp16, gamma = blocks_10_attn_ln_weight_to_fp16, x = x_127_cast); + tensor var_1180_to_fp16 = const()[name = tensor("op_1180_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147985920)))]; + tensor var_1181_to_fp16 = const()[name = tensor("op_1181_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(149165632)))]; + tensor q_41_cast = linear(bias = var_1181_to_fp16, weight = var_1180_to_fp16, x = var_1169_cast); + tensor var_1184_to_fp16 = const()[name = tensor("op_1184_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(149167232)))]; + tensor k_41_bias_0_to_fp16 = const()[name = tensor("k_41_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(150346944)))]; + tensor k_41_cast = linear(bias = k_41_bias_0_to_fp16, weight = var_1184_to_fp16, x = var_1169_cast); + tensor var_1188_to_fp16 = const()[name = tensor("op_1188_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(150348544)))]; + tensor var_1189_to_fp16 = const()[name = tensor("op_1189_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151528256)))]; + tensor v_41_cast = linear(bias = var_1189_to_fp16, weight = var_1188_to_fp16, x = var_1169_cast); + tensor var_1197 = const()[name = tensor("op_1197"), val = tensor([1, 1500, 12, -1])]; + tensor var_1198_cast = reshape(shape = var_1197, x = q_41_cast); + tensor const_104_to_fp16 = const()[name = tensor("const_104_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_43_cast = mul(x = var_1198_cast, y = const_104_to_fp16); + tensor var_1204 = const()[name = tensor("op_1204"), val = tensor([1, 1500, 12, -1])]; + tensor var_1205_cast = reshape(shape = var_1204, x = k_41_cast); + tensor const_105_to_fp16 = const()[name = tensor("const_105_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_43_cast = mul(x = var_1205_cast, y = const_105_to_fp16); + tensor var_1211 = const()[name = tensor("op_1211"), val = tensor([1, 1500, 12, -1])]; + tensor var_1212_cast = reshape(shape = var_1211, x = v_41_cast); + tensor var_1213 = const()[name = tensor("op_1213"), val = tensor([0, 2, 1, 3])]; + tensor qk_21_transpose_x_0 = const()[name = tensor("qk_21_transpose_x_0"), val = tensor(false)]; + tensor qk_21_transpose_y_0 = const()[name = tensor("qk_21_transpose_y_0"), val = tensor(false)]; + tensor transpose_44_perm_0 = const()[name = tensor("transpose_44_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_45_perm_0 = const()[name = tensor("transpose_45_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_53 = transpose(perm = transpose_45_perm_0, x = k_43_cast); + tensor transpose_54 = transpose(perm = transpose_44_perm_0, x = q_43_cast); + tensor qk_21_cast = matmul(transpose_x = qk_21_transpose_x_0, transpose_y = qk_21_transpose_y_0, x = transpose_54, y = transpose_53); + tensor var_1217_cast = softmax(axis = var_1152, x = qk_21_cast); + tensor var_1219_transpose_x_0 = const()[name = tensor("op_1219_transpose_x_0"), val = tensor(false)]; + tensor var_1219_transpose_y_0 = const()[name = tensor("op_1219_transpose_y_0"), val = tensor(false)]; + tensor transpose_55 = transpose(perm = var_1213, x = var_1212_cast); + tensor var_1219_cast = matmul(transpose_x = var_1219_transpose_x_0, transpose_y = var_1219_transpose_y_0, x = var_1217_cast, y = transpose_55); + tensor var_1220 = const()[name = tensor("op_1220"), val = tensor([0, 2, 1, 3])]; + tensor concat_10 = const()[name = tensor("concat_10"), val = tensor([1, 1500, 768])]; + tensor transpose_52 = transpose(perm = var_1220, x = var_1219_cast); + tensor x_131_cast = reshape(shape = concat_10, x = transpose_52); + tensor var_1225_to_fp16 = const()[name = tensor("op_1225_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151529856)))]; + tensor var_1226_to_fp16 = const()[name = tensor("op_1226_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152709568)))]; + tensor var_1227_cast = linear(bias = var_1226_to_fp16, weight = var_1225_to_fp16, x = x_131_cast); + tensor x_133_cast = add(x = x_127_cast, y = var_1227_cast); + tensor var_1233_axes_0 = const()[name = tensor("op_1233_axes_0"), val = tensor([-1])]; + tensor blocks_10_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_10_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152711168)))]; + tensor blocks_10_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_10_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152712768)))]; + tensor var_1233_cast = layer_norm(axes = var_1233_axes_0, beta = blocks_10_mlp_ln_bias_to_fp16, epsilon = var_1158_to_fp16, gamma = blocks_10_mlp_ln_weight_to_fp16, x = x_133_cast); + tensor var_1242_to_fp16 = const()[name = tensor("op_1242_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152714368)))]; + tensor var_1243_to_fp16 = const()[name = tensor("op_1243_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(157433024)))]; + tensor input_89_cast = linear(bias = var_1243_to_fp16, weight = var_1242_to_fp16, x = var_1233_cast); + tensor x_137_mode_0 = const()[name = tensor("x_137_mode_0"), val = tensor("EXACT")]; + tensor x_137_cast = gelu(mode = x_137_mode_0, x = input_89_cast); + tensor var_1248_to_fp16 = const()[name = tensor("op_1248_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(157439232)))]; + tensor var_1249_to_fp16 = const()[name = tensor("op_1249_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162157888)))]; + tensor var_1250_cast = linear(bias = var_1249_to_fp16, weight = var_1248_to_fp16, x = x_137_cast); + tensor x_139_cast = add(x = x_133_cast, y = var_1250_cast); + tensor var_1259 = const()[name = tensor("op_1259"), val = tensor(-1)]; + tensor var_1276_axes_0 = const()[name = tensor("op_1276_axes_0"), val = tensor([-1])]; + tensor blocks_11_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_11_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162159488)))]; + tensor blocks_11_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_11_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162161088)))]; + tensor var_1265_to_fp16 = const()[name = tensor("op_1265_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1276_cast = layer_norm(axes = var_1276_axes_0, beta = blocks_11_attn_ln_bias_to_fp16, epsilon = var_1265_to_fp16, gamma = blocks_11_attn_ln_weight_to_fp16, x = x_139_cast); + tensor var_1287_to_fp16 = const()[name = tensor("op_1287_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162162688)))]; + tensor var_1288_to_fp16 = const()[name = tensor("op_1288_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163342400)))]; + tensor q_45_cast = linear(bias = var_1288_to_fp16, weight = var_1287_to_fp16, x = var_1276_cast); + tensor var_1291_to_fp16 = const()[name = tensor("op_1291_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163344000)))]; + tensor k_45_bias_0_to_fp16 = const()[name = tensor("k_45_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164523712)))]; + tensor k_45_cast = linear(bias = k_45_bias_0_to_fp16, weight = var_1291_to_fp16, x = var_1276_cast); + tensor var_1295_to_fp16 = const()[name = tensor("op_1295_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164525312)))]; + tensor var_1296_to_fp16 = const()[name = tensor("op_1296_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165705024)))]; + tensor v_45_cast = linear(bias = var_1296_to_fp16, weight = var_1295_to_fp16, x = var_1276_cast); + tensor var_1304 = const()[name = tensor("op_1304"), val = tensor([1, 1500, 12, -1])]; + tensor var_1305_cast = reshape(shape = var_1304, x = q_45_cast); + tensor const_106_to_fp16 = const()[name = tensor("const_106_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_cast = mul(x = var_1305_cast, y = const_106_to_fp16); + tensor var_1311 = const()[name = tensor("op_1311"), val = tensor([1, 1500, 12, -1])]; + tensor var_1312_cast = reshape(shape = var_1311, x = k_45_cast); + tensor const_107_to_fp16 = const()[name = tensor("const_107_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_cast = mul(x = var_1312_cast, y = const_107_to_fp16); + tensor var_1318 = const()[name = tensor("op_1318"), val = tensor([1, 1500, 12, -1])]; + tensor var_1319_cast = reshape(shape = var_1318, x = v_45_cast); + tensor var_1320 = const()[name = tensor("op_1320"), val = tensor([0, 2, 1, 3])]; + tensor qk_transpose_x_0 = const()[name = tensor("qk_transpose_x_0"), val = tensor(false)]; + tensor qk_transpose_y_0 = const()[name = tensor("qk_transpose_y_0"), val = tensor(false)]; + tensor transpose_46_perm_0 = const()[name = tensor("transpose_46_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_47_perm_0 = const()[name = tensor("transpose_47_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_49 = transpose(perm = transpose_47_perm_0, x = k_cast); + tensor transpose_50 = transpose(perm = transpose_46_perm_0, x = q_cast); + tensor qk_cast = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_50, y = transpose_49); + tensor var_1324_cast = softmax(axis = var_1259, x = qk_cast); + tensor var_1326_transpose_x_0 = const()[name = tensor("op_1326_transpose_x_0"), val = tensor(false)]; + tensor var_1326_transpose_y_0 = const()[name = tensor("op_1326_transpose_y_0"), val = tensor(false)]; + tensor transpose_51 = transpose(perm = var_1320, x = var_1319_cast); + tensor var_1326_cast = matmul(transpose_x = var_1326_transpose_x_0, transpose_y = var_1326_transpose_y_0, x = var_1324_cast, y = transpose_51); + tensor var_1327 = const()[name = tensor("op_1327"), val = tensor([0, 2, 1, 3])]; + tensor concat_11 = const()[name = tensor("concat_11"), val = tensor([1, 1500, 768])]; + tensor transpose_48 = transpose(perm = var_1327, x = var_1326_cast); + tensor x_143_cast = reshape(shape = concat_11, x = transpose_48); + tensor var_1332_to_fp16 = const()[name = tensor("op_1332_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165706624)))]; + tensor var_1333_to_fp16 = const()[name = tensor("op_1333_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166886336)))]; + tensor var_1334_cast = linear(bias = var_1333_to_fp16, weight = var_1332_to_fp16, x = x_143_cast); + tensor x_145_cast = add(x = x_139_cast, y = var_1334_cast); + tensor var_1340_axes_0 = const()[name = tensor("op_1340_axes_0"), val = tensor([-1])]; + tensor blocks_11_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_11_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166887936)))]; + tensor blocks_11_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_11_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166889536)))]; + tensor var_1340_cast = layer_norm(axes = var_1340_axes_0, beta = blocks_11_mlp_ln_bias_to_fp16, epsilon = var_1265_to_fp16, gamma = blocks_11_mlp_ln_weight_to_fp16, x = x_145_cast); + tensor var_1349_to_fp16 = const()[name = tensor("op_1349_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166891136)))]; + tensor var_1350_to_fp16 = const()[name = tensor("op_1350_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171609792)))]; + tensor input_97_cast = linear(bias = var_1350_to_fp16, weight = var_1349_to_fp16, x = var_1340_cast); + tensor x_149_mode_0 = const()[name = tensor("x_149_mode_0"), val = tensor("EXACT")]; + tensor x_149_cast = gelu(mode = x_149_mode_0, x = input_97_cast); + tensor var_1355_to_fp16 = const()[name = tensor("op_1355_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171616000)))]; + tensor var_1356_to_fp16 = const()[name = tensor("op_1356_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176334656)))]; + tensor var_1357_cast = linear(bias = var_1356_to_fp16, weight = var_1355_to_fp16, x = x_149_cast); + tensor x_cast = add(x = x_145_cast, y = var_1357_cast); + tensor var_1370_axes_0 = const()[name = tensor("op_1370_axes_0"), val = tensor([-1])]; + tensor ln_post_weight_to_fp16 = const()[name = tensor("ln_post_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176336256)))]; + tensor ln_post_bias_to_fp16 = const()[name = tensor("ln_post_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176337856)))]; + tensor var_1361_to_fp16 = const()[name = tensor("op_1361_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1370_cast = layer_norm(axes = var_1370_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_1361_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast); + tensor var_1370_cast_to_fp32_dtype_0 = const()[name = tensor("op_1370_cast_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor output = cast(dtype = var_1370_cast_to_fp32_dtype_0, x = var_1370_cast); + } -> (output); +} \ No newline at end of file diff --git a/ggml-small-encoder.mlmodelc/weights/weight.bin b/ggml-small-encoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..07e9e74a32beb311ae36f6e021b4a0cafb02b44c --- /dev/null +++ b/ggml-small-encoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:87eed4ae76f11a2d4a50786bc7423d4b45c2d0d9ca05577a3bd2557452072eaf +size 176339456 diff --git a/ggml-small.en-encoder.mlmodelc.zip b/ggml-small.en-encoder.mlmodelc.zip index 77bb2f2fd6a33134be5cbf4bdaeeeee07c364f2b..cffcfc748185d20c6fd4cde73e9341ac7e49d6a5 100644 --- a/ggml-small.en-encoder.mlmodelc.zip +++ b/ggml-small.en-encoder.mlmodelc.zip @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:b2ef1c506378b825b4b4341979a93e1656b5d6c129f17114cfb8fb78aabc2f89 +oid sha256:677fad005c458eba9a1d69a8e4a0bef1b129b208585ca90a4dd8ba4f8610eaed size 162952446 diff --git a/ggml-small.en-encoder.mlmodelc/analytics/coremldata.bin b/ggml-small.en-encoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..9af8eb5beab7ec5764639dbc0fead5c8eb711493 --- /dev/null +++ b/ggml-small.en-encoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:18ad2072ae82872c2ba8a187071e1e7d6c1105253685e7aa95138adcf07874e0 +size 207 diff --git a/ggml-small.en-encoder.mlmodelc/coremldata.bin b/ggml-small.en-encoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..c732ed7b8caeac3c15acda1c2836f3e1f525a5a8 --- /dev/null +++ b/ggml-small.en-encoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:05fe28591b40616fa0c34ad7b853133623f5300923ec812acb11459c411acf3b +size 149 diff --git a/ggml-small.en-encoder.mlmodelc/metadata.json b/ggml-small.en-encoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..19154636e803e3816e6f39e22ba5a864da93204e --- /dev/null +++ b/ggml-small.en-encoder.mlmodelc/metadata.json @@ -0,0 +1,64 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float32", + "formattedType" : "MultiArray (Float32)", + "shortDescription" : "", + "shape" : "[]", + "name" : "output", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 6, + "mlProgramOperationTypeHistogram" : { + "Linear" : 72, + "Matmul" : 24, + "Cast" : 2, + "Conv" : 2, + "Softmax" : 12, + "Add" : 25, + "LayerNorm" : 25, + "Mul" : 24, + "Transpose" : 49, + "Gelu" : 14, + "Reshape" : 48 + }, + "computePrecision" : "Mixed (Float16, Float32, Int32)", + "isUpdatable" : "0", + "availability" : { + "macOS" : "12.0", + "tvOS" : "15.0", + "watchOS" : "8.0", + "iOS" : "15.0", + "macCatalyst" : "15.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float32", + "formattedType" : "MultiArray (Float32 1 × 80 × 3000)", + "shortDescription" : "", + "shape" : "[1, 80, 3000]", + "name" : "logmel_data", + "type" : "MultiArray" + } + ], + "generatedClassName" : "coreml_encoder_small_en", + "method" : "predict" + } +] \ No newline at end of file diff --git a/ggml-small.en-encoder.mlmodelc/model.mil b/ggml-small.en-encoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..8149f70d09bf9f27147da1dfaed40fe475e486c2 --- /dev/null +++ b/ggml-small.en-encoder.mlmodelc/model.mil @@ -0,0 +1,747 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "4.28.4"}, {"coremlc-version", "1436.100.10"}})] +{ + func main(tensor logmel_data) { + tensor var_32 = const()[name = tensor("op_32"), val = tensor(1)]; + tensor var_40 = const()[name = tensor("op_40"), val = tensor([1])]; + tensor var_42 = const()[name = tensor("op_42"), val = tensor([1])]; + tensor var_44_pad_type_0 = const()[name = tensor("op_44_pad_type_0"), val = tensor("custom")]; + tensor var_44_pad_0 = const()[name = tensor("op_44_pad_0"), val = tensor([1, 1])]; + tensor logmel_data_to_fp16_dtype_0 = const()[name = tensor("logmel_data_to_fp16_dtype_0"), val = tensor("fp16")]; + tensor weight_3_to_fp16 = const()[name = tensor("weight_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor bias_3_to_fp16 = const()[name = tensor("bias_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(368768)))]; + tensor cast_367 = cast(dtype = logmel_data_to_fp16_dtype_0, x = logmel_data); + tensor var_44_cast = conv(bias = bias_3_to_fp16, dilations = var_42, groups = var_32, pad = var_44_pad_0, pad_type = var_44_pad_type_0, strides = var_40, weight = weight_3_to_fp16, x = cast_367); + tensor input_1_mode_0 = const()[name = tensor("input_1_mode_0"), val = tensor("EXACT")]; + tensor input_1_cast = gelu(mode = input_1_mode_0, x = var_44_cast); + tensor var_48 = const()[name = tensor("op_48"), val = tensor(1)]; + tensor var_57 = const()[name = tensor("op_57"), val = tensor([2])]; + tensor var_59 = const()[name = tensor("op_59"), val = tensor([1])]; + tensor var_61_pad_type_0 = const()[name = tensor("op_61_pad_type_0"), val = tensor("custom")]; + tensor var_61_pad_0 = const()[name = tensor("op_61_pad_0"), val = tensor([1, 1])]; + tensor weight_7_to_fp16 = const()[name = tensor("weight_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(370368)))]; + tensor bias_7_to_fp16 = const()[name = tensor("bias_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3909376)))]; + tensor var_61_cast = conv(bias = bias_7_to_fp16, dilations = var_59, groups = var_48, pad = var_61_pad_0, pad_type = var_61_pad_type_0, strides = var_57, weight = weight_7_to_fp16, x = input_1_cast); + tensor x_3_mode_0 = const()[name = tensor("x_3_mode_0"), val = tensor("EXACT")]; + tensor x_3_cast = gelu(mode = x_3_mode_0, x = var_61_cast); + tensor var_66 = const()[name = tensor("op_66"), val = tensor([0, 2, 1])]; + tensor positional_embedding_to_fp16 = const()[name = tensor("positional_embedding_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3910976)))]; + tensor transpose_96 = transpose(perm = var_66, x = x_3_cast); + tensor var_69_cast = add(x = transpose_96, y = positional_embedding_to_fp16); + tensor var_82 = const()[name = tensor("op_82"), val = tensor(-1)]; + tensor var_99_axes_0 = const()[name = tensor("op_99_axes_0"), val = tensor([-1])]; + tensor blocks_0_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_0_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6215040)))]; + tensor blocks_0_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_0_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6216640)))]; + tensor var_88_to_fp16 = const()[name = tensor("op_88_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_99_cast = layer_norm(axes = var_99_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_88_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_69_cast); + tensor var_110_to_fp16 = const()[name = tensor("op_110_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6218240)))]; + tensor var_111_to_fp16 = const()[name = tensor("op_111_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7397952)))]; + tensor q_1_cast = linear(bias = var_111_to_fp16, weight = var_110_to_fp16, x = var_99_cast); + tensor var_114_to_fp16 = const()[name = tensor("op_114_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7399552)))]; + tensor k_1_bias_0_to_fp16 = const()[name = tensor("k_1_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8579264)))]; + tensor k_1_cast = linear(bias = k_1_bias_0_to_fp16, weight = var_114_to_fp16, x = var_99_cast); + tensor var_118_to_fp16 = const()[name = tensor("op_118_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8580864)))]; + tensor var_119_to_fp16 = const()[name = tensor("op_119_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9760576)))]; + tensor v_1_cast = linear(bias = var_119_to_fp16, weight = var_118_to_fp16, x = var_99_cast); + tensor var_127 = const()[name = tensor("op_127"), val = tensor([1, 1500, 12, -1])]; + tensor var_128_cast = reshape(shape = var_127, x = q_1_cast); + tensor const_84_to_fp16 = const()[name = tensor("const_84_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_3_cast = mul(x = var_128_cast, y = const_84_to_fp16); + tensor var_134 = const()[name = tensor("op_134"), val = tensor([1, 1500, 12, -1])]; + tensor var_135_cast = reshape(shape = var_134, x = k_1_cast); + tensor const_85_to_fp16 = const()[name = tensor("const_85_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_3_cast = mul(x = var_135_cast, y = const_85_to_fp16); + tensor var_141 = const()[name = tensor("op_141"), val = tensor([1, 1500, 12, -1])]; + tensor var_142_cast = reshape(shape = var_141, x = v_1_cast); + tensor var_143 = const()[name = tensor("op_143"), val = tensor([0, 2, 1, 3])]; + tensor qk_1_transpose_x_0 = const()[name = tensor("qk_1_transpose_x_0"), val = tensor(false)]; + tensor qk_1_transpose_y_0 = const()[name = tensor("qk_1_transpose_y_0"), val = tensor(false)]; + tensor transpose_24_perm_0 = const()[name = tensor("transpose_24_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_25_perm_0 = const()[name = tensor("transpose_25_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_93 = transpose(perm = transpose_25_perm_0, x = k_3_cast); + tensor transpose_94 = transpose(perm = transpose_24_perm_0, x = q_3_cast); + tensor qk_1_cast = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_94, y = transpose_93); + tensor var_147_cast = softmax(axis = var_82, x = qk_1_cast); + tensor var_149_transpose_x_0 = const()[name = tensor("op_149_transpose_x_0"), val = tensor(false)]; + tensor var_149_transpose_y_0 = const()[name = tensor("op_149_transpose_y_0"), val = tensor(false)]; + tensor transpose_95 = transpose(perm = var_143, x = var_142_cast); + tensor var_149_cast = matmul(transpose_x = var_149_transpose_x_0, transpose_y = var_149_transpose_y_0, x = var_147_cast, y = transpose_95); + tensor var_150 = const()[name = tensor("op_150"), val = tensor([0, 2, 1, 3])]; + tensor concat_0 = const()[name = tensor("concat_0"), val = tensor([1, 1500, 768])]; + tensor transpose_92 = transpose(perm = var_150, x = var_149_cast); + tensor x_11_cast = reshape(shape = concat_0, x = transpose_92); + tensor var_155_to_fp16 = const()[name = tensor("op_155_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9762176)))]; + tensor var_156_to_fp16 = const()[name = tensor("op_156_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10941888)))]; + tensor var_157_cast = linear(bias = var_156_to_fp16, weight = var_155_to_fp16, x = x_11_cast); + tensor x_13_cast = add(x = var_69_cast, y = var_157_cast); + tensor var_163_axes_0 = const()[name = tensor("op_163_axes_0"), val = tensor([-1])]; + tensor blocks_0_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10943488)))]; + tensor blocks_0_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10945088)))]; + tensor var_163_cast = layer_norm(axes = var_163_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_88_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast); + tensor var_172_to_fp16 = const()[name = tensor("op_172_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10946688)))]; + tensor var_173_to_fp16 = const()[name = tensor("op_173_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15665344)))]; + tensor input_9_cast = linear(bias = var_173_to_fp16, weight = var_172_to_fp16, x = var_163_cast); + tensor x_17_mode_0 = const()[name = tensor("x_17_mode_0"), val = tensor("EXACT")]; + tensor x_17_cast = gelu(mode = x_17_mode_0, x = input_9_cast); + tensor var_178_to_fp16 = const()[name = tensor("op_178_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15671552)))]; + tensor var_179_to_fp16 = const()[name = tensor("op_179_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20390208)))]; + tensor var_180_cast = linear(bias = var_179_to_fp16, weight = var_178_to_fp16, x = x_17_cast); + tensor x_19_cast = add(x = x_13_cast, y = var_180_cast); + tensor var_189 = const()[name = tensor("op_189"), val = tensor(-1)]; + tensor var_206_axes_0 = const()[name = tensor("op_206_axes_0"), val = tensor([-1])]; + tensor blocks_1_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_1_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20391808)))]; + tensor blocks_1_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_1_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20393408)))]; + tensor var_195_to_fp16 = const()[name = tensor("op_195_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_206_cast = layer_norm(axes = var_206_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_195_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast); + tensor var_217_to_fp16 = const()[name = tensor("op_217_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20395008)))]; + tensor var_218_to_fp16 = const()[name = tensor("op_218_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21574720)))]; + tensor q_5_cast = linear(bias = var_218_to_fp16, weight = var_217_to_fp16, x = var_206_cast); + tensor var_221_to_fp16 = const()[name = tensor("op_221_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21576320)))]; + tensor k_5_bias_0_to_fp16 = const()[name = tensor("k_5_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22756032)))]; + tensor k_5_cast = linear(bias = k_5_bias_0_to_fp16, weight = var_221_to_fp16, x = var_206_cast); + tensor var_225_to_fp16 = const()[name = tensor("op_225_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22757632)))]; + tensor var_226_to_fp16 = const()[name = tensor("op_226_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23937344)))]; + tensor v_5_cast = linear(bias = var_226_to_fp16, weight = var_225_to_fp16, x = var_206_cast); + tensor var_234 = const()[name = tensor("op_234"), val = tensor([1, 1500, 12, -1])]; + tensor var_235_cast = reshape(shape = var_234, x = q_5_cast); + tensor const_86_to_fp16 = const()[name = tensor("const_86_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_7_cast = mul(x = var_235_cast, y = const_86_to_fp16); + tensor var_241 = const()[name = tensor("op_241"), val = tensor([1, 1500, 12, -1])]; + tensor var_242_cast = reshape(shape = var_241, x = k_5_cast); + tensor const_87_to_fp16 = const()[name = tensor("const_87_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_7_cast = mul(x = var_242_cast, y = const_87_to_fp16); + tensor var_248 = const()[name = tensor("op_248"), val = tensor([1, 1500, 12, -1])]; + tensor var_249_cast = reshape(shape = var_248, x = v_5_cast); + tensor var_250 = const()[name = tensor("op_250"), val = tensor([0, 2, 1, 3])]; + tensor qk_3_transpose_x_0 = const()[name = tensor("qk_3_transpose_x_0"), val = tensor(false)]; + tensor qk_3_transpose_y_0 = const()[name = tensor("qk_3_transpose_y_0"), val = tensor(false)]; + tensor transpose_26_perm_0 = const()[name = tensor("transpose_26_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_27_perm_0 = const()[name = tensor("transpose_27_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_89 = transpose(perm = transpose_27_perm_0, x = k_7_cast); + tensor transpose_90 = transpose(perm = transpose_26_perm_0, x = q_7_cast); + tensor qk_3_cast = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_90, y = transpose_89); + tensor var_254_cast = softmax(axis = var_189, x = qk_3_cast); + tensor var_256_transpose_x_0 = const()[name = tensor("op_256_transpose_x_0"), val = tensor(false)]; + tensor var_256_transpose_y_0 = const()[name = tensor("op_256_transpose_y_0"), val = tensor(false)]; + tensor transpose_91 = transpose(perm = var_250, x = var_249_cast); + tensor var_256_cast = matmul(transpose_x = var_256_transpose_x_0, transpose_y = var_256_transpose_y_0, x = var_254_cast, y = transpose_91); + tensor var_257 = const()[name = tensor("op_257"), val = tensor([0, 2, 1, 3])]; + tensor concat_1 = const()[name = tensor("concat_1"), val = tensor([1, 1500, 768])]; + tensor transpose_88 = transpose(perm = var_257, x = var_256_cast); + tensor x_23_cast = reshape(shape = concat_1, x = transpose_88); + tensor var_262_to_fp16 = const()[name = tensor("op_262_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23938944)))]; + tensor var_263_to_fp16 = const()[name = tensor("op_263_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25118656)))]; + tensor var_264_cast = linear(bias = var_263_to_fp16, weight = var_262_to_fp16, x = x_23_cast); + tensor x_25_cast = add(x = x_19_cast, y = var_264_cast); + tensor var_270_axes_0 = const()[name = tensor("op_270_axes_0"), val = tensor([-1])]; + tensor blocks_1_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25120256)))]; + tensor blocks_1_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25121856)))]; + tensor var_270_cast = layer_norm(axes = var_270_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_195_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast); + tensor var_279_to_fp16 = const()[name = tensor("op_279_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25123456)))]; + tensor var_280_to_fp16 = const()[name = tensor("op_280_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29842112)))]; + tensor input_17_cast = linear(bias = var_280_to_fp16, weight = var_279_to_fp16, x = var_270_cast); + tensor x_29_mode_0 = const()[name = tensor("x_29_mode_0"), val = tensor("EXACT")]; + tensor x_29_cast = gelu(mode = x_29_mode_0, x = input_17_cast); + tensor var_285_to_fp16 = const()[name = tensor("op_285_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29848320)))]; + tensor var_286_to_fp16 = const()[name = tensor("op_286_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34566976)))]; + tensor var_287_cast = linear(bias = var_286_to_fp16, weight = var_285_to_fp16, x = x_29_cast); + tensor x_31_cast = add(x = x_25_cast, y = var_287_cast); + tensor var_296 = const()[name = tensor("op_296"), val = tensor(-1)]; + tensor var_313_axes_0 = const()[name = tensor("op_313_axes_0"), val = tensor([-1])]; + tensor blocks_2_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_2_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34568576)))]; + tensor blocks_2_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_2_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34570176)))]; + tensor var_302_to_fp16 = const()[name = tensor("op_302_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_313_cast = layer_norm(axes = var_313_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_302_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast); + tensor var_324_to_fp16 = const()[name = tensor("op_324_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34571776)))]; + tensor var_325_to_fp16 = const()[name = tensor("op_325_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35751488)))]; + tensor q_9_cast = linear(bias = var_325_to_fp16, weight = var_324_to_fp16, x = var_313_cast); + tensor var_328_to_fp16 = const()[name = tensor("op_328_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35753088)))]; + tensor k_9_bias_0_to_fp16 = const()[name = tensor("k_9_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36932800)))]; + tensor k_9_cast = linear(bias = k_9_bias_0_to_fp16, weight = var_328_to_fp16, x = var_313_cast); + tensor var_332_to_fp16 = const()[name = tensor("op_332_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36934400)))]; + tensor var_333_to_fp16 = const()[name = tensor("op_333_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38114112)))]; + tensor v_9_cast = linear(bias = var_333_to_fp16, weight = var_332_to_fp16, x = var_313_cast); + tensor var_341 = const()[name = tensor("op_341"), val = tensor([1, 1500, 12, -1])]; + tensor var_342_cast = reshape(shape = var_341, x = q_9_cast); + tensor const_88_to_fp16 = const()[name = tensor("const_88_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_11_cast = mul(x = var_342_cast, y = const_88_to_fp16); + tensor var_348 = const()[name = tensor("op_348"), val = tensor([1, 1500, 12, -1])]; + tensor var_349_cast = reshape(shape = var_348, x = k_9_cast); + tensor const_89_to_fp16 = const()[name = tensor("const_89_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_11_cast = mul(x = var_349_cast, y = const_89_to_fp16); + tensor var_355 = const()[name = tensor("op_355"), val = tensor([1, 1500, 12, -1])]; + tensor var_356_cast = reshape(shape = var_355, x = v_9_cast); + tensor var_357 = const()[name = tensor("op_357"), val = tensor([0, 2, 1, 3])]; + tensor qk_5_transpose_x_0 = const()[name = tensor("qk_5_transpose_x_0"), val = tensor(false)]; + tensor qk_5_transpose_y_0 = const()[name = tensor("qk_5_transpose_y_0"), val = tensor(false)]; + tensor transpose_28_perm_0 = const()[name = tensor("transpose_28_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_29_perm_0 = const()[name = tensor("transpose_29_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_85 = transpose(perm = transpose_29_perm_0, x = k_11_cast); + tensor transpose_86 = transpose(perm = transpose_28_perm_0, x = q_11_cast); + tensor qk_5_cast = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_86, y = transpose_85); + tensor var_361_cast = softmax(axis = var_296, x = qk_5_cast); + tensor var_363_transpose_x_0 = const()[name = tensor("op_363_transpose_x_0"), val = tensor(false)]; + tensor var_363_transpose_y_0 = const()[name = tensor("op_363_transpose_y_0"), val = tensor(false)]; + tensor transpose_87 = transpose(perm = var_357, x = var_356_cast); + tensor var_363_cast = matmul(transpose_x = var_363_transpose_x_0, transpose_y = var_363_transpose_y_0, x = var_361_cast, y = transpose_87); + tensor var_364 = const()[name = tensor("op_364"), val = tensor([0, 2, 1, 3])]; + tensor concat_2 = const()[name = tensor("concat_2"), val = tensor([1, 1500, 768])]; + tensor transpose_84 = transpose(perm = var_364, x = var_363_cast); + tensor x_35_cast = reshape(shape = concat_2, x = transpose_84); + tensor var_369_to_fp16 = const()[name = tensor("op_369_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38115712)))]; + tensor var_370_to_fp16 = const()[name = tensor("op_370_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39295424)))]; + tensor var_371_cast = linear(bias = var_370_to_fp16, weight = var_369_to_fp16, x = x_35_cast); + tensor x_37_cast = add(x = x_31_cast, y = var_371_cast); + tensor var_377_axes_0 = const()[name = tensor("op_377_axes_0"), val = tensor([-1])]; + tensor blocks_2_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39297024)))]; + tensor blocks_2_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39298624)))]; + tensor var_377_cast = layer_norm(axes = var_377_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_302_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast); + tensor var_386_to_fp16 = const()[name = tensor("op_386_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39300224)))]; + tensor var_387_to_fp16 = const()[name = tensor("op_387_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44018880)))]; + tensor input_25_cast = linear(bias = var_387_to_fp16, weight = var_386_to_fp16, x = var_377_cast); + tensor x_41_mode_0 = const()[name = tensor("x_41_mode_0"), val = tensor("EXACT")]; + tensor x_41_cast = gelu(mode = x_41_mode_0, x = input_25_cast); + tensor var_392_to_fp16 = const()[name = tensor("op_392_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44025088)))]; + tensor var_393_to_fp16 = const()[name = tensor("op_393_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48743744)))]; + tensor var_394_cast = linear(bias = var_393_to_fp16, weight = var_392_to_fp16, x = x_41_cast); + tensor x_43_cast = add(x = x_37_cast, y = var_394_cast); + tensor var_403 = const()[name = tensor("op_403"), val = tensor(-1)]; + tensor var_420_axes_0 = const()[name = tensor("op_420_axes_0"), val = tensor([-1])]; + tensor blocks_3_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_3_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48745344)))]; + tensor blocks_3_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_3_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48746944)))]; + tensor var_409_to_fp16 = const()[name = tensor("op_409_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_420_cast = layer_norm(axes = var_420_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_409_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast); + tensor var_431_to_fp16 = const()[name = tensor("op_431_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48748544)))]; + tensor var_432_to_fp16 = const()[name = tensor("op_432_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49928256)))]; + tensor q_13_cast = linear(bias = var_432_to_fp16, weight = var_431_to_fp16, x = var_420_cast); + tensor var_435_to_fp16 = const()[name = tensor("op_435_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49929856)))]; + tensor k_13_bias_0_to_fp16 = const()[name = tensor("k_13_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51109568)))]; + tensor k_13_cast = linear(bias = k_13_bias_0_to_fp16, weight = var_435_to_fp16, x = var_420_cast); + tensor var_439_to_fp16 = const()[name = tensor("op_439_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51111168)))]; + tensor var_440_to_fp16 = const()[name = tensor("op_440_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(52290880)))]; + tensor v_13_cast = linear(bias = var_440_to_fp16, weight = var_439_to_fp16, x = var_420_cast); + tensor var_448 = const()[name = tensor("op_448"), val = tensor([1, 1500, 12, -1])]; + tensor var_449_cast = reshape(shape = var_448, x = q_13_cast); + tensor const_90_to_fp16 = const()[name = tensor("const_90_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_15_cast = mul(x = var_449_cast, y = const_90_to_fp16); + tensor var_455 = const()[name = tensor("op_455"), val = tensor([1, 1500, 12, -1])]; + tensor var_456_cast = reshape(shape = var_455, x = k_13_cast); + tensor const_91_to_fp16 = const()[name = tensor("const_91_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_15_cast = mul(x = var_456_cast, y = const_91_to_fp16); + tensor var_462 = const()[name = tensor("op_462"), val = tensor([1, 1500, 12, -1])]; + tensor var_463_cast = reshape(shape = var_462, x = v_13_cast); + tensor var_464 = const()[name = tensor("op_464"), val = tensor([0, 2, 1, 3])]; + tensor qk_7_transpose_x_0 = const()[name = tensor("qk_7_transpose_x_0"), val = tensor(false)]; + tensor qk_7_transpose_y_0 = const()[name = tensor("qk_7_transpose_y_0"), val = tensor(false)]; + tensor transpose_30_perm_0 = const()[name = tensor("transpose_30_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_31_perm_0 = const()[name = tensor("transpose_31_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_81 = transpose(perm = transpose_31_perm_0, x = k_15_cast); + tensor transpose_82 = transpose(perm = transpose_30_perm_0, x = q_15_cast); + tensor qk_7_cast = matmul(transpose_x = qk_7_transpose_x_0, transpose_y = qk_7_transpose_y_0, x = transpose_82, y = transpose_81); + tensor var_468_cast = softmax(axis = var_403, x = qk_7_cast); + tensor var_470_transpose_x_0 = const()[name = tensor("op_470_transpose_x_0"), val = tensor(false)]; + tensor var_470_transpose_y_0 = const()[name = tensor("op_470_transpose_y_0"), val = tensor(false)]; + tensor transpose_83 = transpose(perm = var_464, x = var_463_cast); + tensor var_470_cast = matmul(transpose_x = var_470_transpose_x_0, transpose_y = var_470_transpose_y_0, x = var_468_cast, y = transpose_83); + tensor var_471 = const()[name = tensor("op_471"), val = tensor([0, 2, 1, 3])]; + tensor concat_3 = const()[name = tensor("concat_3"), val = tensor([1, 1500, 768])]; + tensor transpose_80 = transpose(perm = var_471, x = var_470_cast); + tensor x_47_cast = reshape(shape = concat_3, x = transpose_80); + tensor var_476_to_fp16 = const()[name = tensor("op_476_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(52292480)))]; + tensor var_477_to_fp16 = const()[name = tensor("op_477_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53472192)))]; + tensor var_478_cast = linear(bias = var_477_to_fp16, weight = var_476_to_fp16, x = x_47_cast); + tensor x_49_cast = add(x = x_43_cast, y = var_478_cast); + tensor var_484_axes_0 = const()[name = tensor("op_484_axes_0"), val = tensor([-1])]; + tensor blocks_3_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53473792)))]; + tensor blocks_3_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53475392)))]; + tensor var_484_cast = layer_norm(axes = var_484_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_409_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast); + tensor var_493_to_fp16 = const()[name = tensor("op_493_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53476992)))]; + tensor var_494_to_fp16 = const()[name = tensor("op_494_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(58195648)))]; + tensor input_33_cast = linear(bias = var_494_to_fp16, weight = var_493_to_fp16, x = var_484_cast); + tensor x_53_mode_0 = const()[name = tensor("x_53_mode_0"), val = tensor("EXACT")]; + tensor x_53_cast = gelu(mode = x_53_mode_0, x = input_33_cast); + tensor var_499_to_fp16 = const()[name = tensor("op_499_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(58201856)))]; + tensor var_500_to_fp16 = const()[name = tensor("op_500_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62920512)))]; + tensor var_501_cast = linear(bias = var_500_to_fp16, weight = var_499_to_fp16, x = x_53_cast); + tensor x_55_cast = add(x = x_49_cast, y = var_501_cast); + tensor var_510 = const()[name = tensor("op_510"), val = tensor(-1)]; + tensor var_527_axes_0 = const()[name = tensor("op_527_axes_0"), val = tensor([-1])]; + tensor blocks_4_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_4_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62922112)))]; + tensor blocks_4_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_4_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62923712)))]; + tensor var_516_to_fp16 = const()[name = tensor("op_516_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_527_cast = layer_norm(axes = var_527_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_516_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast); + tensor var_538_to_fp16 = const()[name = tensor("op_538_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62925312)))]; + tensor var_539_to_fp16 = const()[name = tensor("op_539_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64105024)))]; + tensor q_17_cast = linear(bias = var_539_to_fp16, weight = var_538_to_fp16, x = var_527_cast); + tensor var_542_to_fp16 = const()[name = tensor("op_542_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64106624)))]; + tensor k_17_bias_0_to_fp16 = const()[name = tensor("k_17_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65286336)))]; + tensor k_17_cast = linear(bias = k_17_bias_0_to_fp16, weight = var_542_to_fp16, x = var_527_cast); + tensor var_546_to_fp16 = const()[name = tensor("op_546_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65287936)))]; + tensor var_547_to_fp16 = const()[name = tensor("op_547_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66467648)))]; + tensor v_17_cast = linear(bias = var_547_to_fp16, weight = var_546_to_fp16, x = var_527_cast); + tensor var_555 = const()[name = tensor("op_555"), val = tensor([1, 1500, 12, -1])]; + tensor var_556_cast = reshape(shape = var_555, x = q_17_cast); + tensor const_92_to_fp16 = const()[name = tensor("const_92_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_19_cast = mul(x = var_556_cast, y = const_92_to_fp16); + tensor var_562 = const()[name = tensor("op_562"), val = tensor([1, 1500, 12, -1])]; + tensor var_563_cast = reshape(shape = var_562, x = k_17_cast); + tensor const_93_to_fp16 = const()[name = tensor("const_93_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_19_cast = mul(x = var_563_cast, y = const_93_to_fp16); + tensor var_569 = const()[name = tensor("op_569"), val = tensor([1, 1500, 12, -1])]; + tensor var_570_cast = reshape(shape = var_569, x = v_17_cast); + tensor var_571 = const()[name = tensor("op_571"), val = tensor([0, 2, 1, 3])]; + tensor qk_9_transpose_x_0 = const()[name = tensor("qk_9_transpose_x_0"), val = tensor(false)]; + tensor qk_9_transpose_y_0 = const()[name = tensor("qk_9_transpose_y_0"), val = tensor(false)]; + tensor transpose_32_perm_0 = const()[name = tensor("transpose_32_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_33_perm_0 = const()[name = tensor("transpose_33_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_77 = transpose(perm = transpose_33_perm_0, x = k_19_cast); + tensor transpose_78 = transpose(perm = transpose_32_perm_0, x = q_19_cast); + tensor qk_9_cast = matmul(transpose_x = qk_9_transpose_x_0, transpose_y = qk_9_transpose_y_0, x = transpose_78, y = transpose_77); + tensor var_575_cast = softmax(axis = var_510, x = qk_9_cast); + tensor var_577_transpose_x_0 = const()[name = tensor("op_577_transpose_x_0"), val = tensor(false)]; + tensor var_577_transpose_y_0 = const()[name = tensor("op_577_transpose_y_0"), val = tensor(false)]; + tensor transpose_79 = transpose(perm = var_571, x = var_570_cast); + tensor var_577_cast = matmul(transpose_x = var_577_transpose_x_0, transpose_y = var_577_transpose_y_0, x = var_575_cast, y = transpose_79); + tensor var_578 = const()[name = tensor("op_578"), val = tensor([0, 2, 1, 3])]; + tensor concat_4 = const()[name = tensor("concat_4"), val = tensor([1, 1500, 768])]; + tensor transpose_76 = transpose(perm = var_578, x = var_577_cast); + tensor x_59_cast = reshape(shape = concat_4, x = transpose_76); + tensor var_583_to_fp16 = const()[name = tensor("op_583_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66469248)))]; + tensor var_584_to_fp16 = const()[name = tensor("op_584_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67648960)))]; + tensor var_585_cast = linear(bias = var_584_to_fp16, weight = var_583_to_fp16, x = x_59_cast); + tensor x_61_cast = add(x = x_55_cast, y = var_585_cast); + tensor var_591_axes_0 = const()[name = tensor("op_591_axes_0"), val = tensor([-1])]; + tensor blocks_4_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67650560)))]; + tensor blocks_4_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_4_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67652160)))]; + tensor var_591_cast = layer_norm(axes = var_591_axes_0, beta = blocks_4_mlp_ln_bias_to_fp16, epsilon = var_516_to_fp16, gamma = blocks_4_mlp_ln_weight_to_fp16, x = x_61_cast); + tensor var_600_to_fp16 = const()[name = tensor("op_600_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67653760)))]; + tensor var_601_to_fp16 = const()[name = tensor("op_601_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(72372416)))]; + tensor input_41_cast = linear(bias = var_601_to_fp16, weight = var_600_to_fp16, x = var_591_cast); + tensor x_65_mode_0 = const()[name = tensor("x_65_mode_0"), val = tensor("EXACT")]; + tensor x_65_cast = gelu(mode = x_65_mode_0, x = input_41_cast); + tensor var_606_to_fp16 = const()[name = tensor("op_606_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(72378624)))]; + tensor var_607_to_fp16 = const()[name = tensor("op_607_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77097280)))]; + tensor var_608_cast = linear(bias = var_607_to_fp16, weight = var_606_to_fp16, x = x_65_cast); + tensor x_67_cast = add(x = x_61_cast, y = var_608_cast); + tensor var_617 = const()[name = tensor("op_617"), val = tensor(-1)]; + tensor var_634_axes_0 = const()[name = tensor("op_634_axes_0"), val = tensor([-1])]; + tensor blocks_5_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_5_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77098880)))]; + tensor blocks_5_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_5_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77100480)))]; + tensor var_623_to_fp16 = const()[name = tensor("op_623_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_634_cast = layer_norm(axes = var_634_axes_0, beta = blocks_5_attn_ln_bias_to_fp16, epsilon = var_623_to_fp16, gamma = blocks_5_attn_ln_weight_to_fp16, x = x_67_cast); + tensor var_645_to_fp16 = const()[name = tensor("op_645_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77102080)))]; + tensor var_646_to_fp16 = const()[name = tensor("op_646_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78281792)))]; + tensor q_21_cast = linear(bias = var_646_to_fp16, weight = var_645_to_fp16, x = var_634_cast); + tensor var_649_to_fp16 = const()[name = tensor("op_649_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78283392)))]; + tensor k_21_bias_0_to_fp16 = const()[name = tensor("k_21_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79463104)))]; + tensor k_21_cast = linear(bias = k_21_bias_0_to_fp16, weight = var_649_to_fp16, x = var_634_cast); + tensor var_653_to_fp16 = const()[name = tensor("op_653_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79464704)))]; + tensor var_654_to_fp16 = const()[name = tensor("op_654_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80644416)))]; + tensor v_21_cast = linear(bias = var_654_to_fp16, weight = var_653_to_fp16, x = var_634_cast); + tensor var_662 = const()[name = tensor("op_662"), val = tensor([1, 1500, 12, -1])]; + tensor var_663_cast = reshape(shape = var_662, x = q_21_cast); + tensor const_94_to_fp16 = const()[name = tensor("const_94_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_23_cast = mul(x = var_663_cast, y = const_94_to_fp16); + tensor var_669 = const()[name = tensor("op_669"), val = tensor([1, 1500, 12, -1])]; + tensor var_670_cast = reshape(shape = var_669, x = k_21_cast); + tensor const_95_to_fp16 = const()[name = tensor("const_95_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_23_cast = mul(x = var_670_cast, y = const_95_to_fp16); + tensor var_676 = const()[name = tensor("op_676"), val = tensor([1, 1500, 12, -1])]; + tensor var_677_cast = reshape(shape = var_676, x = v_21_cast); + tensor var_678 = const()[name = tensor("op_678"), val = tensor([0, 2, 1, 3])]; + tensor qk_11_transpose_x_0 = const()[name = tensor("qk_11_transpose_x_0"), val = tensor(false)]; + tensor qk_11_transpose_y_0 = const()[name = tensor("qk_11_transpose_y_0"), val = tensor(false)]; + tensor transpose_34_perm_0 = const()[name = tensor("transpose_34_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_35_perm_0 = const()[name = tensor("transpose_35_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_73 = transpose(perm = transpose_35_perm_0, x = k_23_cast); + tensor transpose_74 = transpose(perm = transpose_34_perm_0, x = q_23_cast); + tensor qk_11_cast = matmul(transpose_x = qk_11_transpose_x_0, transpose_y = qk_11_transpose_y_0, x = transpose_74, y = transpose_73); + tensor var_682_cast = softmax(axis = var_617, x = qk_11_cast); + tensor var_684_transpose_x_0 = const()[name = tensor("op_684_transpose_x_0"), val = tensor(false)]; + tensor var_684_transpose_y_0 = const()[name = tensor("op_684_transpose_y_0"), val = tensor(false)]; + tensor transpose_75 = transpose(perm = var_678, x = var_677_cast); + tensor var_684_cast = matmul(transpose_x = var_684_transpose_x_0, transpose_y = var_684_transpose_y_0, x = var_682_cast, y = transpose_75); + tensor var_685 = const()[name = tensor("op_685"), val = tensor([0, 2, 1, 3])]; + tensor concat_5 = const()[name = tensor("concat_5"), val = tensor([1, 1500, 768])]; + tensor transpose_72 = transpose(perm = var_685, x = var_684_cast); + tensor x_71_cast = reshape(shape = concat_5, x = transpose_72); + tensor var_690_to_fp16 = const()[name = tensor("op_690_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80646016)))]; + tensor var_691_to_fp16 = const()[name = tensor("op_691_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81825728)))]; + tensor var_692_cast = linear(bias = var_691_to_fp16, weight = var_690_to_fp16, x = x_71_cast); + tensor x_73_cast = add(x = x_67_cast, y = var_692_cast); + tensor var_698_axes_0 = const()[name = tensor("op_698_axes_0"), val = tensor([-1])]; + tensor blocks_5_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81827328)))]; + tensor blocks_5_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_5_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81828928)))]; + tensor var_698_cast = layer_norm(axes = var_698_axes_0, beta = blocks_5_mlp_ln_bias_to_fp16, epsilon = var_623_to_fp16, gamma = blocks_5_mlp_ln_weight_to_fp16, x = x_73_cast); + tensor var_707_to_fp16 = const()[name = tensor("op_707_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81830528)))]; + tensor var_708_to_fp16 = const()[name = tensor("op_708_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86549184)))]; + tensor input_49_cast = linear(bias = var_708_to_fp16, weight = var_707_to_fp16, x = var_698_cast); + tensor x_77_mode_0 = const()[name = tensor("x_77_mode_0"), val = tensor("EXACT")]; + tensor x_77_cast = gelu(mode = x_77_mode_0, x = input_49_cast); + tensor var_713_to_fp16 = const()[name = tensor("op_713_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86555392)))]; + tensor var_714_to_fp16 = const()[name = tensor("op_714_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91274048)))]; + tensor var_715_cast = linear(bias = var_714_to_fp16, weight = var_713_to_fp16, x = x_77_cast); + tensor x_79_cast = add(x = x_73_cast, y = var_715_cast); + tensor var_724 = const()[name = tensor("op_724"), val = tensor(-1)]; + tensor var_741_axes_0 = const()[name = tensor("op_741_axes_0"), val = tensor([-1])]; + tensor blocks_6_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_6_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91275648)))]; + tensor blocks_6_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_6_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91277248)))]; + tensor var_730_to_fp16 = const()[name = tensor("op_730_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_741_cast = layer_norm(axes = var_741_axes_0, beta = blocks_6_attn_ln_bias_to_fp16, epsilon = var_730_to_fp16, gamma = blocks_6_attn_ln_weight_to_fp16, x = x_79_cast); + tensor var_752_to_fp16 = const()[name = tensor("op_752_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91278848)))]; + tensor var_753_to_fp16 = const()[name = tensor("op_753_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92458560)))]; + tensor q_25_cast = linear(bias = var_753_to_fp16, weight = var_752_to_fp16, x = var_741_cast); + tensor var_756_to_fp16 = const()[name = tensor("op_756_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92460160)))]; + tensor k_25_bias_0_to_fp16 = const()[name = tensor("k_25_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93639872)))]; + tensor k_25_cast = linear(bias = k_25_bias_0_to_fp16, weight = var_756_to_fp16, x = var_741_cast); + tensor var_760_to_fp16 = const()[name = tensor("op_760_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93641472)))]; + tensor var_761_to_fp16 = const()[name = tensor("op_761_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(94821184)))]; + tensor v_25_cast = linear(bias = var_761_to_fp16, weight = var_760_to_fp16, x = var_741_cast); + tensor var_769 = const()[name = tensor("op_769"), val = tensor([1, 1500, 12, -1])]; + tensor var_770_cast = reshape(shape = var_769, x = q_25_cast); + tensor const_96_to_fp16 = const()[name = tensor("const_96_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_27_cast = mul(x = var_770_cast, y = const_96_to_fp16); + tensor var_776 = const()[name = tensor("op_776"), val = tensor([1, 1500, 12, -1])]; + tensor var_777_cast = reshape(shape = var_776, x = k_25_cast); + tensor const_97_to_fp16 = const()[name = tensor("const_97_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_27_cast = mul(x = var_777_cast, y = const_97_to_fp16); + tensor var_783 = const()[name = tensor("op_783"), val = tensor([1, 1500, 12, -1])]; + tensor var_784_cast = reshape(shape = var_783, x = v_25_cast); + tensor var_785 = const()[name = tensor("op_785"), val = tensor([0, 2, 1, 3])]; + tensor qk_13_transpose_x_0 = const()[name = tensor("qk_13_transpose_x_0"), val = tensor(false)]; + tensor qk_13_transpose_y_0 = const()[name = tensor("qk_13_transpose_y_0"), val = tensor(false)]; + tensor transpose_36_perm_0 = const()[name = tensor("transpose_36_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_37_perm_0 = const()[name = tensor("transpose_37_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_69 = transpose(perm = transpose_37_perm_0, x = k_27_cast); + tensor transpose_70 = transpose(perm = transpose_36_perm_0, x = q_27_cast); + tensor qk_13_cast = matmul(transpose_x = qk_13_transpose_x_0, transpose_y = qk_13_transpose_y_0, x = transpose_70, y = transpose_69); + tensor var_789_cast = softmax(axis = var_724, x = qk_13_cast); + tensor var_791_transpose_x_0 = const()[name = tensor("op_791_transpose_x_0"), val = tensor(false)]; + tensor var_791_transpose_y_0 = const()[name = tensor("op_791_transpose_y_0"), val = tensor(false)]; + tensor transpose_71 = transpose(perm = var_785, x = var_784_cast); + tensor var_791_cast = matmul(transpose_x = var_791_transpose_x_0, transpose_y = var_791_transpose_y_0, x = var_789_cast, y = transpose_71); + tensor var_792 = const()[name = tensor("op_792"), val = tensor([0, 2, 1, 3])]; + tensor concat_6 = const()[name = tensor("concat_6"), val = tensor([1, 1500, 768])]; + tensor transpose_68 = transpose(perm = var_792, x = var_791_cast); + tensor x_83_cast = reshape(shape = concat_6, x = transpose_68); + tensor var_797_to_fp16 = const()[name = tensor("op_797_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(94822784)))]; + tensor var_798_to_fp16 = const()[name = tensor("op_798_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96002496)))]; + tensor var_799_cast = linear(bias = var_798_to_fp16, weight = var_797_to_fp16, x = x_83_cast); + tensor x_85_cast = add(x = x_79_cast, y = var_799_cast); + tensor var_805_axes_0 = const()[name = tensor("op_805_axes_0"), val = tensor([-1])]; + tensor blocks_6_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_6_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96004096)))]; + tensor blocks_6_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_6_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96005696)))]; + tensor var_805_cast = layer_norm(axes = var_805_axes_0, beta = blocks_6_mlp_ln_bias_to_fp16, epsilon = var_730_to_fp16, gamma = blocks_6_mlp_ln_weight_to_fp16, x = x_85_cast); + tensor var_814_to_fp16 = const()[name = tensor("op_814_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96007296)))]; + tensor var_815_to_fp16 = const()[name = tensor("op_815_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100725952)))]; + tensor input_57_cast = linear(bias = var_815_to_fp16, weight = var_814_to_fp16, x = var_805_cast); + tensor x_89_mode_0 = const()[name = tensor("x_89_mode_0"), val = tensor("EXACT")]; + tensor x_89_cast = gelu(mode = x_89_mode_0, x = input_57_cast); + tensor var_820_to_fp16 = const()[name = tensor("op_820_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100732160)))]; + tensor var_821_to_fp16 = const()[name = tensor("op_821_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105450816)))]; + tensor var_822_cast = linear(bias = var_821_to_fp16, weight = var_820_to_fp16, x = x_89_cast); + tensor x_91_cast = add(x = x_85_cast, y = var_822_cast); + tensor var_831 = const()[name = tensor("op_831"), val = tensor(-1)]; + tensor var_848_axes_0 = const()[name = tensor("op_848_axes_0"), val = tensor([-1])]; + tensor blocks_7_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_7_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105452416)))]; + tensor blocks_7_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_7_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105454016)))]; + tensor var_837_to_fp16 = const()[name = tensor("op_837_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_848_cast = layer_norm(axes = var_848_axes_0, beta = blocks_7_attn_ln_bias_to_fp16, epsilon = var_837_to_fp16, gamma = blocks_7_attn_ln_weight_to_fp16, x = x_91_cast); + tensor var_859_to_fp16 = const()[name = tensor("op_859_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(105455616)))]; + tensor var_860_to_fp16 = const()[name = tensor("op_860_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106635328)))]; + tensor q_29_cast = linear(bias = var_860_to_fp16, weight = var_859_to_fp16, x = var_848_cast); + tensor var_863_to_fp16 = const()[name = tensor("op_863_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(106636928)))]; + tensor k_29_bias_0_to_fp16 = const()[name = tensor("k_29_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107816640)))]; + tensor k_29_cast = linear(bias = k_29_bias_0_to_fp16, weight = var_863_to_fp16, x = var_848_cast); + tensor var_867_to_fp16 = const()[name = tensor("op_867_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(107818240)))]; + tensor var_868_to_fp16 = const()[name = tensor("op_868_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108997952)))]; + tensor v_29_cast = linear(bias = var_868_to_fp16, weight = var_867_to_fp16, x = var_848_cast); + tensor var_876 = const()[name = tensor("op_876"), val = tensor([1, 1500, 12, -1])]; + tensor var_877_cast = reshape(shape = var_876, x = q_29_cast); + tensor const_98_to_fp16 = const()[name = tensor("const_98_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_31_cast = mul(x = var_877_cast, y = const_98_to_fp16); + tensor var_883 = const()[name = tensor("op_883"), val = tensor([1, 1500, 12, -1])]; + tensor var_884_cast = reshape(shape = var_883, x = k_29_cast); + tensor const_99_to_fp16 = const()[name = tensor("const_99_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_31_cast = mul(x = var_884_cast, y = const_99_to_fp16); + tensor var_890 = const()[name = tensor("op_890"), val = tensor([1, 1500, 12, -1])]; + tensor var_891_cast = reshape(shape = var_890, x = v_29_cast); + tensor var_892 = const()[name = tensor("op_892"), val = tensor([0, 2, 1, 3])]; + tensor qk_15_transpose_x_0 = const()[name = tensor("qk_15_transpose_x_0"), val = tensor(false)]; + tensor qk_15_transpose_y_0 = const()[name = tensor("qk_15_transpose_y_0"), val = tensor(false)]; + tensor transpose_38_perm_0 = const()[name = tensor("transpose_38_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_39_perm_0 = const()[name = tensor("transpose_39_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_65 = transpose(perm = transpose_39_perm_0, x = k_31_cast); + tensor transpose_66 = transpose(perm = transpose_38_perm_0, x = q_31_cast); + tensor qk_15_cast = matmul(transpose_x = qk_15_transpose_x_0, transpose_y = qk_15_transpose_y_0, x = transpose_66, y = transpose_65); + tensor var_896_cast = softmax(axis = var_831, x = qk_15_cast); + tensor var_898_transpose_x_0 = const()[name = tensor("op_898_transpose_x_0"), val = tensor(false)]; + tensor var_898_transpose_y_0 = const()[name = tensor("op_898_transpose_y_0"), val = tensor(false)]; + tensor transpose_67 = transpose(perm = var_892, x = var_891_cast); + tensor var_898_cast = matmul(transpose_x = var_898_transpose_x_0, transpose_y = var_898_transpose_y_0, x = var_896_cast, y = transpose_67); + tensor var_899 = const()[name = tensor("op_899"), val = tensor([0, 2, 1, 3])]; + tensor concat_7 = const()[name = tensor("concat_7"), val = tensor([1, 1500, 768])]; + tensor transpose_64 = transpose(perm = var_899, x = var_898_cast); + tensor x_95_cast = reshape(shape = concat_7, x = transpose_64); + tensor var_904_to_fp16 = const()[name = tensor("op_904_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(108999552)))]; + tensor var_905_to_fp16 = const()[name = tensor("op_905_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110179264)))]; + tensor var_906_cast = linear(bias = var_905_to_fp16, weight = var_904_to_fp16, x = x_95_cast); + tensor x_97_cast = add(x = x_91_cast, y = var_906_cast); + tensor var_912_axes_0 = const()[name = tensor("op_912_axes_0"), val = tensor([-1])]; + tensor blocks_7_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_7_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110180864)))]; + tensor blocks_7_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_7_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110182464)))]; + tensor var_912_cast = layer_norm(axes = var_912_axes_0, beta = blocks_7_mlp_ln_bias_to_fp16, epsilon = var_837_to_fp16, gamma = blocks_7_mlp_ln_weight_to_fp16, x = x_97_cast); + tensor var_921_to_fp16 = const()[name = tensor("op_921_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(110184064)))]; + tensor var_922_to_fp16 = const()[name = tensor("op_922_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114902720)))]; + tensor input_65_cast = linear(bias = var_922_to_fp16, weight = var_921_to_fp16, x = var_912_cast); + tensor x_101_mode_0 = const()[name = tensor("x_101_mode_0"), val = tensor("EXACT")]; + tensor x_101_cast = gelu(mode = x_101_mode_0, x = input_65_cast); + tensor var_927_to_fp16 = const()[name = tensor("op_927_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(114908928)))]; + tensor var_928_to_fp16 = const()[name = tensor("op_928_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119627584)))]; + tensor var_929_cast = linear(bias = var_928_to_fp16, weight = var_927_to_fp16, x = x_101_cast); + tensor x_103_cast = add(x = x_97_cast, y = var_929_cast); + tensor var_938 = const()[name = tensor("op_938"), val = tensor(-1)]; + tensor var_955_axes_0 = const()[name = tensor("op_955_axes_0"), val = tensor([-1])]; + tensor blocks_8_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_8_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119629184)))]; + tensor blocks_8_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_8_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119630784)))]; + tensor var_944_to_fp16 = const()[name = tensor("op_944_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_955_cast = layer_norm(axes = var_955_axes_0, beta = blocks_8_attn_ln_bias_to_fp16, epsilon = var_944_to_fp16, gamma = blocks_8_attn_ln_weight_to_fp16, x = x_103_cast); + tensor var_966_to_fp16 = const()[name = tensor("op_966_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(119632384)))]; + tensor var_967_to_fp16 = const()[name = tensor("op_967_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120812096)))]; + tensor q_33_cast = linear(bias = var_967_to_fp16, weight = var_966_to_fp16, x = var_955_cast); + tensor var_970_to_fp16 = const()[name = tensor("op_970_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(120813696)))]; + tensor k_33_bias_0_to_fp16 = const()[name = tensor("k_33_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121993408)))]; + tensor k_33_cast = linear(bias = k_33_bias_0_to_fp16, weight = var_970_to_fp16, x = var_955_cast); + tensor var_974_to_fp16 = const()[name = tensor("op_974_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(121995008)))]; + tensor var_975_to_fp16 = const()[name = tensor("op_975_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123174720)))]; + tensor v_33_cast = linear(bias = var_975_to_fp16, weight = var_974_to_fp16, x = var_955_cast); + tensor var_983 = const()[name = tensor("op_983"), val = tensor([1, 1500, 12, -1])]; + tensor var_984_cast = reshape(shape = var_983, x = q_33_cast); + tensor const_100_to_fp16 = const()[name = tensor("const_100_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_35_cast = mul(x = var_984_cast, y = const_100_to_fp16); + tensor var_990 = const()[name = tensor("op_990"), val = tensor([1, 1500, 12, -1])]; + tensor var_991_cast = reshape(shape = var_990, x = k_33_cast); + tensor const_101_to_fp16 = const()[name = tensor("const_101_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_35_cast = mul(x = var_991_cast, y = const_101_to_fp16); + tensor var_997 = const()[name = tensor("op_997"), val = tensor([1, 1500, 12, -1])]; + tensor var_998_cast = reshape(shape = var_997, x = v_33_cast); + tensor var_999 = const()[name = tensor("op_999"), val = tensor([0, 2, 1, 3])]; + tensor qk_17_transpose_x_0 = const()[name = tensor("qk_17_transpose_x_0"), val = tensor(false)]; + tensor qk_17_transpose_y_0 = const()[name = tensor("qk_17_transpose_y_0"), val = tensor(false)]; + tensor transpose_40_perm_0 = const()[name = tensor("transpose_40_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_41_perm_0 = const()[name = tensor("transpose_41_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_61 = transpose(perm = transpose_41_perm_0, x = k_35_cast); + tensor transpose_62 = transpose(perm = transpose_40_perm_0, x = q_35_cast); + tensor qk_17_cast = matmul(transpose_x = qk_17_transpose_x_0, transpose_y = qk_17_transpose_y_0, x = transpose_62, y = transpose_61); + tensor var_1003_cast = softmax(axis = var_938, x = qk_17_cast); + tensor var_1005_transpose_x_0 = const()[name = tensor("op_1005_transpose_x_0"), val = tensor(false)]; + tensor var_1005_transpose_y_0 = const()[name = tensor("op_1005_transpose_y_0"), val = tensor(false)]; + tensor transpose_63 = transpose(perm = var_999, x = var_998_cast); + tensor var_1005_cast = matmul(transpose_x = var_1005_transpose_x_0, transpose_y = var_1005_transpose_y_0, x = var_1003_cast, y = transpose_63); + tensor var_1006 = const()[name = tensor("op_1006"), val = tensor([0, 2, 1, 3])]; + tensor concat_8 = const()[name = tensor("concat_8"), val = tensor([1, 1500, 768])]; + tensor transpose_60 = transpose(perm = var_1006, x = var_1005_cast); + tensor x_107_cast = reshape(shape = concat_8, x = transpose_60); + tensor var_1011_to_fp16 = const()[name = tensor("op_1011_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(123176320)))]; + tensor var_1012_to_fp16 = const()[name = tensor("op_1012_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124356032)))]; + tensor var_1013_cast = linear(bias = var_1012_to_fp16, weight = var_1011_to_fp16, x = x_107_cast); + tensor x_109_cast = add(x = x_103_cast, y = var_1013_cast); + tensor var_1019_axes_0 = const()[name = tensor("op_1019_axes_0"), val = tensor([-1])]; + tensor blocks_8_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_8_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124357632)))]; + tensor blocks_8_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_8_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124359232)))]; + tensor var_1019_cast = layer_norm(axes = var_1019_axes_0, beta = blocks_8_mlp_ln_bias_to_fp16, epsilon = var_944_to_fp16, gamma = blocks_8_mlp_ln_weight_to_fp16, x = x_109_cast); + tensor var_1028_to_fp16 = const()[name = tensor("op_1028_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(124360832)))]; + tensor var_1029_to_fp16 = const()[name = tensor("op_1029_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129079488)))]; + tensor input_73_cast = linear(bias = var_1029_to_fp16, weight = var_1028_to_fp16, x = var_1019_cast); + tensor x_113_mode_0 = const()[name = tensor("x_113_mode_0"), val = tensor("EXACT")]; + tensor x_113_cast = gelu(mode = x_113_mode_0, x = input_73_cast); + tensor var_1034_to_fp16 = const()[name = tensor("op_1034_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(129085696)))]; + tensor var_1035_to_fp16 = const()[name = tensor("op_1035_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133804352)))]; + tensor var_1036_cast = linear(bias = var_1035_to_fp16, weight = var_1034_to_fp16, x = x_113_cast); + tensor x_115_cast = add(x = x_109_cast, y = var_1036_cast); + tensor var_1045 = const()[name = tensor("op_1045"), val = tensor(-1)]; + tensor var_1062_axes_0 = const()[name = tensor("op_1062_axes_0"), val = tensor([-1])]; + tensor blocks_9_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_9_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133805952)))]; + tensor blocks_9_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_9_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133807552)))]; + tensor var_1051_to_fp16 = const()[name = tensor("op_1051_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1062_cast = layer_norm(axes = var_1062_axes_0, beta = blocks_9_attn_ln_bias_to_fp16, epsilon = var_1051_to_fp16, gamma = blocks_9_attn_ln_weight_to_fp16, x = x_115_cast); + tensor var_1073_to_fp16 = const()[name = tensor("op_1073_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(133809152)))]; + tensor var_1074_to_fp16 = const()[name = tensor("op_1074_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134988864)))]; + tensor q_37_cast = linear(bias = var_1074_to_fp16, weight = var_1073_to_fp16, x = var_1062_cast); + tensor var_1077_to_fp16 = const()[name = tensor("op_1077_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(134990464)))]; + tensor k_37_bias_0_to_fp16 = const()[name = tensor("k_37_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136170176)))]; + tensor k_37_cast = linear(bias = k_37_bias_0_to_fp16, weight = var_1077_to_fp16, x = var_1062_cast); + tensor var_1081_to_fp16 = const()[name = tensor("op_1081_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(136171776)))]; + tensor var_1082_to_fp16 = const()[name = tensor("op_1082_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137351488)))]; + tensor v_37_cast = linear(bias = var_1082_to_fp16, weight = var_1081_to_fp16, x = var_1062_cast); + tensor var_1090 = const()[name = tensor("op_1090"), val = tensor([1, 1500, 12, -1])]; + tensor var_1091_cast = reshape(shape = var_1090, x = q_37_cast); + tensor const_102_to_fp16 = const()[name = tensor("const_102_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_39_cast = mul(x = var_1091_cast, y = const_102_to_fp16); + tensor var_1097 = const()[name = tensor("op_1097"), val = tensor([1, 1500, 12, -1])]; + tensor var_1098_cast = reshape(shape = var_1097, x = k_37_cast); + tensor const_103_to_fp16 = const()[name = tensor("const_103_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_39_cast = mul(x = var_1098_cast, y = const_103_to_fp16); + tensor var_1104 = const()[name = tensor("op_1104"), val = tensor([1, 1500, 12, -1])]; + tensor var_1105_cast = reshape(shape = var_1104, x = v_37_cast); + tensor var_1106 = const()[name = tensor("op_1106"), val = tensor([0, 2, 1, 3])]; + tensor qk_19_transpose_x_0 = const()[name = tensor("qk_19_transpose_x_0"), val = tensor(false)]; + tensor qk_19_transpose_y_0 = const()[name = tensor("qk_19_transpose_y_0"), val = tensor(false)]; + tensor transpose_42_perm_0 = const()[name = tensor("transpose_42_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_43_perm_0 = const()[name = tensor("transpose_43_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_57 = transpose(perm = transpose_43_perm_0, x = k_39_cast); + tensor transpose_58 = transpose(perm = transpose_42_perm_0, x = q_39_cast); + tensor qk_19_cast = matmul(transpose_x = qk_19_transpose_x_0, transpose_y = qk_19_transpose_y_0, x = transpose_58, y = transpose_57); + tensor var_1110_cast = softmax(axis = var_1045, x = qk_19_cast); + tensor var_1112_transpose_x_0 = const()[name = tensor("op_1112_transpose_x_0"), val = tensor(false)]; + tensor var_1112_transpose_y_0 = const()[name = tensor("op_1112_transpose_y_0"), val = tensor(false)]; + tensor transpose_59 = transpose(perm = var_1106, x = var_1105_cast); + tensor var_1112_cast = matmul(transpose_x = var_1112_transpose_x_0, transpose_y = var_1112_transpose_y_0, x = var_1110_cast, y = transpose_59); + tensor var_1113 = const()[name = tensor("op_1113"), val = tensor([0, 2, 1, 3])]; + tensor concat_9 = const()[name = tensor("concat_9"), val = tensor([1, 1500, 768])]; + tensor transpose_56 = transpose(perm = var_1113, x = var_1112_cast); + tensor x_119_cast = reshape(shape = concat_9, x = transpose_56); + tensor var_1118_to_fp16 = const()[name = tensor("op_1118_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(137353088)))]; + tensor var_1119_to_fp16 = const()[name = tensor("op_1119_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138532800)))]; + tensor var_1120_cast = linear(bias = var_1119_to_fp16, weight = var_1118_to_fp16, x = x_119_cast); + tensor x_121_cast = add(x = x_115_cast, y = var_1120_cast); + tensor var_1126_axes_0 = const()[name = tensor("op_1126_axes_0"), val = tensor([-1])]; + tensor blocks_9_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_9_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138534400)))]; + tensor blocks_9_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_9_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138536000)))]; + tensor var_1126_cast = layer_norm(axes = var_1126_axes_0, beta = blocks_9_mlp_ln_bias_to_fp16, epsilon = var_1051_to_fp16, gamma = blocks_9_mlp_ln_weight_to_fp16, x = x_121_cast); + tensor var_1135_to_fp16 = const()[name = tensor("op_1135_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(138537600)))]; + tensor var_1136_to_fp16 = const()[name = tensor("op_1136_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143256256)))]; + tensor input_81_cast = linear(bias = var_1136_to_fp16, weight = var_1135_to_fp16, x = var_1126_cast); + tensor x_125_mode_0 = const()[name = tensor("x_125_mode_0"), val = tensor("EXACT")]; + tensor x_125_cast = gelu(mode = x_125_mode_0, x = input_81_cast); + tensor var_1141_to_fp16 = const()[name = tensor("op_1141_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(143262464)))]; + tensor var_1142_to_fp16 = const()[name = tensor("op_1142_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147981120)))]; + tensor var_1143_cast = linear(bias = var_1142_to_fp16, weight = var_1141_to_fp16, x = x_125_cast); + tensor x_127_cast = add(x = x_121_cast, y = var_1143_cast); + tensor var_1152 = const()[name = tensor("op_1152"), val = tensor(-1)]; + tensor var_1169_axes_0 = const()[name = tensor("op_1169_axes_0"), val = tensor([-1])]; + tensor blocks_10_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_10_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147982720)))]; + tensor blocks_10_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_10_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147984320)))]; + tensor var_1158_to_fp16 = const()[name = tensor("op_1158_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1169_cast = layer_norm(axes = var_1169_axes_0, beta = blocks_10_attn_ln_bias_to_fp16, epsilon = var_1158_to_fp16, gamma = blocks_10_attn_ln_weight_to_fp16, x = x_127_cast); + tensor var_1180_to_fp16 = const()[name = tensor("op_1180_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(147985920)))]; + tensor var_1181_to_fp16 = const()[name = tensor("op_1181_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(149165632)))]; + tensor q_41_cast = linear(bias = var_1181_to_fp16, weight = var_1180_to_fp16, x = var_1169_cast); + tensor var_1184_to_fp16 = const()[name = tensor("op_1184_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(149167232)))]; + tensor k_41_bias_0_to_fp16 = const()[name = tensor("k_41_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(150346944)))]; + tensor k_41_cast = linear(bias = k_41_bias_0_to_fp16, weight = var_1184_to_fp16, x = var_1169_cast); + tensor var_1188_to_fp16 = const()[name = tensor("op_1188_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(150348544)))]; + tensor var_1189_to_fp16 = const()[name = tensor("op_1189_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151528256)))]; + tensor v_41_cast = linear(bias = var_1189_to_fp16, weight = var_1188_to_fp16, x = var_1169_cast); + tensor var_1197 = const()[name = tensor("op_1197"), val = tensor([1, 1500, 12, -1])]; + tensor var_1198_cast = reshape(shape = var_1197, x = q_41_cast); + tensor const_104_to_fp16 = const()[name = tensor("const_104_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_43_cast = mul(x = var_1198_cast, y = const_104_to_fp16); + tensor var_1204 = const()[name = tensor("op_1204"), val = tensor([1, 1500, 12, -1])]; + tensor var_1205_cast = reshape(shape = var_1204, x = k_41_cast); + tensor const_105_to_fp16 = const()[name = tensor("const_105_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_43_cast = mul(x = var_1205_cast, y = const_105_to_fp16); + tensor var_1211 = const()[name = tensor("op_1211"), val = tensor([1, 1500, 12, -1])]; + tensor var_1212_cast = reshape(shape = var_1211, x = v_41_cast); + tensor var_1213 = const()[name = tensor("op_1213"), val = tensor([0, 2, 1, 3])]; + tensor qk_21_transpose_x_0 = const()[name = tensor("qk_21_transpose_x_0"), val = tensor(false)]; + tensor qk_21_transpose_y_0 = const()[name = tensor("qk_21_transpose_y_0"), val = tensor(false)]; + tensor transpose_44_perm_0 = const()[name = tensor("transpose_44_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_45_perm_0 = const()[name = tensor("transpose_45_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_53 = transpose(perm = transpose_45_perm_0, x = k_43_cast); + tensor transpose_54 = transpose(perm = transpose_44_perm_0, x = q_43_cast); + tensor qk_21_cast = matmul(transpose_x = qk_21_transpose_x_0, transpose_y = qk_21_transpose_y_0, x = transpose_54, y = transpose_53); + tensor var_1217_cast = softmax(axis = var_1152, x = qk_21_cast); + tensor var_1219_transpose_x_0 = const()[name = tensor("op_1219_transpose_x_0"), val = tensor(false)]; + tensor var_1219_transpose_y_0 = const()[name = tensor("op_1219_transpose_y_0"), val = tensor(false)]; + tensor transpose_55 = transpose(perm = var_1213, x = var_1212_cast); + tensor var_1219_cast = matmul(transpose_x = var_1219_transpose_x_0, transpose_y = var_1219_transpose_y_0, x = var_1217_cast, y = transpose_55); + tensor var_1220 = const()[name = tensor("op_1220"), val = tensor([0, 2, 1, 3])]; + tensor concat_10 = const()[name = tensor("concat_10"), val = tensor([1, 1500, 768])]; + tensor transpose_52 = transpose(perm = var_1220, x = var_1219_cast); + tensor x_131_cast = reshape(shape = concat_10, x = transpose_52); + tensor var_1225_to_fp16 = const()[name = tensor("op_1225_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(151529856)))]; + tensor var_1226_to_fp16 = const()[name = tensor("op_1226_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152709568)))]; + tensor var_1227_cast = linear(bias = var_1226_to_fp16, weight = var_1225_to_fp16, x = x_131_cast); + tensor x_133_cast = add(x = x_127_cast, y = var_1227_cast); + tensor var_1233_axes_0 = const()[name = tensor("op_1233_axes_0"), val = tensor([-1])]; + tensor blocks_10_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_10_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152711168)))]; + tensor blocks_10_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_10_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152712768)))]; + tensor var_1233_cast = layer_norm(axes = var_1233_axes_0, beta = blocks_10_mlp_ln_bias_to_fp16, epsilon = var_1158_to_fp16, gamma = blocks_10_mlp_ln_weight_to_fp16, x = x_133_cast); + tensor var_1242_to_fp16 = const()[name = tensor("op_1242_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(152714368)))]; + tensor var_1243_to_fp16 = const()[name = tensor("op_1243_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(157433024)))]; + tensor input_89_cast = linear(bias = var_1243_to_fp16, weight = var_1242_to_fp16, x = var_1233_cast); + tensor x_137_mode_0 = const()[name = tensor("x_137_mode_0"), val = tensor("EXACT")]; + tensor x_137_cast = gelu(mode = x_137_mode_0, x = input_89_cast); + tensor var_1248_to_fp16 = const()[name = tensor("op_1248_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(157439232)))]; + tensor var_1249_to_fp16 = const()[name = tensor("op_1249_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162157888)))]; + tensor var_1250_cast = linear(bias = var_1249_to_fp16, weight = var_1248_to_fp16, x = x_137_cast); + tensor x_139_cast = add(x = x_133_cast, y = var_1250_cast); + tensor var_1259 = const()[name = tensor("op_1259"), val = tensor(-1)]; + tensor var_1276_axes_0 = const()[name = tensor("op_1276_axes_0"), val = tensor([-1])]; + tensor blocks_11_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_11_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162159488)))]; + tensor blocks_11_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_11_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162161088)))]; + tensor var_1265_to_fp16 = const()[name = tensor("op_1265_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1276_cast = layer_norm(axes = var_1276_axes_0, beta = blocks_11_attn_ln_bias_to_fp16, epsilon = var_1265_to_fp16, gamma = blocks_11_attn_ln_weight_to_fp16, x = x_139_cast); + tensor var_1287_to_fp16 = const()[name = tensor("op_1287_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(162162688)))]; + tensor var_1288_to_fp16 = const()[name = tensor("op_1288_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163342400)))]; + tensor q_45_cast = linear(bias = var_1288_to_fp16, weight = var_1287_to_fp16, x = var_1276_cast); + tensor var_1291_to_fp16 = const()[name = tensor("op_1291_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(163344000)))]; + tensor k_45_bias_0_to_fp16 = const()[name = tensor("k_45_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164523712)))]; + tensor k_45_cast = linear(bias = k_45_bias_0_to_fp16, weight = var_1291_to_fp16, x = var_1276_cast); + tensor var_1295_to_fp16 = const()[name = tensor("op_1295_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(164525312)))]; + tensor var_1296_to_fp16 = const()[name = tensor("op_1296_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165705024)))]; + tensor v_45_cast = linear(bias = var_1296_to_fp16, weight = var_1295_to_fp16, x = var_1276_cast); + tensor var_1304 = const()[name = tensor("op_1304"), val = tensor([1, 1500, 12, -1])]; + tensor var_1305_cast = reshape(shape = var_1304, x = q_45_cast); + tensor const_106_to_fp16 = const()[name = tensor("const_106_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_cast = mul(x = var_1305_cast, y = const_106_to_fp16); + tensor var_1311 = const()[name = tensor("op_1311"), val = tensor([1, 1500, 12, -1])]; + tensor var_1312_cast = reshape(shape = var_1311, x = k_45_cast); + tensor const_107_to_fp16 = const()[name = tensor("const_107_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_cast = mul(x = var_1312_cast, y = const_107_to_fp16); + tensor var_1318 = const()[name = tensor("op_1318"), val = tensor([1, 1500, 12, -1])]; + tensor var_1319_cast = reshape(shape = var_1318, x = v_45_cast); + tensor var_1320 = const()[name = tensor("op_1320"), val = tensor([0, 2, 1, 3])]; + tensor qk_transpose_x_0 = const()[name = tensor("qk_transpose_x_0"), val = tensor(false)]; + tensor qk_transpose_y_0 = const()[name = tensor("qk_transpose_y_0"), val = tensor(false)]; + tensor transpose_46_perm_0 = const()[name = tensor("transpose_46_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_47_perm_0 = const()[name = tensor("transpose_47_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_49 = transpose(perm = transpose_47_perm_0, x = k_cast); + tensor transpose_50 = transpose(perm = transpose_46_perm_0, x = q_cast); + tensor qk_cast = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_50, y = transpose_49); + tensor var_1324_cast = softmax(axis = var_1259, x = qk_cast); + tensor var_1326_transpose_x_0 = const()[name = tensor("op_1326_transpose_x_0"), val = tensor(false)]; + tensor var_1326_transpose_y_0 = const()[name = tensor("op_1326_transpose_y_0"), val = tensor(false)]; + tensor transpose_51 = transpose(perm = var_1320, x = var_1319_cast); + tensor var_1326_cast = matmul(transpose_x = var_1326_transpose_x_0, transpose_y = var_1326_transpose_y_0, x = var_1324_cast, y = transpose_51); + tensor var_1327 = const()[name = tensor("op_1327"), val = tensor([0, 2, 1, 3])]; + tensor concat_11 = const()[name = tensor("concat_11"), val = tensor([1, 1500, 768])]; + tensor transpose_48 = transpose(perm = var_1327, x = var_1326_cast); + tensor x_143_cast = reshape(shape = concat_11, x = transpose_48); + tensor var_1332_to_fp16 = const()[name = tensor("op_1332_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(165706624)))]; + tensor var_1333_to_fp16 = const()[name = tensor("op_1333_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166886336)))]; + tensor var_1334_cast = linear(bias = var_1333_to_fp16, weight = var_1332_to_fp16, x = x_143_cast); + tensor x_145_cast = add(x = x_139_cast, y = var_1334_cast); + tensor var_1340_axes_0 = const()[name = tensor("op_1340_axes_0"), val = tensor([-1])]; + tensor blocks_11_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_11_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166887936)))]; + tensor blocks_11_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_11_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166889536)))]; + tensor var_1340_cast = layer_norm(axes = var_1340_axes_0, beta = blocks_11_mlp_ln_bias_to_fp16, epsilon = var_1265_to_fp16, gamma = blocks_11_mlp_ln_weight_to_fp16, x = x_145_cast); + tensor var_1349_to_fp16 = const()[name = tensor("op_1349_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(166891136)))]; + tensor var_1350_to_fp16 = const()[name = tensor("op_1350_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171609792)))]; + tensor input_97_cast = linear(bias = var_1350_to_fp16, weight = var_1349_to_fp16, x = var_1340_cast); + tensor x_149_mode_0 = const()[name = tensor("x_149_mode_0"), val = tensor("EXACT")]; + tensor x_149_cast = gelu(mode = x_149_mode_0, x = input_97_cast); + tensor var_1355_to_fp16 = const()[name = tensor("op_1355_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(171616000)))]; + tensor var_1356_to_fp16 = const()[name = tensor("op_1356_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176334656)))]; + tensor var_1357_cast = linear(bias = var_1356_to_fp16, weight = var_1355_to_fp16, x = x_149_cast); + tensor x_cast = add(x = x_145_cast, y = var_1357_cast); + tensor var_1370_axes_0 = const()[name = tensor("op_1370_axes_0"), val = tensor([-1])]; + tensor ln_post_weight_to_fp16 = const()[name = tensor("ln_post_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176336256)))]; + tensor ln_post_bias_to_fp16 = const()[name = tensor("ln_post_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(176337856)))]; + tensor var_1361_to_fp16 = const()[name = tensor("op_1361_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_1370_cast = layer_norm(axes = var_1370_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_1361_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast); + tensor var_1370_cast_to_fp32_dtype_0 = const()[name = tensor("op_1370_cast_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor output = cast(dtype = var_1370_cast_to_fp32_dtype_0, x = var_1370_cast); + } -> (output); +} \ No newline at end of file diff --git a/ggml-small.en-encoder.mlmodelc/weights/weight.bin b/ggml-small.en-encoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..d1229df486ac8a7a1e9667bd28651b80e7b116ea --- /dev/null +++ b/ggml-small.en-encoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d66ece4975aeea37c628f6b59d85de85d34beb2eb9b5f61f7d3dc7c1aca18143 +size 176339456 diff --git a/ggml-tiny-encoder.mlmodelc.zip b/ggml-tiny-encoder.mlmodelc.zip index f0a50f75adbb3cef51e857f0621e6e28d51b15d0..e671de5caab01d673f9e7bc30c28c0cab0b36b2c 100644 --- a/ggml-tiny-encoder.mlmodelc.zip +++ b/ggml-tiny-encoder.mlmodelc.zip @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:c88cbd2648e1f5415092bcf5256add463a0f19943e6938f46e8d4ffdebd47739 +oid sha256:4ce48c596dae4b732338b7e3f6d4db507fb80dafbba4115e0431cf0af778fab3 size 15037446 diff --git a/ggml-tiny-encoder.mlmodelc/analytics/coremldata.bin b/ggml-tiny-encoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..16ba4c4c65c22b09941cb787a9302b5fc0130c55 --- /dev/null +++ b/ggml-tiny-encoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f2f746e204d04779629b4959dad86d4ada040edd54f0ed1c80aa6f482a0fa72e +size 207 diff --git a/ggml-tiny-encoder.mlmodelc/coremldata.bin b/ggml-tiny-encoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..c732ed7b8caeac3c15acda1c2836f3e1f525a5a8 --- /dev/null +++ b/ggml-tiny-encoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:05fe28591b40616fa0c34ad7b853133623f5300923ec812acb11459c411acf3b +size 149 diff --git a/ggml-tiny-encoder.mlmodelc/metadata.json b/ggml-tiny-encoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..01f1ff45fbbba4fa68d9d81e145c8dd405299f11 --- /dev/null +++ b/ggml-tiny-encoder.mlmodelc/metadata.json @@ -0,0 +1,64 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float32", + "formattedType" : "MultiArray (Float32)", + "shortDescription" : "", + "shape" : "[]", + "name" : "output", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 6, + "mlProgramOperationTypeHistogram" : { + "Linear" : 24, + "Matmul" : 8, + "Cast" : 2, + "Conv" : 2, + "Softmax" : 4, + "Add" : 9, + "LayerNorm" : 9, + "Mul" : 8, + "Transpose" : 17, + "Gelu" : 6, + "Reshape" : 16 + }, + "computePrecision" : "Mixed (Float16, Float32, Int32)", + "isUpdatable" : "0", + "availability" : { + "macOS" : "12.0", + "tvOS" : "15.0", + "watchOS" : "8.0", + "iOS" : "15.0", + "macCatalyst" : "15.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float32", + "formattedType" : "MultiArray (Float32 1 × 80 × 3000)", + "shortDescription" : "", + "shape" : "[1, 80, 3000]", + "name" : "logmel_data", + "type" : "MultiArray" + } + ], + "generatedClassName" : "coreml_encoder_tiny", + "method" : "predict" + } +] \ No newline at end of file diff --git a/ggml-tiny-encoder.mlmodelc/model.mil b/ggml-tiny-encoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..ccedc27eec99a20a310efef71e5bbc4d874439cc --- /dev/null +++ b/ggml-tiny-encoder.mlmodelc/model.mil @@ -0,0 +1,275 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "4.28.4"}, {"coremlc-version", "1436.100.10"}})] +{ + func main(tensor logmel_data) { + tensor var_16 = const()[name = tensor("op_16"), val = tensor(1)]; + tensor var_24 = const()[name = tensor("op_24"), val = tensor([1])]; + tensor var_26 = const()[name = tensor("op_26"), val = tensor([1])]; + tensor var_28_pad_type_0 = const()[name = tensor("op_28_pad_type_0"), val = tensor("custom")]; + tensor var_28_pad_0 = const()[name = tensor("op_28_pad_0"), val = tensor([1, 1])]; + tensor logmel_data_to_fp16_dtype_0 = const()[name = tensor("logmel_data_to_fp16_dtype_0"), val = tensor("fp16")]; + tensor weight_3_to_fp16 = const()[name = tensor("weight_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor bias_3_to_fp16 = const()[name = tensor("bias_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184448)))]; + tensor cast_127 = cast(dtype = logmel_data_to_fp16_dtype_0, x = logmel_data); + tensor var_28_cast = conv(bias = bias_3_to_fp16, dilations = var_26, groups = var_16, pad = var_28_pad_0, pad_type = var_28_pad_type_0, strides = var_24, weight = weight_3_to_fp16, x = cast_127); + tensor input_1_mode_0 = const()[name = tensor("input_1_mode_0"), val = tensor("EXACT")]; + tensor input_1_cast = gelu(mode = input_1_mode_0, x = var_28_cast); + tensor var_32 = const()[name = tensor("op_32"), val = tensor(1)]; + tensor var_41 = const()[name = tensor("op_41"), val = tensor([2])]; + tensor var_43 = const()[name = tensor("op_43"), val = tensor([1])]; + tensor var_45_pad_type_0 = const()[name = tensor("op_45_pad_type_0"), val = tensor("custom")]; + tensor var_45_pad_0 = const()[name = tensor("op_45_pad_0"), val = tensor([1, 1])]; + tensor weight_7_to_fp16 = const()[name = tensor("weight_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(185280)))]; + tensor bias_7_to_fp16 = const()[name = tensor("bias_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1070080)))]; + tensor var_45_cast = conv(bias = bias_7_to_fp16, dilations = var_43, groups = var_32, pad = var_45_pad_0, pad_type = var_45_pad_type_0, strides = var_41, weight = weight_7_to_fp16, x = input_1_cast); + tensor x_3_mode_0 = const()[name = tensor("x_3_mode_0"), val = tensor("EXACT")]; + tensor x_3_cast = gelu(mode = x_3_mode_0, x = var_45_cast); + tensor var_50 = const()[name = tensor("op_50"), val = tensor([0, 2, 1])]; + tensor positional_embedding_to_fp16 = const()[name = tensor("positional_embedding_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1070912)))]; + tensor transpose_32 = transpose(perm = var_50, x = x_3_cast); + tensor var_53_cast = add(x = transpose_32, y = positional_embedding_to_fp16); + tensor var_65 = const()[name = tensor("op_65"), val = tensor(-1)]; + tensor var_82_axes_0 = const()[name = tensor("op_82_axes_0"), val = tensor([-1])]; + tensor blocks_0_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_0_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2222976)))]; + tensor blocks_0_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_0_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2223808)))]; + tensor var_71_to_fp16 = const()[name = tensor("op_71_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_82_cast = layer_norm(axes = var_82_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_71_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_53_cast); + tensor var_93_to_fp16 = const()[name = tensor("op_93_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2224640)))]; + tensor var_94_to_fp16 = const()[name = tensor("op_94_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2519616)))]; + tensor q_1_cast = linear(bias = var_94_to_fp16, weight = var_93_to_fp16, x = var_82_cast); + tensor var_97_to_fp16 = const()[name = tensor("op_97_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2520448)))]; + tensor k_1_bias_0_to_fp16 = const()[name = tensor("k_1_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2815424)))]; + tensor k_1_cast = linear(bias = k_1_bias_0_to_fp16, weight = var_97_to_fp16, x = var_82_cast); + tensor var_101_to_fp16 = const()[name = tensor("op_101_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2816256)))]; + tensor var_102_to_fp16 = const()[name = tensor("op_102_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3111232)))]; + tensor v_1_cast = linear(bias = var_102_to_fp16, weight = var_101_to_fp16, x = var_82_cast); + tensor var_110 = const()[name = tensor("op_110"), val = tensor([1, 1500, 6, -1])]; + tensor var_111_cast = reshape(shape = var_110, x = q_1_cast); + tensor const_28_to_fp16 = const()[name = tensor("const_28_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_3_cast = mul(x = var_111_cast, y = const_28_to_fp16); + tensor var_117 = const()[name = tensor("op_117"), val = tensor([1, 1500, 6, -1])]; + tensor var_118_cast = reshape(shape = var_117, x = k_1_cast); + tensor const_29_to_fp16 = const()[name = tensor("const_29_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_3_cast = mul(x = var_118_cast, y = const_29_to_fp16); + tensor var_124 = const()[name = tensor("op_124"), val = tensor([1, 1500, 6, -1])]; + tensor var_125_cast = reshape(shape = var_124, x = v_1_cast); + tensor var_126 = const()[name = tensor("op_126"), val = tensor([0, 2, 1, 3])]; + tensor qk_1_transpose_x_0 = const()[name = tensor("qk_1_transpose_x_0"), val = tensor(false)]; + tensor qk_1_transpose_y_0 = const()[name = tensor("qk_1_transpose_y_0"), val = tensor(false)]; + tensor transpose_8_perm_0 = const()[name = tensor("transpose_8_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_9_perm_0 = const()[name = tensor("transpose_9_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_29 = transpose(perm = transpose_9_perm_0, x = k_3_cast); + tensor transpose_30 = transpose(perm = transpose_8_perm_0, x = q_3_cast); + tensor qk_1_cast = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_30, y = transpose_29); + tensor var_130_cast = softmax(axis = var_65, x = qk_1_cast); + tensor var_132_transpose_x_0 = const()[name = tensor("op_132_transpose_x_0"), val = tensor(false)]; + tensor var_132_transpose_y_0 = const()[name = tensor("op_132_transpose_y_0"), val = tensor(false)]; + tensor transpose_31 = transpose(perm = var_126, x = var_125_cast); + tensor var_132_cast = matmul(transpose_x = var_132_transpose_x_0, transpose_y = var_132_transpose_y_0, x = var_130_cast, y = transpose_31); + tensor var_133 = const()[name = tensor("op_133"), val = tensor([0, 2, 1, 3])]; + tensor concat_0 = const()[name = tensor("concat_0"), val = tensor([1, 1500, 384])]; + tensor transpose_28 = transpose(perm = var_133, x = var_132_cast); + tensor x_11_cast = reshape(shape = concat_0, x = transpose_28); + tensor var_138_to_fp16 = const()[name = tensor("op_138_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3112064)))]; + tensor var_139_to_fp16 = const()[name = tensor("op_139_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3407040)))]; + tensor var_140_cast = linear(bias = var_139_to_fp16, weight = var_138_to_fp16, x = x_11_cast); + tensor x_13_cast = add(x = var_53_cast, y = var_140_cast); + tensor var_146_axes_0 = const()[name = tensor("op_146_axes_0"), val = tensor([-1])]; + tensor blocks_0_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3407872)))]; + tensor blocks_0_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3408704)))]; + tensor var_146_cast = layer_norm(axes = var_146_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_71_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast); + tensor var_155_to_fp16 = const()[name = tensor("op_155_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3409536)))]; + tensor var_156_to_fp16 = const()[name = tensor("op_156_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4589248)))]; + tensor input_9_cast = linear(bias = var_156_to_fp16, weight = var_155_to_fp16, x = var_146_cast); + tensor x_17_mode_0 = const()[name = tensor("x_17_mode_0"), val = tensor("EXACT")]; + tensor x_17_cast = gelu(mode = x_17_mode_0, x = input_9_cast); + tensor var_161_to_fp16 = const()[name = tensor("op_161_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4592384)))]; + tensor var_162_to_fp16 = const()[name = tensor("op_162_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5772096)))]; + tensor var_163_cast = linear(bias = var_162_to_fp16, weight = var_161_to_fp16, x = x_17_cast); + tensor x_19_cast = add(x = x_13_cast, y = var_163_cast); + tensor var_171 = const()[name = tensor("op_171"), val = tensor(-1)]; + tensor var_188_axes_0 = const()[name = tensor("op_188_axes_0"), val = tensor([-1])]; + tensor blocks_1_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_1_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5772928)))]; + tensor blocks_1_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_1_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5773760)))]; + tensor var_177_to_fp16 = const()[name = tensor("op_177_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_188_cast = layer_norm(axes = var_188_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_177_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast); + tensor var_199_to_fp16 = const()[name = tensor("op_199_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5774592)))]; + tensor var_200_to_fp16 = const()[name = tensor("op_200_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6069568)))]; + tensor q_5_cast = linear(bias = var_200_to_fp16, weight = var_199_to_fp16, x = var_188_cast); + tensor var_203_to_fp16 = const()[name = tensor("op_203_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6070400)))]; + tensor k_5_bias_0_to_fp16 = const()[name = tensor("k_5_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6365376)))]; + tensor k_5_cast = linear(bias = k_5_bias_0_to_fp16, weight = var_203_to_fp16, x = var_188_cast); + tensor var_207_to_fp16 = const()[name = tensor("op_207_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6366208)))]; + tensor var_208_to_fp16 = const()[name = tensor("op_208_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6661184)))]; + tensor v_5_cast = linear(bias = var_208_to_fp16, weight = var_207_to_fp16, x = var_188_cast); + tensor var_216 = const()[name = tensor("op_216"), val = tensor([1, 1500, 6, -1])]; + tensor var_217_cast = reshape(shape = var_216, x = q_5_cast); + tensor const_30_to_fp16 = const()[name = tensor("const_30_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_7_cast = mul(x = var_217_cast, y = const_30_to_fp16); + tensor var_223 = const()[name = tensor("op_223"), val = tensor([1, 1500, 6, -1])]; + tensor var_224_cast = reshape(shape = var_223, x = k_5_cast); + tensor const_31_to_fp16 = const()[name = tensor("const_31_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_7_cast = mul(x = var_224_cast, y = const_31_to_fp16); + tensor var_230 = const()[name = tensor("op_230"), val = tensor([1, 1500, 6, -1])]; + tensor var_231_cast = reshape(shape = var_230, x = v_5_cast); + tensor var_232 = const()[name = tensor("op_232"), val = tensor([0, 2, 1, 3])]; + tensor qk_3_transpose_x_0 = const()[name = tensor("qk_3_transpose_x_0"), val = tensor(false)]; + tensor qk_3_transpose_y_0 = const()[name = tensor("qk_3_transpose_y_0"), val = tensor(false)]; + tensor transpose_10_perm_0 = const()[name = tensor("transpose_10_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_11_perm_0 = const()[name = tensor("transpose_11_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_25 = transpose(perm = transpose_11_perm_0, x = k_7_cast); + tensor transpose_26 = transpose(perm = transpose_10_perm_0, x = q_7_cast); + tensor qk_3_cast = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_26, y = transpose_25); + tensor var_236_cast = softmax(axis = var_171, x = qk_3_cast); + tensor var_238_transpose_x_0 = const()[name = tensor("op_238_transpose_x_0"), val = tensor(false)]; + tensor var_238_transpose_y_0 = const()[name = tensor("op_238_transpose_y_0"), val = tensor(false)]; + tensor transpose_27 = transpose(perm = var_232, x = var_231_cast); + tensor var_238_cast = matmul(transpose_x = var_238_transpose_x_0, transpose_y = var_238_transpose_y_0, x = var_236_cast, y = transpose_27); + tensor var_239 = const()[name = tensor("op_239"), val = tensor([0, 2, 1, 3])]; + tensor concat_1 = const()[name = tensor("concat_1"), val = tensor([1, 1500, 384])]; + tensor transpose_24 = transpose(perm = var_239, x = var_238_cast); + tensor x_23_cast = reshape(shape = concat_1, x = transpose_24); + tensor var_244_to_fp16 = const()[name = tensor("op_244_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6662016)))]; + tensor var_245_to_fp16 = const()[name = tensor("op_245_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6956992)))]; + tensor var_246_cast = linear(bias = var_245_to_fp16, weight = var_244_to_fp16, x = x_23_cast); + tensor x_25_cast = add(x = x_19_cast, y = var_246_cast); + tensor var_252_axes_0 = const()[name = tensor("op_252_axes_0"), val = tensor([-1])]; + tensor blocks_1_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6957824)))]; + tensor blocks_1_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6958656)))]; + tensor var_252_cast = layer_norm(axes = var_252_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_177_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast); + tensor var_261_to_fp16 = const()[name = tensor("op_261_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6959488)))]; + tensor var_262_to_fp16 = const()[name = tensor("op_262_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8139200)))]; + tensor input_17_cast = linear(bias = var_262_to_fp16, weight = var_261_to_fp16, x = var_252_cast); + tensor x_29_mode_0 = const()[name = tensor("x_29_mode_0"), val = tensor("EXACT")]; + tensor x_29_cast = gelu(mode = x_29_mode_0, x = input_17_cast); + tensor var_267_to_fp16 = const()[name = tensor("op_267_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8142336)))]; + tensor var_268_to_fp16 = const()[name = tensor("op_268_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9322048)))]; + tensor var_269_cast = linear(bias = var_268_to_fp16, weight = var_267_to_fp16, x = x_29_cast); + tensor x_31_cast = add(x = x_25_cast, y = var_269_cast); + tensor var_277 = const()[name = tensor("op_277"), val = tensor(-1)]; + tensor var_294_axes_0 = const()[name = tensor("op_294_axes_0"), val = tensor([-1])]; + tensor blocks_2_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_2_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9322880)))]; + tensor blocks_2_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_2_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9323712)))]; + tensor var_283_to_fp16 = const()[name = tensor("op_283_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_294_cast = layer_norm(axes = var_294_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_283_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast); + tensor var_305_to_fp16 = const()[name = tensor("op_305_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9324544)))]; + tensor var_306_to_fp16 = const()[name = tensor("op_306_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9619520)))]; + tensor q_9_cast = linear(bias = var_306_to_fp16, weight = var_305_to_fp16, x = var_294_cast); + tensor var_309_to_fp16 = const()[name = tensor("op_309_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9620352)))]; + tensor k_9_bias_0_to_fp16 = const()[name = tensor("k_9_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9915328)))]; + tensor k_9_cast = linear(bias = k_9_bias_0_to_fp16, weight = var_309_to_fp16, x = var_294_cast); + tensor var_313_to_fp16 = const()[name = tensor("op_313_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9916160)))]; + tensor var_314_to_fp16 = const()[name = tensor("op_314_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10211136)))]; + tensor v_9_cast = linear(bias = var_314_to_fp16, weight = var_313_to_fp16, x = var_294_cast); + tensor var_322 = const()[name = tensor("op_322"), val = tensor([1, 1500, 6, -1])]; + tensor var_323_cast = reshape(shape = var_322, x = q_9_cast); + tensor const_32_to_fp16 = const()[name = tensor("const_32_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_11_cast = mul(x = var_323_cast, y = const_32_to_fp16); + tensor var_329 = const()[name = tensor("op_329"), val = tensor([1, 1500, 6, -1])]; + tensor var_330_cast = reshape(shape = var_329, x = k_9_cast); + tensor const_33_to_fp16 = const()[name = tensor("const_33_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_11_cast = mul(x = var_330_cast, y = const_33_to_fp16); + tensor var_336 = const()[name = tensor("op_336"), val = tensor([1, 1500, 6, -1])]; + tensor var_337_cast = reshape(shape = var_336, x = v_9_cast); + tensor var_338 = const()[name = tensor("op_338"), val = tensor([0, 2, 1, 3])]; + tensor qk_5_transpose_x_0 = const()[name = tensor("qk_5_transpose_x_0"), val = tensor(false)]; + tensor qk_5_transpose_y_0 = const()[name = tensor("qk_5_transpose_y_0"), val = tensor(false)]; + tensor transpose_12_perm_0 = const()[name = tensor("transpose_12_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_13_perm_0 = const()[name = tensor("transpose_13_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_21 = transpose(perm = transpose_13_perm_0, x = k_11_cast); + tensor transpose_22 = transpose(perm = transpose_12_perm_0, x = q_11_cast); + tensor qk_5_cast = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_22, y = transpose_21); + tensor var_342_cast = softmax(axis = var_277, x = qk_5_cast); + tensor var_344_transpose_x_0 = const()[name = tensor("op_344_transpose_x_0"), val = tensor(false)]; + tensor var_344_transpose_y_0 = const()[name = tensor("op_344_transpose_y_0"), val = tensor(false)]; + tensor transpose_23 = transpose(perm = var_338, x = var_337_cast); + tensor var_344_cast = matmul(transpose_x = var_344_transpose_x_0, transpose_y = var_344_transpose_y_0, x = var_342_cast, y = transpose_23); + tensor var_345 = const()[name = tensor("op_345"), val = tensor([0, 2, 1, 3])]; + tensor concat_2 = const()[name = tensor("concat_2"), val = tensor([1, 1500, 384])]; + tensor transpose_20 = transpose(perm = var_345, x = var_344_cast); + tensor x_35_cast = reshape(shape = concat_2, x = transpose_20); + tensor var_350_to_fp16 = const()[name = tensor("op_350_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10211968)))]; + tensor var_351_to_fp16 = const()[name = tensor("op_351_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10506944)))]; + tensor var_352_cast = linear(bias = var_351_to_fp16, weight = var_350_to_fp16, x = x_35_cast); + tensor x_37_cast = add(x = x_31_cast, y = var_352_cast); + tensor var_358_axes_0 = const()[name = tensor("op_358_axes_0"), val = tensor([-1])]; + tensor blocks_2_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10507776)))]; + tensor blocks_2_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10508608)))]; + tensor var_358_cast = layer_norm(axes = var_358_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_283_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast); + tensor var_367_to_fp16 = const()[name = tensor("op_367_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10509440)))]; + tensor var_368_to_fp16 = const()[name = tensor("op_368_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11689152)))]; + tensor input_25_cast = linear(bias = var_368_to_fp16, weight = var_367_to_fp16, x = var_358_cast); + tensor x_41_mode_0 = const()[name = tensor("x_41_mode_0"), val = tensor("EXACT")]; + tensor x_41_cast = gelu(mode = x_41_mode_0, x = input_25_cast); + tensor var_373_to_fp16 = const()[name = tensor("op_373_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11692288)))]; + tensor var_374_to_fp16 = const()[name = tensor("op_374_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12872000)))]; + tensor var_375_cast = linear(bias = var_374_to_fp16, weight = var_373_to_fp16, x = x_41_cast); + tensor x_43_cast = add(x = x_37_cast, y = var_375_cast); + tensor var_383 = const()[name = tensor("op_383"), val = tensor(-1)]; + tensor var_400_axes_0 = const()[name = tensor("op_400_axes_0"), val = tensor([-1])]; + tensor blocks_3_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_3_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12872832)))]; + tensor blocks_3_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_3_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12873664)))]; + tensor var_389_to_fp16 = const()[name = tensor("op_389_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_400_cast = layer_norm(axes = var_400_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_389_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast); + tensor var_411_to_fp16 = const()[name = tensor("op_411_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12874496)))]; + tensor var_412_to_fp16 = const()[name = tensor("op_412_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13169472)))]; + tensor q_13_cast = linear(bias = var_412_to_fp16, weight = var_411_to_fp16, x = var_400_cast); + tensor var_415_to_fp16 = const()[name = tensor("op_415_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13170304)))]; + tensor k_13_bias_0_to_fp16 = const()[name = tensor("k_13_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13465280)))]; + tensor k_13_cast = linear(bias = k_13_bias_0_to_fp16, weight = var_415_to_fp16, x = var_400_cast); + tensor var_419_to_fp16 = const()[name = tensor("op_419_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13466112)))]; + tensor var_420_to_fp16 = const()[name = tensor("op_420_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13761088)))]; + tensor v_13_cast = linear(bias = var_420_to_fp16, weight = var_419_to_fp16, x = var_400_cast); + tensor var_428 = const()[name = tensor("op_428"), val = tensor([1, 1500, 6, -1])]; + tensor var_429_cast = reshape(shape = var_428, x = q_13_cast); + tensor const_34_to_fp16 = const()[name = tensor("const_34_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_cast = mul(x = var_429_cast, y = const_34_to_fp16); + tensor var_435 = const()[name = tensor("op_435"), val = tensor([1, 1500, 6, -1])]; + tensor var_436_cast = reshape(shape = var_435, x = k_13_cast); + tensor const_35_to_fp16 = const()[name = tensor("const_35_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_cast = mul(x = var_436_cast, y = const_35_to_fp16); + tensor var_442 = const()[name = tensor("op_442"), val = tensor([1, 1500, 6, -1])]; + tensor var_443_cast = reshape(shape = var_442, x = v_13_cast); + tensor var_444 = const()[name = tensor("op_444"), val = tensor([0, 2, 1, 3])]; + tensor qk_transpose_x_0 = const()[name = tensor("qk_transpose_x_0"), val = tensor(false)]; + tensor qk_transpose_y_0 = const()[name = tensor("qk_transpose_y_0"), val = tensor(false)]; + tensor transpose_14_perm_0 = const()[name = tensor("transpose_14_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_15_perm_0 = const()[name = tensor("transpose_15_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_17 = transpose(perm = transpose_15_perm_0, x = k_cast); + tensor transpose_18 = transpose(perm = transpose_14_perm_0, x = q_cast); + tensor qk_cast = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_18, y = transpose_17); + tensor var_448_cast = softmax(axis = var_383, x = qk_cast); + tensor var_450_transpose_x_0 = const()[name = tensor("op_450_transpose_x_0"), val = tensor(false)]; + tensor var_450_transpose_y_0 = const()[name = tensor("op_450_transpose_y_0"), val = tensor(false)]; + tensor transpose_19 = transpose(perm = var_444, x = var_443_cast); + tensor var_450_cast = matmul(transpose_x = var_450_transpose_x_0, transpose_y = var_450_transpose_y_0, x = var_448_cast, y = transpose_19); + tensor var_451 = const()[name = tensor("op_451"), val = tensor([0, 2, 1, 3])]; + tensor concat_3 = const()[name = tensor("concat_3"), val = tensor([1, 1500, 384])]; + tensor transpose_16 = transpose(perm = var_451, x = var_450_cast); + tensor x_47_cast = reshape(shape = concat_3, x = transpose_16); + tensor var_456_to_fp16 = const()[name = tensor("op_456_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13761920)))]; + tensor var_457_to_fp16 = const()[name = tensor("op_457_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14056896)))]; + tensor var_458_cast = linear(bias = var_457_to_fp16, weight = var_456_to_fp16, x = x_47_cast); + tensor x_49_cast = add(x = x_43_cast, y = var_458_cast); + tensor var_464_axes_0 = const()[name = tensor("op_464_axes_0"), val = tensor([-1])]; + tensor blocks_3_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14057728)))]; + tensor blocks_3_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14058560)))]; + tensor var_464_cast = layer_norm(axes = var_464_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_389_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast); + tensor var_473_to_fp16 = const()[name = tensor("op_473_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14059392)))]; + tensor var_474_to_fp16 = const()[name = tensor("op_474_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15239104)))]; + tensor input_33_cast = linear(bias = var_474_to_fp16, weight = var_473_to_fp16, x = var_464_cast); + tensor x_53_mode_0 = const()[name = tensor("x_53_mode_0"), val = tensor("EXACT")]; + tensor x_53_cast = gelu(mode = x_53_mode_0, x = input_33_cast); + tensor var_479_to_fp16 = const()[name = tensor("op_479_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15242240)))]; + tensor var_480_to_fp16 = const()[name = tensor("op_480_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16421952)))]; + tensor var_481_cast = linear(bias = var_480_to_fp16, weight = var_479_to_fp16, x = x_53_cast); + tensor x_cast = add(x = x_49_cast, y = var_481_cast); + tensor var_494_axes_0 = const()[name = tensor("op_494_axes_0"), val = tensor([-1])]; + tensor ln_post_weight_to_fp16 = const()[name = tensor("ln_post_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16422784)))]; + tensor ln_post_bias_to_fp16 = const()[name = tensor("ln_post_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16423616)))]; + tensor var_485_to_fp16 = const()[name = tensor("op_485_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_494_cast = layer_norm(axes = var_494_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_485_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast); + tensor var_494_cast_to_fp32_dtype_0 = const()[name = tensor("op_494_cast_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor output = cast(dtype = var_494_cast_to_fp32_dtype_0, x = var_494_cast); + } -> (output); +} \ No newline at end of file diff --git a/ggml-tiny-encoder.mlmodelc/weights/weight.bin b/ggml-tiny-encoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..1e6c109974ed71f6326eca3e75c025a1c6234b26 --- /dev/null +++ b/ggml-tiny-encoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:5d3e4113a57206004a0ca73454ac9fcc7aed518a1729a83402629ccec54a022c +size 16424448 diff --git a/ggml-tiny.en-encoder.mlmodelc.zip b/ggml-tiny.en-encoder.mlmodelc.zip index 3eeeb7dbc71af46e4c0cc48d72d64cbb9b84f1a5..d3326df33ea675808ce5c7f9785448df4aa8bd09 100644 --- a/ggml-tiny.en-encoder.mlmodelc.zip +++ b/ggml-tiny.en-encoder.mlmodelc.zip @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:82b32eef73c94bb0c432a776a047b757d9525c26d84038a15d8798d7c8d1ee58 +oid sha256:4a0c60546b99bbc34389d7e43becc1c6ccd2b24b06bbe82fc45721bfe13ebcbb size 15034655 diff --git a/ggml-tiny.en-encoder.mlmodelc/analytics/coremldata.bin b/ggml-tiny.en-encoder.mlmodelc/analytics/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..16ba4c4c65c22b09941cb787a9302b5fc0130c55 --- /dev/null +++ b/ggml-tiny.en-encoder.mlmodelc/analytics/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f2f746e204d04779629b4959dad86d4ada040edd54f0ed1c80aa6f482a0fa72e +size 207 diff --git a/ggml-tiny.en-encoder.mlmodelc/coremldata.bin b/ggml-tiny.en-encoder.mlmodelc/coremldata.bin new file mode 100644 index 0000000000000000000000000000000000000000..c732ed7b8caeac3c15acda1c2836f3e1f525a5a8 --- /dev/null +++ b/ggml-tiny.en-encoder.mlmodelc/coremldata.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:05fe28591b40616fa0c34ad7b853133623f5300923ec812acb11459c411acf3b +size 149 diff --git a/ggml-tiny.en-encoder.mlmodelc/metadata.json b/ggml-tiny.en-encoder.mlmodelc/metadata.json new file mode 100644 index 0000000000000000000000000000000000000000..de3f1ce87c6a39a44201cfdcb1bdd865eb81e4d3 --- /dev/null +++ b/ggml-tiny.en-encoder.mlmodelc/metadata.json @@ -0,0 +1,64 @@ +[ + { + "metadataOutputVersion" : "3.0", + "storagePrecision" : "Float16", + "outputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float32", + "formattedType" : "MultiArray (Float32)", + "shortDescription" : "", + "shape" : "[]", + "name" : "output", + "type" : "MultiArray" + } + ], + "modelParameters" : [ + + ], + "specificationVersion" : 6, + "mlProgramOperationTypeHistogram" : { + "Linear" : 24, + "Matmul" : 8, + "Cast" : 2, + "Conv" : 2, + "Softmax" : 4, + "Add" : 9, + "LayerNorm" : 9, + "Mul" : 8, + "Transpose" : 17, + "Gelu" : 6, + "Reshape" : 16 + }, + "computePrecision" : "Mixed (Float16, Float32, Int32)", + "isUpdatable" : "0", + "availability" : { + "macOS" : "12.0", + "tvOS" : "15.0", + "watchOS" : "8.0", + "iOS" : "15.0", + "macCatalyst" : "15.0" + }, + "modelType" : { + "name" : "MLModelType_mlProgram" + }, + "userDefinedMetadata" : { + + }, + "inputSchema" : [ + { + "hasShapeFlexibility" : "0", + "isOptional" : "0", + "dataType" : "Float32", + "formattedType" : "MultiArray (Float32 1 × 80 × 3000)", + "shortDescription" : "", + "shape" : "[1, 80, 3000]", + "name" : "logmel_data", + "type" : "MultiArray" + } + ], + "generatedClassName" : "coreml_encoder_tiny_en", + "method" : "predict" + } +] \ No newline at end of file diff --git a/ggml-tiny.en-encoder.mlmodelc/model.mil b/ggml-tiny.en-encoder.mlmodelc/model.mil new file mode 100644 index 0000000000000000000000000000000000000000..ccedc27eec99a20a310efef71e5bbc4d874439cc --- /dev/null +++ b/ggml-tiny.en-encoder.mlmodelc/model.mil @@ -0,0 +1,275 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "4.28.4"}, {"coremlc-version", "1436.100.10"}})] +{ + func main(tensor logmel_data) { + tensor var_16 = const()[name = tensor("op_16"), val = tensor(1)]; + tensor var_24 = const()[name = tensor("op_24"), val = tensor([1])]; + tensor var_26 = const()[name = tensor("op_26"), val = tensor([1])]; + tensor var_28_pad_type_0 = const()[name = tensor("op_28_pad_type_0"), val = tensor("custom")]; + tensor var_28_pad_0 = const()[name = tensor("op_28_pad_0"), val = tensor([1, 1])]; + tensor logmel_data_to_fp16_dtype_0 = const()[name = tensor("logmel_data_to_fp16_dtype_0"), val = tensor("fp16")]; + tensor weight_3_to_fp16 = const()[name = tensor("weight_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor bias_3_to_fp16 = const()[name = tensor("bias_3_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(184448)))]; + tensor cast_127 = cast(dtype = logmel_data_to_fp16_dtype_0, x = logmel_data); + tensor var_28_cast = conv(bias = bias_3_to_fp16, dilations = var_26, groups = var_16, pad = var_28_pad_0, pad_type = var_28_pad_type_0, strides = var_24, weight = weight_3_to_fp16, x = cast_127); + tensor input_1_mode_0 = const()[name = tensor("input_1_mode_0"), val = tensor("EXACT")]; + tensor input_1_cast = gelu(mode = input_1_mode_0, x = var_28_cast); + tensor var_32 = const()[name = tensor("op_32"), val = tensor(1)]; + tensor var_41 = const()[name = tensor("op_41"), val = tensor([2])]; + tensor var_43 = const()[name = tensor("op_43"), val = tensor([1])]; + tensor var_45_pad_type_0 = const()[name = tensor("op_45_pad_type_0"), val = tensor("custom")]; + tensor var_45_pad_0 = const()[name = tensor("op_45_pad_0"), val = tensor([1, 1])]; + tensor weight_7_to_fp16 = const()[name = tensor("weight_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(185280)))]; + tensor bias_7_to_fp16 = const()[name = tensor("bias_7_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1070080)))]; + tensor var_45_cast = conv(bias = bias_7_to_fp16, dilations = var_43, groups = var_32, pad = var_45_pad_0, pad_type = var_45_pad_type_0, strides = var_41, weight = weight_7_to_fp16, x = input_1_cast); + tensor x_3_mode_0 = const()[name = tensor("x_3_mode_0"), val = tensor("EXACT")]; + tensor x_3_cast = gelu(mode = x_3_mode_0, x = var_45_cast); + tensor var_50 = const()[name = tensor("op_50"), val = tensor([0, 2, 1])]; + tensor positional_embedding_to_fp16 = const()[name = tensor("positional_embedding_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1070912)))]; + tensor transpose_32 = transpose(perm = var_50, x = x_3_cast); + tensor var_53_cast = add(x = transpose_32, y = positional_embedding_to_fp16); + tensor var_65 = const()[name = tensor("op_65"), val = tensor(-1)]; + tensor var_82_axes_0 = const()[name = tensor("op_82_axes_0"), val = tensor([-1])]; + tensor blocks_0_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_0_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2222976)))]; + tensor blocks_0_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_0_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2223808)))]; + tensor var_71_to_fp16 = const()[name = tensor("op_71_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_82_cast = layer_norm(axes = var_82_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_71_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_53_cast); + tensor var_93_to_fp16 = const()[name = tensor("op_93_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2224640)))]; + tensor var_94_to_fp16 = const()[name = tensor("op_94_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2519616)))]; + tensor q_1_cast = linear(bias = var_94_to_fp16, weight = var_93_to_fp16, x = var_82_cast); + tensor var_97_to_fp16 = const()[name = tensor("op_97_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2520448)))]; + tensor k_1_bias_0_to_fp16 = const()[name = tensor("k_1_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2815424)))]; + tensor k_1_cast = linear(bias = k_1_bias_0_to_fp16, weight = var_97_to_fp16, x = var_82_cast); + tensor var_101_to_fp16 = const()[name = tensor("op_101_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2816256)))]; + tensor var_102_to_fp16 = const()[name = tensor("op_102_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3111232)))]; + tensor v_1_cast = linear(bias = var_102_to_fp16, weight = var_101_to_fp16, x = var_82_cast); + tensor var_110 = const()[name = tensor("op_110"), val = tensor([1, 1500, 6, -1])]; + tensor var_111_cast = reshape(shape = var_110, x = q_1_cast); + tensor const_28_to_fp16 = const()[name = tensor("const_28_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_3_cast = mul(x = var_111_cast, y = const_28_to_fp16); + tensor var_117 = const()[name = tensor("op_117"), val = tensor([1, 1500, 6, -1])]; + tensor var_118_cast = reshape(shape = var_117, x = k_1_cast); + tensor const_29_to_fp16 = const()[name = tensor("const_29_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_3_cast = mul(x = var_118_cast, y = const_29_to_fp16); + tensor var_124 = const()[name = tensor("op_124"), val = tensor([1, 1500, 6, -1])]; + tensor var_125_cast = reshape(shape = var_124, x = v_1_cast); + tensor var_126 = const()[name = tensor("op_126"), val = tensor([0, 2, 1, 3])]; + tensor qk_1_transpose_x_0 = const()[name = tensor("qk_1_transpose_x_0"), val = tensor(false)]; + tensor qk_1_transpose_y_0 = const()[name = tensor("qk_1_transpose_y_0"), val = tensor(false)]; + tensor transpose_8_perm_0 = const()[name = tensor("transpose_8_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_9_perm_0 = const()[name = tensor("transpose_9_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_29 = transpose(perm = transpose_9_perm_0, x = k_3_cast); + tensor transpose_30 = transpose(perm = transpose_8_perm_0, x = q_3_cast); + tensor qk_1_cast = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_30, y = transpose_29); + tensor var_130_cast = softmax(axis = var_65, x = qk_1_cast); + tensor var_132_transpose_x_0 = const()[name = tensor("op_132_transpose_x_0"), val = tensor(false)]; + tensor var_132_transpose_y_0 = const()[name = tensor("op_132_transpose_y_0"), val = tensor(false)]; + tensor transpose_31 = transpose(perm = var_126, x = var_125_cast); + tensor var_132_cast = matmul(transpose_x = var_132_transpose_x_0, transpose_y = var_132_transpose_y_0, x = var_130_cast, y = transpose_31); + tensor var_133 = const()[name = tensor("op_133"), val = tensor([0, 2, 1, 3])]; + tensor concat_0 = const()[name = tensor("concat_0"), val = tensor([1, 1500, 384])]; + tensor transpose_28 = transpose(perm = var_133, x = var_132_cast); + tensor x_11_cast = reshape(shape = concat_0, x = transpose_28); + tensor var_138_to_fp16 = const()[name = tensor("op_138_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3112064)))]; + tensor var_139_to_fp16 = const()[name = tensor("op_139_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3407040)))]; + tensor var_140_cast = linear(bias = var_139_to_fp16, weight = var_138_to_fp16, x = x_11_cast); + tensor x_13_cast = add(x = var_53_cast, y = var_140_cast); + tensor var_146_axes_0 = const()[name = tensor("op_146_axes_0"), val = tensor([-1])]; + tensor blocks_0_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3407872)))]; + tensor blocks_0_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_0_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3408704)))]; + tensor var_146_cast = layer_norm(axes = var_146_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_71_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast); + tensor var_155_to_fp16 = const()[name = tensor("op_155_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3409536)))]; + tensor var_156_to_fp16 = const()[name = tensor("op_156_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4589248)))]; + tensor input_9_cast = linear(bias = var_156_to_fp16, weight = var_155_to_fp16, x = var_146_cast); + tensor x_17_mode_0 = const()[name = tensor("x_17_mode_0"), val = tensor("EXACT")]; + tensor x_17_cast = gelu(mode = x_17_mode_0, x = input_9_cast); + tensor var_161_to_fp16 = const()[name = tensor("op_161_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4592384)))]; + tensor var_162_to_fp16 = const()[name = tensor("op_162_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5772096)))]; + tensor var_163_cast = linear(bias = var_162_to_fp16, weight = var_161_to_fp16, x = x_17_cast); + tensor x_19_cast = add(x = x_13_cast, y = var_163_cast); + tensor var_171 = const()[name = tensor("op_171"), val = tensor(-1)]; + tensor var_188_axes_0 = const()[name = tensor("op_188_axes_0"), val = tensor([-1])]; + tensor blocks_1_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_1_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5772928)))]; + tensor blocks_1_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_1_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5773760)))]; + tensor var_177_to_fp16 = const()[name = tensor("op_177_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_188_cast = layer_norm(axes = var_188_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_177_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast); + tensor var_199_to_fp16 = const()[name = tensor("op_199_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5774592)))]; + tensor var_200_to_fp16 = const()[name = tensor("op_200_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6069568)))]; + tensor q_5_cast = linear(bias = var_200_to_fp16, weight = var_199_to_fp16, x = var_188_cast); + tensor var_203_to_fp16 = const()[name = tensor("op_203_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6070400)))]; + tensor k_5_bias_0_to_fp16 = const()[name = tensor("k_5_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6365376)))]; + tensor k_5_cast = linear(bias = k_5_bias_0_to_fp16, weight = var_203_to_fp16, x = var_188_cast); + tensor var_207_to_fp16 = const()[name = tensor("op_207_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6366208)))]; + tensor var_208_to_fp16 = const()[name = tensor("op_208_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6661184)))]; + tensor v_5_cast = linear(bias = var_208_to_fp16, weight = var_207_to_fp16, x = var_188_cast); + tensor var_216 = const()[name = tensor("op_216"), val = tensor([1, 1500, 6, -1])]; + tensor var_217_cast = reshape(shape = var_216, x = q_5_cast); + tensor const_30_to_fp16 = const()[name = tensor("const_30_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_7_cast = mul(x = var_217_cast, y = const_30_to_fp16); + tensor var_223 = const()[name = tensor("op_223"), val = tensor([1, 1500, 6, -1])]; + tensor var_224_cast = reshape(shape = var_223, x = k_5_cast); + tensor const_31_to_fp16 = const()[name = tensor("const_31_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_7_cast = mul(x = var_224_cast, y = const_31_to_fp16); + tensor var_230 = const()[name = tensor("op_230"), val = tensor([1, 1500, 6, -1])]; + tensor var_231_cast = reshape(shape = var_230, x = v_5_cast); + tensor var_232 = const()[name = tensor("op_232"), val = tensor([0, 2, 1, 3])]; + tensor qk_3_transpose_x_0 = const()[name = tensor("qk_3_transpose_x_0"), val = tensor(false)]; + tensor qk_3_transpose_y_0 = const()[name = tensor("qk_3_transpose_y_0"), val = tensor(false)]; + tensor transpose_10_perm_0 = const()[name = tensor("transpose_10_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_11_perm_0 = const()[name = tensor("transpose_11_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_25 = transpose(perm = transpose_11_perm_0, x = k_7_cast); + tensor transpose_26 = transpose(perm = transpose_10_perm_0, x = q_7_cast); + tensor qk_3_cast = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_26, y = transpose_25); + tensor var_236_cast = softmax(axis = var_171, x = qk_3_cast); + tensor var_238_transpose_x_0 = const()[name = tensor("op_238_transpose_x_0"), val = tensor(false)]; + tensor var_238_transpose_y_0 = const()[name = tensor("op_238_transpose_y_0"), val = tensor(false)]; + tensor transpose_27 = transpose(perm = var_232, x = var_231_cast); + tensor var_238_cast = matmul(transpose_x = var_238_transpose_x_0, transpose_y = var_238_transpose_y_0, x = var_236_cast, y = transpose_27); + tensor var_239 = const()[name = tensor("op_239"), val = tensor([0, 2, 1, 3])]; + tensor concat_1 = const()[name = tensor("concat_1"), val = tensor([1, 1500, 384])]; + tensor transpose_24 = transpose(perm = var_239, x = var_238_cast); + tensor x_23_cast = reshape(shape = concat_1, x = transpose_24); + tensor var_244_to_fp16 = const()[name = tensor("op_244_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6662016)))]; + tensor var_245_to_fp16 = const()[name = tensor("op_245_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6956992)))]; + tensor var_246_cast = linear(bias = var_245_to_fp16, weight = var_244_to_fp16, x = x_23_cast); + tensor x_25_cast = add(x = x_19_cast, y = var_246_cast); + tensor var_252_axes_0 = const()[name = tensor("op_252_axes_0"), val = tensor([-1])]; + tensor blocks_1_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6957824)))]; + tensor blocks_1_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_1_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6958656)))]; + tensor var_252_cast = layer_norm(axes = var_252_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_177_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast); + tensor var_261_to_fp16 = const()[name = tensor("op_261_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6959488)))]; + tensor var_262_to_fp16 = const()[name = tensor("op_262_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8139200)))]; + tensor input_17_cast = linear(bias = var_262_to_fp16, weight = var_261_to_fp16, x = var_252_cast); + tensor x_29_mode_0 = const()[name = tensor("x_29_mode_0"), val = tensor("EXACT")]; + tensor x_29_cast = gelu(mode = x_29_mode_0, x = input_17_cast); + tensor var_267_to_fp16 = const()[name = tensor("op_267_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8142336)))]; + tensor var_268_to_fp16 = const()[name = tensor("op_268_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9322048)))]; + tensor var_269_cast = linear(bias = var_268_to_fp16, weight = var_267_to_fp16, x = x_29_cast); + tensor x_31_cast = add(x = x_25_cast, y = var_269_cast); + tensor var_277 = const()[name = tensor("op_277"), val = tensor(-1)]; + tensor var_294_axes_0 = const()[name = tensor("op_294_axes_0"), val = tensor([-1])]; + tensor blocks_2_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_2_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9322880)))]; + tensor blocks_2_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_2_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9323712)))]; + tensor var_283_to_fp16 = const()[name = tensor("op_283_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_294_cast = layer_norm(axes = var_294_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_283_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast); + tensor var_305_to_fp16 = const()[name = tensor("op_305_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9324544)))]; + tensor var_306_to_fp16 = const()[name = tensor("op_306_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9619520)))]; + tensor q_9_cast = linear(bias = var_306_to_fp16, weight = var_305_to_fp16, x = var_294_cast); + tensor var_309_to_fp16 = const()[name = tensor("op_309_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9620352)))]; + tensor k_9_bias_0_to_fp16 = const()[name = tensor("k_9_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9915328)))]; + tensor k_9_cast = linear(bias = k_9_bias_0_to_fp16, weight = var_309_to_fp16, x = var_294_cast); + tensor var_313_to_fp16 = const()[name = tensor("op_313_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9916160)))]; + tensor var_314_to_fp16 = const()[name = tensor("op_314_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10211136)))]; + tensor v_9_cast = linear(bias = var_314_to_fp16, weight = var_313_to_fp16, x = var_294_cast); + tensor var_322 = const()[name = tensor("op_322"), val = tensor([1, 1500, 6, -1])]; + tensor var_323_cast = reshape(shape = var_322, x = q_9_cast); + tensor const_32_to_fp16 = const()[name = tensor("const_32_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_11_cast = mul(x = var_323_cast, y = const_32_to_fp16); + tensor var_329 = const()[name = tensor("op_329"), val = tensor([1, 1500, 6, -1])]; + tensor var_330_cast = reshape(shape = var_329, x = k_9_cast); + tensor const_33_to_fp16 = const()[name = tensor("const_33_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_11_cast = mul(x = var_330_cast, y = const_33_to_fp16); + tensor var_336 = const()[name = tensor("op_336"), val = tensor([1, 1500, 6, -1])]; + tensor var_337_cast = reshape(shape = var_336, x = v_9_cast); + tensor var_338 = const()[name = tensor("op_338"), val = tensor([0, 2, 1, 3])]; + tensor qk_5_transpose_x_0 = const()[name = tensor("qk_5_transpose_x_0"), val = tensor(false)]; + tensor qk_5_transpose_y_0 = const()[name = tensor("qk_5_transpose_y_0"), val = tensor(false)]; + tensor transpose_12_perm_0 = const()[name = tensor("transpose_12_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_13_perm_0 = const()[name = tensor("transpose_13_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_21 = transpose(perm = transpose_13_perm_0, x = k_11_cast); + tensor transpose_22 = transpose(perm = transpose_12_perm_0, x = q_11_cast); + tensor qk_5_cast = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_22, y = transpose_21); + tensor var_342_cast = softmax(axis = var_277, x = qk_5_cast); + tensor var_344_transpose_x_0 = const()[name = tensor("op_344_transpose_x_0"), val = tensor(false)]; + tensor var_344_transpose_y_0 = const()[name = tensor("op_344_transpose_y_0"), val = tensor(false)]; + tensor transpose_23 = transpose(perm = var_338, x = var_337_cast); + tensor var_344_cast = matmul(transpose_x = var_344_transpose_x_0, transpose_y = var_344_transpose_y_0, x = var_342_cast, y = transpose_23); + tensor var_345 = const()[name = tensor("op_345"), val = tensor([0, 2, 1, 3])]; + tensor concat_2 = const()[name = tensor("concat_2"), val = tensor([1, 1500, 384])]; + tensor transpose_20 = transpose(perm = var_345, x = var_344_cast); + tensor x_35_cast = reshape(shape = concat_2, x = transpose_20); + tensor var_350_to_fp16 = const()[name = tensor("op_350_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10211968)))]; + tensor var_351_to_fp16 = const()[name = tensor("op_351_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10506944)))]; + tensor var_352_cast = linear(bias = var_351_to_fp16, weight = var_350_to_fp16, x = x_35_cast); + tensor x_37_cast = add(x = x_31_cast, y = var_352_cast); + tensor var_358_axes_0 = const()[name = tensor("op_358_axes_0"), val = tensor([-1])]; + tensor blocks_2_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10507776)))]; + tensor blocks_2_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_2_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10508608)))]; + tensor var_358_cast = layer_norm(axes = var_358_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_283_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast); + tensor var_367_to_fp16 = const()[name = tensor("op_367_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10509440)))]; + tensor var_368_to_fp16 = const()[name = tensor("op_368_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11689152)))]; + tensor input_25_cast = linear(bias = var_368_to_fp16, weight = var_367_to_fp16, x = var_358_cast); + tensor x_41_mode_0 = const()[name = tensor("x_41_mode_0"), val = tensor("EXACT")]; + tensor x_41_cast = gelu(mode = x_41_mode_0, x = input_25_cast); + tensor var_373_to_fp16 = const()[name = tensor("op_373_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11692288)))]; + tensor var_374_to_fp16 = const()[name = tensor("op_374_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12872000)))]; + tensor var_375_cast = linear(bias = var_374_to_fp16, weight = var_373_to_fp16, x = x_41_cast); + tensor x_43_cast = add(x = x_37_cast, y = var_375_cast); + tensor var_383 = const()[name = tensor("op_383"), val = tensor(-1)]; + tensor var_400_axes_0 = const()[name = tensor("op_400_axes_0"), val = tensor([-1])]; + tensor blocks_3_attn_ln_weight_to_fp16 = const()[name = tensor("blocks_3_attn_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12872832)))]; + tensor blocks_3_attn_ln_bias_to_fp16 = const()[name = tensor("blocks_3_attn_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12873664)))]; + tensor var_389_to_fp16 = const()[name = tensor("op_389_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_400_cast = layer_norm(axes = var_400_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_389_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast); + tensor var_411_to_fp16 = const()[name = tensor("op_411_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12874496)))]; + tensor var_412_to_fp16 = const()[name = tensor("op_412_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13169472)))]; + tensor q_13_cast = linear(bias = var_412_to_fp16, weight = var_411_to_fp16, x = var_400_cast); + tensor var_415_to_fp16 = const()[name = tensor("op_415_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13170304)))]; + tensor k_13_bias_0_to_fp16 = const()[name = tensor("k_13_bias_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13465280)))]; + tensor k_13_cast = linear(bias = k_13_bias_0_to_fp16, weight = var_415_to_fp16, x = var_400_cast); + tensor var_419_to_fp16 = const()[name = tensor("op_419_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13466112)))]; + tensor var_420_to_fp16 = const()[name = tensor("op_420_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13761088)))]; + tensor v_13_cast = linear(bias = var_420_to_fp16, weight = var_419_to_fp16, x = var_400_cast); + tensor var_428 = const()[name = tensor("op_428"), val = tensor([1, 1500, 6, -1])]; + tensor var_429_cast = reshape(shape = var_428, x = q_13_cast); + tensor const_34_to_fp16 = const()[name = tensor("const_34_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor q_cast = mul(x = var_429_cast, y = const_34_to_fp16); + tensor var_435 = const()[name = tensor("op_435"), val = tensor([1, 1500, 6, -1])]; + tensor var_436_cast = reshape(shape = var_435, x = k_13_cast); + tensor const_35_to_fp16 = const()[name = tensor("const_35_to_fp16"), val = tensor([[[[0x1.6ap-2]]]])]; + tensor k_cast = mul(x = var_436_cast, y = const_35_to_fp16); + tensor var_442 = const()[name = tensor("op_442"), val = tensor([1, 1500, 6, -1])]; + tensor var_443_cast = reshape(shape = var_442, x = v_13_cast); + tensor var_444 = const()[name = tensor("op_444"), val = tensor([0, 2, 1, 3])]; + tensor qk_transpose_x_0 = const()[name = tensor("qk_transpose_x_0"), val = tensor(false)]; + tensor qk_transpose_y_0 = const()[name = tensor("qk_transpose_y_0"), val = tensor(false)]; + tensor transpose_14_perm_0 = const()[name = tensor("transpose_14_perm_0"), val = tensor([0, 2, 1, 3])]; + tensor transpose_15_perm_0 = const()[name = tensor("transpose_15_perm_0"), val = tensor([0, 2, 3, 1])]; + tensor transpose_17 = transpose(perm = transpose_15_perm_0, x = k_cast); + tensor transpose_18 = transpose(perm = transpose_14_perm_0, x = q_cast); + tensor qk_cast = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_18, y = transpose_17); + tensor var_448_cast = softmax(axis = var_383, x = qk_cast); + tensor var_450_transpose_x_0 = const()[name = tensor("op_450_transpose_x_0"), val = tensor(false)]; + tensor var_450_transpose_y_0 = const()[name = tensor("op_450_transpose_y_0"), val = tensor(false)]; + tensor transpose_19 = transpose(perm = var_444, x = var_443_cast); + tensor var_450_cast = matmul(transpose_x = var_450_transpose_x_0, transpose_y = var_450_transpose_y_0, x = var_448_cast, y = transpose_19); + tensor var_451 = const()[name = tensor("op_451"), val = tensor([0, 2, 1, 3])]; + tensor concat_3 = const()[name = tensor("concat_3"), val = tensor([1, 1500, 384])]; + tensor transpose_16 = transpose(perm = var_451, x = var_450_cast); + tensor x_47_cast = reshape(shape = concat_3, x = transpose_16); + tensor var_456_to_fp16 = const()[name = tensor("op_456_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13761920)))]; + tensor var_457_to_fp16 = const()[name = tensor("op_457_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14056896)))]; + tensor var_458_cast = linear(bias = var_457_to_fp16, weight = var_456_to_fp16, x = x_47_cast); + tensor x_49_cast = add(x = x_43_cast, y = var_458_cast); + tensor var_464_axes_0 = const()[name = tensor("op_464_axes_0"), val = tensor([-1])]; + tensor blocks_3_mlp_ln_weight_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14057728)))]; + tensor blocks_3_mlp_ln_bias_to_fp16 = const()[name = tensor("blocks_3_mlp_ln_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14058560)))]; + tensor var_464_cast = layer_norm(axes = var_464_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_389_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast); + tensor var_473_to_fp16 = const()[name = tensor("op_473_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14059392)))]; + tensor var_474_to_fp16 = const()[name = tensor("op_474_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15239104)))]; + tensor input_33_cast = linear(bias = var_474_to_fp16, weight = var_473_to_fp16, x = var_464_cast); + tensor x_53_mode_0 = const()[name = tensor("x_53_mode_0"), val = tensor("EXACT")]; + tensor x_53_cast = gelu(mode = x_53_mode_0, x = input_33_cast); + tensor var_479_to_fp16 = const()[name = tensor("op_479_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15242240)))]; + tensor var_480_to_fp16 = const()[name = tensor("op_480_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16421952)))]; + tensor var_481_cast = linear(bias = var_480_to_fp16, weight = var_479_to_fp16, x = x_53_cast); + tensor x_cast = add(x = x_49_cast, y = var_481_cast); + tensor var_494_axes_0 = const()[name = tensor("op_494_axes_0"), val = tensor([-1])]; + tensor ln_post_weight_to_fp16 = const()[name = tensor("ln_post_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16422784)))]; + tensor ln_post_bias_to_fp16 = const()[name = tensor("ln_post_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16423616)))]; + tensor var_485_to_fp16 = const()[name = tensor("op_485_to_fp16"), val = tensor(0x1.5p-17)]; + tensor var_494_cast = layer_norm(axes = var_494_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_485_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast); + tensor var_494_cast_to_fp32_dtype_0 = const()[name = tensor("op_494_cast_to_fp32_dtype_0"), val = tensor("fp32")]; + tensor output = cast(dtype = var_494_cast_to_fp32_dtype_0, x = var_494_cast); + } -> (output); +} \ No newline at end of file diff --git a/ggml-tiny.en-encoder.mlmodelc/weights/weight.bin b/ggml-tiny.en-encoder.mlmodelc/weights/weight.bin new file mode 100644 index 0000000000000000000000000000000000000000..6c5a335a3ef0654b49359ba603cc1449c90a7e1e --- /dev/null +++ b/ggml-tiny.en-encoder.mlmodelc/weights/weight.bin @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c59faa13544fde24098d9594a3f955d0515b4aa8f3e8a60cd1c022e76bba8d31 +size 16424448