| program(1.0) | |
| [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3304.5.2"}, {"coremlc-version", "3304.6.2"}, {"coremltools-component-torch", "2.2.2"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "7.2"}})] | |
| { | |
| func main<ios17>(tensor<fp16, [1, 80, 3000]> logmel_data) { | |
| tensor<int32, []> var_16 = const()[name = tensor<string, []>("op_16"), val = tensor<int32, []>(1)]; | |
| tensor<int32, [1]> var_24 = const()[name = tensor<string, []>("op_24"), val = tensor<int32, [1]>([1])]; | |
| tensor<int32, [1]> var_26 = const()[name = tensor<string, []>("op_26"), val = tensor<int32, [1]>([1])]; | |
| tensor<string, []> var_28_pad_type_0 = const()[name = tensor<string, []>("op_28_pad_type_0"), val = tensor<string, []>("custom")]; | |
| tensor<int32, [2]> var_28_pad_0 = const()[name = tensor<string, []>("op_28_pad_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<fp16, [384, 80, 3]> weight_3_to_fp16 = const()[name = tensor<string, []>("weight_3_to_fp16"), val = tensor<fp16, [384, 80, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))]; | |
| tensor<fp16, [384]> bias_3_to_fp16 = const()[name = tensor<string, []>("bias_3_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(184448)))]; | |
| tensor<fp16, [1, 384, 3000]> var_28_cast_fp16 = 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 = logmel_data)[name = tensor<string, []>("op_28_cast_fp16")]; | |
| tensor<string, []> input_1_mode_0 = const()[name = tensor<string, []>("input_1_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 384, 3000]> input_1_cast_fp16 = gelu(mode = input_1_mode_0, x = var_28_cast_fp16)[name = tensor<string, []>("input_1_cast_fp16")]; | |
| tensor<int32, []> var_33 = const()[name = tensor<string, []>("op_33"), val = tensor<int32, []>(1)]; | |
| tensor<int32, [1]> var_42 = const()[name = tensor<string, []>("op_42"), val = tensor<int32, [1]>([2])]; | |
| tensor<int32, [1]> var_44 = const()[name = tensor<string, []>("op_44"), val = tensor<int32, [1]>([1])]; | |
| tensor<string, []> var_46_pad_type_0 = const()[name = tensor<string, []>("op_46_pad_type_0"), val = tensor<string, []>("custom")]; | |
| tensor<int32, [2]> var_46_pad_0 = const()[name = tensor<string, []>("op_46_pad_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<fp16, [384, 384, 3]> weight_7_to_fp16 = const()[name = tensor<string, []>("weight_7_to_fp16"), val = tensor<fp16, [384, 384, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(185280)))]; | |
| tensor<fp16, [384]> bias_7_to_fp16 = const()[name = tensor<string, []>("bias_7_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1070080)))]; | |
| tensor<fp16, [1, 384, 1500]> var_46_cast_fp16 = conv(bias = bias_7_to_fp16, dilations = var_44, groups = var_33, pad = var_46_pad_0, pad_type = var_46_pad_type_0, strides = var_42, weight = weight_7_to_fp16, x = input_1_cast_fp16)[name = tensor<string, []>("op_46_cast_fp16")]; | |
| tensor<string, []> x_3_mode_0 = const()[name = tensor<string, []>("x_3_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 384, 1500]> x_3_cast_fp16 = gelu(mode = x_3_mode_0, x = var_46_cast_fp16)[name = tensor<string, []>("x_3_cast_fp16")]; | |
| tensor<int32, [3]> var_52 = const()[name = tensor<string, []>("op_52"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<fp16, [1500, 384]> positional_embedding_to_fp16 = const()[name = tensor<string, []>("positional_embedding_to_fp16"), val = tensor<fp16, [1500, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1070912)))]; | |
| tensor<fp16, [1, 1500, 384]> transpose_40 = transpose(perm = var_52, x = x_3_cast_fp16)[name = tensor<string, []>("transpose_40")]; | |
| tensor<fp16, [1, 1500, 384]> var_55_cast_fp16 = add(x = transpose_40, y = positional_embedding_to_fp16)[name = tensor<string, []>("op_55_cast_fp16")]; | |
| tensor<int32, []> var_67 = const()[name = tensor<string, []>("op_67"), val = tensor<int32, []>(-1)]; | |
| tensor<int32, [1]> var_83_axes_0 = const()[name = tensor<string, []>("op_83_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [384]> blocks_0_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_0_attn_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2222976)))]; | |
| tensor<fp16, [384]> blocks_0_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_0_attn_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2223808)))]; | |
| tensor<fp16, []> var_73_to_fp16 = const()[name = tensor<string, []>("op_73_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 384]> var_83_cast_fp16 = layer_norm(axes = var_83_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_73_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_55_cast_fp16)[name = tensor<string, []>("op_83_cast_fp16")]; | |
| tensor<fp16, [384, 384]> var_94_to_fp16 = const()[name = tensor<string, []>("op_94_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2224640)))]; | |
| tensor<fp16, [384]> var_95_to_fp16 = const()[name = tensor<string, []>("op_95_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2519616)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_0_cast_fp16 = linear(bias = var_95_to_fp16, weight = var_94_to_fp16, x = var_83_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")]; | |
| tensor<fp16, [384, 384]> var_98_to_fp16 = const()[name = tensor<string, []>("op_98_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2520448)))]; | |
| tensor<fp16, [384]> linear_1_bias_0_to_fp16 = const()[name = tensor<string, []>("linear_1_bias_0_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2815424)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_98_to_fp16, x = var_83_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")]; | |
| tensor<fp16, [384, 384]> var_102_to_fp16 = const()[name = tensor<string, []>("op_102_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2816256)))]; | |
| tensor<fp16, [384]> var_103_to_fp16 = const()[name = tensor<string, []>("op_103_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3111232)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_2_cast_fp16 = linear(bias = var_103_to_fp16, weight = var_102_to_fp16, x = var_83_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")]; | |
| tensor<int32, [4]> var_111 = const()[name = tensor<string, []>("op_111"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_112_cast_fp16 = reshape(shape = var_111, x = linear_0_cast_fp16)[name = tensor<string, []>("op_112_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_28_to_fp16 = const()[name = tensor<string, []>("const_28_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 6, 64]> q_3_cast_fp16 = mul(x = var_112_cast_fp16, y = const_28_to_fp16)[name = tensor<string, []>("q_3_cast_fp16")]; | |
| tensor<int32, [4]> var_118 = const()[name = tensor<string, []>("op_118"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_119_cast_fp16 = reshape(shape = var_118, x = linear_1_cast_fp16)[name = tensor<string, []>("op_119_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_29_to_fp16 = const()[name = tensor<string, []>("const_29_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 6, 64]> k_3_cast_fp16 = mul(x = var_119_cast_fp16, y = const_29_to_fp16)[name = tensor<string, []>("k_3_cast_fp16")]; | |
| tensor<int32, [4]> var_125 = const()[name = tensor<string, []>("op_125"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_126_cast_fp16 = reshape(shape = var_125, x = linear_2_cast_fp16)[name = tensor<string, []>("op_126_cast_fp16")]; | |
| tensor<int32, [4]> var_127 = const()[name = tensor<string, []>("op_127"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<bool, []> qk_1_transpose_x_0 = const()[name = tensor<string, []>("qk_1_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> qk_1_transpose_y_0 = const()[name = tensor<string, []>("qk_1_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<int32, [4]> transpose_16_perm_0 = const()[name = tensor<string, []>("transpose_16_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_17_perm_0 = const()[name = tensor<string, []>("transpose_17_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 6, 64, 1500]> transpose_37 = transpose(perm = transpose_17_perm_0, x = k_3_cast_fp16)[name = tensor<string, []>("transpose_37")]; | |
| tensor<fp16, [1, 6, 1500, 64]> transpose_38 = transpose(perm = transpose_16_perm_0, x = q_3_cast_fp16)[name = tensor<string, []>("transpose_38")]; | |
| tensor<fp16, [1, 6, 1500, 1500]> qk_1_cast_fp16 = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_38, y = transpose_37)[name = tensor<string, []>("qk_1_cast_fp16")]; | |
| tensor<fp16, [1, 6, 1500, 1500]> var_131_cast_fp16 = softmax(axis = var_67, x = qk_1_cast_fp16)[name = tensor<string, []>("op_131_cast_fp16")]; | |
| tensor<bool, []> var_133_transpose_x_0 = const()[name = tensor<string, []>("op_133_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> var_133_transpose_y_0 = const()[name = tensor<string, []>("op_133_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 6, 1500, 64]> transpose_39 = transpose(perm = var_127, x = var_126_cast_fp16)[name = tensor<string, []>("transpose_39")]; | |
| tensor<fp16, [1, 6, 1500, 64]> var_133_cast_fp16 = matmul(transpose_x = var_133_transpose_x_0, transpose_y = var_133_transpose_y_0, x = var_131_cast_fp16, y = transpose_39)[name = tensor<string, []>("op_133_cast_fp16")]; | |
| tensor<int32, [4]> var_134 = const()[name = tensor<string, []>("op_134"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_0 = const()[name = tensor<string, []>("concat_0"), val = tensor<int32, [3]>([1, 1500, 384])]; | |
| tensor<fp16, [1, 1500, 6, 64]> transpose_36 = transpose(perm = var_134, x = var_133_cast_fp16)[name = tensor<string, []>("transpose_36")]; | |
| tensor<fp16, [1, 1500, 384]> x_11_cast_fp16 = reshape(shape = concat_0, x = transpose_36)[name = tensor<string, []>("x_11_cast_fp16")]; | |
| tensor<fp16, [384, 384]> var_139_to_fp16 = const()[name = tensor<string, []>("op_139_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3112064)))]; | |
| tensor<fp16, [384]> var_140_to_fp16 = const()[name = tensor<string, []>("op_140_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3407040)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_3_cast_fp16 = linear(bias = var_140_to_fp16, weight = var_139_to_fp16, x = x_11_cast_fp16)[name = tensor<string, []>("linear_3_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 384]> x_13_cast_fp16 = add(x = var_55_cast_fp16, y = linear_3_cast_fp16)[name = tensor<string, []>("x_13_cast_fp16")]; | |
| tensor<int32, [1]> var_147_axes_0 = const()[name = tensor<string, []>("op_147_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [384]> blocks_0_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_0_mlp_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3407872)))]; | |
| tensor<fp16, [384]> blocks_0_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_0_mlp_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3408704)))]; | |
| tensor<fp16, [1, 1500, 384]> var_147_cast_fp16 = layer_norm(axes = var_147_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_73_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast_fp16)[name = tensor<string, []>("op_147_cast_fp16")]; | |
| tensor<fp16, [1536, 384]> var_156_to_fp16 = const()[name = tensor<string, []>("op_156_to_fp16"), val = tensor<fp16, [1536, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3409536)))]; | |
| tensor<fp16, [1536]> var_157_to_fp16 = const()[name = tensor<string, []>("op_157_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4589248)))]; | |
| tensor<fp16, [1, 1500, 1536]> linear_4_cast_fp16 = linear(bias = var_157_to_fp16, weight = var_156_to_fp16, x = var_147_cast_fp16)[name = tensor<string, []>("linear_4_cast_fp16")]; | |
| tensor<string, []> x_17_mode_0 = const()[name = tensor<string, []>("x_17_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 1500, 1536]> x_17_cast_fp16 = gelu(mode = x_17_mode_0, x = linear_4_cast_fp16)[name = tensor<string, []>("x_17_cast_fp16")]; | |
| tensor<fp16, [384, 1536]> var_162_to_fp16 = const()[name = tensor<string, []>("op_162_to_fp16"), val = tensor<fp16, [384, 1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4592384)))]; | |
| tensor<fp16, [384]> var_163_to_fp16 = const()[name = tensor<string, []>("op_163_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5772096)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_5_cast_fp16 = linear(bias = var_163_to_fp16, weight = var_162_to_fp16, x = x_17_cast_fp16)[name = tensor<string, []>("linear_5_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 384]> x_19_cast_fp16 = add(x = x_13_cast_fp16, y = linear_5_cast_fp16)[name = tensor<string, []>("x_19_cast_fp16")]; | |
| tensor<int32, []> var_172 = const()[name = tensor<string, []>("op_172"), val = tensor<int32, []>(-1)]; | |
| tensor<int32, [1]> var_188_axes_0 = const()[name = tensor<string, []>("op_188_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [384]> blocks_1_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_1_attn_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5772928)))]; | |
| tensor<fp16, [384]> blocks_1_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_1_attn_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5773760)))]; | |
| tensor<fp16, []> var_178_to_fp16 = const()[name = tensor<string, []>("op_178_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 384]> var_188_cast_fp16 = layer_norm(axes = var_188_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_178_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast_fp16)[name = tensor<string, []>("op_188_cast_fp16")]; | |
| tensor<fp16, [384, 384]> var_199_to_fp16 = const()[name = tensor<string, []>("op_199_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5774592)))]; | |
| tensor<fp16, [384]> var_200_to_fp16 = const()[name = tensor<string, []>("op_200_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6069568)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_6_cast_fp16 = linear(bias = var_200_to_fp16, weight = var_199_to_fp16, x = var_188_cast_fp16)[name = tensor<string, []>("linear_6_cast_fp16")]; | |
| tensor<fp16, [384, 384]> var_203_to_fp16 = const()[name = tensor<string, []>("op_203_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6070400)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_7_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_203_to_fp16, x = var_188_cast_fp16)[name = tensor<string, []>("linear_7_cast_fp16")]; | |
| tensor<fp16, [384, 384]> var_207_to_fp16 = const()[name = tensor<string, []>("op_207_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6365376)))]; | |
| tensor<fp16, [384]> var_208_to_fp16 = const()[name = tensor<string, []>("op_208_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6660352)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_8_cast_fp16 = linear(bias = var_208_to_fp16, weight = var_207_to_fp16, x = var_188_cast_fp16)[name = tensor<string, []>("linear_8_cast_fp16")]; | |
| tensor<int32, [4]> var_216 = const()[name = tensor<string, []>("op_216"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_217_cast_fp16 = reshape(shape = var_216, x = linear_6_cast_fp16)[name = tensor<string, []>("op_217_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_30_to_fp16 = const()[name = tensor<string, []>("const_30_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 6, 64]> q_7_cast_fp16 = mul(x = var_217_cast_fp16, y = const_30_to_fp16)[name = tensor<string, []>("q_7_cast_fp16")]; | |
| tensor<int32, [4]> var_223 = const()[name = tensor<string, []>("op_223"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_224_cast_fp16 = reshape(shape = var_223, x = linear_7_cast_fp16)[name = tensor<string, []>("op_224_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_31_to_fp16 = const()[name = tensor<string, []>("const_31_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 6, 64]> k_7_cast_fp16 = mul(x = var_224_cast_fp16, y = const_31_to_fp16)[name = tensor<string, []>("k_7_cast_fp16")]; | |
| tensor<int32, [4]> var_230 = const()[name = tensor<string, []>("op_230"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_231_cast_fp16 = reshape(shape = var_230, x = linear_8_cast_fp16)[name = tensor<string, []>("op_231_cast_fp16")]; | |
| tensor<int32, [4]> var_232 = const()[name = tensor<string, []>("op_232"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<bool, []> qk_3_transpose_x_0 = const()[name = tensor<string, []>("qk_3_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> qk_3_transpose_y_0 = const()[name = tensor<string, []>("qk_3_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<int32, [4]> transpose_18_perm_0 = const()[name = tensor<string, []>("transpose_18_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_19_perm_0 = const()[name = tensor<string, []>("transpose_19_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 6, 64, 1500]> transpose_33 = transpose(perm = transpose_19_perm_0, x = k_7_cast_fp16)[name = tensor<string, []>("transpose_33")]; | |
| tensor<fp16, [1, 6, 1500, 64]> transpose_34 = transpose(perm = transpose_18_perm_0, x = q_7_cast_fp16)[name = tensor<string, []>("transpose_34")]; | |
| tensor<fp16, [1, 6, 1500, 1500]> qk_3_cast_fp16 = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_34, y = transpose_33)[name = tensor<string, []>("qk_3_cast_fp16")]; | |
| tensor<fp16, [1, 6, 1500, 1500]> var_236_cast_fp16 = softmax(axis = var_172, x = qk_3_cast_fp16)[name = tensor<string, []>("op_236_cast_fp16")]; | |
| tensor<bool, []> var_238_transpose_x_0 = const()[name = tensor<string, []>("op_238_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> var_238_transpose_y_0 = const()[name = tensor<string, []>("op_238_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 6, 1500, 64]> transpose_35 = transpose(perm = var_232, x = var_231_cast_fp16)[name = tensor<string, []>("transpose_35")]; | |
| tensor<fp16, [1, 6, 1500, 64]> var_238_cast_fp16 = matmul(transpose_x = var_238_transpose_x_0, transpose_y = var_238_transpose_y_0, x = var_236_cast_fp16, y = transpose_35)[name = tensor<string, []>("op_238_cast_fp16")]; | |
| tensor<int32, [4]> var_239 = const()[name = tensor<string, []>("op_239"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_1 = const()[name = tensor<string, []>("concat_1"), val = tensor<int32, [3]>([1, 1500, 384])]; | |
| tensor<fp16, [1, 1500, 6, 64]> transpose_32 = transpose(perm = var_239, x = var_238_cast_fp16)[name = tensor<string, []>("transpose_32")]; | |
| tensor<fp16, [1, 1500, 384]> x_23_cast_fp16 = reshape(shape = concat_1, x = transpose_32)[name = tensor<string, []>("x_23_cast_fp16")]; | |
| tensor<fp16, [384, 384]> var_244_to_fp16 = const()[name = tensor<string, []>("op_244_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6661184)))]; | |
| tensor<fp16, [384]> var_245_to_fp16 = const()[name = tensor<string, []>("op_245_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6956160)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_9_cast_fp16 = linear(bias = var_245_to_fp16, weight = var_244_to_fp16, x = x_23_cast_fp16)[name = tensor<string, []>("linear_9_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 384]> x_25_cast_fp16 = add(x = x_19_cast_fp16, y = linear_9_cast_fp16)[name = tensor<string, []>("x_25_cast_fp16")]; | |
| tensor<int32, [1]> var_252_axes_0 = const()[name = tensor<string, []>("op_252_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [384]> blocks_1_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_1_mlp_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6956992)))]; | |
| tensor<fp16, [384]> blocks_1_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_1_mlp_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6957824)))]; | |
| tensor<fp16, [1, 1500, 384]> var_252_cast_fp16 = layer_norm(axes = var_252_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_178_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast_fp16)[name = tensor<string, []>("op_252_cast_fp16")]; | |
| tensor<fp16, [1536, 384]> var_261_to_fp16 = const()[name = tensor<string, []>("op_261_to_fp16"), val = tensor<fp16, [1536, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6958656)))]; | |
| tensor<fp16, [1536]> var_262_to_fp16 = const()[name = tensor<string, []>("op_262_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8138368)))]; | |
| tensor<fp16, [1, 1500, 1536]> linear_10_cast_fp16 = linear(bias = var_262_to_fp16, weight = var_261_to_fp16, x = var_252_cast_fp16)[name = tensor<string, []>("linear_10_cast_fp16")]; | |
| tensor<string, []> x_29_mode_0 = const()[name = tensor<string, []>("x_29_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 1500, 1536]> x_29_cast_fp16 = gelu(mode = x_29_mode_0, x = linear_10_cast_fp16)[name = tensor<string, []>("x_29_cast_fp16")]; | |
| tensor<fp16, [384, 1536]> var_267_to_fp16 = const()[name = tensor<string, []>("op_267_to_fp16"), val = tensor<fp16, [384, 1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8141504)))]; | |
| tensor<fp16, [384]> var_268_to_fp16 = const()[name = tensor<string, []>("op_268_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9321216)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_11_cast_fp16 = linear(bias = var_268_to_fp16, weight = var_267_to_fp16, x = x_29_cast_fp16)[name = tensor<string, []>("linear_11_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 384]> x_31_cast_fp16 = add(x = x_25_cast_fp16, y = linear_11_cast_fp16)[name = tensor<string, []>("x_31_cast_fp16")]; | |
| tensor<int32, []> var_277 = const()[name = tensor<string, []>("op_277"), val = tensor<int32, []>(-1)]; | |
| tensor<int32, [1]> var_293_axes_0 = const()[name = tensor<string, []>("op_293_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [384]> blocks_2_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_2_attn_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9322048)))]; | |
| tensor<fp16, [384]> blocks_2_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_2_attn_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9322880)))]; | |
| tensor<fp16, []> var_283_to_fp16 = const()[name = tensor<string, []>("op_283_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 384]> var_293_cast_fp16 = layer_norm(axes = var_293_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_fp16)[name = tensor<string, []>("op_293_cast_fp16")]; | |
| tensor<fp16, [384, 384]> var_304_to_fp16 = const()[name = tensor<string, []>("op_304_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9323712)))]; | |
| tensor<fp16, [384]> var_305_to_fp16 = const()[name = tensor<string, []>("op_305_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9618688)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_12_cast_fp16 = linear(bias = var_305_to_fp16, weight = var_304_to_fp16, x = var_293_cast_fp16)[name = tensor<string, []>("linear_12_cast_fp16")]; | |
| tensor<fp16, [384, 384]> var_308_to_fp16 = const()[name = tensor<string, []>("op_308_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9619520)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_13_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_308_to_fp16, x = var_293_cast_fp16)[name = tensor<string, []>("linear_13_cast_fp16")]; | |
| tensor<fp16, [384, 384]> var_312_to_fp16 = const()[name = tensor<string, []>("op_312_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9914496)))]; | |
| tensor<fp16, [384]> var_313_to_fp16 = const()[name = tensor<string, []>("op_313_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10209472)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_14_cast_fp16 = linear(bias = var_313_to_fp16, weight = var_312_to_fp16, x = var_293_cast_fp16)[name = tensor<string, []>("linear_14_cast_fp16")]; | |
| tensor<int32, [4]> var_321 = const()[name = tensor<string, []>("op_321"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_322_cast_fp16 = reshape(shape = var_321, x = linear_12_cast_fp16)[name = tensor<string, []>("op_322_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_32_to_fp16 = const()[name = tensor<string, []>("const_32_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 6, 64]> q_11_cast_fp16 = mul(x = var_322_cast_fp16, y = const_32_to_fp16)[name = tensor<string, []>("q_11_cast_fp16")]; | |
| tensor<int32, [4]> var_328 = const()[name = tensor<string, []>("op_328"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_329_cast_fp16 = reshape(shape = var_328, x = linear_13_cast_fp16)[name = tensor<string, []>("op_329_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_33_to_fp16 = const()[name = tensor<string, []>("const_33_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 6, 64]> k_11_cast_fp16 = mul(x = var_329_cast_fp16, y = const_33_to_fp16)[name = tensor<string, []>("k_11_cast_fp16")]; | |
| tensor<int32, [4]> var_335 = const()[name = tensor<string, []>("op_335"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_336_cast_fp16 = reshape(shape = var_335, x = linear_14_cast_fp16)[name = tensor<string, []>("op_336_cast_fp16")]; | |
| tensor<int32, [4]> var_337 = const()[name = tensor<string, []>("op_337"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<bool, []> qk_5_transpose_x_0 = const()[name = tensor<string, []>("qk_5_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> qk_5_transpose_y_0 = const()[name = tensor<string, []>("qk_5_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<int32, [4]> transpose_20_perm_0 = const()[name = tensor<string, []>("transpose_20_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_21_perm_0 = const()[name = tensor<string, []>("transpose_21_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 6, 64, 1500]> transpose_29 = transpose(perm = transpose_21_perm_0, x = k_11_cast_fp16)[name = tensor<string, []>("transpose_29")]; | |
| tensor<fp16, [1, 6, 1500, 64]> transpose_30 = transpose(perm = transpose_20_perm_0, x = q_11_cast_fp16)[name = tensor<string, []>("transpose_30")]; | |
| tensor<fp16, [1, 6, 1500, 1500]> qk_5_cast_fp16 = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_30, y = transpose_29)[name = tensor<string, []>("qk_5_cast_fp16")]; | |
| tensor<fp16, [1, 6, 1500, 1500]> var_341_cast_fp16 = softmax(axis = var_277, x = qk_5_cast_fp16)[name = tensor<string, []>("op_341_cast_fp16")]; | |
| tensor<bool, []> var_343_transpose_x_0 = const()[name = tensor<string, []>("op_343_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> var_343_transpose_y_0 = const()[name = tensor<string, []>("op_343_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 6, 1500, 64]> transpose_31 = transpose(perm = var_337, x = var_336_cast_fp16)[name = tensor<string, []>("transpose_31")]; | |
| tensor<fp16, [1, 6, 1500, 64]> var_343_cast_fp16 = matmul(transpose_x = var_343_transpose_x_0, transpose_y = var_343_transpose_y_0, x = var_341_cast_fp16, y = transpose_31)[name = tensor<string, []>("op_343_cast_fp16")]; | |
| tensor<int32, [4]> var_344 = const()[name = tensor<string, []>("op_344"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_2 = const()[name = tensor<string, []>("concat_2"), val = tensor<int32, [3]>([1, 1500, 384])]; | |
| tensor<fp16, [1, 1500, 6, 64]> transpose_28 = transpose(perm = var_344, x = var_343_cast_fp16)[name = tensor<string, []>("transpose_28")]; | |
| tensor<fp16, [1, 1500, 384]> x_35_cast_fp16 = reshape(shape = concat_2, x = transpose_28)[name = tensor<string, []>("x_35_cast_fp16")]; | |
| tensor<fp16, [384, 384]> var_349_to_fp16 = const()[name = tensor<string, []>("op_349_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10210304)))]; | |
| tensor<fp16, [384]> var_350_to_fp16 = const()[name = tensor<string, []>("op_350_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10505280)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_15_cast_fp16 = linear(bias = var_350_to_fp16, weight = var_349_to_fp16, x = x_35_cast_fp16)[name = tensor<string, []>("linear_15_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 384]> x_37_cast_fp16 = add(x = x_31_cast_fp16, y = linear_15_cast_fp16)[name = tensor<string, []>("x_37_cast_fp16")]; | |
| tensor<int32, [1]> var_357_axes_0 = const()[name = tensor<string, []>("op_357_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [384]> blocks_2_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_2_mlp_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10506112)))]; | |
| tensor<fp16, [384]> blocks_2_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_2_mlp_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10506944)))]; | |
| tensor<fp16, [1, 1500, 384]> var_357_cast_fp16 = layer_norm(axes = var_357_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_fp16)[name = tensor<string, []>("op_357_cast_fp16")]; | |
| tensor<fp16, [1536, 384]> var_366_to_fp16 = const()[name = tensor<string, []>("op_366_to_fp16"), val = tensor<fp16, [1536, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10507776)))]; | |
| tensor<fp16, [1536]> var_367_to_fp16 = const()[name = tensor<string, []>("op_367_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11687488)))]; | |
| tensor<fp16, [1, 1500, 1536]> linear_16_cast_fp16 = linear(bias = var_367_to_fp16, weight = var_366_to_fp16, x = var_357_cast_fp16)[name = tensor<string, []>("linear_16_cast_fp16")]; | |
| tensor<string, []> x_41_mode_0 = const()[name = tensor<string, []>("x_41_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 1500, 1536]> x_41_cast_fp16 = gelu(mode = x_41_mode_0, x = linear_16_cast_fp16)[name = tensor<string, []>("x_41_cast_fp16")]; | |
| tensor<fp16, [384, 1536]> var_372_to_fp16 = const()[name = tensor<string, []>("op_372_to_fp16"), val = tensor<fp16, [384, 1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11690624)))]; | |
| tensor<fp16, [384]> var_373_to_fp16 = const()[name = tensor<string, []>("op_373_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12870336)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_17_cast_fp16 = linear(bias = var_373_to_fp16, weight = var_372_to_fp16, x = x_41_cast_fp16)[name = tensor<string, []>("linear_17_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 384]> x_43_cast_fp16 = add(x = x_37_cast_fp16, y = linear_17_cast_fp16)[name = tensor<string, []>("x_43_cast_fp16")]; | |
| tensor<int32, []> var_382 = const()[name = tensor<string, []>("op_382"), val = tensor<int32, []>(-1)]; | |
| tensor<int32, [1]> var_398_axes_0 = const()[name = tensor<string, []>("op_398_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [384]> blocks_3_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_3_attn_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12871168)))]; | |
| tensor<fp16, [384]> blocks_3_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_3_attn_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12872000)))]; | |
| tensor<fp16, []> var_388_to_fp16 = const()[name = tensor<string, []>("op_388_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 384]> var_398_cast_fp16 = layer_norm(axes = var_398_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_388_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast_fp16)[name = tensor<string, []>("op_398_cast_fp16")]; | |
| tensor<fp16, [384, 384]> var_409_to_fp16 = const()[name = tensor<string, []>("op_409_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12872832)))]; | |
| tensor<fp16, [384]> var_410_to_fp16 = const()[name = tensor<string, []>("op_410_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13167808)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_18_cast_fp16 = linear(bias = var_410_to_fp16, weight = var_409_to_fp16, x = var_398_cast_fp16)[name = tensor<string, []>("linear_18_cast_fp16")]; | |
| tensor<fp16, [384, 384]> var_413_to_fp16 = const()[name = tensor<string, []>("op_413_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13168640)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_413_to_fp16, x = var_398_cast_fp16)[name = tensor<string, []>("linear_19_cast_fp16")]; | |
| tensor<fp16, [384, 384]> var_417_to_fp16 = const()[name = tensor<string, []>("op_417_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13463616)))]; | |
| tensor<fp16, [384]> var_418_to_fp16 = const()[name = tensor<string, []>("op_418_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13758592)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_20_cast_fp16 = linear(bias = var_418_to_fp16, weight = var_417_to_fp16, x = var_398_cast_fp16)[name = tensor<string, []>("linear_20_cast_fp16")]; | |
| tensor<int32, [4]> var_426 = const()[name = tensor<string, []>("op_426"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_427_cast_fp16 = reshape(shape = var_426, x = linear_18_cast_fp16)[name = tensor<string, []>("op_427_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_34_to_fp16 = const()[name = tensor<string, []>("const_34_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 6, 64]> q_cast_fp16 = mul(x = var_427_cast_fp16, y = const_34_to_fp16)[name = tensor<string, []>("q_cast_fp16")]; | |
| tensor<int32, [4]> var_433 = const()[name = tensor<string, []>("op_433"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_434_cast_fp16 = reshape(shape = var_433, x = linear_19_cast_fp16)[name = tensor<string, []>("op_434_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_35_to_fp16 = const()[name = tensor<string, []>("const_35_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 6, 64]> k_cast_fp16 = mul(x = var_434_cast_fp16, y = const_35_to_fp16)[name = tensor<string, []>("k_cast_fp16")]; | |
| tensor<int32, [4]> var_440 = const()[name = tensor<string, []>("op_440"), val = tensor<int32, [4]>([1, 1500, 6, -1])]; | |
| tensor<fp16, [1, 1500, 6, 64]> var_441_cast_fp16 = reshape(shape = var_440, x = linear_20_cast_fp16)[name = tensor<string, []>("op_441_cast_fp16")]; | |
| tensor<int32, [4]> var_442 = const()[name = tensor<string, []>("op_442"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<bool, []> qk_transpose_x_0 = const()[name = tensor<string, []>("qk_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> qk_transpose_y_0 = const()[name = tensor<string, []>("qk_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<int32, [4]> transpose_22_perm_0 = const()[name = tensor<string, []>("transpose_22_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_23_perm_0 = const()[name = tensor<string, []>("transpose_23_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 6, 64, 1500]> transpose_25 = transpose(perm = transpose_23_perm_0, x = k_cast_fp16)[name = tensor<string, []>("transpose_25")]; | |
| tensor<fp16, [1, 6, 1500, 64]> transpose_26 = transpose(perm = transpose_22_perm_0, x = q_cast_fp16)[name = tensor<string, []>("transpose_26")]; | |
| tensor<fp16, [1, 6, 1500, 1500]> qk_cast_fp16 = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_26, y = transpose_25)[name = tensor<string, []>("qk_cast_fp16")]; | |
| tensor<fp16, [1, 6, 1500, 1500]> var_446_cast_fp16 = softmax(axis = var_382, x = qk_cast_fp16)[name = tensor<string, []>("op_446_cast_fp16")]; | |
| tensor<bool, []> var_448_transpose_x_0 = const()[name = tensor<string, []>("op_448_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> var_448_transpose_y_0 = const()[name = tensor<string, []>("op_448_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 6, 1500, 64]> transpose_27 = transpose(perm = var_442, x = var_441_cast_fp16)[name = tensor<string, []>("transpose_27")]; | |
| tensor<fp16, [1, 6, 1500, 64]> var_448_cast_fp16 = matmul(transpose_x = var_448_transpose_x_0, transpose_y = var_448_transpose_y_0, x = var_446_cast_fp16, y = transpose_27)[name = tensor<string, []>("op_448_cast_fp16")]; | |
| tensor<int32, [4]> var_449 = const()[name = tensor<string, []>("op_449"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_3 = const()[name = tensor<string, []>("concat_3"), val = tensor<int32, [3]>([1, 1500, 384])]; | |
| tensor<fp16, [1, 1500, 6, 64]> transpose_24 = transpose(perm = var_449, x = var_448_cast_fp16)[name = tensor<string, []>("transpose_24")]; | |
| tensor<fp16, [1, 1500, 384]> x_47_cast_fp16 = reshape(shape = concat_3, x = transpose_24)[name = tensor<string, []>("x_47_cast_fp16")]; | |
| tensor<fp16, [384, 384]> var_454_to_fp16 = const()[name = tensor<string, []>("op_454_to_fp16"), val = tensor<fp16, [384, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13759424)))]; | |
| tensor<fp16, [384]> var_455_to_fp16 = const()[name = tensor<string, []>("op_455_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14054400)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_21_cast_fp16 = linear(bias = var_455_to_fp16, weight = var_454_to_fp16, x = x_47_cast_fp16)[name = tensor<string, []>("linear_21_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 384]> x_49_cast_fp16 = add(x = x_43_cast_fp16, y = linear_21_cast_fp16)[name = tensor<string, []>("x_49_cast_fp16")]; | |
| tensor<int32, [1]> var_462_axes_0 = const()[name = tensor<string, []>("op_462_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [384]> blocks_3_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_3_mlp_ln_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14055232)))]; | |
| tensor<fp16, [384]> blocks_3_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_3_mlp_ln_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14056064)))]; | |
| tensor<fp16, [1, 1500, 384]> var_462_cast_fp16 = layer_norm(axes = var_462_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_388_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast_fp16)[name = tensor<string, []>("op_462_cast_fp16")]; | |
| tensor<fp16, [1536, 384]> var_471_to_fp16 = const()[name = tensor<string, []>("op_471_to_fp16"), val = tensor<fp16, [1536, 384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14056896)))]; | |
| tensor<fp16, [1536]> var_472_to_fp16 = const()[name = tensor<string, []>("op_472_to_fp16"), val = tensor<fp16, [1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15236608)))]; | |
| tensor<fp16, [1, 1500, 1536]> linear_22_cast_fp16 = linear(bias = var_472_to_fp16, weight = var_471_to_fp16, x = var_462_cast_fp16)[name = tensor<string, []>("linear_22_cast_fp16")]; | |
| tensor<string, []> x_53_mode_0 = const()[name = tensor<string, []>("x_53_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 1500, 1536]> x_53_cast_fp16 = gelu(mode = x_53_mode_0, x = linear_22_cast_fp16)[name = tensor<string, []>("x_53_cast_fp16")]; | |
| tensor<fp16, [384, 1536]> var_477_to_fp16 = const()[name = tensor<string, []>("op_477_to_fp16"), val = tensor<fp16, [384, 1536]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15239744)))]; | |
| tensor<fp16, [384]> var_478_to_fp16 = const()[name = tensor<string, []>("op_478_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16419456)))]; | |
| tensor<fp16, [1, 1500, 384]> linear_23_cast_fp16 = linear(bias = var_478_to_fp16, weight = var_477_to_fp16, x = x_53_cast_fp16)[name = tensor<string, []>("linear_23_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 384]> x_cast_fp16 = add(x = x_49_cast_fp16, y = linear_23_cast_fp16)[name = tensor<string, []>("x_cast_fp16")]; | |
| tensor<int32, [1]> var_491_axes_0 = const()[name = tensor<string, []>("op_491_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [384]> ln_post_weight_to_fp16 = const()[name = tensor<string, []>("ln_post_weight_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16420288)))]; | |
| tensor<fp16, [384]> ln_post_bias_to_fp16 = const()[name = tensor<string, []>("ln_post_bias_to_fp16"), val = tensor<fp16, [384]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16421120)))]; | |
| tensor<fp16, []> var_482_to_fp16 = const()[name = tensor<string, []>("op_482_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 384]> output = layer_norm(axes = var_491_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_482_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast_fp16)[name = tensor<string, []>("op_491_cast_fp16")]; | |
| } -> (output); | |
| } |