| program(1.3) | |
| [buildInfo = dict<string, string>({{"coremlc-component-MIL", "3400.43.1"}, {"coremlc-version", "3400.58.2"}, {"coremltools-component-torch", "2.4.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "8.0"}})] | |
| { | |
| func main<ios18>(tensor<fp16, [1, 80, 3000]> logmel_data) { | |
| string var_44_pad_type_0 = const()[name = string("op_44_pad_type_0"), val = string("custom")]; | |
| tensor<int32, [2]> var_44_pad_0 = const()[name = string("op_44_pad_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [1]> var_44_strides_0 = const()[name = string("op_44_strides_0"), val = tensor<int32, [1]>([1])]; | |
| tensor<int32, [1]> var_44_dilations_0 = const()[name = string("op_44_dilations_0"), val = tensor<int32, [1]>([1])]; | |
| int32 var_44_groups_0 = const()[name = string("op_44_groups_0"), val = int32(1)]; | |
| tensor<fp16, [768, 80, 3]> weight_3_to_fp16 = const()[name = string("weight_3_to_fp16"), val = tensor<fp16, [768, 80, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; | |
| tensor<fp16, [768]> bias_3_to_fp16 = const()[name = string("bias_3_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368768)))]; | |
| tensor<fp16, [1, 768, 3000]> var_44_cast_fp16 = conv(bias = bias_3_to_fp16, dilations = var_44_dilations_0, groups = var_44_groups_0, pad = var_44_pad_0, pad_type = var_44_pad_type_0, strides = var_44_strides_0, weight = weight_3_to_fp16, x = logmel_data)[name = string("op_44_cast_fp16")]; | |
| string input_1_mode_0 = const()[name = string("input_1_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 768, 3000]> input_1_cast_fp16 = gelu(mode = input_1_mode_0, x = var_44_cast_fp16)[name = string("input_1_cast_fp16")]; | |
| string var_62_pad_type_0 = const()[name = string("op_62_pad_type_0"), val = string("custom")]; | |
| tensor<int32, [2]> var_62_pad_0 = const()[name = string("op_62_pad_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<int32, [1]> var_62_strides_0 = const()[name = string("op_62_strides_0"), val = tensor<int32, [1]>([2])]; | |
| tensor<int32, [1]> var_62_dilations_0 = const()[name = string("op_62_dilations_0"), val = tensor<int32, [1]>([1])]; | |
| int32 var_62_groups_0 = const()[name = string("op_62_groups_0"), val = int32(1)]; | |
| tensor<fp16, [768, 768, 3]> weight_7_to_fp16 = const()[name = string("weight_7_to_fp16"), val = tensor<fp16, [768, 768, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370368)))]; | |
| tensor<fp16, [768]> bias_7_to_fp16 = const()[name = string("bias_7_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3909376)))]; | |
| tensor<fp16, [1, 768, 1500]> var_62_cast_fp16 = conv(bias = bias_7_to_fp16, dilations = var_62_dilations_0, groups = var_62_groups_0, pad = var_62_pad_0, pad_type = var_62_pad_type_0, strides = var_62_strides_0, weight = weight_7_to_fp16, x = input_1_cast_fp16)[name = string("op_62_cast_fp16")]; | |
| string x_3_mode_0 = const()[name = string("x_3_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 768, 1500]> x_3_cast_fp16 = gelu(mode = x_3_mode_0, x = var_62_cast_fp16)[name = string("x_3_cast_fp16")]; | |
| tensor<int32, [3]> var_68 = const()[name = string("op_68"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<fp16, [1500, 768]> positional_embedding_to_fp16 = const()[name = string("positional_embedding_to_fp16"), val = tensor<fp16, [1500, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3910976)))]; | |
| tensor<fp16, [1, 1500, 768]> x_5_cast_fp16 = transpose(perm = var_68, x = x_3_cast_fp16)[name = string("transpose_120")]; | |
| tensor<fp16, [1, 1500, 768]> var_71_cast_fp16 = add(x = x_5_cast_fp16, y = positional_embedding_to_fp16)[name = string("op_71_cast_fp16")]; | |
| int32 var_84 = const()[name = string("op_84"), val = int32(-1)]; | |
| tensor<int32, [1]> var_100_axes_0 = const()[name = string("op_100_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_0_attn_ln_weight_to_fp16 = const()[name = string("blocks_0_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6215040)))]; | |
| tensor<fp16, [768]> blocks_0_attn_ln_bias_to_fp16 = const()[name = string("blocks_0_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6216640)))]; | |
| fp16 var_90_to_fp16 = const()[name = string("op_90_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_100_cast_fp16 = layer_norm(axes = var_100_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_90_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_71_cast_fp16)[name = string("op_100_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_111_to_fp16 = const()[name = string("op_111_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6218240)))]; | |
| tensor<fp16, [768]> var_112_to_fp16 = const()[name = string("op_112_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7397952)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_0_cast_fp16 = linear(bias = var_112_to_fp16, weight = var_111_to_fp16, x = var_100_cast_fp16)[name = string("linear_0_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_115_to_fp16 = const()[name = string("op_115_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(7399552)))]; | |
| tensor<fp16, [768]> linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8579264)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_115_to_fp16, x = var_100_cast_fp16)[name = string("linear_1_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_119_to_fp16 = const()[name = string("op_119_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8580864)))]; | |
| tensor<fp16, [768]> var_120_to_fp16 = const()[name = string("op_120_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9760576)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_2_cast_fp16 = linear(bias = var_120_to_fp16, weight = var_119_to_fp16, x = var_100_cast_fp16)[name = string("linear_2_cast_fp16")]; | |
| tensor<int32, [4]> var_128 = const()[name = string("op_128"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_129_cast_fp16 = reshape(shape = var_128, x = linear_0_cast_fp16)[name = string("op_129_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_84_to_fp16 = const()[name = string("const_84_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_3_cast_fp16 = mul(x = var_129_cast_fp16, y = const_84_to_fp16)[name = string("q_3_cast_fp16")]; | |
| tensor<int32, [4]> var_135 = const()[name = string("op_135"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_136_cast_fp16 = reshape(shape = var_135, x = linear_1_cast_fp16)[name = string("op_136_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_85_to_fp16 = const()[name = string("const_85_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_3_cast_fp16 = mul(x = var_136_cast_fp16, y = const_85_to_fp16)[name = string("k_3_cast_fp16")]; | |
| tensor<int32, [4]> var_142 = const()[name = string("op_142"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_143_cast_fp16 = reshape(shape = var_142, x = linear_2_cast_fp16)[name = string("op_143_cast_fp16")]; | |
| tensor<int32, [4]> var_144 = const()[name = string("op_144"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_1_transpose_x_0 = const()[name = string("qk_1_transpose_x_0"), val = bool(false)]; | |
| bool qk_1_transpose_y_0 = const()[name = string("qk_1_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_48_perm_0 = const()[name = string("transpose_48_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_49_perm_0 = const()[name = string("transpose_49_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_49 = transpose(perm = transpose_49_perm_0, x = k_3_cast_fp16)[name = string("transpose_117")]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_48 = transpose(perm = transpose_48_perm_0, x = q_3_cast_fp16)[name = string("transpose_118")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_1_cast_fp16 = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_48, y = transpose_49)[name = string("qk_1_cast_fp16")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> var_148_cast_fp16 = softmax(axis = var_84, x = qk_1_cast_fp16)[name = string("op_148_cast_fp16")]; | |
| bool var_150_transpose_x_0 = const()[name = string("op_150_transpose_x_0"), val = bool(false)]; | |
| bool var_150_transpose_y_0 = const()[name = string("op_150_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> v_3_cast_fp16 = transpose(perm = var_144, x = var_143_cast_fp16)[name = string("transpose_119")]; | |
| tensor<fp16, [1, 12, 1500, 64]> var_150_cast_fp16 = matmul(transpose_x = var_150_transpose_x_0, transpose_y = var_150_transpose_y_0, x = var_148_cast_fp16, y = v_3_cast_fp16)[name = string("op_150_cast_fp16")]; | |
| tensor<int32, [4]> var_151 = const()[name = string("op_151"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_0 = const()[name = string("concat_0"), val = tensor<int32, [3]>([1, 1500, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_152_cast_fp16 = transpose(perm = var_151, x = var_150_cast_fp16)[name = string("transpose_116")]; | |
| tensor<fp16, [1, 1500, 768]> x_11_cast_fp16 = reshape(shape = concat_0, x = var_152_cast_fp16)[name = string("x_11_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_156_to_fp16 = const()[name = string("op_156_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9762176)))]; | |
| tensor<fp16, [768]> var_157_to_fp16 = const()[name = string("op_157_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10941888)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_3_cast_fp16 = linear(bias = var_157_to_fp16, weight = var_156_to_fp16, x = x_11_cast_fp16)[name = string("linear_3_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_13_cast_fp16 = add(x = var_71_cast_fp16, y = linear_3_cast_fp16)[name = string("x_13_cast_fp16")]; | |
| tensor<int32, [1]> var_164_axes_0 = const()[name = string("op_164_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_0_mlp_ln_weight_to_fp16 = const()[name = string("blocks_0_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10943488)))]; | |
| tensor<fp16, [768]> blocks_0_mlp_ln_bias_to_fp16 = const()[name = string("blocks_0_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10945088)))]; | |
| tensor<fp16, [1, 1500, 768]> var_164_cast_fp16 = layer_norm(axes = var_164_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_90_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast_fp16)[name = string("op_164_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> var_173_to_fp16 = const()[name = string("op_173_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(10946688)))]; | |
| tensor<fp16, [3072]> var_174_to_fp16 = const()[name = string("op_174_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15665344)))]; | |
| tensor<fp16, [1, 1500, 3072]> linear_4_cast_fp16 = linear(bias = var_174_to_fp16, weight = var_173_to_fp16, x = var_164_cast_fp16)[name = string("linear_4_cast_fp16")]; | |
| string x_17_mode_0 = const()[name = string("x_17_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_17_cast_fp16 = gelu(mode = x_17_mode_0, x = linear_4_cast_fp16)[name = string("x_17_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> var_179_to_fp16 = const()[name = string("op_179_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(15671552)))]; | |
| tensor<fp16, [768]> var_180_to_fp16 = const()[name = string("op_180_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20390208)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_5_cast_fp16 = linear(bias = var_180_to_fp16, weight = var_179_to_fp16, x = x_17_cast_fp16)[name = string("linear_5_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_19_cast_fp16 = add(x = x_13_cast_fp16, y = linear_5_cast_fp16)[name = string("x_19_cast_fp16")]; | |
| int32 var_190 = const()[name = string("op_190"), val = int32(-1)]; | |
| tensor<int32, [1]> var_206_axes_0 = const()[name = string("op_206_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_1_attn_ln_weight_to_fp16 = const()[name = string("blocks_1_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20391808)))]; | |
| tensor<fp16, [768]> blocks_1_attn_ln_bias_to_fp16 = const()[name = string("blocks_1_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20393408)))]; | |
| fp16 var_196_to_fp16 = const()[name = string("op_196_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_206_cast_fp16 = layer_norm(axes = var_206_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_196_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast_fp16)[name = string("op_206_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_217_to_fp16 = const()[name = string("op_217_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(20395008)))]; | |
| tensor<fp16, [768]> var_218_to_fp16 = const()[name = string("op_218_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21574720)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_6_cast_fp16 = linear(bias = var_218_to_fp16, weight = var_217_to_fp16, x = var_206_cast_fp16)[name = string("linear_6_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_221_to_fp16 = const()[name = string("op_221_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(21576320)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_7_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_221_to_fp16, x = var_206_cast_fp16)[name = string("linear_7_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_225_to_fp16 = const()[name = string("op_225_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22756032)))]; | |
| tensor<fp16, [768]> var_226_to_fp16 = const()[name = string("op_226_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23935744)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_8_cast_fp16 = linear(bias = var_226_to_fp16, weight = var_225_to_fp16, x = var_206_cast_fp16)[name = string("linear_8_cast_fp16")]; | |
| tensor<int32, [4]> var_234 = const()[name = string("op_234"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_235_cast_fp16 = reshape(shape = var_234, x = linear_6_cast_fp16)[name = string("op_235_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_86_to_fp16 = const()[name = string("const_86_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_7_cast_fp16 = mul(x = var_235_cast_fp16, y = const_86_to_fp16)[name = string("q_7_cast_fp16")]; | |
| tensor<int32, [4]> var_241 = const()[name = string("op_241"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_242_cast_fp16 = reshape(shape = var_241, x = linear_7_cast_fp16)[name = string("op_242_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_87_to_fp16 = const()[name = string("const_87_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_7_cast_fp16 = mul(x = var_242_cast_fp16, y = const_87_to_fp16)[name = string("k_7_cast_fp16")]; | |
| tensor<int32, [4]> var_248 = const()[name = string("op_248"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_249_cast_fp16 = reshape(shape = var_248, x = linear_8_cast_fp16)[name = string("op_249_cast_fp16")]; | |
| tensor<int32, [4]> var_250 = const()[name = string("op_250"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_3_transpose_x_0 = const()[name = string("qk_3_transpose_x_0"), val = bool(false)]; | |
| bool qk_3_transpose_y_0 = const()[name = string("qk_3_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_50_perm_0 = const()[name = string("transpose_50_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_51_perm_0 = const()[name = string("transpose_51_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_51 = transpose(perm = transpose_51_perm_0, x = k_7_cast_fp16)[name = string("transpose_113")]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_50 = transpose(perm = transpose_50_perm_0, x = q_7_cast_fp16)[name = string("transpose_114")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_3_cast_fp16 = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_50, y = transpose_51)[name = string("qk_3_cast_fp16")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> var_254_cast_fp16 = softmax(axis = var_190, x = qk_3_cast_fp16)[name = string("op_254_cast_fp16")]; | |
| bool var_256_transpose_x_0 = const()[name = string("op_256_transpose_x_0"), val = bool(false)]; | |
| bool var_256_transpose_y_0 = const()[name = string("op_256_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> v_7_cast_fp16 = transpose(perm = var_250, x = var_249_cast_fp16)[name = string("transpose_115")]; | |
| tensor<fp16, [1, 12, 1500, 64]> var_256_cast_fp16 = matmul(transpose_x = var_256_transpose_x_0, transpose_y = var_256_transpose_y_0, x = var_254_cast_fp16, y = v_7_cast_fp16)[name = string("op_256_cast_fp16")]; | |
| tensor<int32, [4]> var_257 = const()[name = string("op_257"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_1 = const()[name = string("concat_1"), val = tensor<int32, [3]>([1, 1500, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_258_cast_fp16 = transpose(perm = var_257, x = var_256_cast_fp16)[name = string("transpose_112")]; | |
| tensor<fp16, [1, 1500, 768]> x_23_cast_fp16 = reshape(shape = concat_1, x = var_258_cast_fp16)[name = string("x_23_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_262_to_fp16 = const()[name = string("op_262_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23937344)))]; | |
| tensor<fp16, [768]> var_263_to_fp16 = const()[name = string("op_263_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25117056)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_9_cast_fp16 = linear(bias = var_263_to_fp16, weight = var_262_to_fp16, x = x_23_cast_fp16)[name = string("linear_9_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_25_cast_fp16 = add(x = x_19_cast_fp16, y = linear_9_cast_fp16)[name = string("x_25_cast_fp16")]; | |
| tensor<int32, [1]> var_270_axes_0 = const()[name = string("op_270_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_1_mlp_ln_weight_to_fp16 = const()[name = string("blocks_1_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25118656)))]; | |
| tensor<fp16, [768]> blocks_1_mlp_ln_bias_to_fp16 = const()[name = string("blocks_1_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25120256)))]; | |
| tensor<fp16, [1, 1500, 768]> var_270_cast_fp16 = layer_norm(axes = var_270_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_196_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast_fp16)[name = string("op_270_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> var_279_to_fp16 = const()[name = string("op_279_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25121856)))]; | |
| tensor<fp16, [3072]> var_280_to_fp16 = const()[name = string("op_280_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29840512)))]; | |
| tensor<fp16, [1, 1500, 3072]> linear_10_cast_fp16 = linear(bias = var_280_to_fp16, weight = var_279_to_fp16, x = var_270_cast_fp16)[name = string("linear_10_cast_fp16")]; | |
| string x_29_mode_0 = const()[name = string("x_29_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_29_cast_fp16 = gelu(mode = x_29_mode_0, x = linear_10_cast_fp16)[name = string("x_29_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> var_285_to_fp16 = const()[name = string("op_285_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(29846720)))]; | |
| tensor<fp16, [768]> var_286_to_fp16 = const()[name = string("op_286_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34565376)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_11_cast_fp16 = linear(bias = var_286_to_fp16, weight = var_285_to_fp16, x = x_29_cast_fp16)[name = string("linear_11_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_31_cast_fp16 = add(x = x_25_cast_fp16, y = linear_11_cast_fp16)[name = string("x_31_cast_fp16")]; | |
| int32 var_296 = const()[name = string("op_296"), val = int32(-1)]; | |
| tensor<int32, [1]> var_312_axes_0 = const()[name = string("op_312_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_2_attn_ln_weight_to_fp16 = const()[name = string("blocks_2_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34566976)))]; | |
| tensor<fp16, [768]> blocks_2_attn_ln_bias_to_fp16 = const()[name = string("blocks_2_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34568576)))]; | |
| fp16 var_302_to_fp16 = const()[name = string("op_302_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_312_cast_fp16 = layer_norm(axes = var_312_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_fp16)[name = string("op_312_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_323_to_fp16 = const()[name = string("op_323_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34570176)))]; | |
| tensor<fp16, [768]> var_324_to_fp16 = const()[name = string("op_324_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35749888)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_12_cast_fp16 = linear(bias = var_324_to_fp16, weight = var_323_to_fp16, x = var_312_cast_fp16)[name = string("linear_12_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_327_to_fp16 = const()[name = string("op_327_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35751488)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_13_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_327_to_fp16, x = var_312_cast_fp16)[name = string("linear_13_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_331_to_fp16 = const()[name = string("op_331_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(36931200)))]; | |
| tensor<fp16, [768]> var_332_to_fp16 = const()[name = string("op_332_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38110912)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_14_cast_fp16 = linear(bias = var_332_to_fp16, weight = var_331_to_fp16, x = var_312_cast_fp16)[name = string("linear_14_cast_fp16")]; | |
| tensor<int32, [4]> var_340 = const()[name = string("op_340"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_341_cast_fp16 = reshape(shape = var_340, x = linear_12_cast_fp16)[name = string("op_341_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_88_to_fp16 = const()[name = string("const_88_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_11_cast_fp16 = mul(x = var_341_cast_fp16, y = const_88_to_fp16)[name = string("q_11_cast_fp16")]; | |
| tensor<int32, [4]> var_347 = const()[name = string("op_347"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_348_cast_fp16 = reshape(shape = var_347, x = linear_13_cast_fp16)[name = string("op_348_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_89_to_fp16 = const()[name = string("const_89_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_11_cast_fp16 = mul(x = var_348_cast_fp16, y = const_89_to_fp16)[name = string("k_11_cast_fp16")]; | |
| tensor<int32, [4]> var_354 = const()[name = string("op_354"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_355_cast_fp16 = reshape(shape = var_354, x = linear_14_cast_fp16)[name = string("op_355_cast_fp16")]; | |
| tensor<int32, [4]> var_356 = const()[name = string("op_356"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_5_transpose_x_0 = const()[name = string("qk_5_transpose_x_0"), val = bool(false)]; | |
| bool qk_5_transpose_y_0 = const()[name = string("qk_5_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_52_perm_0 = const()[name = string("transpose_52_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_53_perm_0 = const()[name = string("transpose_53_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_53 = transpose(perm = transpose_53_perm_0, x = k_11_cast_fp16)[name = string("transpose_109")]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_52 = transpose(perm = transpose_52_perm_0, x = q_11_cast_fp16)[name = string("transpose_110")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_5_cast_fp16 = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_52, y = transpose_53)[name = string("qk_5_cast_fp16")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> var_360_cast_fp16 = softmax(axis = var_296, x = qk_5_cast_fp16)[name = string("op_360_cast_fp16")]; | |
| bool var_362_transpose_x_0 = const()[name = string("op_362_transpose_x_0"), val = bool(false)]; | |
| bool var_362_transpose_y_0 = const()[name = string("op_362_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> v_11_cast_fp16 = transpose(perm = var_356, x = var_355_cast_fp16)[name = string("transpose_111")]; | |
| tensor<fp16, [1, 12, 1500, 64]> var_362_cast_fp16 = matmul(transpose_x = var_362_transpose_x_0, transpose_y = var_362_transpose_y_0, x = var_360_cast_fp16, y = v_11_cast_fp16)[name = string("op_362_cast_fp16")]; | |
| tensor<int32, [4]> var_363 = const()[name = string("op_363"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_2 = const()[name = string("concat_2"), val = tensor<int32, [3]>([1, 1500, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_364_cast_fp16 = transpose(perm = var_363, x = var_362_cast_fp16)[name = string("transpose_108")]; | |
| tensor<fp16, [1, 1500, 768]> x_35_cast_fp16 = reshape(shape = concat_2, x = var_364_cast_fp16)[name = string("x_35_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_368_to_fp16 = const()[name = string("op_368_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(38112512)))]; | |
| tensor<fp16, [768]> var_369_to_fp16 = const()[name = string("op_369_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39292224)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_15_cast_fp16 = linear(bias = var_369_to_fp16, weight = var_368_to_fp16, x = x_35_cast_fp16)[name = string("linear_15_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_37_cast_fp16 = add(x = x_31_cast_fp16, y = linear_15_cast_fp16)[name = string("x_37_cast_fp16")]; | |
| tensor<int32, [1]> var_376_axes_0 = const()[name = string("op_376_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_2_mlp_ln_weight_to_fp16 = const()[name = string("blocks_2_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39293824)))]; | |
| tensor<fp16, [768]> blocks_2_mlp_ln_bias_to_fp16 = const()[name = string("blocks_2_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39295424)))]; | |
| tensor<fp16, [1, 1500, 768]> var_376_cast_fp16 = layer_norm(axes = var_376_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_fp16)[name = string("op_376_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> var_385_to_fp16 = const()[name = string("op_385_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39297024)))]; | |
| tensor<fp16, [3072]> var_386_to_fp16 = const()[name = string("op_386_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44015680)))]; | |
| tensor<fp16, [1, 1500, 3072]> linear_16_cast_fp16 = linear(bias = var_386_to_fp16, weight = var_385_to_fp16, x = var_376_cast_fp16)[name = string("linear_16_cast_fp16")]; | |
| string x_41_mode_0 = const()[name = string("x_41_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_41_cast_fp16 = gelu(mode = x_41_mode_0, x = linear_16_cast_fp16)[name = string("x_41_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> var_391_to_fp16 = const()[name = string("op_391_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44021888)))]; | |
| tensor<fp16, [768]> var_392_to_fp16 = const()[name = string("op_392_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48740544)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_17_cast_fp16 = linear(bias = var_392_to_fp16, weight = var_391_to_fp16, x = x_41_cast_fp16)[name = string("linear_17_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_43_cast_fp16 = add(x = x_37_cast_fp16, y = linear_17_cast_fp16)[name = string("x_43_cast_fp16")]; | |
| int32 var_402 = const()[name = string("op_402"), val = int32(-1)]; | |
| tensor<int32, [1]> var_418_axes_0 = const()[name = string("op_418_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_3_attn_ln_weight_to_fp16 = const()[name = string("blocks_3_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48742144)))]; | |
| tensor<fp16, [768]> blocks_3_attn_ln_bias_to_fp16 = const()[name = string("blocks_3_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48743744)))]; | |
| fp16 var_408_to_fp16 = const()[name = string("op_408_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_418_cast_fp16 = layer_norm(axes = var_418_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_408_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast_fp16)[name = string("op_418_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_429_to_fp16 = const()[name = string("op_429_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(48745344)))]; | |
| tensor<fp16, [768]> var_430_to_fp16 = const()[name = string("op_430_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49925056)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_18_cast_fp16 = linear(bias = var_430_to_fp16, weight = var_429_to_fp16, x = var_418_cast_fp16)[name = string("linear_18_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_433_to_fp16 = const()[name = string("op_433_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(49926656)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_433_to_fp16, x = var_418_cast_fp16)[name = string("linear_19_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_437_to_fp16 = const()[name = string("op_437_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51106368)))]; | |
| tensor<fp16, [768]> var_438_to_fp16 = const()[name = string("op_438_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52286080)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_20_cast_fp16 = linear(bias = var_438_to_fp16, weight = var_437_to_fp16, x = var_418_cast_fp16)[name = string("linear_20_cast_fp16")]; | |
| tensor<int32, [4]> var_446 = const()[name = string("op_446"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_447_cast_fp16 = reshape(shape = var_446, x = linear_18_cast_fp16)[name = string("op_447_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_90_to_fp16 = const()[name = string("const_90_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_15_cast_fp16 = mul(x = var_447_cast_fp16, y = const_90_to_fp16)[name = string("q_15_cast_fp16")]; | |
| tensor<int32, [4]> var_453 = const()[name = string("op_453"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_454_cast_fp16 = reshape(shape = var_453, x = linear_19_cast_fp16)[name = string("op_454_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_91_to_fp16 = const()[name = string("const_91_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_15_cast_fp16 = mul(x = var_454_cast_fp16, y = const_91_to_fp16)[name = string("k_15_cast_fp16")]; | |
| tensor<int32, [4]> var_460 = const()[name = string("op_460"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_461_cast_fp16 = reshape(shape = var_460, x = linear_20_cast_fp16)[name = string("op_461_cast_fp16")]; | |
| tensor<int32, [4]> var_462 = const()[name = string("op_462"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_7_transpose_x_0 = const()[name = string("qk_7_transpose_x_0"), val = bool(false)]; | |
| bool qk_7_transpose_y_0 = const()[name = string("qk_7_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_54_perm_0 = const()[name = string("transpose_54_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_55_perm_0 = const()[name = string("transpose_55_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_55 = transpose(perm = transpose_55_perm_0, x = k_15_cast_fp16)[name = string("transpose_105")]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_54 = transpose(perm = transpose_54_perm_0, x = q_15_cast_fp16)[name = string("transpose_106")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_7_cast_fp16 = matmul(transpose_x = qk_7_transpose_x_0, transpose_y = qk_7_transpose_y_0, x = transpose_54, y = transpose_55)[name = string("qk_7_cast_fp16")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> var_466_cast_fp16 = softmax(axis = var_402, x = qk_7_cast_fp16)[name = string("op_466_cast_fp16")]; | |
| bool var_468_transpose_x_0 = const()[name = string("op_468_transpose_x_0"), val = bool(false)]; | |
| bool var_468_transpose_y_0 = const()[name = string("op_468_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> v_15_cast_fp16 = transpose(perm = var_462, x = var_461_cast_fp16)[name = string("transpose_107")]; | |
| tensor<fp16, [1, 12, 1500, 64]> var_468_cast_fp16 = matmul(transpose_x = var_468_transpose_x_0, transpose_y = var_468_transpose_y_0, x = var_466_cast_fp16, y = v_15_cast_fp16)[name = string("op_468_cast_fp16")]; | |
| tensor<int32, [4]> var_469 = const()[name = string("op_469"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_3 = const()[name = string("concat_3"), val = tensor<int32, [3]>([1, 1500, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_470_cast_fp16 = transpose(perm = var_469, x = var_468_cast_fp16)[name = string("transpose_104")]; | |
| tensor<fp16, [1, 1500, 768]> x_47_cast_fp16 = reshape(shape = concat_3, x = var_470_cast_fp16)[name = string("x_47_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_474_to_fp16 = const()[name = string("op_474_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(52287680)))]; | |
| tensor<fp16, [768]> var_475_to_fp16 = const()[name = string("op_475_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53467392)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_21_cast_fp16 = linear(bias = var_475_to_fp16, weight = var_474_to_fp16, x = x_47_cast_fp16)[name = string("linear_21_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_49_cast_fp16 = add(x = x_43_cast_fp16, y = linear_21_cast_fp16)[name = string("x_49_cast_fp16")]; | |
| tensor<int32, [1]> var_482_axes_0 = const()[name = string("op_482_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_3_mlp_ln_weight_to_fp16 = const()[name = string("blocks_3_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53468992)))]; | |
| tensor<fp16, [768]> blocks_3_mlp_ln_bias_to_fp16 = const()[name = string("blocks_3_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53470592)))]; | |
| tensor<fp16, [1, 1500, 768]> var_482_cast_fp16 = layer_norm(axes = var_482_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_408_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast_fp16)[name = string("op_482_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> var_491_to_fp16 = const()[name = string("op_491_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(53472192)))]; | |
| tensor<fp16, [3072]> var_492_to_fp16 = const()[name = string("op_492_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58190848)))]; | |
| tensor<fp16, [1, 1500, 3072]> linear_22_cast_fp16 = linear(bias = var_492_to_fp16, weight = var_491_to_fp16, x = var_482_cast_fp16)[name = string("linear_22_cast_fp16")]; | |
| string x_53_mode_0 = const()[name = string("x_53_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_53_cast_fp16 = gelu(mode = x_53_mode_0, x = linear_22_cast_fp16)[name = string("x_53_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> var_497_to_fp16 = const()[name = string("op_497_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(58197056)))]; | |
| tensor<fp16, [768]> var_498_to_fp16 = const()[name = string("op_498_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62915712)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_23_cast_fp16 = linear(bias = var_498_to_fp16, weight = var_497_to_fp16, x = x_53_cast_fp16)[name = string("linear_23_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_55_cast_fp16 = add(x = x_49_cast_fp16, y = linear_23_cast_fp16)[name = string("x_55_cast_fp16")]; | |
| int32 var_508 = const()[name = string("op_508"), val = int32(-1)]; | |
| tensor<int32, [1]> var_524_axes_0 = const()[name = string("op_524_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_4_attn_ln_weight_to_fp16 = const()[name = string("blocks_4_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62917312)))]; | |
| tensor<fp16, [768]> blocks_4_attn_ln_bias_to_fp16 = const()[name = string("blocks_4_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62918912)))]; | |
| fp16 var_514_to_fp16 = const()[name = string("op_514_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_524_cast_fp16 = layer_norm(axes = var_524_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_514_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast_fp16)[name = string("op_524_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_535_to_fp16 = const()[name = string("op_535_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62920512)))]; | |
| tensor<fp16, [768]> var_536_to_fp16 = const()[name = string("op_536_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64100224)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_24_cast_fp16 = linear(bias = var_536_to_fp16, weight = var_535_to_fp16, x = var_524_cast_fp16)[name = string("linear_24_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_539_to_fp16 = const()[name = string("op_539_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64101824)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_25_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_539_to_fp16, x = var_524_cast_fp16)[name = string("linear_25_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_543_to_fp16 = const()[name = string("op_543_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65281536)))]; | |
| tensor<fp16, [768]> var_544_to_fp16 = const()[name = string("op_544_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66461248)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_26_cast_fp16 = linear(bias = var_544_to_fp16, weight = var_543_to_fp16, x = var_524_cast_fp16)[name = string("linear_26_cast_fp16")]; | |
| tensor<int32, [4]> var_552 = const()[name = string("op_552"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_553_cast_fp16 = reshape(shape = var_552, x = linear_24_cast_fp16)[name = string("op_553_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_92_to_fp16 = const()[name = string("const_92_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_19_cast_fp16 = mul(x = var_553_cast_fp16, y = const_92_to_fp16)[name = string("q_19_cast_fp16")]; | |
| tensor<int32, [4]> var_559 = const()[name = string("op_559"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_560_cast_fp16 = reshape(shape = var_559, x = linear_25_cast_fp16)[name = string("op_560_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_93_to_fp16 = const()[name = string("const_93_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_19_cast_fp16 = mul(x = var_560_cast_fp16, y = const_93_to_fp16)[name = string("k_19_cast_fp16")]; | |
| tensor<int32, [4]> var_566 = const()[name = string("op_566"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_567_cast_fp16 = reshape(shape = var_566, x = linear_26_cast_fp16)[name = string("op_567_cast_fp16")]; | |
| tensor<int32, [4]> var_568 = const()[name = string("op_568"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_9_transpose_x_0 = const()[name = string("qk_9_transpose_x_0"), val = bool(false)]; | |
| bool qk_9_transpose_y_0 = const()[name = string("qk_9_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_56_perm_0 = const()[name = string("transpose_56_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_57_perm_0 = const()[name = string("transpose_57_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_57 = transpose(perm = transpose_57_perm_0, x = k_19_cast_fp16)[name = string("transpose_101")]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_56 = transpose(perm = transpose_56_perm_0, x = q_19_cast_fp16)[name = string("transpose_102")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_9_cast_fp16 = matmul(transpose_x = qk_9_transpose_x_0, transpose_y = qk_9_transpose_y_0, x = transpose_56, y = transpose_57)[name = string("qk_9_cast_fp16")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> var_572_cast_fp16 = softmax(axis = var_508, x = qk_9_cast_fp16)[name = string("op_572_cast_fp16")]; | |
| bool var_574_transpose_x_0 = const()[name = string("op_574_transpose_x_0"), val = bool(false)]; | |
| bool var_574_transpose_y_0 = const()[name = string("op_574_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> v_19_cast_fp16 = transpose(perm = var_568, x = var_567_cast_fp16)[name = string("transpose_103")]; | |
| tensor<fp16, [1, 12, 1500, 64]> var_574_cast_fp16 = matmul(transpose_x = var_574_transpose_x_0, transpose_y = var_574_transpose_y_0, x = var_572_cast_fp16, y = v_19_cast_fp16)[name = string("op_574_cast_fp16")]; | |
| tensor<int32, [4]> var_575 = const()[name = string("op_575"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_4 = const()[name = string("concat_4"), val = tensor<int32, [3]>([1, 1500, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_576_cast_fp16 = transpose(perm = var_575, x = var_574_cast_fp16)[name = string("transpose_100")]; | |
| tensor<fp16, [1, 1500, 768]> x_59_cast_fp16 = reshape(shape = concat_4, x = var_576_cast_fp16)[name = string("x_59_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_580_to_fp16 = const()[name = string("op_580_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66462848)))]; | |
| tensor<fp16, [768]> var_581_to_fp16 = const()[name = string("op_581_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67642560)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_27_cast_fp16 = linear(bias = var_581_to_fp16, weight = var_580_to_fp16, x = x_59_cast_fp16)[name = string("linear_27_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_61_cast_fp16 = add(x = x_55_cast_fp16, y = linear_27_cast_fp16)[name = string("x_61_cast_fp16")]; | |
| tensor<int32, [1]> var_588_axes_0 = const()[name = string("op_588_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_4_mlp_ln_weight_to_fp16 = const()[name = string("blocks_4_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67644160)))]; | |
| tensor<fp16, [768]> blocks_4_mlp_ln_bias_to_fp16 = const()[name = string("blocks_4_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67645760)))]; | |
| tensor<fp16, [1, 1500, 768]> var_588_cast_fp16 = layer_norm(axes = var_588_axes_0, beta = blocks_4_mlp_ln_bias_to_fp16, epsilon = var_514_to_fp16, gamma = blocks_4_mlp_ln_weight_to_fp16, x = x_61_cast_fp16)[name = string("op_588_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> var_597_to_fp16 = const()[name = string("op_597_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67647360)))]; | |
| tensor<fp16, [3072]> var_598_to_fp16 = const()[name = string("op_598_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72366016)))]; | |
| tensor<fp16, [1, 1500, 3072]> linear_28_cast_fp16 = linear(bias = var_598_to_fp16, weight = var_597_to_fp16, x = var_588_cast_fp16)[name = string("linear_28_cast_fp16")]; | |
| string x_65_mode_0 = const()[name = string("x_65_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_65_cast_fp16 = gelu(mode = x_65_mode_0, x = linear_28_cast_fp16)[name = string("x_65_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> var_603_to_fp16 = const()[name = string("op_603_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(72372224)))]; | |
| tensor<fp16, [768]> var_604_to_fp16 = const()[name = string("op_604_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77090880)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_29_cast_fp16 = linear(bias = var_604_to_fp16, weight = var_603_to_fp16, x = x_65_cast_fp16)[name = string("linear_29_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_67_cast_fp16 = add(x = x_61_cast_fp16, y = linear_29_cast_fp16)[name = string("x_67_cast_fp16")]; | |
| int32 var_614 = const()[name = string("op_614"), val = int32(-1)]; | |
| tensor<int32, [1]> var_630_axes_0 = const()[name = string("op_630_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_5_attn_ln_weight_to_fp16 = const()[name = string("blocks_5_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77092480)))]; | |
| tensor<fp16, [768]> blocks_5_attn_ln_bias_to_fp16 = const()[name = string("blocks_5_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77094080)))]; | |
| fp16 var_620_to_fp16 = const()[name = string("op_620_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_630_cast_fp16 = layer_norm(axes = var_630_axes_0, beta = blocks_5_attn_ln_bias_to_fp16, epsilon = var_620_to_fp16, gamma = blocks_5_attn_ln_weight_to_fp16, x = x_67_cast_fp16)[name = string("op_630_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_641_to_fp16 = const()[name = string("op_641_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77095680)))]; | |
| tensor<fp16, [768]> var_642_to_fp16 = const()[name = string("op_642_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78275392)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_30_cast_fp16 = linear(bias = var_642_to_fp16, weight = var_641_to_fp16, x = var_630_cast_fp16)[name = string("linear_30_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_645_to_fp16 = const()[name = string("op_645_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(78276992)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_31_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_645_to_fp16, x = var_630_cast_fp16)[name = string("linear_31_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_649_to_fp16 = const()[name = string("op_649_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79456704)))]; | |
| tensor<fp16, [768]> var_650_to_fp16 = const()[name = string("op_650_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80636416)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_32_cast_fp16 = linear(bias = var_650_to_fp16, weight = var_649_to_fp16, x = var_630_cast_fp16)[name = string("linear_32_cast_fp16")]; | |
| tensor<int32, [4]> var_658 = const()[name = string("op_658"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_659_cast_fp16 = reshape(shape = var_658, x = linear_30_cast_fp16)[name = string("op_659_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_94_to_fp16 = const()[name = string("const_94_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_23_cast_fp16 = mul(x = var_659_cast_fp16, y = const_94_to_fp16)[name = string("q_23_cast_fp16")]; | |
| tensor<int32, [4]> var_665 = const()[name = string("op_665"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_666_cast_fp16 = reshape(shape = var_665, x = linear_31_cast_fp16)[name = string("op_666_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_95_to_fp16 = const()[name = string("const_95_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_23_cast_fp16 = mul(x = var_666_cast_fp16, y = const_95_to_fp16)[name = string("k_23_cast_fp16")]; | |
| tensor<int32, [4]> var_672 = const()[name = string("op_672"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_673_cast_fp16 = reshape(shape = var_672, x = linear_32_cast_fp16)[name = string("op_673_cast_fp16")]; | |
| tensor<int32, [4]> var_674 = const()[name = string("op_674"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_11_transpose_x_0 = const()[name = string("qk_11_transpose_x_0"), val = bool(false)]; | |
| bool qk_11_transpose_y_0 = const()[name = string("qk_11_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_58_perm_0 = const()[name = string("transpose_58_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_59_perm_0 = const()[name = string("transpose_59_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_59 = transpose(perm = transpose_59_perm_0, x = k_23_cast_fp16)[name = string("transpose_97")]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_58 = transpose(perm = transpose_58_perm_0, x = q_23_cast_fp16)[name = string("transpose_98")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_11_cast_fp16 = matmul(transpose_x = qk_11_transpose_x_0, transpose_y = qk_11_transpose_y_0, x = transpose_58, y = transpose_59)[name = string("qk_11_cast_fp16")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> var_678_cast_fp16 = softmax(axis = var_614, x = qk_11_cast_fp16)[name = string("op_678_cast_fp16")]; | |
| bool var_680_transpose_x_0 = const()[name = string("op_680_transpose_x_0"), val = bool(false)]; | |
| bool var_680_transpose_y_0 = const()[name = string("op_680_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> v_23_cast_fp16 = transpose(perm = var_674, x = var_673_cast_fp16)[name = string("transpose_99")]; | |
| tensor<fp16, [1, 12, 1500, 64]> var_680_cast_fp16 = matmul(transpose_x = var_680_transpose_x_0, transpose_y = var_680_transpose_y_0, x = var_678_cast_fp16, y = v_23_cast_fp16)[name = string("op_680_cast_fp16")]; | |
| tensor<int32, [4]> var_681 = const()[name = string("op_681"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_5 = const()[name = string("concat_5"), val = tensor<int32, [3]>([1, 1500, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_682_cast_fp16 = transpose(perm = var_681, x = var_680_cast_fp16)[name = string("transpose_96")]; | |
| tensor<fp16, [1, 1500, 768]> x_71_cast_fp16 = reshape(shape = concat_5, x = var_682_cast_fp16)[name = string("x_71_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_686_to_fp16 = const()[name = string("op_686_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(80638016)))]; | |
| tensor<fp16, [768]> var_687_to_fp16 = const()[name = string("op_687_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81817728)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_33_cast_fp16 = linear(bias = var_687_to_fp16, weight = var_686_to_fp16, x = x_71_cast_fp16)[name = string("linear_33_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_73_cast_fp16 = add(x = x_67_cast_fp16, y = linear_33_cast_fp16)[name = string("x_73_cast_fp16")]; | |
| tensor<int32, [1]> var_694_axes_0 = const()[name = string("op_694_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_5_mlp_ln_weight_to_fp16 = const()[name = string("blocks_5_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81819328)))]; | |
| tensor<fp16, [768]> blocks_5_mlp_ln_bias_to_fp16 = const()[name = string("blocks_5_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81820928)))]; | |
| tensor<fp16, [1, 1500, 768]> var_694_cast_fp16 = layer_norm(axes = var_694_axes_0, beta = blocks_5_mlp_ln_bias_to_fp16, epsilon = var_620_to_fp16, gamma = blocks_5_mlp_ln_weight_to_fp16, x = x_73_cast_fp16)[name = string("op_694_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> var_703_to_fp16 = const()[name = string("op_703_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81822528)))]; | |
| tensor<fp16, [3072]> var_704_to_fp16 = const()[name = string("op_704_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86541184)))]; | |
| tensor<fp16, [1, 1500, 3072]> linear_34_cast_fp16 = linear(bias = var_704_to_fp16, weight = var_703_to_fp16, x = var_694_cast_fp16)[name = string("linear_34_cast_fp16")]; | |
| string x_77_mode_0 = const()[name = string("x_77_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_77_cast_fp16 = gelu(mode = x_77_mode_0, x = linear_34_cast_fp16)[name = string("x_77_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> var_709_to_fp16 = const()[name = string("op_709_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86547392)))]; | |
| tensor<fp16, [768]> var_710_to_fp16 = const()[name = string("op_710_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91266048)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_35_cast_fp16 = linear(bias = var_710_to_fp16, weight = var_709_to_fp16, x = x_77_cast_fp16)[name = string("linear_35_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_79_cast_fp16 = add(x = x_73_cast_fp16, y = linear_35_cast_fp16)[name = string("x_79_cast_fp16")]; | |
| int32 var_720 = const()[name = string("op_720"), val = int32(-1)]; | |
| tensor<int32, [1]> var_736_axes_0 = const()[name = string("op_736_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_6_attn_ln_weight_to_fp16 = const()[name = string("blocks_6_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91267648)))]; | |
| tensor<fp16, [768]> blocks_6_attn_ln_bias_to_fp16 = const()[name = string("blocks_6_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91269248)))]; | |
| fp16 var_726_to_fp16 = const()[name = string("op_726_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_736_cast_fp16 = layer_norm(axes = var_736_axes_0, beta = blocks_6_attn_ln_bias_to_fp16, epsilon = var_726_to_fp16, gamma = blocks_6_attn_ln_weight_to_fp16, x = x_79_cast_fp16)[name = string("op_736_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_747_to_fp16 = const()[name = string("op_747_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91270848)))]; | |
| tensor<fp16, [768]> var_748_to_fp16 = const()[name = string("op_748_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92450560)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_36_cast_fp16 = linear(bias = var_748_to_fp16, weight = var_747_to_fp16, x = var_736_cast_fp16)[name = string("linear_36_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_751_to_fp16 = const()[name = string("op_751_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92452160)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_37_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_751_to_fp16, x = var_736_cast_fp16)[name = string("linear_37_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_755_to_fp16 = const()[name = string("op_755_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93631872)))]; | |
| tensor<fp16, [768]> var_756_to_fp16 = const()[name = string("op_756_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94811584)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_38_cast_fp16 = linear(bias = var_756_to_fp16, weight = var_755_to_fp16, x = var_736_cast_fp16)[name = string("linear_38_cast_fp16")]; | |
| tensor<int32, [4]> var_764 = const()[name = string("op_764"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_765_cast_fp16 = reshape(shape = var_764, x = linear_36_cast_fp16)[name = string("op_765_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_96_to_fp16 = const()[name = string("const_96_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_27_cast_fp16 = mul(x = var_765_cast_fp16, y = const_96_to_fp16)[name = string("q_27_cast_fp16")]; | |
| tensor<int32, [4]> var_771 = const()[name = string("op_771"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_772_cast_fp16 = reshape(shape = var_771, x = linear_37_cast_fp16)[name = string("op_772_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_97_to_fp16 = const()[name = string("const_97_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_27_cast_fp16 = mul(x = var_772_cast_fp16, y = const_97_to_fp16)[name = string("k_27_cast_fp16")]; | |
| tensor<int32, [4]> var_778 = const()[name = string("op_778"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_779_cast_fp16 = reshape(shape = var_778, x = linear_38_cast_fp16)[name = string("op_779_cast_fp16")]; | |
| tensor<int32, [4]> var_780 = const()[name = string("op_780"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_13_transpose_x_0 = const()[name = string("qk_13_transpose_x_0"), val = bool(false)]; | |
| bool qk_13_transpose_y_0 = const()[name = string("qk_13_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_60_perm_0 = const()[name = string("transpose_60_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_61_perm_0 = const()[name = string("transpose_61_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_61 = transpose(perm = transpose_61_perm_0, x = k_27_cast_fp16)[name = string("transpose_93")]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_60 = transpose(perm = transpose_60_perm_0, x = q_27_cast_fp16)[name = string("transpose_94")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_13_cast_fp16 = matmul(transpose_x = qk_13_transpose_x_0, transpose_y = qk_13_transpose_y_0, x = transpose_60, y = transpose_61)[name = string("qk_13_cast_fp16")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> var_784_cast_fp16 = softmax(axis = var_720, x = qk_13_cast_fp16)[name = string("op_784_cast_fp16")]; | |
| bool var_786_transpose_x_0 = const()[name = string("op_786_transpose_x_0"), val = bool(false)]; | |
| bool var_786_transpose_y_0 = const()[name = string("op_786_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> v_27_cast_fp16 = transpose(perm = var_780, x = var_779_cast_fp16)[name = string("transpose_95")]; | |
| tensor<fp16, [1, 12, 1500, 64]> var_786_cast_fp16 = matmul(transpose_x = var_786_transpose_x_0, transpose_y = var_786_transpose_y_0, x = var_784_cast_fp16, y = v_27_cast_fp16)[name = string("op_786_cast_fp16")]; | |
| tensor<int32, [4]> var_787 = const()[name = string("op_787"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_6 = const()[name = string("concat_6"), val = tensor<int32, [3]>([1, 1500, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_788_cast_fp16 = transpose(perm = var_787, x = var_786_cast_fp16)[name = string("transpose_92")]; | |
| tensor<fp16, [1, 1500, 768]> x_83_cast_fp16 = reshape(shape = concat_6, x = var_788_cast_fp16)[name = string("x_83_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_792_to_fp16 = const()[name = string("op_792_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(94813184)))]; | |
| tensor<fp16, [768]> var_793_to_fp16 = const()[name = string("op_793_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95992896)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_39_cast_fp16 = linear(bias = var_793_to_fp16, weight = var_792_to_fp16, x = x_83_cast_fp16)[name = string("linear_39_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_85_cast_fp16 = add(x = x_79_cast_fp16, y = linear_39_cast_fp16)[name = string("x_85_cast_fp16")]; | |
| tensor<int32, [1]> var_800_axes_0 = const()[name = string("op_800_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_6_mlp_ln_weight_to_fp16 = const()[name = string("blocks_6_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95994496)))]; | |
| tensor<fp16, [768]> blocks_6_mlp_ln_bias_to_fp16 = const()[name = string("blocks_6_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95996096)))]; | |
| tensor<fp16, [1, 1500, 768]> var_800_cast_fp16 = layer_norm(axes = var_800_axes_0, beta = blocks_6_mlp_ln_bias_to_fp16, epsilon = var_726_to_fp16, gamma = blocks_6_mlp_ln_weight_to_fp16, x = x_85_cast_fp16)[name = string("op_800_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> var_809_to_fp16 = const()[name = string("op_809_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(95997696)))]; | |
| tensor<fp16, [3072]> var_810_to_fp16 = const()[name = string("op_810_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100716352)))]; | |
| tensor<fp16, [1, 1500, 3072]> linear_40_cast_fp16 = linear(bias = var_810_to_fp16, weight = var_809_to_fp16, x = var_800_cast_fp16)[name = string("linear_40_cast_fp16")]; | |
| string x_89_mode_0 = const()[name = string("x_89_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_89_cast_fp16 = gelu(mode = x_89_mode_0, x = linear_40_cast_fp16)[name = string("x_89_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> var_815_to_fp16 = const()[name = string("op_815_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100722560)))]; | |
| tensor<fp16, [768]> var_816_to_fp16 = const()[name = string("op_816_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105441216)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_41_cast_fp16 = linear(bias = var_816_to_fp16, weight = var_815_to_fp16, x = x_89_cast_fp16)[name = string("linear_41_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_91_cast_fp16 = add(x = x_85_cast_fp16, y = linear_41_cast_fp16)[name = string("x_91_cast_fp16")]; | |
| int32 var_826 = const()[name = string("op_826"), val = int32(-1)]; | |
| tensor<int32, [1]> var_842_axes_0 = const()[name = string("op_842_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_7_attn_ln_weight_to_fp16 = const()[name = string("blocks_7_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105442816)))]; | |
| tensor<fp16, [768]> blocks_7_attn_ln_bias_to_fp16 = const()[name = string("blocks_7_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105444416)))]; | |
| fp16 var_832_to_fp16 = const()[name = string("op_832_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_842_cast_fp16 = layer_norm(axes = var_842_axes_0, beta = blocks_7_attn_ln_bias_to_fp16, epsilon = var_832_to_fp16, gamma = blocks_7_attn_ln_weight_to_fp16, x = x_91_cast_fp16)[name = string("op_842_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_853_to_fp16 = const()[name = string("op_853_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105446016)))]; | |
| tensor<fp16, [768]> var_854_to_fp16 = const()[name = string("op_854_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106625728)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_42_cast_fp16 = linear(bias = var_854_to_fp16, weight = var_853_to_fp16, x = var_842_cast_fp16)[name = string("linear_42_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_857_to_fp16 = const()[name = string("op_857_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(106627328)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_43_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_857_to_fp16, x = var_842_cast_fp16)[name = string("linear_43_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_861_to_fp16 = const()[name = string("op_861_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(107807040)))]; | |
| tensor<fp16, [768]> var_862_to_fp16 = const()[name = string("op_862_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108986752)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_44_cast_fp16 = linear(bias = var_862_to_fp16, weight = var_861_to_fp16, x = var_842_cast_fp16)[name = string("linear_44_cast_fp16")]; | |
| tensor<int32, [4]> var_870 = const()[name = string("op_870"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_871_cast_fp16 = reshape(shape = var_870, x = linear_42_cast_fp16)[name = string("op_871_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_98_to_fp16 = const()[name = string("const_98_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_31_cast_fp16 = mul(x = var_871_cast_fp16, y = const_98_to_fp16)[name = string("q_31_cast_fp16")]; | |
| tensor<int32, [4]> var_877 = const()[name = string("op_877"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_878_cast_fp16 = reshape(shape = var_877, x = linear_43_cast_fp16)[name = string("op_878_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_99_to_fp16 = const()[name = string("const_99_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_31_cast_fp16 = mul(x = var_878_cast_fp16, y = const_99_to_fp16)[name = string("k_31_cast_fp16")]; | |
| tensor<int32, [4]> var_884 = const()[name = string("op_884"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_885_cast_fp16 = reshape(shape = var_884, x = linear_44_cast_fp16)[name = string("op_885_cast_fp16")]; | |
| tensor<int32, [4]> var_886 = const()[name = string("op_886"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_15_transpose_x_0 = const()[name = string("qk_15_transpose_x_0"), val = bool(false)]; | |
| bool qk_15_transpose_y_0 = const()[name = string("qk_15_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_62_perm_0 = const()[name = string("transpose_62_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_63_perm_0 = const()[name = string("transpose_63_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_63 = transpose(perm = transpose_63_perm_0, x = k_31_cast_fp16)[name = string("transpose_89")]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_62 = transpose(perm = transpose_62_perm_0, x = q_31_cast_fp16)[name = string("transpose_90")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_15_cast_fp16 = matmul(transpose_x = qk_15_transpose_x_0, transpose_y = qk_15_transpose_y_0, x = transpose_62, y = transpose_63)[name = string("qk_15_cast_fp16")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> var_890_cast_fp16 = softmax(axis = var_826, x = qk_15_cast_fp16)[name = string("op_890_cast_fp16")]; | |
| bool var_892_transpose_x_0 = const()[name = string("op_892_transpose_x_0"), val = bool(false)]; | |
| bool var_892_transpose_y_0 = const()[name = string("op_892_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> v_31_cast_fp16 = transpose(perm = var_886, x = var_885_cast_fp16)[name = string("transpose_91")]; | |
| tensor<fp16, [1, 12, 1500, 64]> var_892_cast_fp16 = matmul(transpose_x = var_892_transpose_x_0, transpose_y = var_892_transpose_y_0, x = var_890_cast_fp16, y = v_31_cast_fp16)[name = string("op_892_cast_fp16")]; | |
| tensor<int32, [4]> var_893 = const()[name = string("op_893"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_7 = const()[name = string("concat_7"), val = tensor<int32, [3]>([1, 1500, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_894_cast_fp16 = transpose(perm = var_893, x = var_892_cast_fp16)[name = string("transpose_88")]; | |
| tensor<fp16, [1, 1500, 768]> x_95_cast_fp16 = reshape(shape = concat_7, x = var_894_cast_fp16)[name = string("x_95_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_898_to_fp16 = const()[name = string("op_898_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(108988352)))]; | |
| tensor<fp16, [768]> var_899_to_fp16 = const()[name = string("op_899_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110168064)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_45_cast_fp16 = linear(bias = var_899_to_fp16, weight = var_898_to_fp16, x = x_95_cast_fp16)[name = string("linear_45_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_97_cast_fp16 = add(x = x_91_cast_fp16, y = linear_45_cast_fp16)[name = string("x_97_cast_fp16")]; | |
| tensor<int32, [1]> var_906_axes_0 = const()[name = string("op_906_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_7_mlp_ln_weight_to_fp16 = const()[name = string("blocks_7_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110169664)))]; | |
| tensor<fp16, [768]> blocks_7_mlp_ln_bias_to_fp16 = const()[name = string("blocks_7_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110171264)))]; | |
| tensor<fp16, [1, 1500, 768]> var_906_cast_fp16 = layer_norm(axes = var_906_axes_0, beta = blocks_7_mlp_ln_bias_to_fp16, epsilon = var_832_to_fp16, gamma = blocks_7_mlp_ln_weight_to_fp16, x = x_97_cast_fp16)[name = string("op_906_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> var_915_to_fp16 = const()[name = string("op_915_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110172864)))]; | |
| tensor<fp16, [3072]> var_916_to_fp16 = const()[name = string("op_916_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114891520)))]; | |
| tensor<fp16, [1, 1500, 3072]> linear_46_cast_fp16 = linear(bias = var_916_to_fp16, weight = var_915_to_fp16, x = var_906_cast_fp16)[name = string("linear_46_cast_fp16")]; | |
| string x_101_mode_0 = const()[name = string("x_101_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_101_cast_fp16 = gelu(mode = x_101_mode_0, x = linear_46_cast_fp16)[name = string("x_101_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> var_921_to_fp16 = const()[name = string("op_921_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114897728)))]; | |
| tensor<fp16, [768]> var_922_to_fp16 = const()[name = string("op_922_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119616384)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_47_cast_fp16 = linear(bias = var_922_to_fp16, weight = var_921_to_fp16, x = x_101_cast_fp16)[name = string("linear_47_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_103_cast_fp16 = add(x = x_97_cast_fp16, y = linear_47_cast_fp16)[name = string("x_103_cast_fp16")]; | |
| int32 var_932 = const()[name = string("op_932"), val = int32(-1)]; | |
| tensor<int32, [1]> var_948_axes_0 = const()[name = string("op_948_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_8_attn_ln_weight_to_fp16 = const()[name = string("blocks_8_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119617984)))]; | |
| tensor<fp16, [768]> blocks_8_attn_ln_bias_to_fp16 = const()[name = string("blocks_8_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119619584)))]; | |
| fp16 var_938_to_fp16 = const()[name = string("op_938_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_948_cast_fp16 = layer_norm(axes = var_948_axes_0, beta = blocks_8_attn_ln_bias_to_fp16, epsilon = var_938_to_fp16, gamma = blocks_8_attn_ln_weight_to_fp16, x = x_103_cast_fp16)[name = string("op_948_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_959_to_fp16 = const()[name = string("op_959_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119621184)))]; | |
| tensor<fp16, [768]> var_960_to_fp16 = const()[name = string("op_960_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120800896)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_48_cast_fp16 = linear(bias = var_960_to_fp16, weight = var_959_to_fp16, x = var_948_cast_fp16)[name = string("linear_48_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_963_to_fp16 = const()[name = string("op_963_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120802496)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_49_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_963_to_fp16, x = var_948_cast_fp16)[name = string("linear_49_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_967_to_fp16 = const()[name = string("op_967_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121982208)))]; | |
| tensor<fp16, [768]> var_968_to_fp16 = const()[name = string("op_968_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123161920)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_50_cast_fp16 = linear(bias = var_968_to_fp16, weight = var_967_to_fp16, x = var_948_cast_fp16)[name = string("linear_50_cast_fp16")]; | |
| tensor<int32, [4]> var_976 = const()[name = string("op_976"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_977_cast_fp16 = reshape(shape = var_976, x = linear_48_cast_fp16)[name = string("op_977_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_100_to_fp16 = const()[name = string("const_100_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_35_cast_fp16 = mul(x = var_977_cast_fp16, y = const_100_to_fp16)[name = string("q_35_cast_fp16")]; | |
| tensor<int32, [4]> var_983 = const()[name = string("op_983"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_984_cast_fp16 = reshape(shape = var_983, x = linear_49_cast_fp16)[name = string("op_984_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_101_to_fp16 = const()[name = string("const_101_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_35_cast_fp16 = mul(x = var_984_cast_fp16, y = const_101_to_fp16)[name = string("k_35_cast_fp16")]; | |
| tensor<int32, [4]> var_990 = const()[name = string("op_990"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_991_cast_fp16 = reshape(shape = var_990, x = linear_50_cast_fp16)[name = string("op_991_cast_fp16")]; | |
| tensor<int32, [4]> var_992 = const()[name = string("op_992"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_17_transpose_x_0 = const()[name = string("qk_17_transpose_x_0"), val = bool(false)]; | |
| bool qk_17_transpose_y_0 = const()[name = string("qk_17_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_64_perm_0 = const()[name = string("transpose_64_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_65_perm_0 = const()[name = string("transpose_65_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_65 = transpose(perm = transpose_65_perm_0, x = k_35_cast_fp16)[name = string("transpose_85")]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_64 = transpose(perm = transpose_64_perm_0, x = q_35_cast_fp16)[name = string("transpose_86")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_17_cast_fp16 = matmul(transpose_x = qk_17_transpose_x_0, transpose_y = qk_17_transpose_y_0, x = transpose_64, y = transpose_65)[name = string("qk_17_cast_fp16")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> var_996_cast_fp16 = softmax(axis = var_932, x = qk_17_cast_fp16)[name = string("op_996_cast_fp16")]; | |
| bool var_998_transpose_x_0 = const()[name = string("op_998_transpose_x_0"), val = bool(false)]; | |
| bool var_998_transpose_y_0 = const()[name = string("op_998_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> v_35_cast_fp16 = transpose(perm = var_992, x = var_991_cast_fp16)[name = string("transpose_87")]; | |
| tensor<fp16, [1, 12, 1500, 64]> var_998_cast_fp16 = matmul(transpose_x = var_998_transpose_x_0, transpose_y = var_998_transpose_y_0, x = var_996_cast_fp16, y = v_35_cast_fp16)[name = string("op_998_cast_fp16")]; | |
| tensor<int32, [4]> var_999 = const()[name = string("op_999"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_8 = const()[name = string("concat_8"), val = tensor<int32, [3]>([1, 1500, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_1000_cast_fp16 = transpose(perm = var_999, x = var_998_cast_fp16)[name = string("transpose_84")]; | |
| tensor<fp16, [1, 1500, 768]> x_107_cast_fp16 = reshape(shape = concat_8, x = var_1000_cast_fp16)[name = string("x_107_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_1004_to_fp16 = const()[name = string("op_1004_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123163520)))]; | |
| tensor<fp16, [768]> var_1005_to_fp16 = const()[name = string("op_1005_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124343232)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_51_cast_fp16 = linear(bias = var_1005_to_fp16, weight = var_1004_to_fp16, x = x_107_cast_fp16)[name = string("linear_51_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_109_cast_fp16 = add(x = x_103_cast_fp16, y = linear_51_cast_fp16)[name = string("x_109_cast_fp16")]; | |
| tensor<int32, [1]> var_1012_axes_0 = const()[name = string("op_1012_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_8_mlp_ln_weight_to_fp16 = const()[name = string("blocks_8_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124344832)))]; | |
| tensor<fp16, [768]> blocks_8_mlp_ln_bias_to_fp16 = const()[name = string("blocks_8_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124346432)))]; | |
| tensor<fp16, [1, 1500, 768]> var_1012_cast_fp16 = layer_norm(axes = var_1012_axes_0, beta = blocks_8_mlp_ln_bias_to_fp16, epsilon = var_938_to_fp16, gamma = blocks_8_mlp_ln_weight_to_fp16, x = x_109_cast_fp16)[name = string("op_1012_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> var_1021_to_fp16 = const()[name = string("op_1021_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124348032)))]; | |
| tensor<fp16, [3072]> var_1022_to_fp16 = const()[name = string("op_1022_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129066688)))]; | |
| tensor<fp16, [1, 1500, 3072]> linear_52_cast_fp16 = linear(bias = var_1022_to_fp16, weight = var_1021_to_fp16, x = var_1012_cast_fp16)[name = string("linear_52_cast_fp16")]; | |
| string x_113_mode_0 = const()[name = string("x_113_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_113_cast_fp16 = gelu(mode = x_113_mode_0, x = linear_52_cast_fp16)[name = string("x_113_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> var_1027_to_fp16 = const()[name = string("op_1027_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(129072896)))]; | |
| tensor<fp16, [768]> var_1028_to_fp16 = const()[name = string("op_1028_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133791552)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_53_cast_fp16 = linear(bias = var_1028_to_fp16, weight = var_1027_to_fp16, x = x_113_cast_fp16)[name = string("linear_53_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_115_cast_fp16 = add(x = x_109_cast_fp16, y = linear_53_cast_fp16)[name = string("x_115_cast_fp16")]; | |
| int32 var_1038 = const()[name = string("op_1038"), val = int32(-1)]; | |
| tensor<int32, [1]> var_1054_axes_0 = const()[name = string("op_1054_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_9_attn_ln_weight_to_fp16 = const()[name = string("blocks_9_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133793152)))]; | |
| tensor<fp16, [768]> blocks_9_attn_ln_bias_to_fp16 = const()[name = string("blocks_9_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133794752)))]; | |
| fp16 var_1044_to_fp16 = const()[name = string("op_1044_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_1054_cast_fp16 = layer_norm(axes = var_1054_axes_0, beta = blocks_9_attn_ln_bias_to_fp16, epsilon = var_1044_to_fp16, gamma = blocks_9_attn_ln_weight_to_fp16, x = x_115_cast_fp16)[name = string("op_1054_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_1065_to_fp16 = const()[name = string("op_1065_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(133796352)))]; | |
| tensor<fp16, [768]> var_1066_to_fp16 = const()[name = string("op_1066_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134976064)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_54_cast_fp16 = linear(bias = var_1066_to_fp16, weight = var_1065_to_fp16, x = var_1054_cast_fp16)[name = string("linear_54_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_1069_to_fp16 = const()[name = string("op_1069_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(134977664)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_55_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1069_to_fp16, x = var_1054_cast_fp16)[name = string("linear_55_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_1073_to_fp16 = const()[name = string("op_1073_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(136157376)))]; | |
| tensor<fp16, [768]> var_1074_to_fp16 = const()[name = string("op_1074_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137337088)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_56_cast_fp16 = linear(bias = var_1074_to_fp16, weight = var_1073_to_fp16, x = var_1054_cast_fp16)[name = string("linear_56_cast_fp16")]; | |
| tensor<int32, [4]> var_1082 = const()[name = string("op_1082"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_1083_cast_fp16 = reshape(shape = var_1082, x = linear_54_cast_fp16)[name = string("op_1083_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_102_to_fp16 = const()[name = string("const_102_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_39_cast_fp16 = mul(x = var_1083_cast_fp16, y = const_102_to_fp16)[name = string("q_39_cast_fp16")]; | |
| tensor<int32, [4]> var_1089 = const()[name = string("op_1089"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_1090_cast_fp16 = reshape(shape = var_1089, x = linear_55_cast_fp16)[name = string("op_1090_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_103_to_fp16 = const()[name = string("const_103_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_39_cast_fp16 = mul(x = var_1090_cast_fp16, y = const_103_to_fp16)[name = string("k_39_cast_fp16")]; | |
| tensor<int32, [4]> var_1096 = const()[name = string("op_1096"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_1097_cast_fp16 = reshape(shape = var_1096, x = linear_56_cast_fp16)[name = string("op_1097_cast_fp16")]; | |
| tensor<int32, [4]> var_1098 = const()[name = string("op_1098"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_19_transpose_x_0 = const()[name = string("qk_19_transpose_x_0"), val = bool(false)]; | |
| bool qk_19_transpose_y_0 = const()[name = string("qk_19_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_66_perm_0 = const()[name = string("transpose_66_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_67_perm_0 = const()[name = string("transpose_67_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_67 = transpose(perm = transpose_67_perm_0, x = k_39_cast_fp16)[name = string("transpose_81")]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_66 = transpose(perm = transpose_66_perm_0, x = q_39_cast_fp16)[name = string("transpose_82")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_19_cast_fp16 = matmul(transpose_x = qk_19_transpose_x_0, transpose_y = qk_19_transpose_y_0, x = transpose_66, y = transpose_67)[name = string("qk_19_cast_fp16")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> var_1102_cast_fp16 = softmax(axis = var_1038, x = qk_19_cast_fp16)[name = string("op_1102_cast_fp16")]; | |
| bool var_1104_transpose_x_0 = const()[name = string("op_1104_transpose_x_0"), val = bool(false)]; | |
| bool var_1104_transpose_y_0 = const()[name = string("op_1104_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> v_39_cast_fp16 = transpose(perm = var_1098, x = var_1097_cast_fp16)[name = string("transpose_83")]; | |
| tensor<fp16, [1, 12, 1500, 64]> var_1104_cast_fp16 = matmul(transpose_x = var_1104_transpose_x_0, transpose_y = var_1104_transpose_y_0, x = var_1102_cast_fp16, y = v_39_cast_fp16)[name = string("op_1104_cast_fp16")]; | |
| tensor<int32, [4]> var_1105 = const()[name = string("op_1105"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_9 = const()[name = string("concat_9"), val = tensor<int32, [3]>([1, 1500, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_1106_cast_fp16 = transpose(perm = var_1105, x = var_1104_cast_fp16)[name = string("transpose_80")]; | |
| tensor<fp16, [1, 1500, 768]> x_119_cast_fp16 = reshape(shape = concat_9, x = var_1106_cast_fp16)[name = string("x_119_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_1110_to_fp16 = const()[name = string("op_1110_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137338688)))]; | |
| tensor<fp16, [768]> var_1111_to_fp16 = const()[name = string("op_1111_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138518400)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_57_cast_fp16 = linear(bias = var_1111_to_fp16, weight = var_1110_to_fp16, x = x_119_cast_fp16)[name = string("linear_57_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_121_cast_fp16 = add(x = x_115_cast_fp16, y = linear_57_cast_fp16)[name = string("x_121_cast_fp16")]; | |
| tensor<int32, [1]> var_1118_axes_0 = const()[name = string("op_1118_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_9_mlp_ln_weight_to_fp16 = const()[name = string("blocks_9_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138520000)))]; | |
| tensor<fp16, [768]> blocks_9_mlp_ln_bias_to_fp16 = const()[name = string("blocks_9_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138521600)))]; | |
| tensor<fp16, [1, 1500, 768]> var_1118_cast_fp16 = layer_norm(axes = var_1118_axes_0, beta = blocks_9_mlp_ln_bias_to_fp16, epsilon = var_1044_to_fp16, gamma = blocks_9_mlp_ln_weight_to_fp16, x = x_121_cast_fp16)[name = string("op_1118_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> var_1127_to_fp16 = const()[name = string("op_1127_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(138523200)))]; | |
| tensor<fp16, [3072]> var_1128_to_fp16 = const()[name = string("op_1128_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143241856)))]; | |
| tensor<fp16, [1, 1500, 3072]> linear_58_cast_fp16 = linear(bias = var_1128_to_fp16, weight = var_1127_to_fp16, x = var_1118_cast_fp16)[name = string("linear_58_cast_fp16")]; | |
| string x_125_mode_0 = const()[name = string("x_125_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_125_cast_fp16 = gelu(mode = x_125_mode_0, x = linear_58_cast_fp16)[name = string("x_125_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> var_1133_to_fp16 = const()[name = string("op_1133_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(143248064)))]; | |
| tensor<fp16, [768]> var_1134_to_fp16 = const()[name = string("op_1134_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147966720)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_59_cast_fp16 = linear(bias = var_1134_to_fp16, weight = var_1133_to_fp16, x = x_125_cast_fp16)[name = string("linear_59_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_127_cast_fp16 = add(x = x_121_cast_fp16, y = linear_59_cast_fp16)[name = string("x_127_cast_fp16")]; | |
| int32 var_1144 = const()[name = string("op_1144"), val = int32(-1)]; | |
| tensor<int32, [1]> var_1160_axes_0 = const()[name = string("op_1160_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_10_attn_ln_weight_to_fp16 = const()[name = string("blocks_10_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147968320)))]; | |
| tensor<fp16, [768]> blocks_10_attn_ln_bias_to_fp16 = const()[name = string("blocks_10_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147969920)))]; | |
| fp16 var_1150_to_fp16 = const()[name = string("op_1150_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_1160_cast_fp16 = layer_norm(axes = var_1160_axes_0, beta = blocks_10_attn_ln_bias_to_fp16, epsilon = var_1150_to_fp16, gamma = blocks_10_attn_ln_weight_to_fp16, x = x_127_cast_fp16)[name = string("op_1160_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_1171_to_fp16 = const()[name = string("op_1171_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(147971520)))]; | |
| tensor<fp16, [768]> var_1172_to_fp16 = const()[name = string("op_1172_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149151232)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_60_cast_fp16 = linear(bias = var_1172_to_fp16, weight = var_1171_to_fp16, x = var_1160_cast_fp16)[name = string("linear_60_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_1175_to_fp16 = const()[name = string("op_1175_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(149152832)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_61_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1175_to_fp16, x = var_1160_cast_fp16)[name = string("linear_61_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_1179_to_fp16 = const()[name = string("op_1179_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150332544)))]; | |
| tensor<fp16, [768]> var_1180_to_fp16 = const()[name = string("op_1180_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151512256)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_62_cast_fp16 = linear(bias = var_1180_to_fp16, weight = var_1179_to_fp16, x = var_1160_cast_fp16)[name = string("linear_62_cast_fp16")]; | |
| tensor<int32, [4]> var_1188 = const()[name = string("op_1188"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_1189_cast_fp16 = reshape(shape = var_1188, x = linear_60_cast_fp16)[name = string("op_1189_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_104_to_fp16 = const()[name = string("const_104_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_43_cast_fp16 = mul(x = var_1189_cast_fp16, y = const_104_to_fp16)[name = string("q_43_cast_fp16")]; | |
| tensor<int32, [4]> var_1195 = const()[name = string("op_1195"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_1196_cast_fp16 = reshape(shape = var_1195, x = linear_61_cast_fp16)[name = string("op_1196_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_105_to_fp16 = const()[name = string("const_105_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_43_cast_fp16 = mul(x = var_1196_cast_fp16, y = const_105_to_fp16)[name = string("k_43_cast_fp16")]; | |
| tensor<int32, [4]> var_1202 = const()[name = string("op_1202"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_1203_cast_fp16 = reshape(shape = var_1202, x = linear_62_cast_fp16)[name = string("op_1203_cast_fp16")]; | |
| tensor<int32, [4]> var_1204 = const()[name = string("op_1204"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_21_transpose_x_0 = const()[name = string("qk_21_transpose_x_0"), val = bool(false)]; | |
| bool qk_21_transpose_y_0 = const()[name = string("qk_21_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_68_perm_0 = const()[name = string("transpose_68_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_69_perm_0 = const()[name = string("transpose_69_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_69 = transpose(perm = transpose_69_perm_0, x = k_43_cast_fp16)[name = string("transpose_77")]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_68 = transpose(perm = transpose_68_perm_0, x = q_43_cast_fp16)[name = string("transpose_78")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_21_cast_fp16 = matmul(transpose_x = qk_21_transpose_x_0, transpose_y = qk_21_transpose_y_0, x = transpose_68, y = transpose_69)[name = string("qk_21_cast_fp16")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> var_1208_cast_fp16 = softmax(axis = var_1144, x = qk_21_cast_fp16)[name = string("op_1208_cast_fp16")]; | |
| bool var_1210_transpose_x_0 = const()[name = string("op_1210_transpose_x_0"), val = bool(false)]; | |
| bool var_1210_transpose_y_0 = const()[name = string("op_1210_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> v_43_cast_fp16 = transpose(perm = var_1204, x = var_1203_cast_fp16)[name = string("transpose_79")]; | |
| tensor<fp16, [1, 12, 1500, 64]> var_1210_cast_fp16 = matmul(transpose_x = var_1210_transpose_x_0, transpose_y = var_1210_transpose_y_0, x = var_1208_cast_fp16, y = v_43_cast_fp16)[name = string("op_1210_cast_fp16")]; | |
| tensor<int32, [4]> var_1211 = const()[name = string("op_1211"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_10 = const()[name = string("concat_10"), val = tensor<int32, [3]>([1, 1500, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_1212_cast_fp16 = transpose(perm = var_1211, x = var_1210_cast_fp16)[name = string("transpose_76")]; | |
| tensor<fp16, [1, 1500, 768]> x_131_cast_fp16 = reshape(shape = concat_10, x = var_1212_cast_fp16)[name = string("x_131_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_1216_to_fp16 = const()[name = string("op_1216_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(151513856)))]; | |
| tensor<fp16, [768]> var_1217_to_fp16 = const()[name = string("op_1217_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152693568)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_63_cast_fp16 = linear(bias = var_1217_to_fp16, weight = var_1216_to_fp16, x = x_131_cast_fp16)[name = string("linear_63_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_133_cast_fp16 = add(x = x_127_cast_fp16, y = linear_63_cast_fp16)[name = string("x_133_cast_fp16")]; | |
| tensor<int32, [1]> var_1224_axes_0 = const()[name = string("op_1224_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_10_mlp_ln_weight_to_fp16 = const()[name = string("blocks_10_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152695168)))]; | |
| tensor<fp16, [768]> blocks_10_mlp_ln_bias_to_fp16 = const()[name = string("blocks_10_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152696768)))]; | |
| tensor<fp16, [1, 1500, 768]> var_1224_cast_fp16 = layer_norm(axes = var_1224_axes_0, beta = blocks_10_mlp_ln_bias_to_fp16, epsilon = var_1150_to_fp16, gamma = blocks_10_mlp_ln_weight_to_fp16, x = x_133_cast_fp16)[name = string("op_1224_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> var_1233_to_fp16 = const()[name = string("op_1233_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152698368)))]; | |
| tensor<fp16, [3072]> var_1234_to_fp16 = const()[name = string("op_1234_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157417024)))]; | |
| tensor<fp16, [1, 1500, 3072]> linear_64_cast_fp16 = linear(bias = var_1234_to_fp16, weight = var_1233_to_fp16, x = var_1224_cast_fp16)[name = string("linear_64_cast_fp16")]; | |
| string x_137_mode_0 = const()[name = string("x_137_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_137_cast_fp16 = gelu(mode = x_137_mode_0, x = linear_64_cast_fp16)[name = string("x_137_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> var_1239_to_fp16 = const()[name = string("op_1239_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157423232)))]; | |
| tensor<fp16, [768]> var_1240_to_fp16 = const()[name = string("op_1240_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162141888)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_65_cast_fp16 = linear(bias = var_1240_to_fp16, weight = var_1239_to_fp16, x = x_137_cast_fp16)[name = string("linear_65_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_139_cast_fp16 = add(x = x_133_cast_fp16, y = linear_65_cast_fp16)[name = string("x_139_cast_fp16")]; | |
| int32 var_1250 = const()[name = string("op_1250"), val = int32(-1)]; | |
| tensor<int32, [1]> var_1266_axes_0 = const()[name = string("op_1266_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_11_attn_ln_weight_to_fp16 = const()[name = string("blocks_11_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162143488)))]; | |
| tensor<fp16, [768]> blocks_11_attn_ln_bias_to_fp16 = const()[name = string("blocks_11_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162145088)))]; | |
| fp16 var_1256_to_fp16 = const()[name = string("op_1256_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_1266_cast_fp16 = layer_norm(axes = var_1266_axes_0, beta = blocks_11_attn_ln_bias_to_fp16, epsilon = var_1256_to_fp16, gamma = blocks_11_attn_ln_weight_to_fp16, x = x_139_cast_fp16)[name = string("op_1266_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_1277_to_fp16 = const()[name = string("op_1277_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(162146688)))]; | |
| tensor<fp16, [768]> var_1278_to_fp16 = const()[name = string("op_1278_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163326400)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_66_cast_fp16 = linear(bias = var_1278_to_fp16, weight = var_1277_to_fp16, x = var_1266_cast_fp16)[name = string("linear_66_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_1281_to_fp16 = const()[name = string("op_1281_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163328000)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_67_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1281_to_fp16, x = var_1266_cast_fp16)[name = string("linear_67_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_1285_to_fp16 = const()[name = string("op_1285_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(164507712)))]; | |
| tensor<fp16, [768]> var_1286_to_fp16 = const()[name = string("op_1286_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165687424)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_68_cast_fp16 = linear(bias = var_1286_to_fp16, weight = var_1285_to_fp16, x = var_1266_cast_fp16)[name = string("linear_68_cast_fp16")]; | |
| tensor<int32, [4]> var_1294 = const()[name = string("op_1294"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_1295_cast_fp16 = reshape(shape = var_1294, x = linear_66_cast_fp16)[name = string("op_1295_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_106_to_fp16 = const()[name = string("const_106_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_cast_fp16 = mul(x = var_1295_cast_fp16, y = const_106_to_fp16)[name = string("q_cast_fp16")]; | |
| tensor<int32, [4]> var_1301 = const()[name = string("op_1301"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_1302_cast_fp16 = reshape(shape = var_1301, x = linear_67_cast_fp16)[name = string("op_1302_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_107_to_fp16 = const()[name = string("const_107_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_cast_fp16 = mul(x = var_1302_cast_fp16, y = const_107_to_fp16)[name = string("k_cast_fp16")]; | |
| tensor<int32, [4]> var_1308 = const()[name = string("op_1308"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_1309_cast_fp16 = reshape(shape = var_1308, x = linear_68_cast_fp16)[name = string("op_1309_cast_fp16")]; | |
| tensor<int32, [4]> var_1310 = const()[name = string("op_1310"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| bool qk_transpose_x_0 = const()[name = string("qk_transpose_x_0"), val = bool(false)]; | |
| bool qk_transpose_y_0 = const()[name = string("qk_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_70_perm_0 = const()[name = string("transpose_70_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_71_perm_0 = const()[name = string("transpose_71_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_71 = transpose(perm = transpose_71_perm_0, x = k_cast_fp16)[name = string("transpose_73")]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_70 = transpose(perm = transpose_70_perm_0, x = q_cast_fp16)[name = string("transpose_74")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_cast_fp16 = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_70, y = transpose_71)[name = string("qk_cast_fp16")]; | |
| tensor<fp16, [1, 12, 1500, 1500]> var_1314_cast_fp16 = softmax(axis = var_1250, x = qk_cast_fp16)[name = string("op_1314_cast_fp16")]; | |
| bool var_1316_transpose_x_0 = const()[name = string("op_1316_transpose_x_0"), val = bool(false)]; | |
| bool var_1316_transpose_y_0 = const()[name = string("op_1316_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> v_cast_fp16 = transpose(perm = var_1310, x = var_1309_cast_fp16)[name = string("transpose_75")]; | |
| tensor<fp16, [1, 12, 1500, 64]> var_1316_cast_fp16 = matmul(transpose_x = var_1316_transpose_x_0, transpose_y = var_1316_transpose_y_0, x = var_1314_cast_fp16, y = v_cast_fp16)[name = string("op_1316_cast_fp16")]; | |
| tensor<int32, [4]> var_1317 = const()[name = string("op_1317"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_11 = const()[name = string("concat_11"), val = tensor<int32, [3]>([1, 1500, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_1318_cast_fp16 = transpose(perm = var_1317, x = var_1316_cast_fp16)[name = string("transpose_72")]; | |
| tensor<fp16, [1, 1500, 768]> x_143_cast_fp16 = reshape(shape = concat_11, x = var_1318_cast_fp16)[name = string("x_143_cast_fp16")]; | |
| tensor<fp16, [768, 768]> var_1322_to_fp16 = const()[name = string("op_1322_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165689024)))]; | |
| tensor<fp16, [768]> var_1323_to_fp16 = const()[name = string("op_1323_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166868736)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_69_cast_fp16 = linear(bias = var_1323_to_fp16, weight = var_1322_to_fp16, x = x_143_cast_fp16)[name = string("linear_69_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_145_cast_fp16 = add(x = x_139_cast_fp16, y = linear_69_cast_fp16)[name = string("x_145_cast_fp16")]; | |
| tensor<int32, [1]> var_1330_axes_0 = const()[name = string("op_1330_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_11_mlp_ln_weight_to_fp16 = const()[name = string("blocks_11_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166870336)))]; | |
| tensor<fp16, [768]> blocks_11_mlp_ln_bias_to_fp16 = const()[name = string("blocks_11_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166871936)))]; | |
| tensor<fp16, [1, 1500, 768]> var_1330_cast_fp16 = layer_norm(axes = var_1330_axes_0, beta = blocks_11_mlp_ln_bias_to_fp16, epsilon = var_1256_to_fp16, gamma = blocks_11_mlp_ln_weight_to_fp16, x = x_145_cast_fp16)[name = string("op_1330_cast_fp16")]; | |
| tensor<fp16, [3072, 768]> var_1339_to_fp16 = const()[name = string("op_1339_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(166873536)))]; | |
| tensor<fp16, [3072]> var_1340_to_fp16 = const()[name = string("op_1340_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171592192)))]; | |
| tensor<fp16, [1, 1500, 3072]> linear_70_cast_fp16 = linear(bias = var_1340_to_fp16, weight = var_1339_to_fp16, x = var_1330_cast_fp16)[name = string("linear_70_cast_fp16")]; | |
| string x_149_mode_0 = const()[name = string("x_149_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_149_cast_fp16 = gelu(mode = x_149_mode_0, x = linear_70_cast_fp16)[name = string("x_149_cast_fp16")]; | |
| tensor<fp16, [768, 3072]> var_1345_to_fp16 = const()[name = string("op_1345_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(171598400)))]; | |
| tensor<fp16, [768]> var_1346_to_fp16 = const()[name = string("op_1346_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176317056)))]; | |
| tensor<fp16, [1, 1500, 768]> linear_71_cast_fp16 = linear(bias = var_1346_to_fp16, weight = var_1345_to_fp16, x = x_149_cast_fp16)[name = string("linear_71_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 768]> x_cast_fp16 = add(x = x_145_cast_fp16, y = linear_71_cast_fp16)[name = string("x_cast_fp16")]; | |
| tensor<int32, [1]> var_1359_axes_0 = const()[name = string("op_1359_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> ln_post_weight_to_fp16 = const()[name = string("ln_post_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176318656)))]; | |
| tensor<fp16, [768]> ln_post_bias_to_fp16 = const()[name = string("ln_post_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176320256)))]; | |
| fp16 var_1350_to_fp16 = const()[name = string("op_1350_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> output = layer_norm(axes = var_1359_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_1350_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast_fp16)[name = string("op_1359_cast_fp16")]; | |
| } -> (output); | |
| } |