| program(1.3) | |
| [buildInfo = dict<string, string>({{"coremlc-component-MIL", "3400.43.1"}, {"coremlc-version", "3400.58.2"}})] | |
| { | |
| func main<ios18>(tensor<fp32, [1, 1500, 1280]> linear_95_cast_fp16, tensor<fp32, [1, 1500, 1280]> x_193_cast_fp16) { | |
| tensor<fp16, [1280]> linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(64)))]; | |
| string cast_1_dtype_0 = const()[name = string("cast_1_dtype_0"), val = string("fp16")]; | |
| string cast_0_dtype_0 = const()[name = string("cast_0_dtype_0"), val = string("fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> cast_0 = cast(dtype = cast_0_dtype_0, x = linear_95_cast_fp16)[name = string("cast_0")]; | |
| tensor<fp16, [1, 1500, 1280]> cast_1 = cast(dtype = cast_1_dtype_0, x = x_193_cast_fp16)[name = string("cast_1")]; | |
| tensor<fp16, [1, 1500, 1280]> x_199_cast_fp16 = add(x = cast_1, y = cast_0)[name = string("x_199_cast_fp16")]; | |
| int32 var_1820 = const()[name = string("op_1820"), val = int32(-1)]; | |
| tensor<int32, [1]> var_1836_axes_0 = const()[name = string("op_1836_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_16_attn_ln_weight_to_fp16 = const()[name = string("blocks_16_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(2688)))]; | |
| tensor<fp16, [1280]> blocks_16_attn_ln_bias_to_fp16 = const()[name = string("blocks_16_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(5312)))]; | |
| fp16 var_1826_to_fp16 = const()[name = string("op_1826_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1280]> var_1836_cast_fp16 = layer_norm(axes = var_1836_axes_0, beta = blocks_16_attn_ln_bias_to_fp16, epsilon = var_1826_to_fp16, gamma = blocks_16_attn_ln_weight_to_fp16, x = x_199_cast_fp16)[name = string("op_1836_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_1847_to_fp16 = const()[name = string("op_1847_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(7936)))]; | |
| tensor<fp16, [1280]> var_1848_to_fp16 = const()[name = string("op_1848_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(3284800)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_96_cast_fp16 = linear(bias = var_1848_to_fp16, weight = var_1847_to_fp16, x = var_1836_cast_fp16)[name = string("linear_96_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_1851_to_fp16 = const()[name = string("op_1851_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(3287424)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_97_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1851_to_fp16, x = var_1836_cast_fp16)[name = string("linear_97_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_1855_to_fp16 = const()[name = string("op_1855_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(6564288)))]; | |
| tensor<fp16, [1280]> var_1856_to_fp16 = const()[name = string("op_1856_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(9841152)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_98_cast_fp16 = linear(bias = var_1856_to_fp16, weight = var_1855_to_fp16, x = var_1836_cast_fp16)[name = string("linear_98_cast_fp16")]; | |
| tensor<int32, [4]> var_1864 = const()[name = string("op_1864"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_1865_cast_fp16 = reshape(shape = var_1864, x = linear_96_cast_fp16)[name = string("op_1865_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_256_to_fp16 = const()[name = string("const_256_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> q_67_cast_fp16 = mul(x = var_1865_cast_fp16, y = const_256_to_fp16)[name = string("q_67_cast_fp16")]; | |
| tensor<int32, [4]> var_1871 = const()[name = string("op_1871"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_1872_cast_fp16 = reshape(shape = var_1871, x = linear_97_cast_fp16)[name = string("op_1872_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_257_to_fp16 = const()[name = string("const_257_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> k_67_cast_fp16 = mul(x = var_1872_cast_fp16, y = const_257_to_fp16)[name = string("k_67_cast_fp16")]; | |
| tensor<int32, [4]> var_1878 = const()[name = string("op_1878"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_1879_cast_fp16 = reshape(shape = var_1878, x = linear_98_cast_fp16)[name = string("op_1879_cast_fp16")]; | |
| tensor<int32, [4]> var_1880 = const()[name = string("op_1880"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| bool qk_33_transpose_x_0 = const()[name = string("qk_33_transpose_x_0"), val = bool(false)]; | |
| bool qk_33_transpose_y_0 = const()[name = string("qk_33_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, 20, 64, 1500]> transpose_65 = transpose(perm = transpose_65_perm_0, x = k_67_cast_fp16)[name = string("transpose_158")]; | |
| tensor<fp16, [1, 20, 1500, 64]> transpose_64 = transpose(perm = transpose_64_perm_0, x = q_67_cast_fp16)[name = string("transpose_159")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> qk_33_cast_fp16 = matmul(transpose_x = qk_33_transpose_x_0, transpose_y = qk_33_transpose_y_0, x = transpose_64, y = transpose_65)[name = string("qk_33_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> var_1884_cast_fp16 = softmax(axis = var_1820, x = qk_33_cast_fp16)[name = string("op_1884_cast_fp16")]; | |
| bool var_1886_transpose_x_0 = const()[name = string("op_1886_transpose_x_0"), val = bool(false)]; | |
| bool var_1886_transpose_y_0 = const()[name = string("op_1886_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 20, 1500, 64]> v_67_cast_fp16 = transpose(perm = var_1880, x = var_1879_cast_fp16)[name = string("transpose_157")]; | |
| tensor<fp16, [1, 20, 1500, 64]> var_1886_cast_fp16 = matmul(transpose_x = var_1886_transpose_x_0, transpose_y = var_1886_transpose_y_0, x = var_1884_cast_fp16, y = v_67_cast_fp16)[name = string("op_1886_cast_fp16")]; | |
| tensor<int32, [4]> var_1887 = const()[name = string("op_1887"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_16 = const()[name = string("concat_16"), val = tensor<int32, [3]>([1, 1500, 1280])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_1888_cast_fp16 = transpose(perm = var_1887, x = var_1886_cast_fp16)[name = string("transpose_156")]; | |
| tensor<fp16, [1, 1500, 1280]> x_203_cast_fp16 = reshape(shape = concat_16, x = var_1888_cast_fp16)[name = string("x_203_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_1892_to_fp16 = const()[name = string("op_1892_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(9843776)))]; | |
| tensor<fp16, [1280]> var_1893_to_fp16 = const()[name = string("op_1893_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(13120640)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_99_cast_fp16 = linear(bias = var_1893_to_fp16, weight = var_1892_to_fp16, x = x_203_cast_fp16)[name = string("linear_99_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_205_cast_fp16 = add(x = x_199_cast_fp16, y = linear_99_cast_fp16)[name = string("x_205_cast_fp16")]; | |
| tensor<int32, [1]> var_1900_axes_0 = const()[name = string("op_1900_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_16_mlp_ln_weight_to_fp16 = const()[name = string("blocks_16_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(13123264)))]; | |
| tensor<fp16, [1280]> blocks_16_mlp_ln_bias_to_fp16 = const()[name = string("blocks_16_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(13125888)))]; | |
| tensor<fp16, [1, 1500, 1280]> var_1900_cast_fp16 = layer_norm(axes = var_1900_axes_0, beta = blocks_16_mlp_ln_bias_to_fp16, epsilon = var_1826_to_fp16, gamma = blocks_16_mlp_ln_weight_to_fp16, x = x_205_cast_fp16)[name = string("op_1900_cast_fp16")]; | |
| tensor<fp16, [5120, 1280]> var_1909_to_fp16 = const()[name = string("op_1909_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(13128512)))]; | |
| tensor<fp16, [5120]> var_1910_to_fp16 = const()[name = string("op_1910_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(26235776)))]; | |
| tensor<fp16, [1, 1500, 5120]> linear_100_cast_fp16 = linear(bias = var_1910_to_fp16, weight = var_1909_to_fp16, x = var_1900_cast_fp16)[name = string("linear_100_cast_fp16")]; | |
| string x_209_mode_0 = const()[name = string("x_209_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 5120]> x_209_cast_fp16 = gelu(mode = x_209_mode_0, x = linear_100_cast_fp16)[name = string("x_209_cast_fp16")]; | |
| tensor<fp16, [1280, 5120]> var_1915_to_fp16 = const()[name = string("op_1915_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(26246080)))]; | |
| tensor<fp16, [1280]> var_1916_to_fp16 = const()[name = string("op_1916_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(39353344)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_101_cast_fp16 = linear(bias = var_1916_to_fp16, weight = var_1915_to_fp16, x = x_209_cast_fp16)[name = string("linear_101_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_211_cast_fp16 = add(x = x_205_cast_fp16, y = linear_101_cast_fp16)[name = string("x_211_cast_fp16")]; | |
| int32 var_1926 = const()[name = string("op_1926"), val = int32(-1)]; | |
| tensor<int32, [1]> var_1942_axes_0 = const()[name = string("op_1942_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_17_attn_ln_weight_to_fp16 = const()[name = string("blocks_17_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(39355968)))]; | |
| tensor<fp16, [1280]> blocks_17_attn_ln_bias_to_fp16 = const()[name = string("blocks_17_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(39358592)))]; | |
| fp16 var_1932_to_fp16 = const()[name = string("op_1932_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1280]> var_1942_cast_fp16 = layer_norm(axes = var_1942_axes_0, beta = blocks_17_attn_ln_bias_to_fp16, epsilon = var_1932_to_fp16, gamma = blocks_17_attn_ln_weight_to_fp16, x = x_211_cast_fp16)[name = string("op_1942_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_1953_to_fp16 = const()[name = string("op_1953_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(39361216)))]; | |
| tensor<fp16, [1280]> var_1954_to_fp16 = const()[name = string("op_1954_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(42638080)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_102_cast_fp16 = linear(bias = var_1954_to_fp16, weight = var_1953_to_fp16, x = var_1942_cast_fp16)[name = string("linear_102_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_1957_to_fp16 = const()[name = string("op_1957_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(42640704)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_103_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_1957_to_fp16, x = var_1942_cast_fp16)[name = string("linear_103_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_1961_to_fp16 = const()[name = string("op_1961_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(45917568)))]; | |
| tensor<fp16, [1280]> var_1962_to_fp16 = const()[name = string("op_1962_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(49194432)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_104_cast_fp16 = linear(bias = var_1962_to_fp16, weight = var_1961_to_fp16, x = var_1942_cast_fp16)[name = string("linear_104_cast_fp16")]; | |
| tensor<int32, [4]> var_1970 = const()[name = string("op_1970"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_1971_cast_fp16 = reshape(shape = var_1970, x = linear_102_cast_fp16)[name = string("op_1971_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_258_to_fp16 = const()[name = string("const_258_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> q_71_cast_fp16 = mul(x = var_1971_cast_fp16, y = const_258_to_fp16)[name = string("q_71_cast_fp16")]; | |
| tensor<int32, [4]> var_1977 = const()[name = string("op_1977"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_1978_cast_fp16 = reshape(shape = var_1977, x = linear_103_cast_fp16)[name = string("op_1978_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_259_to_fp16 = const()[name = string("const_259_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> k_71_cast_fp16 = mul(x = var_1978_cast_fp16, y = const_259_to_fp16)[name = string("k_71_cast_fp16")]; | |
| tensor<int32, [4]> var_1984 = const()[name = string("op_1984"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_1985_cast_fp16 = reshape(shape = var_1984, x = linear_104_cast_fp16)[name = string("op_1985_cast_fp16")]; | |
| tensor<int32, [4]> var_1986 = const()[name = string("op_1986"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| bool qk_35_transpose_x_0 = const()[name = string("qk_35_transpose_x_0"), val = bool(false)]; | |
| bool qk_35_transpose_y_0 = const()[name = string("qk_35_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, 20, 64, 1500]> transpose_67 = transpose(perm = transpose_67_perm_0, x = k_71_cast_fp16)[name = string("transpose_154")]; | |
| tensor<fp16, [1, 20, 1500, 64]> transpose_66 = transpose(perm = transpose_66_perm_0, x = q_71_cast_fp16)[name = string("transpose_155")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> qk_35_cast_fp16 = matmul(transpose_x = qk_35_transpose_x_0, transpose_y = qk_35_transpose_y_0, x = transpose_66, y = transpose_67)[name = string("qk_35_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> var_1990_cast_fp16 = softmax(axis = var_1926, x = qk_35_cast_fp16)[name = string("op_1990_cast_fp16")]; | |
| bool var_1992_transpose_x_0 = const()[name = string("op_1992_transpose_x_0"), val = bool(false)]; | |
| bool var_1992_transpose_y_0 = const()[name = string("op_1992_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 20, 1500, 64]> v_71_cast_fp16 = transpose(perm = var_1986, x = var_1985_cast_fp16)[name = string("transpose_153")]; | |
| tensor<fp16, [1, 20, 1500, 64]> var_1992_cast_fp16 = matmul(transpose_x = var_1992_transpose_x_0, transpose_y = var_1992_transpose_y_0, x = var_1990_cast_fp16, y = v_71_cast_fp16)[name = string("op_1992_cast_fp16")]; | |
| tensor<int32, [4]> var_1993 = const()[name = string("op_1993"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_17 = const()[name = string("concat_17"), val = tensor<int32, [3]>([1, 1500, 1280])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_1994_cast_fp16 = transpose(perm = var_1993, x = var_1992_cast_fp16)[name = string("transpose_152")]; | |
| tensor<fp16, [1, 1500, 1280]> x_215_cast_fp16 = reshape(shape = concat_17, x = var_1994_cast_fp16)[name = string("x_215_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_1998_to_fp16 = const()[name = string("op_1998_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(49197056)))]; | |
| tensor<fp16, [1280]> var_1999_to_fp16 = const()[name = string("op_1999_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(52473920)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_105_cast_fp16 = linear(bias = var_1999_to_fp16, weight = var_1998_to_fp16, x = x_215_cast_fp16)[name = string("linear_105_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_217_cast_fp16 = add(x = x_211_cast_fp16, y = linear_105_cast_fp16)[name = string("x_217_cast_fp16")]; | |
| tensor<int32, [1]> var_2006_axes_0 = const()[name = string("op_2006_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_17_mlp_ln_weight_to_fp16 = const()[name = string("blocks_17_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(52476544)))]; | |
| tensor<fp16, [1280]> blocks_17_mlp_ln_bias_to_fp16 = const()[name = string("blocks_17_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(52479168)))]; | |
| tensor<fp16, [1, 1500, 1280]> var_2006_cast_fp16 = layer_norm(axes = var_2006_axes_0, beta = blocks_17_mlp_ln_bias_to_fp16, epsilon = var_1932_to_fp16, gamma = blocks_17_mlp_ln_weight_to_fp16, x = x_217_cast_fp16)[name = string("op_2006_cast_fp16")]; | |
| tensor<fp16, [5120, 1280]> var_2015_to_fp16 = const()[name = string("op_2015_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(52481792)))]; | |
| tensor<fp16, [5120]> var_2016_to_fp16 = const()[name = string("op_2016_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(65589056)))]; | |
| tensor<fp16, [1, 1500, 5120]> linear_106_cast_fp16 = linear(bias = var_2016_to_fp16, weight = var_2015_to_fp16, x = var_2006_cast_fp16)[name = string("linear_106_cast_fp16")]; | |
| string x_221_mode_0 = const()[name = string("x_221_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 5120]> x_221_cast_fp16 = gelu(mode = x_221_mode_0, x = linear_106_cast_fp16)[name = string("x_221_cast_fp16")]; | |
| tensor<fp16, [1280, 5120]> var_2021_to_fp16 = const()[name = string("op_2021_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(65599360)))]; | |
| tensor<fp16, [1280]> var_2022_to_fp16 = const()[name = string("op_2022_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(78706624)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_107_cast_fp16 = linear(bias = var_2022_to_fp16, weight = var_2021_to_fp16, x = x_221_cast_fp16)[name = string("linear_107_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_223_cast_fp16 = add(x = x_217_cast_fp16, y = linear_107_cast_fp16)[name = string("x_223_cast_fp16")]; | |
| int32 var_2032 = const()[name = string("op_2032"), val = int32(-1)]; | |
| tensor<int32, [1]> var_2048_axes_0 = const()[name = string("op_2048_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_18_attn_ln_weight_to_fp16 = const()[name = string("blocks_18_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(78709248)))]; | |
| tensor<fp16, [1280]> blocks_18_attn_ln_bias_to_fp16 = const()[name = string("blocks_18_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(78711872)))]; | |
| fp16 var_2038_to_fp16 = const()[name = string("op_2038_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1280]> var_2048_cast_fp16 = layer_norm(axes = var_2048_axes_0, beta = blocks_18_attn_ln_bias_to_fp16, epsilon = var_2038_to_fp16, gamma = blocks_18_attn_ln_weight_to_fp16, x = x_223_cast_fp16)[name = string("op_2048_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2059_to_fp16 = const()[name = string("op_2059_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(78714496)))]; | |
| tensor<fp16, [1280]> var_2060_to_fp16 = const()[name = string("op_2060_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(81991360)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_108_cast_fp16 = linear(bias = var_2060_to_fp16, weight = var_2059_to_fp16, x = var_2048_cast_fp16)[name = string("linear_108_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2063_to_fp16 = const()[name = string("op_2063_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(81993984)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_109_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2063_to_fp16, x = var_2048_cast_fp16)[name = string("linear_109_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2067_to_fp16 = const()[name = string("op_2067_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(85270848)))]; | |
| tensor<fp16, [1280]> var_2068_to_fp16 = const()[name = string("op_2068_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(88547712)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_110_cast_fp16 = linear(bias = var_2068_to_fp16, weight = var_2067_to_fp16, x = var_2048_cast_fp16)[name = string("linear_110_cast_fp16")]; | |
| tensor<int32, [4]> var_2076 = const()[name = string("op_2076"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2077_cast_fp16 = reshape(shape = var_2076, x = linear_108_cast_fp16)[name = string("op_2077_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_260_to_fp16 = const()[name = string("const_260_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> q_75_cast_fp16 = mul(x = var_2077_cast_fp16, y = const_260_to_fp16)[name = string("q_75_cast_fp16")]; | |
| tensor<int32, [4]> var_2083 = const()[name = string("op_2083"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2084_cast_fp16 = reshape(shape = var_2083, x = linear_109_cast_fp16)[name = string("op_2084_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_261_to_fp16 = const()[name = string("const_261_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> k_75_cast_fp16 = mul(x = var_2084_cast_fp16, y = const_261_to_fp16)[name = string("k_75_cast_fp16")]; | |
| tensor<int32, [4]> var_2090 = const()[name = string("op_2090"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2091_cast_fp16 = reshape(shape = var_2090, x = linear_110_cast_fp16)[name = string("op_2091_cast_fp16")]; | |
| tensor<int32, [4]> var_2092 = const()[name = string("op_2092"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| bool qk_37_transpose_x_0 = const()[name = string("qk_37_transpose_x_0"), val = bool(false)]; | |
| bool qk_37_transpose_y_0 = const()[name = string("qk_37_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, 20, 64, 1500]> transpose_69 = transpose(perm = transpose_69_perm_0, x = k_75_cast_fp16)[name = string("transpose_150")]; | |
| tensor<fp16, [1, 20, 1500, 64]> transpose_68 = transpose(perm = transpose_68_perm_0, x = q_75_cast_fp16)[name = string("transpose_151")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> qk_37_cast_fp16 = matmul(transpose_x = qk_37_transpose_x_0, transpose_y = qk_37_transpose_y_0, x = transpose_68, y = transpose_69)[name = string("qk_37_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> var_2096_cast_fp16 = softmax(axis = var_2032, x = qk_37_cast_fp16)[name = string("op_2096_cast_fp16")]; | |
| bool var_2098_transpose_x_0 = const()[name = string("op_2098_transpose_x_0"), val = bool(false)]; | |
| bool var_2098_transpose_y_0 = const()[name = string("op_2098_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 20, 1500, 64]> v_75_cast_fp16 = transpose(perm = var_2092, x = var_2091_cast_fp16)[name = string("transpose_149")]; | |
| tensor<fp16, [1, 20, 1500, 64]> var_2098_cast_fp16 = matmul(transpose_x = var_2098_transpose_x_0, transpose_y = var_2098_transpose_y_0, x = var_2096_cast_fp16, y = v_75_cast_fp16)[name = string("op_2098_cast_fp16")]; | |
| tensor<int32, [4]> var_2099 = const()[name = string("op_2099"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_18 = const()[name = string("concat_18"), val = tensor<int32, [3]>([1, 1500, 1280])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2100_cast_fp16 = transpose(perm = var_2099, x = var_2098_cast_fp16)[name = string("transpose_148")]; | |
| tensor<fp16, [1, 1500, 1280]> x_227_cast_fp16 = reshape(shape = concat_18, x = var_2100_cast_fp16)[name = string("x_227_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2104_to_fp16 = const()[name = string("op_2104_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(88550336)))]; | |
| tensor<fp16, [1280]> var_2105_to_fp16 = const()[name = string("op_2105_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(91827200)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_111_cast_fp16 = linear(bias = var_2105_to_fp16, weight = var_2104_to_fp16, x = x_227_cast_fp16)[name = string("linear_111_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_229_cast_fp16 = add(x = x_223_cast_fp16, y = linear_111_cast_fp16)[name = string("x_229_cast_fp16")]; | |
| tensor<int32, [1]> var_2112_axes_0 = const()[name = string("op_2112_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_18_mlp_ln_weight_to_fp16 = const()[name = string("blocks_18_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(91829824)))]; | |
| tensor<fp16, [1280]> blocks_18_mlp_ln_bias_to_fp16 = const()[name = string("blocks_18_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(91832448)))]; | |
| tensor<fp16, [1, 1500, 1280]> var_2112_cast_fp16 = layer_norm(axes = var_2112_axes_0, beta = blocks_18_mlp_ln_bias_to_fp16, epsilon = var_2038_to_fp16, gamma = blocks_18_mlp_ln_weight_to_fp16, x = x_229_cast_fp16)[name = string("op_2112_cast_fp16")]; | |
| tensor<fp16, [5120, 1280]> var_2121_to_fp16 = const()[name = string("op_2121_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(91835072)))]; | |
| tensor<fp16, [5120]> var_2122_to_fp16 = const()[name = string("op_2122_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(104942336)))]; | |
| tensor<fp16, [1, 1500, 5120]> linear_112_cast_fp16 = linear(bias = var_2122_to_fp16, weight = var_2121_to_fp16, x = var_2112_cast_fp16)[name = string("linear_112_cast_fp16")]; | |
| string x_233_mode_0 = const()[name = string("x_233_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 5120]> x_233_cast_fp16 = gelu(mode = x_233_mode_0, x = linear_112_cast_fp16)[name = string("x_233_cast_fp16")]; | |
| tensor<fp16, [1280, 5120]> var_2127_to_fp16 = const()[name = string("op_2127_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(104952640)))]; | |
| tensor<fp16, [1280]> var_2128_to_fp16 = const()[name = string("op_2128_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(118059904)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_113_cast_fp16 = linear(bias = var_2128_to_fp16, weight = var_2127_to_fp16, x = x_233_cast_fp16)[name = string("linear_113_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_235_cast_fp16 = add(x = x_229_cast_fp16, y = linear_113_cast_fp16)[name = string("x_235_cast_fp16")]; | |
| int32 var_2138 = const()[name = string("op_2138"), val = int32(-1)]; | |
| tensor<int32, [1]> var_2154_axes_0 = const()[name = string("op_2154_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_19_attn_ln_weight_to_fp16 = const()[name = string("blocks_19_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(118062528)))]; | |
| tensor<fp16, [1280]> blocks_19_attn_ln_bias_to_fp16 = const()[name = string("blocks_19_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(118065152)))]; | |
| fp16 var_2144_to_fp16 = const()[name = string("op_2144_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1280]> var_2154_cast_fp16 = layer_norm(axes = var_2154_axes_0, beta = blocks_19_attn_ln_bias_to_fp16, epsilon = var_2144_to_fp16, gamma = blocks_19_attn_ln_weight_to_fp16, x = x_235_cast_fp16)[name = string("op_2154_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2165_to_fp16 = const()[name = string("op_2165_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(118067776)))]; | |
| tensor<fp16, [1280]> var_2166_to_fp16 = const()[name = string("op_2166_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(121344640)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_114_cast_fp16 = linear(bias = var_2166_to_fp16, weight = var_2165_to_fp16, x = var_2154_cast_fp16)[name = string("linear_114_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2169_to_fp16 = const()[name = string("op_2169_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(121347264)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_115_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2169_to_fp16, x = var_2154_cast_fp16)[name = string("linear_115_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2173_to_fp16 = const()[name = string("op_2173_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(124624128)))]; | |
| tensor<fp16, [1280]> var_2174_to_fp16 = const()[name = string("op_2174_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(127900992)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_116_cast_fp16 = linear(bias = var_2174_to_fp16, weight = var_2173_to_fp16, x = var_2154_cast_fp16)[name = string("linear_116_cast_fp16")]; | |
| tensor<int32, [4]> var_2182 = const()[name = string("op_2182"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2183_cast_fp16 = reshape(shape = var_2182, x = linear_114_cast_fp16)[name = string("op_2183_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_262_to_fp16 = const()[name = string("const_262_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> q_79_cast_fp16 = mul(x = var_2183_cast_fp16, y = const_262_to_fp16)[name = string("q_79_cast_fp16")]; | |
| tensor<int32, [4]> var_2189 = const()[name = string("op_2189"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2190_cast_fp16 = reshape(shape = var_2189, x = linear_115_cast_fp16)[name = string("op_2190_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_263_to_fp16 = const()[name = string("const_263_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> k_79_cast_fp16 = mul(x = var_2190_cast_fp16, y = const_263_to_fp16)[name = string("k_79_cast_fp16")]; | |
| tensor<int32, [4]> var_2196 = const()[name = string("op_2196"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2197_cast_fp16 = reshape(shape = var_2196, x = linear_116_cast_fp16)[name = string("op_2197_cast_fp16")]; | |
| tensor<int32, [4]> var_2198 = const()[name = string("op_2198"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| bool qk_39_transpose_x_0 = const()[name = string("qk_39_transpose_x_0"), val = bool(false)]; | |
| bool qk_39_transpose_y_0 = const()[name = string("qk_39_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, 20, 64, 1500]> transpose_71 = transpose(perm = transpose_71_perm_0, x = k_79_cast_fp16)[name = string("transpose_146")]; | |
| tensor<fp16, [1, 20, 1500, 64]> transpose_70 = transpose(perm = transpose_70_perm_0, x = q_79_cast_fp16)[name = string("transpose_147")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> qk_39_cast_fp16 = matmul(transpose_x = qk_39_transpose_x_0, transpose_y = qk_39_transpose_y_0, x = transpose_70, y = transpose_71)[name = string("qk_39_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> var_2202_cast_fp16 = softmax(axis = var_2138, x = qk_39_cast_fp16)[name = string("op_2202_cast_fp16")]; | |
| bool var_2204_transpose_x_0 = const()[name = string("op_2204_transpose_x_0"), val = bool(false)]; | |
| bool var_2204_transpose_y_0 = const()[name = string("op_2204_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 20, 1500, 64]> v_79_cast_fp16 = transpose(perm = var_2198, x = var_2197_cast_fp16)[name = string("transpose_145")]; | |
| tensor<fp16, [1, 20, 1500, 64]> var_2204_cast_fp16 = matmul(transpose_x = var_2204_transpose_x_0, transpose_y = var_2204_transpose_y_0, x = var_2202_cast_fp16, y = v_79_cast_fp16)[name = string("op_2204_cast_fp16")]; | |
| tensor<int32, [4]> var_2205 = const()[name = string("op_2205"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_19 = const()[name = string("concat_19"), val = tensor<int32, [3]>([1, 1500, 1280])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2206_cast_fp16 = transpose(perm = var_2205, x = var_2204_cast_fp16)[name = string("transpose_144")]; | |
| tensor<fp16, [1, 1500, 1280]> x_239_cast_fp16 = reshape(shape = concat_19, x = var_2206_cast_fp16)[name = string("x_239_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2210_to_fp16 = const()[name = string("op_2210_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(127903616)))]; | |
| tensor<fp16, [1280]> var_2211_to_fp16 = const()[name = string("op_2211_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(131180480)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_117_cast_fp16 = linear(bias = var_2211_to_fp16, weight = var_2210_to_fp16, x = x_239_cast_fp16)[name = string("linear_117_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_241_cast_fp16 = add(x = x_235_cast_fp16, y = linear_117_cast_fp16)[name = string("x_241_cast_fp16")]; | |
| tensor<int32, [1]> var_2218_axes_0 = const()[name = string("op_2218_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_19_mlp_ln_weight_to_fp16 = const()[name = string("blocks_19_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(131183104)))]; | |
| tensor<fp16, [1280]> blocks_19_mlp_ln_bias_to_fp16 = const()[name = string("blocks_19_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(131185728)))]; | |
| tensor<fp16, [1, 1500, 1280]> var_2218_cast_fp16 = layer_norm(axes = var_2218_axes_0, beta = blocks_19_mlp_ln_bias_to_fp16, epsilon = var_2144_to_fp16, gamma = blocks_19_mlp_ln_weight_to_fp16, x = x_241_cast_fp16)[name = string("op_2218_cast_fp16")]; | |
| tensor<fp16, [5120, 1280]> var_2227_to_fp16 = const()[name = string("op_2227_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(131188352)))]; | |
| tensor<fp16, [5120]> var_2228_to_fp16 = const()[name = string("op_2228_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(144295616)))]; | |
| tensor<fp16, [1, 1500, 5120]> linear_118_cast_fp16 = linear(bias = var_2228_to_fp16, weight = var_2227_to_fp16, x = var_2218_cast_fp16)[name = string("linear_118_cast_fp16")]; | |
| string x_245_mode_0 = const()[name = string("x_245_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 5120]> x_245_cast_fp16 = gelu(mode = x_245_mode_0, x = linear_118_cast_fp16)[name = string("x_245_cast_fp16")]; | |
| tensor<fp16, [1280, 5120]> var_2233_to_fp16 = const()[name = string("op_2233_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(144305920)))]; | |
| tensor<fp16, [1280]> var_2234_to_fp16 = const()[name = string("op_2234_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(157413184)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_119_cast_fp16 = linear(bias = var_2234_to_fp16, weight = var_2233_to_fp16, x = x_245_cast_fp16)[name = string("linear_119_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_247_cast_fp16 = add(x = x_241_cast_fp16, y = linear_119_cast_fp16)[name = string("x_247_cast_fp16")]; | |
| int32 var_2244 = const()[name = string("op_2244"), val = int32(-1)]; | |
| tensor<int32, [1]> var_2260_axes_0 = const()[name = string("op_2260_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_20_attn_ln_weight_to_fp16 = const()[name = string("blocks_20_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(157415808)))]; | |
| tensor<fp16, [1280]> blocks_20_attn_ln_bias_to_fp16 = const()[name = string("blocks_20_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(157418432)))]; | |
| fp16 var_2250_to_fp16 = const()[name = string("op_2250_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1280]> var_2260_cast_fp16 = layer_norm(axes = var_2260_axes_0, beta = blocks_20_attn_ln_bias_to_fp16, epsilon = var_2250_to_fp16, gamma = blocks_20_attn_ln_weight_to_fp16, x = x_247_cast_fp16)[name = string("op_2260_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2271_to_fp16 = const()[name = string("op_2271_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(157421056)))]; | |
| tensor<fp16, [1280]> var_2272_to_fp16 = const()[name = string("op_2272_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(160697920)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_120_cast_fp16 = linear(bias = var_2272_to_fp16, weight = var_2271_to_fp16, x = var_2260_cast_fp16)[name = string("linear_120_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2275_to_fp16 = const()[name = string("op_2275_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(160700544)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_121_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2275_to_fp16, x = var_2260_cast_fp16)[name = string("linear_121_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2279_to_fp16 = const()[name = string("op_2279_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(163977408)))]; | |
| tensor<fp16, [1280]> var_2280_to_fp16 = const()[name = string("op_2280_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(167254272)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_122_cast_fp16 = linear(bias = var_2280_to_fp16, weight = var_2279_to_fp16, x = var_2260_cast_fp16)[name = string("linear_122_cast_fp16")]; | |
| tensor<int32, [4]> var_2288 = const()[name = string("op_2288"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2289_cast_fp16 = reshape(shape = var_2288, x = linear_120_cast_fp16)[name = string("op_2289_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_264_to_fp16 = const()[name = string("const_264_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> q_83_cast_fp16 = mul(x = var_2289_cast_fp16, y = const_264_to_fp16)[name = string("q_83_cast_fp16")]; | |
| tensor<int32, [4]> var_2295 = const()[name = string("op_2295"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2296_cast_fp16 = reshape(shape = var_2295, x = linear_121_cast_fp16)[name = string("op_2296_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_265_to_fp16 = const()[name = string("const_265_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> k_83_cast_fp16 = mul(x = var_2296_cast_fp16, y = const_265_to_fp16)[name = string("k_83_cast_fp16")]; | |
| tensor<int32, [4]> var_2302 = const()[name = string("op_2302"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2303_cast_fp16 = reshape(shape = var_2302, x = linear_122_cast_fp16)[name = string("op_2303_cast_fp16")]; | |
| tensor<int32, [4]> var_2304 = const()[name = string("op_2304"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| bool qk_41_transpose_x_0 = const()[name = string("qk_41_transpose_x_0"), val = bool(false)]; | |
| bool qk_41_transpose_y_0 = const()[name = string("qk_41_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_72_perm_0 = const()[name = string("transpose_72_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_73_perm_0 = const()[name = string("transpose_73_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 20, 64, 1500]> transpose_73 = transpose(perm = transpose_73_perm_0, x = k_83_cast_fp16)[name = string("transpose_142")]; | |
| tensor<fp16, [1, 20, 1500, 64]> transpose_72 = transpose(perm = transpose_72_perm_0, x = q_83_cast_fp16)[name = string("transpose_143")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> qk_41_cast_fp16 = matmul(transpose_x = qk_41_transpose_x_0, transpose_y = qk_41_transpose_y_0, x = transpose_72, y = transpose_73)[name = string("qk_41_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> var_2308_cast_fp16 = softmax(axis = var_2244, x = qk_41_cast_fp16)[name = string("op_2308_cast_fp16")]; | |
| bool var_2310_transpose_x_0 = const()[name = string("op_2310_transpose_x_0"), val = bool(false)]; | |
| bool var_2310_transpose_y_0 = const()[name = string("op_2310_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 20, 1500, 64]> v_83_cast_fp16 = transpose(perm = var_2304, x = var_2303_cast_fp16)[name = string("transpose_141")]; | |
| tensor<fp16, [1, 20, 1500, 64]> var_2310_cast_fp16 = matmul(transpose_x = var_2310_transpose_x_0, transpose_y = var_2310_transpose_y_0, x = var_2308_cast_fp16, y = v_83_cast_fp16)[name = string("op_2310_cast_fp16")]; | |
| tensor<int32, [4]> var_2311 = const()[name = string("op_2311"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_20 = const()[name = string("concat_20"), val = tensor<int32, [3]>([1, 1500, 1280])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2312_cast_fp16 = transpose(perm = var_2311, x = var_2310_cast_fp16)[name = string("transpose_140")]; | |
| tensor<fp16, [1, 1500, 1280]> x_251_cast_fp16 = reshape(shape = concat_20, x = var_2312_cast_fp16)[name = string("x_251_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2316_to_fp16 = const()[name = string("op_2316_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(167256896)))]; | |
| tensor<fp16, [1280]> var_2317_to_fp16 = const()[name = string("op_2317_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(170533760)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_123_cast_fp16 = linear(bias = var_2317_to_fp16, weight = var_2316_to_fp16, x = x_251_cast_fp16)[name = string("linear_123_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_253_cast_fp16 = add(x = x_247_cast_fp16, y = linear_123_cast_fp16)[name = string("x_253_cast_fp16")]; | |
| tensor<int32, [1]> var_2324_axes_0 = const()[name = string("op_2324_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_20_mlp_ln_weight_to_fp16 = const()[name = string("blocks_20_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(170536384)))]; | |
| tensor<fp16, [1280]> blocks_20_mlp_ln_bias_to_fp16 = const()[name = string("blocks_20_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(170539008)))]; | |
| tensor<fp16, [1, 1500, 1280]> var_2324_cast_fp16 = layer_norm(axes = var_2324_axes_0, beta = blocks_20_mlp_ln_bias_to_fp16, epsilon = var_2250_to_fp16, gamma = blocks_20_mlp_ln_weight_to_fp16, x = x_253_cast_fp16)[name = string("op_2324_cast_fp16")]; | |
| tensor<fp16, [5120, 1280]> var_2333_to_fp16 = const()[name = string("op_2333_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(170541632)))]; | |
| tensor<fp16, [5120]> var_2334_to_fp16 = const()[name = string("op_2334_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(183648896)))]; | |
| tensor<fp16, [1, 1500, 5120]> linear_124_cast_fp16 = linear(bias = var_2334_to_fp16, weight = var_2333_to_fp16, x = var_2324_cast_fp16)[name = string("linear_124_cast_fp16")]; | |
| string x_257_mode_0 = const()[name = string("x_257_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 5120]> x_257_cast_fp16 = gelu(mode = x_257_mode_0, x = linear_124_cast_fp16)[name = string("x_257_cast_fp16")]; | |
| tensor<fp16, [1280, 5120]> var_2339_to_fp16 = const()[name = string("op_2339_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(183659200)))]; | |
| tensor<fp16, [1280]> var_2340_to_fp16 = const()[name = string("op_2340_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(196766464)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_125_cast_fp16 = linear(bias = var_2340_to_fp16, weight = var_2339_to_fp16, x = x_257_cast_fp16)[name = string("linear_125_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_259_cast_fp16 = add(x = x_253_cast_fp16, y = linear_125_cast_fp16)[name = string("x_259_cast_fp16")]; | |
| int32 var_2350 = const()[name = string("op_2350"), val = int32(-1)]; | |
| tensor<int32, [1]> var_2366_axes_0 = const()[name = string("op_2366_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_21_attn_ln_weight_to_fp16 = const()[name = string("blocks_21_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(196769088)))]; | |
| tensor<fp16, [1280]> blocks_21_attn_ln_bias_to_fp16 = const()[name = string("blocks_21_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(196771712)))]; | |
| fp16 var_2356_to_fp16 = const()[name = string("op_2356_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1280]> var_2366_cast_fp16 = layer_norm(axes = var_2366_axes_0, beta = blocks_21_attn_ln_bias_to_fp16, epsilon = var_2356_to_fp16, gamma = blocks_21_attn_ln_weight_to_fp16, x = x_259_cast_fp16)[name = string("op_2366_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2377_to_fp16 = const()[name = string("op_2377_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(196774336)))]; | |
| tensor<fp16, [1280]> var_2378_to_fp16 = const()[name = string("op_2378_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(200051200)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_126_cast_fp16 = linear(bias = var_2378_to_fp16, weight = var_2377_to_fp16, x = var_2366_cast_fp16)[name = string("linear_126_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2381_to_fp16 = const()[name = string("op_2381_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(200053824)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_127_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2381_to_fp16, x = var_2366_cast_fp16)[name = string("linear_127_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2385_to_fp16 = const()[name = string("op_2385_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(203330688)))]; | |
| tensor<fp16, [1280]> var_2386_to_fp16 = const()[name = string("op_2386_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(206607552)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_128_cast_fp16 = linear(bias = var_2386_to_fp16, weight = var_2385_to_fp16, x = var_2366_cast_fp16)[name = string("linear_128_cast_fp16")]; | |
| tensor<int32, [4]> var_2394 = const()[name = string("op_2394"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2395_cast_fp16 = reshape(shape = var_2394, x = linear_126_cast_fp16)[name = string("op_2395_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_266_to_fp16 = const()[name = string("const_266_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> q_87_cast_fp16 = mul(x = var_2395_cast_fp16, y = const_266_to_fp16)[name = string("q_87_cast_fp16")]; | |
| tensor<int32, [4]> var_2401 = const()[name = string("op_2401"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2402_cast_fp16 = reshape(shape = var_2401, x = linear_127_cast_fp16)[name = string("op_2402_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_267_to_fp16 = const()[name = string("const_267_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> k_87_cast_fp16 = mul(x = var_2402_cast_fp16, y = const_267_to_fp16)[name = string("k_87_cast_fp16")]; | |
| tensor<int32, [4]> var_2408 = const()[name = string("op_2408"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2409_cast_fp16 = reshape(shape = var_2408, x = linear_128_cast_fp16)[name = string("op_2409_cast_fp16")]; | |
| tensor<int32, [4]> var_2410 = const()[name = string("op_2410"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| bool qk_43_transpose_x_0 = const()[name = string("qk_43_transpose_x_0"), val = bool(false)]; | |
| bool qk_43_transpose_y_0 = const()[name = string("qk_43_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_74_perm_0 = const()[name = string("transpose_74_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_75_perm_0 = const()[name = string("transpose_75_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 20, 64, 1500]> transpose_75 = transpose(perm = transpose_75_perm_0, x = k_87_cast_fp16)[name = string("transpose_138")]; | |
| tensor<fp16, [1, 20, 1500, 64]> transpose_74 = transpose(perm = transpose_74_perm_0, x = q_87_cast_fp16)[name = string("transpose_139")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> qk_43_cast_fp16 = matmul(transpose_x = qk_43_transpose_x_0, transpose_y = qk_43_transpose_y_0, x = transpose_74, y = transpose_75)[name = string("qk_43_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> var_2414_cast_fp16 = softmax(axis = var_2350, x = qk_43_cast_fp16)[name = string("op_2414_cast_fp16")]; | |
| bool var_2416_transpose_x_0 = const()[name = string("op_2416_transpose_x_0"), val = bool(false)]; | |
| bool var_2416_transpose_y_0 = const()[name = string("op_2416_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 20, 1500, 64]> v_87_cast_fp16 = transpose(perm = var_2410, x = var_2409_cast_fp16)[name = string("transpose_137")]; | |
| tensor<fp16, [1, 20, 1500, 64]> var_2416_cast_fp16 = matmul(transpose_x = var_2416_transpose_x_0, transpose_y = var_2416_transpose_y_0, x = var_2414_cast_fp16, y = v_87_cast_fp16)[name = string("op_2416_cast_fp16")]; | |
| tensor<int32, [4]> var_2417 = const()[name = string("op_2417"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_21 = const()[name = string("concat_21"), val = tensor<int32, [3]>([1, 1500, 1280])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2418_cast_fp16 = transpose(perm = var_2417, x = var_2416_cast_fp16)[name = string("transpose_136")]; | |
| tensor<fp16, [1, 1500, 1280]> x_263_cast_fp16 = reshape(shape = concat_21, x = var_2418_cast_fp16)[name = string("x_263_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2422_to_fp16 = const()[name = string("op_2422_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(206610176)))]; | |
| tensor<fp16, [1280]> var_2423_to_fp16 = const()[name = string("op_2423_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(209887040)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_129_cast_fp16 = linear(bias = var_2423_to_fp16, weight = var_2422_to_fp16, x = x_263_cast_fp16)[name = string("linear_129_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_265_cast_fp16 = add(x = x_259_cast_fp16, y = linear_129_cast_fp16)[name = string("x_265_cast_fp16")]; | |
| tensor<int32, [1]> var_2430_axes_0 = const()[name = string("op_2430_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_21_mlp_ln_weight_to_fp16 = const()[name = string("blocks_21_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(209889664)))]; | |
| tensor<fp16, [1280]> blocks_21_mlp_ln_bias_to_fp16 = const()[name = string("blocks_21_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(209892288)))]; | |
| tensor<fp16, [1, 1500, 1280]> var_2430_cast_fp16 = layer_norm(axes = var_2430_axes_0, beta = blocks_21_mlp_ln_bias_to_fp16, epsilon = var_2356_to_fp16, gamma = blocks_21_mlp_ln_weight_to_fp16, x = x_265_cast_fp16)[name = string("op_2430_cast_fp16")]; | |
| tensor<fp16, [5120, 1280]> var_2439_to_fp16 = const()[name = string("op_2439_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(209894912)))]; | |
| tensor<fp16, [5120]> var_2440_to_fp16 = const()[name = string("op_2440_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(223002176)))]; | |
| tensor<fp16, [1, 1500, 5120]> linear_130_cast_fp16 = linear(bias = var_2440_to_fp16, weight = var_2439_to_fp16, x = var_2430_cast_fp16)[name = string("linear_130_cast_fp16")]; | |
| string x_269_mode_0 = const()[name = string("x_269_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 5120]> x_269_cast_fp16 = gelu(mode = x_269_mode_0, x = linear_130_cast_fp16)[name = string("x_269_cast_fp16")]; | |
| tensor<fp16, [1280, 5120]> var_2445_to_fp16 = const()[name = string("op_2445_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(223012480)))]; | |
| tensor<fp16, [1280]> var_2446_to_fp16 = const()[name = string("op_2446_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(236119744)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_131_cast_fp16 = linear(bias = var_2446_to_fp16, weight = var_2445_to_fp16, x = x_269_cast_fp16)[name = string("linear_131_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_271_cast_fp16 = add(x = x_265_cast_fp16, y = linear_131_cast_fp16)[name = string("x_271_cast_fp16")]; | |
| int32 var_2456 = const()[name = string("op_2456"), val = int32(-1)]; | |
| tensor<int32, [1]> var_2472_axes_0 = const()[name = string("op_2472_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_22_attn_ln_weight_to_fp16 = const()[name = string("blocks_22_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(236122368)))]; | |
| tensor<fp16, [1280]> blocks_22_attn_ln_bias_to_fp16 = const()[name = string("blocks_22_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(236124992)))]; | |
| fp16 var_2462_to_fp16 = const()[name = string("op_2462_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1280]> var_2472_cast_fp16 = layer_norm(axes = var_2472_axes_0, beta = blocks_22_attn_ln_bias_to_fp16, epsilon = var_2462_to_fp16, gamma = blocks_22_attn_ln_weight_to_fp16, x = x_271_cast_fp16)[name = string("op_2472_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2483_to_fp16 = const()[name = string("op_2483_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(236127616)))]; | |
| tensor<fp16, [1280]> var_2484_to_fp16 = const()[name = string("op_2484_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(239404480)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_132_cast_fp16 = linear(bias = var_2484_to_fp16, weight = var_2483_to_fp16, x = var_2472_cast_fp16)[name = string("linear_132_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2487_to_fp16 = const()[name = string("op_2487_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(239407104)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_133_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2487_to_fp16, x = var_2472_cast_fp16)[name = string("linear_133_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2491_to_fp16 = const()[name = string("op_2491_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(242683968)))]; | |
| tensor<fp16, [1280]> var_2492_to_fp16 = const()[name = string("op_2492_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(245960832)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_134_cast_fp16 = linear(bias = var_2492_to_fp16, weight = var_2491_to_fp16, x = var_2472_cast_fp16)[name = string("linear_134_cast_fp16")]; | |
| tensor<int32, [4]> var_2500 = const()[name = string("op_2500"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2501_cast_fp16 = reshape(shape = var_2500, x = linear_132_cast_fp16)[name = string("op_2501_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_268_to_fp16 = const()[name = string("const_268_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> q_91_cast_fp16 = mul(x = var_2501_cast_fp16, y = const_268_to_fp16)[name = string("q_91_cast_fp16")]; | |
| tensor<int32, [4]> var_2507 = const()[name = string("op_2507"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2508_cast_fp16 = reshape(shape = var_2507, x = linear_133_cast_fp16)[name = string("op_2508_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_269_to_fp16 = const()[name = string("const_269_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> k_91_cast_fp16 = mul(x = var_2508_cast_fp16, y = const_269_to_fp16)[name = string("k_91_cast_fp16")]; | |
| tensor<int32, [4]> var_2514 = const()[name = string("op_2514"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2515_cast_fp16 = reshape(shape = var_2514, x = linear_134_cast_fp16)[name = string("op_2515_cast_fp16")]; | |
| tensor<int32, [4]> var_2516 = const()[name = string("op_2516"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| bool qk_45_transpose_x_0 = const()[name = string("qk_45_transpose_x_0"), val = bool(false)]; | |
| bool qk_45_transpose_y_0 = const()[name = string("qk_45_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_76_perm_0 = const()[name = string("transpose_76_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_77_perm_0 = const()[name = string("transpose_77_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 20, 64, 1500]> transpose_77 = transpose(perm = transpose_77_perm_0, x = k_91_cast_fp16)[name = string("transpose_134")]; | |
| tensor<fp16, [1, 20, 1500, 64]> transpose_76 = transpose(perm = transpose_76_perm_0, x = q_91_cast_fp16)[name = string("transpose_135")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> qk_45_cast_fp16 = matmul(transpose_x = qk_45_transpose_x_0, transpose_y = qk_45_transpose_y_0, x = transpose_76, y = transpose_77)[name = string("qk_45_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> var_2520_cast_fp16 = softmax(axis = var_2456, x = qk_45_cast_fp16)[name = string("op_2520_cast_fp16")]; | |
| bool var_2522_transpose_x_0 = const()[name = string("op_2522_transpose_x_0"), val = bool(false)]; | |
| bool var_2522_transpose_y_0 = const()[name = string("op_2522_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 20, 1500, 64]> v_91_cast_fp16 = transpose(perm = var_2516, x = var_2515_cast_fp16)[name = string("transpose_133")]; | |
| tensor<fp16, [1, 20, 1500, 64]> var_2522_cast_fp16 = matmul(transpose_x = var_2522_transpose_x_0, transpose_y = var_2522_transpose_y_0, x = var_2520_cast_fp16, y = v_91_cast_fp16)[name = string("op_2522_cast_fp16")]; | |
| tensor<int32, [4]> var_2523 = const()[name = string("op_2523"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_22 = const()[name = string("concat_22"), val = tensor<int32, [3]>([1, 1500, 1280])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2524_cast_fp16 = transpose(perm = var_2523, x = var_2522_cast_fp16)[name = string("transpose_132")]; | |
| tensor<fp16, [1, 1500, 1280]> x_275_cast_fp16 = reshape(shape = concat_22, x = var_2524_cast_fp16)[name = string("x_275_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2528_to_fp16 = const()[name = string("op_2528_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(245963456)))]; | |
| tensor<fp16, [1280]> var_2529_to_fp16 = const()[name = string("op_2529_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(249240320)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_135_cast_fp16 = linear(bias = var_2529_to_fp16, weight = var_2528_to_fp16, x = x_275_cast_fp16)[name = string("linear_135_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_277_cast_fp16 = add(x = x_271_cast_fp16, y = linear_135_cast_fp16)[name = string("x_277_cast_fp16")]; | |
| tensor<int32, [1]> var_2536_axes_0 = const()[name = string("op_2536_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_22_mlp_ln_weight_to_fp16 = const()[name = string("blocks_22_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(249242944)))]; | |
| tensor<fp16, [1280]> blocks_22_mlp_ln_bias_to_fp16 = const()[name = string("blocks_22_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(249245568)))]; | |
| tensor<fp16, [1, 1500, 1280]> var_2536_cast_fp16 = layer_norm(axes = var_2536_axes_0, beta = blocks_22_mlp_ln_bias_to_fp16, epsilon = var_2462_to_fp16, gamma = blocks_22_mlp_ln_weight_to_fp16, x = x_277_cast_fp16)[name = string("op_2536_cast_fp16")]; | |
| tensor<fp16, [5120, 1280]> var_2545_to_fp16 = const()[name = string("op_2545_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(249248192)))]; | |
| tensor<fp16, [5120]> var_2546_to_fp16 = const()[name = string("op_2546_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(262355456)))]; | |
| tensor<fp16, [1, 1500, 5120]> linear_136_cast_fp16 = linear(bias = var_2546_to_fp16, weight = var_2545_to_fp16, x = var_2536_cast_fp16)[name = string("linear_136_cast_fp16")]; | |
| string x_281_mode_0 = const()[name = string("x_281_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 5120]> x_281_cast_fp16 = gelu(mode = x_281_mode_0, x = linear_136_cast_fp16)[name = string("x_281_cast_fp16")]; | |
| tensor<fp16, [1280, 5120]> var_2551_to_fp16 = const()[name = string("op_2551_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(262365760)))]; | |
| tensor<fp16, [1280]> var_2552_to_fp16 = const()[name = string("op_2552_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(275473024)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_137_cast_fp16 = linear(bias = var_2552_to_fp16, weight = var_2551_to_fp16, x = x_281_cast_fp16)[name = string("linear_137_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_283_cast_fp16 = add(x = x_277_cast_fp16, y = linear_137_cast_fp16)[name = string("x_283_cast_fp16")]; | |
| int32 var_2562 = const()[name = string("op_2562"), val = int32(-1)]; | |
| tensor<int32, [1]> var_2578_axes_0 = const()[name = string("op_2578_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_23_attn_ln_weight_to_fp16 = const()[name = string("blocks_23_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(275475648)))]; | |
| tensor<fp16, [1280]> blocks_23_attn_ln_bias_to_fp16 = const()[name = string("blocks_23_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(275478272)))]; | |
| fp16 var_2568_to_fp16 = const()[name = string("op_2568_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1280]> var_2578_cast_fp16 = layer_norm(axes = var_2578_axes_0, beta = blocks_23_attn_ln_bias_to_fp16, epsilon = var_2568_to_fp16, gamma = blocks_23_attn_ln_weight_to_fp16, x = x_283_cast_fp16)[name = string("op_2578_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2589_to_fp16 = const()[name = string("op_2589_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(275480896)))]; | |
| tensor<fp16, [1280]> var_2590_to_fp16 = const()[name = string("op_2590_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(278757760)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_138_cast_fp16 = linear(bias = var_2590_to_fp16, weight = var_2589_to_fp16, x = var_2578_cast_fp16)[name = string("linear_138_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2593_to_fp16 = const()[name = string("op_2593_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(278760384)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_139_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2593_to_fp16, x = var_2578_cast_fp16)[name = string("linear_139_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2597_to_fp16 = const()[name = string("op_2597_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(282037248)))]; | |
| tensor<fp16, [1280]> var_2598_to_fp16 = const()[name = string("op_2598_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(285314112)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_140_cast_fp16 = linear(bias = var_2598_to_fp16, weight = var_2597_to_fp16, x = var_2578_cast_fp16)[name = string("linear_140_cast_fp16")]; | |
| tensor<int32, [4]> var_2606 = const()[name = string("op_2606"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2607_cast_fp16 = reshape(shape = var_2606, x = linear_138_cast_fp16)[name = string("op_2607_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_270_to_fp16 = const()[name = string("const_270_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> q_95_cast_fp16 = mul(x = var_2607_cast_fp16, y = const_270_to_fp16)[name = string("q_95_cast_fp16")]; | |
| tensor<int32, [4]> var_2613 = const()[name = string("op_2613"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2614_cast_fp16 = reshape(shape = var_2613, x = linear_139_cast_fp16)[name = string("op_2614_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_271_to_fp16 = const()[name = string("const_271_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> k_95_cast_fp16 = mul(x = var_2614_cast_fp16, y = const_271_to_fp16)[name = string("k_95_cast_fp16")]; | |
| tensor<int32, [4]> var_2620 = const()[name = string("op_2620"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2621_cast_fp16 = reshape(shape = var_2620, x = linear_140_cast_fp16)[name = string("op_2621_cast_fp16")]; | |
| tensor<int32, [4]> var_2622 = const()[name = string("op_2622"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| bool qk_47_transpose_x_0 = const()[name = string("qk_47_transpose_x_0"), val = bool(false)]; | |
| bool qk_47_transpose_y_0 = const()[name = string("qk_47_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_78_perm_0 = const()[name = string("transpose_78_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_79_perm_0 = const()[name = string("transpose_79_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 20, 64, 1500]> transpose_79 = transpose(perm = transpose_79_perm_0, x = k_95_cast_fp16)[name = string("transpose_130")]; | |
| tensor<fp16, [1, 20, 1500, 64]> transpose_78 = transpose(perm = transpose_78_perm_0, x = q_95_cast_fp16)[name = string("transpose_131")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> qk_47_cast_fp16 = matmul(transpose_x = qk_47_transpose_x_0, transpose_y = qk_47_transpose_y_0, x = transpose_78, y = transpose_79)[name = string("qk_47_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> var_2626_cast_fp16 = softmax(axis = var_2562, x = qk_47_cast_fp16)[name = string("op_2626_cast_fp16")]; | |
| bool var_2628_transpose_x_0 = const()[name = string("op_2628_transpose_x_0"), val = bool(false)]; | |
| bool var_2628_transpose_y_0 = const()[name = string("op_2628_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 20, 1500, 64]> v_95_cast_fp16 = transpose(perm = var_2622, x = var_2621_cast_fp16)[name = string("transpose_129")]; | |
| tensor<fp16, [1, 20, 1500, 64]> var_2628_cast_fp16 = matmul(transpose_x = var_2628_transpose_x_0, transpose_y = var_2628_transpose_y_0, x = var_2626_cast_fp16, y = v_95_cast_fp16)[name = string("op_2628_cast_fp16")]; | |
| tensor<int32, [4]> var_2629 = const()[name = string("op_2629"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_23 = const()[name = string("concat_23"), val = tensor<int32, [3]>([1, 1500, 1280])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2630_cast_fp16 = transpose(perm = var_2629, x = var_2628_cast_fp16)[name = string("transpose_128")]; | |
| tensor<fp16, [1, 1500, 1280]> x_287_cast_fp16 = reshape(shape = concat_23, x = var_2630_cast_fp16)[name = string("x_287_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2634_to_fp16 = const()[name = string("op_2634_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(285316736)))]; | |
| tensor<fp16, [1280]> var_2635_to_fp16 = const()[name = string("op_2635_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(288593600)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_141_cast_fp16 = linear(bias = var_2635_to_fp16, weight = var_2634_to_fp16, x = x_287_cast_fp16)[name = string("linear_141_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_289_cast_fp16 = add(x = x_283_cast_fp16, y = linear_141_cast_fp16)[name = string("x_289_cast_fp16")]; | |
| tensor<int32, [1]> var_2642_axes_0 = const()[name = string("op_2642_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_23_mlp_ln_weight_to_fp16 = const()[name = string("blocks_23_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(288596224)))]; | |
| tensor<fp16, [1280]> blocks_23_mlp_ln_bias_to_fp16 = const()[name = string("blocks_23_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(288598848)))]; | |
| tensor<fp16, [1, 1500, 1280]> var_2642_cast_fp16 = layer_norm(axes = var_2642_axes_0, beta = blocks_23_mlp_ln_bias_to_fp16, epsilon = var_2568_to_fp16, gamma = blocks_23_mlp_ln_weight_to_fp16, x = x_289_cast_fp16)[name = string("op_2642_cast_fp16")]; | |
| tensor<fp16, [5120, 1280]> var_2651_to_fp16 = const()[name = string("op_2651_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(288601472)))]; | |
| tensor<fp16, [5120]> var_2652_to_fp16 = const()[name = string("op_2652_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(301708736)))]; | |
| tensor<fp16, [1, 1500, 5120]> linear_142_cast_fp16 = linear(bias = var_2652_to_fp16, weight = var_2651_to_fp16, x = var_2642_cast_fp16)[name = string("linear_142_cast_fp16")]; | |
| string x_293_mode_0 = const()[name = string("x_293_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 5120]> x_293_cast_fp16 = gelu(mode = x_293_mode_0, x = linear_142_cast_fp16)[name = string("x_293_cast_fp16")]; | |
| tensor<fp16, [1280, 5120]> var_2657_to_fp16 = const()[name = string("op_2657_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(301719040)))]; | |
| tensor<fp16, [1280]> var_2658_to_fp16 = const()[name = string("op_2658_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(314826304)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_143_cast_fp16 = linear(bias = var_2658_to_fp16, weight = var_2657_to_fp16, x = x_293_cast_fp16)[name = string("linear_143_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_295_cast_fp16 = add(x = x_289_cast_fp16, y = linear_143_cast_fp16)[name = string("x_295_cast_fp16")]; | |
| int32 var_2668 = const()[name = string("op_2668"), val = int32(-1)]; | |
| tensor<int32, [1]> var_2684_axes_0 = const()[name = string("op_2684_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_24_attn_ln_weight_to_fp16 = const()[name = string("blocks_24_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(314828928)))]; | |
| tensor<fp16, [1280]> blocks_24_attn_ln_bias_to_fp16 = const()[name = string("blocks_24_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(314831552)))]; | |
| fp16 var_2674_to_fp16 = const()[name = string("op_2674_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1280]> var_2684_cast_fp16 = layer_norm(axes = var_2684_axes_0, beta = blocks_24_attn_ln_bias_to_fp16, epsilon = var_2674_to_fp16, gamma = blocks_24_attn_ln_weight_to_fp16, x = x_295_cast_fp16)[name = string("op_2684_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2695_to_fp16 = const()[name = string("op_2695_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(314834176)))]; | |
| tensor<fp16, [1280]> var_2696_to_fp16 = const()[name = string("op_2696_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(318111040)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_144_cast_fp16 = linear(bias = var_2696_to_fp16, weight = var_2695_to_fp16, x = var_2684_cast_fp16)[name = string("linear_144_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2699_to_fp16 = const()[name = string("op_2699_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(318113664)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_145_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2699_to_fp16, x = var_2684_cast_fp16)[name = string("linear_145_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2703_to_fp16 = const()[name = string("op_2703_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(321390528)))]; | |
| tensor<fp16, [1280]> var_2704_to_fp16 = const()[name = string("op_2704_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(324667392)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_146_cast_fp16 = linear(bias = var_2704_to_fp16, weight = var_2703_to_fp16, x = var_2684_cast_fp16)[name = string("linear_146_cast_fp16")]; | |
| tensor<int32, [4]> var_2712 = const()[name = string("op_2712"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2713_cast_fp16 = reshape(shape = var_2712, x = linear_144_cast_fp16)[name = string("op_2713_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_272_to_fp16 = const()[name = string("const_272_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> q_99_cast_fp16 = mul(x = var_2713_cast_fp16, y = const_272_to_fp16)[name = string("q_99_cast_fp16")]; | |
| tensor<int32, [4]> var_2719 = const()[name = string("op_2719"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2720_cast_fp16 = reshape(shape = var_2719, x = linear_145_cast_fp16)[name = string("op_2720_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_273_to_fp16 = const()[name = string("const_273_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> k_99_cast_fp16 = mul(x = var_2720_cast_fp16, y = const_273_to_fp16)[name = string("k_99_cast_fp16")]; | |
| tensor<int32, [4]> var_2726 = const()[name = string("op_2726"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2727_cast_fp16 = reshape(shape = var_2726, x = linear_146_cast_fp16)[name = string("op_2727_cast_fp16")]; | |
| tensor<int32, [4]> var_2728 = const()[name = string("op_2728"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| bool qk_49_transpose_x_0 = const()[name = string("qk_49_transpose_x_0"), val = bool(false)]; | |
| bool qk_49_transpose_y_0 = const()[name = string("qk_49_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_80_perm_0 = const()[name = string("transpose_80_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_81_perm_0 = const()[name = string("transpose_81_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 20, 64, 1500]> transpose_81 = transpose(perm = transpose_81_perm_0, x = k_99_cast_fp16)[name = string("transpose_126")]; | |
| tensor<fp16, [1, 20, 1500, 64]> transpose_80 = transpose(perm = transpose_80_perm_0, x = q_99_cast_fp16)[name = string("transpose_127")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> qk_49_cast_fp16 = matmul(transpose_x = qk_49_transpose_x_0, transpose_y = qk_49_transpose_y_0, x = transpose_80, y = transpose_81)[name = string("qk_49_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> var_2732_cast_fp16 = softmax(axis = var_2668, x = qk_49_cast_fp16)[name = string("op_2732_cast_fp16")]; | |
| bool var_2734_transpose_x_0 = const()[name = string("op_2734_transpose_x_0"), val = bool(false)]; | |
| bool var_2734_transpose_y_0 = const()[name = string("op_2734_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 20, 1500, 64]> v_99_cast_fp16 = transpose(perm = var_2728, x = var_2727_cast_fp16)[name = string("transpose_125")]; | |
| tensor<fp16, [1, 20, 1500, 64]> var_2734_cast_fp16 = matmul(transpose_x = var_2734_transpose_x_0, transpose_y = var_2734_transpose_y_0, x = var_2732_cast_fp16, y = v_99_cast_fp16)[name = string("op_2734_cast_fp16")]; | |
| tensor<int32, [4]> var_2735 = const()[name = string("op_2735"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_24 = const()[name = string("concat_24"), val = tensor<int32, [3]>([1, 1500, 1280])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2736_cast_fp16 = transpose(perm = var_2735, x = var_2734_cast_fp16)[name = string("transpose_124")]; | |
| tensor<fp16, [1, 1500, 1280]> x_299_cast_fp16 = reshape(shape = concat_24, x = var_2736_cast_fp16)[name = string("x_299_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2740_to_fp16 = const()[name = string("op_2740_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(324670016)))]; | |
| tensor<fp16, [1280]> var_2741_to_fp16 = const()[name = string("op_2741_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(327946880)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_147_cast_fp16 = linear(bias = var_2741_to_fp16, weight = var_2740_to_fp16, x = x_299_cast_fp16)[name = string("linear_147_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_301_cast_fp16 = add(x = x_295_cast_fp16, y = linear_147_cast_fp16)[name = string("x_301_cast_fp16")]; | |
| tensor<int32, [1]> var_2748_axes_0 = const()[name = string("op_2748_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_24_mlp_ln_weight_to_fp16 = const()[name = string("blocks_24_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(327949504)))]; | |
| tensor<fp16, [1280]> blocks_24_mlp_ln_bias_to_fp16 = const()[name = string("blocks_24_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(327952128)))]; | |
| tensor<fp16, [1, 1500, 1280]> var_2748_cast_fp16 = layer_norm(axes = var_2748_axes_0, beta = blocks_24_mlp_ln_bias_to_fp16, epsilon = var_2674_to_fp16, gamma = blocks_24_mlp_ln_weight_to_fp16, x = x_301_cast_fp16)[name = string("op_2748_cast_fp16")]; | |
| tensor<fp16, [5120, 1280]> var_2757_to_fp16 = const()[name = string("op_2757_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(327954752)))]; | |
| tensor<fp16, [5120]> var_2758_to_fp16 = const()[name = string("op_2758_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(341062016)))]; | |
| tensor<fp16, [1, 1500, 5120]> linear_148_cast_fp16 = linear(bias = var_2758_to_fp16, weight = var_2757_to_fp16, x = var_2748_cast_fp16)[name = string("linear_148_cast_fp16")]; | |
| string x_305_mode_0 = const()[name = string("x_305_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 5120]> x_305_cast_fp16 = gelu(mode = x_305_mode_0, x = linear_148_cast_fp16)[name = string("x_305_cast_fp16")]; | |
| tensor<fp16, [1280, 5120]> var_2763_to_fp16 = const()[name = string("op_2763_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(341072320)))]; | |
| tensor<fp16, [1280]> var_2764_to_fp16 = const()[name = string("op_2764_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(354179584)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_149_cast_fp16 = linear(bias = var_2764_to_fp16, weight = var_2763_to_fp16, x = x_305_cast_fp16)[name = string("linear_149_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_307_cast_fp16 = add(x = x_301_cast_fp16, y = linear_149_cast_fp16)[name = string("x_307_cast_fp16")]; | |
| int32 var_2774 = const()[name = string("op_2774"), val = int32(-1)]; | |
| tensor<int32, [1]> var_2790_axes_0 = const()[name = string("op_2790_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_25_attn_ln_weight_to_fp16 = const()[name = string("blocks_25_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(354182208)))]; | |
| tensor<fp16, [1280]> blocks_25_attn_ln_bias_to_fp16 = const()[name = string("blocks_25_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(354184832)))]; | |
| fp16 var_2780_to_fp16 = const()[name = string("op_2780_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1280]> var_2790_cast_fp16 = layer_norm(axes = var_2790_axes_0, beta = blocks_25_attn_ln_bias_to_fp16, epsilon = var_2780_to_fp16, gamma = blocks_25_attn_ln_weight_to_fp16, x = x_307_cast_fp16)[name = string("op_2790_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2801_to_fp16 = const()[name = string("op_2801_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(354187456)))]; | |
| tensor<fp16, [1280]> var_2802_to_fp16 = const()[name = string("op_2802_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(357464320)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_150_cast_fp16 = linear(bias = var_2802_to_fp16, weight = var_2801_to_fp16, x = var_2790_cast_fp16)[name = string("linear_150_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2805_to_fp16 = const()[name = string("op_2805_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(357466944)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_151_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2805_to_fp16, x = var_2790_cast_fp16)[name = string("linear_151_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2809_to_fp16 = const()[name = string("op_2809_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(360743808)))]; | |
| tensor<fp16, [1280]> var_2810_to_fp16 = const()[name = string("op_2810_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(364020672)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_152_cast_fp16 = linear(bias = var_2810_to_fp16, weight = var_2809_to_fp16, x = var_2790_cast_fp16)[name = string("linear_152_cast_fp16")]; | |
| tensor<int32, [4]> var_2818 = const()[name = string("op_2818"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2819_cast_fp16 = reshape(shape = var_2818, x = linear_150_cast_fp16)[name = string("op_2819_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_274_to_fp16 = const()[name = string("const_274_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> q_103_cast_fp16 = mul(x = var_2819_cast_fp16, y = const_274_to_fp16)[name = string("q_103_cast_fp16")]; | |
| tensor<int32, [4]> var_2825 = const()[name = string("op_2825"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2826_cast_fp16 = reshape(shape = var_2825, x = linear_151_cast_fp16)[name = string("op_2826_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_275_to_fp16 = const()[name = string("const_275_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> k_103_cast_fp16 = mul(x = var_2826_cast_fp16, y = const_275_to_fp16)[name = string("k_103_cast_fp16")]; | |
| tensor<int32, [4]> var_2832 = const()[name = string("op_2832"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2833_cast_fp16 = reshape(shape = var_2832, x = linear_152_cast_fp16)[name = string("op_2833_cast_fp16")]; | |
| tensor<int32, [4]> var_2834 = const()[name = string("op_2834"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| bool qk_51_transpose_x_0 = const()[name = string("qk_51_transpose_x_0"), val = bool(false)]; | |
| bool qk_51_transpose_y_0 = const()[name = string("qk_51_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_82_perm_0 = const()[name = string("transpose_82_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_83_perm_0 = const()[name = string("transpose_83_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 20, 64, 1500]> transpose_83 = transpose(perm = transpose_83_perm_0, x = k_103_cast_fp16)[name = string("transpose_122")]; | |
| tensor<fp16, [1, 20, 1500, 64]> transpose_82 = transpose(perm = transpose_82_perm_0, x = q_103_cast_fp16)[name = string("transpose_123")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> qk_51_cast_fp16 = matmul(transpose_x = qk_51_transpose_x_0, transpose_y = qk_51_transpose_y_0, x = transpose_82, y = transpose_83)[name = string("qk_51_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> var_2838_cast_fp16 = softmax(axis = var_2774, x = qk_51_cast_fp16)[name = string("op_2838_cast_fp16")]; | |
| bool var_2840_transpose_x_0 = const()[name = string("op_2840_transpose_x_0"), val = bool(false)]; | |
| bool var_2840_transpose_y_0 = const()[name = string("op_2840_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 20, 1500, 64]> v_103_cast_fp16 = transpose(perm = var_2834, x = var_2833_cast_fp16)[name = string("transpose_121")]; | |
| tensor<fp16, [1, 20, 1500, 64]> var_2840_cast_fp16 = matmul(transpose_x = var_2840_transpose_x_0, transpose_y = var_2840_transpose_y_0, x = var_2838_cast_fp16, y = v_103_cast_fp16)[name = string("op_2840_cast_fp16")]; | |
| tensor<int32, [4]> var_2841 = const()[name = string("op_2841"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_25 = const()[name = string("concat_25"), val = tensor<int32, [3]>([1, 1500, 1280])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2842_cast_fp16 = transpose(perm = var_2841, x = var_2840_cast_fp16)[name = string("transpose_120")]; | |
| tensor<fp16, [1, 1500, 1280]> x_311_cast_fp16 = reshape(shape = concat_25, x = var_2842_cast_fp16)[name = string("x_311_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2846_to_fp16 = const()[name = string("op_2846_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(364023296)))]; | |
| tensor<fp16, [1280]> var_2847_to_fp16 = const()[name = string("op_2847_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(367300160)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_153_cast_fp16 = linear(bias = var_2847_to_fp16, weight = var_2846_to_fp16, x = x_311_cast_fp16)[name = string("linear_153_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_313_cast_fp16 = add(x = x_307_cast_fp16, y = linear_153_cast_fp16)[name = string("x_313_cast_fp16")]; | |
| tensor<int32, [1]> var_2854_axes_0 = const()[name = string("op_2854_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_25_mlp_ln_weight_to_fp16 = const()[name = string("blocks_25_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(367302784)))]; | |
| tensor<fp16, [1280]> blocks_25_mlp_ln_bias_to_fp16 = const()[name = string("blocks_25_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(367305408)))]; | |
| tensor<fp16, [1, 1500, 1280]> var_2854_cast_fp16 = layer_norm(axes = var_2854_axes_0, beta = blocks_25_mlp_ln_bias_to_fp16, epsilon = var_2780_to_fp16, gamma = blocks_25_mlp_ln_weight_to_fp16, x = x_313_cast_fp16)[name = string("op_2854_cast_fp16")]; | |
| tensor<fp16, [5120, 1280]> var_2863_to_fp16 = const()[name = string("op_2863_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(367308032)))]; | |
| tensor<fp16, [5120]> var_2864_to_fp16 = const()[name = string("op_2864_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(380415296)))]; | |
| tensor<fp16, [1, 1500, 5120]> linear_154_cast_fp16 = linear(bias = var_2864_to_fp16, weight = var_2863_to_fp16, x = var_2854_cast_fp16)[name = string("linear_154_cast_fp16")]; | |
| string x_317_mode_0 = const()[name = string("x_317_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 5120]> x_317_cast_fp16 = gelu(mode = x_317_mode_0, x = linear_154_cast_fp16)[name = string("x_317_cast_fp16")]; | |
| tensor<fp16, [1280, 5120]> var_2869_to_fp16 = const()[name = string("op_2869_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(380425600)))]; | |
| tensor<fp16, [1280]> var_2870_to_fp16 = const()[name = string("op_2870_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(393532864)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_155_cast_fp16 = linear(bias = var_2870_to_fp16, weight = var_2869_to_fp16, x = x_317_cast_fp16)[name = string("linear_155_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_319_cast_fp16 = add(x = x_313_cast_fp16, y = linear_155_cast_fp16)[name = string("x_319_cast_fp16")]; | |
| int32 var_2880 = const()[name = string("op_2880"), val = int32(-1)]; | |
| tensor<int32, [1]> var_2896_axes_0 = const()[name = string("op_2896_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_26_attn_ln_weight_to_fp16 = const()[name = string("blocks_26_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(393535488)))]; | |
| tensor<fp16, [1280]> blocks_26_attn_ln_bias_to_fp16 = const()[name = string("blocks_26_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(393538112)))]; | |
| fp16 var_2886_to_fp16 = const()[name = string("op_2886_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1280]> var_2896_cast_fp16 = layer_norm(axes = var_2896_axes_0, beta = blocks_26_attn_ln_bias_to_fp16, epsilon = var_2886_to_fp16, gamma = blocks_26_attn_ln_weight_to_fp16, x = x_319_cast_fp16)[name = string("op_2896_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2907_to_fp16 = const()[name = string("op_2907_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(393540736)))]; | |
| tensor<fp16, [1280]> var_2908_to_fp16 = const()[name = string("op_2908_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(396817600)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_156_cast_fp16 = linear(bias = var_2908_to_fp16, weight = var_2907_to_fp16, x = var_2896_cast_fp16)[name = string("linear_156_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2911_to_fp16 = const()[name = string("op_2911_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(396820224)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_157_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_2911_to_fp16, x = var_2896_cast_fp16)[name = string("linear_157_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2915_to_fp16 = const()[name = string("op_2915_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(400097088)))]; | |
| tensor<fp16, [1280]> var_2916_to_fp16 = const()[name = string("op_2916_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(403373952)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_158_cast_fp16 = linear(bias = var_2916_to_fp16, weight = var_2915_to_fp16, x = var_2896_cast_fp16)[name = string("linear_158_cast_fp16")]; | |
| tensor<int32, [4]> var_2924 = const()[name = string("op_2924"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2925_cast_fp16 = reshape(shape = var_2924, x = linear_156_cast_fp16)[name = string("op_2925_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_276_to_fp16 = const()[name = string("const_276_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> q_107_cast_fp16 = mul(x = var_2925_cast_fp16, y = const_276_to_fp16)[name = string("q_107_cast_fp16")]; | |
| tensor<int32, [4]> var_2931 = const()[name = string("op_2931"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2932_cast_fp16 = reshape(shape = var_2931, x = linear_157_cast_fp16)[name = string("op_2932_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_277_to_fp16 = const()[name = string("const_277_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> k_107_cast_fp16 = mul(x = var_2932_cast_fp16, y = const_277_to_fp16)[name = string("k_107_cast_fp16")]; | |
| tensor<int32, [4]> var_2938 = const()[name = string("op_2938"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2939_cast_fp16 = reshape(shape = var_2938, x = linear_158_cast_fp16)[name = string("op_2939_cast_fp16")]; | |
| tensor<int32, [4]> var_2940 = const()[name = string("op_2940"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| bool qk_53_transpose_x_0 = const()[name = string("qk_53_transpose_x_0"), val = bool(false)]; | |
| bool qk_53_transpose_y_0 = const()[name = string("qk_53_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_84_perm_0 = const()[name = string("transpose_84_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_85_perm_0 = const()[name = string("transpose_85_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 20, 64, 1500]> transpose_85 = transpose(perm = transpose_85_perm_0, x = k_107_cast_fp16)[name = string("transpose_118")]; | |
| tensor<fp16, [1, 20, 1500, 64]> transpose_84 = transpose(perm = transpose_84_perm_0, x = q_107_cast_fp16)[name = string("transpose_119")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> qk_53_cast_fp16 = matmul(transpose_x = qk_53_transpose_x_0, transpose_y = qk_53_transpose_y_0, x = transpose_84, y = transpose_85)[name = string("qk_53_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> var_2944_cast_fp16 = softmax(axis = var_2880, x = qk_53_cast_fp16)[name = string("op_2944_cast_fp16")]; | |
| bool var_2946_transpose_x_0 = const()[name = string("op_2946_transpose_x_0"), val = bool(false)]; | |
| bool var_2946_transpose_y_0 = const()[name = string("op_2946_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 20, 1500, 64]> v_107_cast_fp16 = transpose(perm = var_2940, x = var_2939_cast_fp16)[name = string("transpose_117")]; | |
| tensor<fp16, [1, 20, 1500, 64]> var_2946_cast_fp16 = matmul(transpose_x = var_2946_transpose_x_0, transpose_y = var_2946_transpose_y_0, x = var_2944_cast_fp16, y = v_107_cast_fp16)[name = string("op_2946_cast_fp16")]; | |
| tensor<int32, [4]> var_2947 = const()[name = string("op_2947"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_26 = const()[name = string("concat_26"), val = tensor<int32, [3]>([1, 1500, 1280])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_2948_cast_fp16 = transpose(perm = var_2947, x = var_2946_cast_fp16)[name = string("transpose_116")]; | |
| tensor<fp16, [1, 1500, 1280]> x_323_cast_fp16 = reshape(shape = concat_26, x = var_2948_cast_fp16)[name = string("x_323_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_2952_to_fp16 = const()[name = string("op_2952_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(403376576)))]; | |
| tensor<fp16, [1280]> var_2953_to_fp16 = const()[name = string("op_2953_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(406653440)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_159_cast_fp16 = linear(bias = var_2953_to_fp16, weight = var_2952_to_fp16, x = x_323_cast_fp16)[name = string("linear_159_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_325_cast_fp16 = add(x = x_319_cast_fp16, y = linear_159_cast_fp16)[name = string("x_325_cast_fp16")]; | |
| tensor<int32, [1]> var_2960_axes_0 = const()[name = string("op_2960_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_26_mlp_ln_weight_to_fp16 = const()[name = string("blocks_26_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(406656064)))]; | |
| tensor<fp16, [1280]> blocks_26_mlp_ln_bias_to_fp16 = const()[name = string("blocks_26_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(406658688)))]; | |
| tensor<fp16, [1, 1500, 1280]> var_2960_cast_fp16 = layer_norm(axes = var_2960_axes_0, beta = blocks_26_mlp_ln_bias_to_fp16, epsilon = var_2886_to_fp16, gamma = blocks_26_mlp_ln_weight_to_fp16, x = x_325_cast_fp16)[name = string("op_2960_cast_fp16")]; | |
| tensor<fp16, [5120, 1280]> var_2969_to_fp16 = const()[name = string("op_2969_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(406661312)))]; | |
| tensor<fp16, [5120]> var_2970_to_fp16 = const()[name = string("op_2970_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(419768576)))]; | |
| tensor<fp16, [1, 1500, 5120]> linear_160_cast_fp16 = linear(bias = var_2970_to_fp16, weight = var_2969_to_fp16, x = var_2960_cast_fp16)[name = string("linear_160_cast_fp16")]; | |
| string x_329_mode_0 = const()[name = string("x_329_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 5120]> x_329_cast_fp16 = gelu(mode = x_329_mode_0, x = linear_160_cast_fp16)[name = string("x_329_cast_fp16")]; | |
| tensor<fp16, [1280, 5120]> var_2975_to_fp16 = const()[name = string("op_2975_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(419778880)))]; | |
| tensor<fp16, [1280]> var_2976_to_fp16 = const()[name = string("op_2976_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(432886144)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_161_cast_fp16 = linear(bias = var_2976_to_fp16, weight = var_2975_to_fp16, x = x_329_cast_fp16)[name = string("linear_161_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_331_cast_fp16 = add(x = x_325_cast_fp16, y = linear_161_cast_fp16)[name = string("x_331_cast_fp16")]; | |
| int32 var_2986 = const()[name = string("op_2986"), val = int32(-1)]; | |
| tensor<int32, [1]> var_3002_axes_0 = const()[name = string("op_3002_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_27_attn_ln_weight_to_fp16 = const()[name = string("blocks_27_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(432888768)))]; | |
| tensor<fp16, [1280]> blocks_27_attn_ln_bias_to_fp16 = const()[name = string("blocks_27_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(432891392)))]; | |
| fp16 var_2992_to_fp16 = const()[name = string("op_2992_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1280]> var_3002_cast_fp16 = layer_norm(axes = var_3002_axes_0, beta = blocks_27_attn_ln_bias_to_fp16, epsilon = var_2992_to_fp16, gamma = blocks_27_attn_ln_weight_to_fp16, x = x_331_cast_fp16)[name = string("op_3002_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_3013_to_fp16 = const()[name = string("op_3013_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(432894016)))]; | |
| tensor<fp16, [1280]> var_3014_to_fp16 = const()[name = string("op_3014_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(436170880)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_162_cast_fp16 = linear(bias = var_3014_to_fp16, weight = var_3013_to_fp16, x = var_3002_cast_fp16)[name = string("linear_162_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_3017_to_fp16 = const()[name = string("op_3017_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(436173504)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_163_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3017_to_fp16, x = var_3002_cast_fp16)[name = string("linear_163_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_3021_to_fp16 = const()[name = string("op_3021_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(439450368)))]; | |
| tensor<fp16, [1280]> var_3022_to_fp16 = const()[name = string("op_3022_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(442727232)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_164_cast_fp16 = linear(bias = var_3022_to_fp16, weight = var_3021_to_fp16, x = var_3002_cast_fp16)[name = string("linear_164_cast_fp16")]; | |
| tensor<int32, [4]> var_3030 = const()[name = string("op_3030"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_3031_cast_fp16 = reshape(shape = var_3030, x = linear_162_cast_fp16)[name = string("op_3031_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_278_to_fp16 = const()[name = string("const_278_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> q_111_cast_fp16 = mul(x = var_3031_cast_fp16, y = const_278_to_fp16)[name = string("q_111_cast_fp16")]; | |
| tensor<int32, [4]> var_3037 = const()[name = string("op_3037"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_3038_cast_fp16 = reshape(shape = var_3037, x = linear_163_cast_fp16)[name = string("op_3038_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_279_to_fp16 = const()[name = string("const_279_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> k_111_cast_fp16 = mul(x = var_3038_cast_fp16, y = const_279_to_fp16)[name = string("k_111_cast_fp16")]; | |
| tensor<int32, [4]> var_3044 = const()[name = string("op_3044"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_3045_cast_fp16 = reshape(shape = var_3044, x = linear_164_cast_fp16)[name = string("op_3045_cast_fp16")]; | |
| tensor<int32, [4]> var_3046 = const()[name = string("op_3046"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| bool qk_55_transpose_x_0 = const()[name = string("qk_55_transpose_x_0"), val = bool(false)]; | |
| bool qk_55_transpose_y_0 = const()[name = string("qk_55_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_86_perm_0 = const()[name = string("transpose_86_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_87_perm_0 = const()[name = string("transpose_87_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 20, 64, 1500]> transpose_87 = transpose(perm = transpose_87_perm_0, x = k_111_cast_fp16)[name = string("transpose_114")]; | |
| tensor<fp16, [1, 20, 1500, 64]> transpose_86 = transpose(perm = transpose_86_perm_0, x = q_111_cast_fp16)[name = string("transpose_115")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> qk_55_cast_fp16 = matmul(transpose_x = qk_55_transpose_x_0, transpose_y = qk_55_transpose_y_0, x = transpose_86, y = transpose_87)[name = string("qk_55_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> var_3050_cast_fp16 = softmax(axis = var_2986, x = qk_55_cast_fp16)[name = string("op_3050_cast_fp16")]; | |
| bool var_3052_transpose_x_0 = const()[name = string("op_3052_transpose_x_0"), val = bool(false)]; | |
| bool var_3052_transpose_y_0 = const()[name = string("op_3052_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 20, 1500, 64]> v_111_cast_fp16 = transpose(perm = var_3046, x = var_3045_cast_fp16)[name = string("transpose_113")]; | |
| tensor<fp16, [1, 20, 1500, 64]> var_3052_cast_fp16 = matmul(transpose_x = var_3052_transpose_x_0, transpose_y = var_3052_transpose_y_0, x = var_3050_cast_fp16, y = v_111_cast_fp16)[name = string("op_3052_cast_fp16")]; | |
| tensor<int32, [4]> var_3053 = const()[name = string("op_3053"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_27 = const()[name = string("concat_27"), val = tensor<int32, [3]>([1, 1500, 1280])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_3054_cast_fp16 = transpose(perm = var_3053, x = var_3052_cast_fp16)[name = string("transpose_112")]; | |
| tensor<fp16, [1, 1500, 1280]> x_335_cast_fp16 = reshape(shape = concat_27, x = var_3054_cast_fp16)[name = string("x_335_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_3058_to_fp16 = const()[name = string("op_3058_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(442729856)))]; | |
| tensor<fp16, [1280]> var_3059_to_fp16 = const()[name = string("op_3059_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(446006720)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_165_cast_fp16 = linear(bias = var_3059_to_fp16, weight = var_3058_to_fp16, x = x_335_cast_fp16)[name = string("linear_165_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_337_cast_fp16 = add(x = x_331_cast_fp16, y = linear_165_cast_fp16)[name = string("x_337_cast_fp16")]; | |
| tensor<int32, [1]> var_3066_axes_0 = const()[name = string("op_3066_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_27_mlp_ln_weight_to_fp16 = const()[name = string("blocks_27_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(446009344)))]; | |
| tensor<fp16, [1280]> blocks_27_mlp_ln_bias_to_fp16 = const()[name = string("blocks_27_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(446011968)))]; | |
| tensor<fp16, [1, 1500, 1280]> var_3066_cast_fp16 = layer_norm(axes = var_3066_axes_0, beta = blocks_27_mlp_ln_bias_to_fp16, epsilon = var_2992_to_fp16, gamma = blocks_27_mlp_ln_weight_to_fp16, x = x_337_cast_fp16)[name = string("op_3066_cast_fp16")]; | |
| tensor<fp16, [5120, 1280]> var_3075_to_fp16 = const()[name = string("op_3075_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(446014592)))]; | |
| tensor<fp16, [5120]> var_3076_to_fp16 = const()[name = string("op_3076_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(459121856)))]; | |
| tensor<fp16, [1, 1500, 5120]> linear_166_cast_fp16 = linear(bias = var_3076_to_fp16, weight = var_3075_to_fp16, x = var_3066_cast_fp16)[name = string("linear_166_cast_fp16")]; | |
| string x_341_mode_0 = const()[name = string("x_341_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 5120]> x_341_cast_fp16 = gelu(mode = x_341_mode_0, x = linear_166_cast_fp16)[name = string("x_341_cast_fp16")]; | |
| tensor<fp16, [1280, 5120]> var_3081_to_fp16 = const()[name = string("op_3081_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(459132160)))]; | |
| tensor<fp16, [1280]> var_3082_to_fp16 = const()[name = string("op_3082_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(472239424)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_167_cast_fp16 = linear(bias = var_3082_to_fp16, weight = var_3081_to_fp16, x = x_341_cast_fp16)[name = string("linear_167_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_343_cast_fp16 = add(x = x_337_cast_fp16, y = linear_167_cast_fp16)[name = string("x_343_cast_fp16")]; | |
| int32 var_3092 = const()[name = string("op_3092"), val = int32(-1)]; | |
| tensor<int32, [1]> var_3108_axes_0 = const()[name = string("op_3108_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_28_attn_ln_weight_to_fp16 = const()[name = string("blocks_28_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(472242048)))]; | |
| tensor<fp16, [1280]> blocks_28_attn_ln_bias_to_fp16 = const()[name = string("blocks_28_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(472244672)))]; | |
| fp16 var_3098_to_fp16 = const()[name = string("op_3098_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1280]> var_3108_cast_fp16 = layer_norm(axes = var_3108_axes_0, beta = blocks_28_attn_ln_bias_to_fp16, epsilon = var_3098_to_fp16, gamma = blocks_28_attn_ln_weight_to_fp16, x = x_343_cast_fp16)[name = string("op_3108_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_3119_to_fp16 = const()[name = string("op_3119_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(472247296)))]; | |
| tensor<fp16, [1280]> var_3120_to_fp16 = const()[name = string("op_3120_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(475524160)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_168_cast_fp16 = linear(bias = var_3120_to_fp16, weight = var_3119_to_fp16, x = var_3108_cast_fp16)[name = string("linear_168_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_3123_to_fp16 = const()[name = string("op_3123_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(475526784)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_169_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3123_to_fp16, x = var_3108_cast_fp16)[name = string("linear_169_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_3127_to_fp16 = const()[name = string("op_3127_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(478803648)))]; | |
| tensor<fp16, [1280]> var_3128_to_fp16 = const()[name = string("op_3128_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(482080512)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_170_cast_fp16 = linear(bias = var_3128_to_fp16, weight = var_3127_to_fp16, x = var_3108_cast_fp16)[name = string("linear_170_cast_fp16")]; | |
| tensor<int32, [4]> var_3136 = const()[name = string("op_3136"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_3137_cast_fp16 = reshape(shape = var_3136, x = linear_168_cast_fp16)[name = string("op_3137_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_280_to_fp16 = const()[name = string("const_280_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> q_115_cast_fp16 = mul(x = var_3137_cast_fp16, y = const_280_to_fp16)[name = string("q_115_cast_fp16")]; | |
| tensor<int32, [4]> var_3143 = const()[name = string("op_3143"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_3144_cast_fp16 = reshape(shape = var_3143, x = linear_169_cast_fp16)[name = string("op_3144_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_281_to_fp16 = const()[name = string("const_281_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> k_115_cast_fp16 = mul(x = var_3144_cast_fp16, y = const_281_to_fp16)[name = string("k_115_cast_fp16")]; | |
| tensor<int32, [4]> var_3150 = const()[name = string("op_3150"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_3151_cast_fp16 = reshape(shape = var_3150, x = linear_170_cast_fp16)[name = string("op_3151_cast_fp16")]; | |
| tensor<int32, [4]> var_3152 = const()[name = string("op_3152"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| bool qk_57_transpose_x_0 = const()[name = string("qk_57_transpose_x_0"), val = bool(false)]; | |
| bool qk_57_transpose_y_0 = const()[name = string("qk_57_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_88_perm_0 = const()[name = string("transpose_88_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_89_perm_0 = const()[name = string("transpose_89_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 20, 64, 1500]> transpose_89 = transpose(perm = transpose_89_perm_0, x = k_115_cast_fp16)[name = string("transpose_110")]; | |
| tensor<fp16, [1, 20, 1500, 64]> transpose_88 = transpose(perm = transpose_88_perm_0, x = q_115_cast_fp16)[name = string("transpose_111")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> qk_57_cast_fp16 = matmul(transpose_x = qk_57_transpose_x_0, transpose_y = qk_57_transpose_y_0, x = transpose_88, y = transpose_89)[name = string("qk_57_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> var_3156_cast_fp16 = softmax(axis = var_3092, x = qk_57_cast_fp16)[name = string("op_3156_cast_fp16")]; | |
| bool var_3158_transpose_x_0 = const()[name = string("op_3158_transpose_x_0"), val = bool(false)]; | |
| bool var_3158_transpose_y_0 = const()[name = string("op_3158_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 20, 1500, 64]> v_115_cast_fp16 = transpose(perm = var_3152, x = var_3151_cast_fp16)[name = string("transpose_109")]; | |
| tensor<fp16, [1, 20, 1500, 64]> var_3158_cast_fp16 = matmul(transpose_x = var_3158_transpose_x_0, transpose_y = var_3158_transpose_y_0, x = var_3156_cast_fp16, y = v_115_cast_fp16)[name = string("op_3158_cast_fp16")]; | |
| tensor<int32, [4]> var_3159 = const()[name = string("op_3159"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_28 = const()[name = string("concat_28"), val = tensor<int32, [3]>([1, 1500, 1280])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_3160_cast_fp16 = transpose(perm = var_3159, x = var_3158_cast_fp16)[name = string("transpose_108")]; | |
| tensor<fp16, [1, 1500, 1280]> x_347_cast_fp16 = reshape(shape = concat_28, x = var_3160_cast_fp16)[name = string("x_347_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_3164_to_fp16 = const()[name = string("op_3164_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(482083136)))]; | |
| tensor<fp16, [1280]> var_3165_to_fp16 = const()[name = string("op_3165_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(485360000)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_171_cast_fp16 = linear(bias = var_3165_to_fp16, weight = var_3164_to_fp16, x = x_347_cast_fp16)[name = string("linear_171_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_349_cast_fp16 = add(x = x_343_cast_fp16, y = linear_171_cast_fp16)[name = string("x_349_cast_fp16")]; | |
| tensor<int32, [1]> var_3172_axes_0 = const()[name = string("op_3172_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_28_mlp_ln_weight_to_fp16 = const()[name = string("blocks_28_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(485362624)))]; | |
| tensor<fp16, [1280]> blocks_28_mlp_ln_bias_to_fp16 = const()[name = string("blocks_28_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(485365248)))]; | |
| tensor<fp16, [1, 1500, 1280]> var_3172_cast_fp16 = layer_norm(axes = var_3172_axes_0, beta = blocks_28_mlp_ln_bias_to_fp16, epsilon = var_3098_to_fp16, gamma = blocks_28_mlp_ln_weight_to_fp16, x = x_349_cast_fp16)[name = string("op_3172_cast_fp16")]; | |
| tensor<fp16, [5120, 1280]> var_3181_to_fp16 = const()[name = string("op_3181_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(485367872)))]; | |
| tensor<fp16, [5120]> var_3182_to_fp16 = const()[name = string("op_3182_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(498475136)))]; | |
| tensor<fp16, [1, 1500, 5120]> linear_172_cast_fp16 = linear(bias = var_3182_to_fp16, weight = var_3181_to_fp16, x = var_3172_cast_fp16)[name = string("linear_172_cast_fp16")]; | |
| string x_353_mode_0 = const()[name = string("x_353_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 5120]> x_353_cast_fp16 = gelu(mode = x_353_mode_0, x = linear_172_cast_fp16)[name = string("x_353_cast_fp16")]; | |
| tensor<fp16, [1280, 5120]> var_3187_to_fp16 = const()[name = string("op_3187_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(498485440)))]; | |
| tensor<fp16, [1280]> var_3188_to_fp16 = const()[name = string("op_3188_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(511592704)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_173_cast_fp16 = linear(bias = var_3188_to_fp16, weight = var_3187_to_fp16, x = x_353_cast_fp16)[name = string("linear_173_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_355_cast_fp16 = add(x = x_349_cast_fp16, y = linear_173_cast_fp16)[name = string("x_355_cast_fp16")]; | |
| int32 var_3198 = const()[name = string("op_3198"), val = int32(-1)]; | |
| tensor<int32, [1]> var_3214_axes_0 = const()[name = string("op_3214_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_29_attn_ln_weight_to_fp16 = const()[name = string("blocks_29_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(511595328)))]; | |
| tensor<fp16, [1280]> blocks_29_attn_ln_bias_to_fp16 = const()[name = string("blocks_29_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(511597952)))]; | |
| fp16 var_3204_to_fp16 = const()[name = string("op_3204_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1280]> var_3214_cast_fp16 = layer_norm(axes = var_3214_axes_0, beta = blocks_29_attn_ln_bias_to_fp16, epsilon = var_3204_to_fp16, gamma = blocks_29_attn_ln_weight_to_fp16, x = x_355_cast_fp16)[name = string("op_3214_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_3225_to_fp16 = const()[name = string("op_3225_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(511600576)))]; | |
| tensor<fp16, [1280]> var_3226_to_fp16 = const()[name = string("op_3226_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(514877440)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_174_cast_fp16 = linear(bias = var_3226_to_fp16, weight = var_3225_to_fp16, x = var_3214_cast_fp16)[name = string("linear_174_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_3229_to_fp16 = const()[name = string("op_3229_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(514880064)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_175_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3229_to_fp16, x = var_3214_cast_fp16)[name = string("linear_175_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_3233_to_fp16 = const()[name = string("op_3233_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(518156928)))]; | |
| tensor<fp16, [1280]> var_3234_to_fp16 = const()[name = string("op_3234_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(521433792)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_176_cast_fp16 = linear(bias = var_3234_to_fp16, weight = var_3233_to_fp16, x = var_3214_cast_fp16)[name = string("linear_176_cast_fp16")]; | |
| tensor<int32, [4]> var_3242 = const()[name = string("op_3242"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_3243_cast_fp16 = reshape(shape = var_3242, x = linear_174_cast_fp16)[name = string("op_3243_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_282_to_fp16 = const()[name = string("const_282_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> q_119_cast_fp16 = mul(x = var_3243_cast_fp16, y = const_282_to_fp16)[name = string("q_119_cast_fp16")]; | |
| tensor<int32, [4]> var_3249 = const()[name = string("op_3249"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_3250_cast_fp16 = reshape(shape = var_3249, x = linear_175_cast_fp16)[name = string("op_3250_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_283_to_fp16 = const()[name = string("const_283_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> k_119_cast_fp16 = mul(x = var_3250_cast_fp16, y = const_283_to_fp16)[name = string("k_119_cast_fp16")]; | |
| tensor<int32, [4]> var_3256 = const()[name = string("op_3256"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_3257_cast_fp16 = reshape(shape = var_3256, x = linear_176_cast_fp16)[name = string("op_3257_cast_fp16")]; | |
| tensor<int32, [4]> var_3258 = const()[name = string("op_3258"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| bool qk_59_transpose_x_0 = const()[name = string("qk_59_transpose_x_0"), val = bool(false)]; | |
| bool qk_59_transpose_y_0 = const()[name = string("qk_59_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_90_perm_0 = const()[name = string("transpose_90_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_91_perm_0 = const()[name = string("transpose_91_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 20, 64, 1500]> transpose_91 = transpose(perm = transpose_91_perm_0, x = k_119_cast_fp16)[name = string("transpose_106")]; | |
| tensor<fp16, [1, 20, 1500, 64]> transpose_90 = transpose(perm = transpose_90_perm_0, x = q_119_cast_fp16)[name = string("transpose_107")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> qk_59_cast_fp16 = matmul(transpose_x = qk_59_transpose_x_0, transpose_y = qk_59_transpose_y_0, x = transpose_90, y = transpose_91)[name = string("qk_59_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> var_3262_cast_fp16 = softmax(axis = var_3198, x = qk_59_cast_fp16)[name = string("op_3262_cast_fp16")]; | |
| bool var_3264_transpose_x_0 = const()[name = string("op_3264_transpose_x_0"), val = bool(false)]; | |
| bool var_3264_transpose_y_0 = const()[name = string("op_3264_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 20, 1500, 64]> v_119_cast_fp16 = transpose(perm = var_3258, x = var_3257_cast_fp16)[name = string("transpose_105")]; | |
| tensor<fp16, [1, 20, 1500, 64]> var_3264_cast_fp16 = matmul(transpose_x = var_3264_transpose_x_0, transpose_y = var_3264_transpose_y_0, x = var_3262_cast_fp16, y = v_119_cast_fp16)[name = string("op_3264_cast_fp16")]; | |
| tensor<int32, [4]> var_3265 = const()[name = string("op_3265"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_29 = const()[name = string("concat_29"), val = tensor<int32, [3]>([1, 1500, 1280])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_3266_cast_fp16 = transpose(perm = var_3265, x = var_3264_cast_fp16)[name = string("transpose_104")]; | |
| tensor<fp16, [1, 1500, 1280]> x_359_cast_fp16 = reshape(shape = concat_29, x = var_3266_cast_fp16)[name = string("x_359_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_3270_to_fp16 = const()[name = string("op_3270_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(521436416)))]; | |
| tensor<fp16, [1280]> var_3271_to_fp16 = const()[name = string("op_3271_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(524713280)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_177_cast_fp16 = linear(bias = var_3271_to_fp16, weight = var_3270_to_fp16, x = x_359_cast_fp16)[name = string("linear_177_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_361_cast_fp16 = add(x = x_355_cast_fp16, y = linear_177_cast_fp16)[name = string("x_361_cast_fp16")]; | |
| tensor<int32, [1]> var_3278_axes_0 = const()[name = string("op_3278_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_29_mlp_ln_weight_to_fp16 = const()[name = string("blocks_29_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(524715904)))]; | |
| tensor<fp16, [1280]> blocks_29_mlp_ln_bias_to_fp16 = const()[name = string("blocks_29_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(524718528)))]; | |
| tensor<fp16, [1, 1500, 1280]> var_3278_cast_fp16 = layer_norm(axes = var_3278_axes_0, beta = blocks_29_mlp_ln_bias_to_fp16, epsilon = var_3204_to_fp16, gamma = blocks_29_mlp_ln_weight_to_fp16, x = x_361_cast_fp16)[name = string("op_3278_cast_fp16")]; | |
| tensor<fp16, [5120, 1280]> var_3287_to_fp16 = const()[name = string("op_3287_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(524721152)))]; | |
| tensor<fp16, [5120]> var_3288_to_fp16 = const()[name = string("op_3288_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(537828416)))]; | |
| tensor<fp16, [1, 1500, 5120]> linear_178_cast_fp16 = linear(bias = var_3288_to_fp16, weight = var_3287_to_fp16, x = var_3278_cast_fp16)[name = string("linear_178_cast_fp16")]; | |
| string x_365_mode_0 = const()[name = string("x_365_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 5120]> x_365_cast_fp16 = gelu(mode = x_365_mode_0, x = linear_178_cast_fp16)[name = string("x_365_cast_fp16")]; | |
| tensor<fp16, [1280, 5120]> var_3293_to_fp16 = const()[name = string("op_3293_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(537838720)))]; | |
| tensor<fp16, [1280]> var_3294_to_fp16 = const()[name = string("op_3294_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(550945984)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_179_cast_fp16 = linear(bias = var_3294_to_fp16, weight = var_3293_to_fp16, x = x_365_cast_fp16)[name = string("linear_179_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_367_cast_fp16 = add(x = x_361_cast_fp16, y = linear_179_cast_fp16)[name = string("x_367_cast_fp16")]; | |
| int32 var_3304 = const()[name = string("op_3304"), val = int32(-1)]; | |
| tensor<int32, [1]> var_3320_axes_0 = const()[name = string("op_3320_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_30_attn_ln_weight_to_fp16 = const()[name = string("blocks_30_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(550948608)))]; | |
| tensor<fp16, [1280]> blocks_30_attn_ln_bias_to_fp16 = const()[name = string("blocks_30_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(550951232)))]; | |
| fp16 var_3310_to_fp16 = const()[name = string("op_3310_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1280]> var_3320_cast_fp16 = layer_norm(axes = var_3320_axes_0, beta = blocks_30_attn_ln_bias_to_fp16, epsilon = var_3310_to_fp16, gamma = blocks_30_attn_ln_weight_to_fp16, x = x_367_cast_fp16)[name = string("op_3320_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_3331_to_fp16 = const()[name = string("op_3331_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(550953856)))]; | |
| tensor<fp16, [1280]> var_3332_to_fp16 = const()[name = string("op_3332_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(554230720)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_180_cast_fp16 = linear(bias = var_3332_to_fp16, weight = var_3331_to_fp16, x = var_3320_cast_fp16)[name = string("linear_180_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_3335_to_fp16 = const()[name = string("op_3335_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(554233344)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_181_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3335_to_fp16, x = var_3320_cast_fp16)[name = string("linear_181_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_3339_to_fp16 = const()[name = string("op_3339_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(557510208)))]; | |
| tensor<fp16, [1280]> var_3340_to_fp16 = const()[name = string("op_3340_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(560787072)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_182_cast_fp16 = linear(bias = var_3340_to_fp16, weight = var_3339_to_fp16, x = var_3320_cast_fp16)[name = string("linear_182_cast_fp16")]; | |
| tensor<int32, [4]> var_3348 = const()[name = string("op_3348"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_3349_cast_fp16 = reshape(shape = var_3348, x = linear_180_cast_fp16)[name = string("op_3349_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_284_to_fp16 = const()[name = string("const_284_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> q_123_cast_fp16 = mul(x = var_3349_cast_fp16, y = const_284_to_fp16)[name = string("q_123_cast_fp16")]; | |
| tensor<int32, [4]> var_3355 = const()[name = string("op_3355"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_3356_cast_fp16 = reshape(shape = var_3355, x = linear_181_cast_fp16)[name = string("op_3356_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_285_to_fp16 = const()[name = string("const_285_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> k_123_cast_fp16 = mul(x = var_3356_cast_fp16, y = const_285_to_fp16)[name = string("k_123_cast_fp16")]; | |
| tensor<int32, [4]> var_3362 = const()[name = string("op_3362"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_3363_cast_fp16 = reshape(shape = var_3362, x = linear_182_cast_fp16)[name = string("op_3363_cast_fp16")]; | |
| tensor<int32, [4]> var_3364 = const()[name = string("op_3364"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| bool qk_61_transpose_x_0 = const()[name = string("qk_61_transpose_x_0"), val = bool(false)]; | |
| bool qk_61_transpose_y_0 = const()[name = string("qk_61_transpose_y_0"), val = bool(false)]; | |
| tensor<int32, [4]> transpose_92_perm_0 = const()[name = string("transpose_92_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_93_perm_0 = const()[name = string("transpose_93_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 20, 64, 1500]> transpose_93 = transpose(perm = transpose_93_perm_0, x = k_123_cast_fp16)[name = string("transpose_102")]; | |
| tensor<fp16, [1, 20, 1500, 64]> transpose_92 = transpose(perm = transpose_92_perm_0, x = q_123_cast_fp16)[name = string("transpose_103")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> qk_61_cast_fp16 = matmul(transpose_x = qk_61_transpose_x_0, transpose_y = qk_61_transpose_y_0, x = transpose_92, y = transpose_93)[name = string("qk_61_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> var_3368_cast_fp16 = softmax(axis = var_3304, x = qk_61_cast_fp16)[name = string("op_3368_cast_fp16")]; | |
| bool var_3370_transpose_x_0 = const()[name = string("op_3370_transpose_x_0"), val = bool(false)]; | |
| bool var_3370_transpose_y_0 = const()[name = string("op_3370_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 20, 1500, 64]> v_123_cast_fp16 = transpose(perm = var_3364, x = var_3363_cast_fp16)[name = string("transpose_101")]; | |
| tensor<fp16, [1, 20, 1500, 64]> var_3370_cast_fp16 = matmul(transpose_x = var_3370_transpose_x_0, transpose_y = var_3370_transpose_y_0, x = var_3368_cast_fp16, y = v_123_cast_fp16)[name = string("op_3370_cast_fp16")]; | |
| tensor<int32, [4]> var_3371 = const()[name = string("op_3371"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_30 = const()[name = string("concat_30"), val = tensor<int32, [3]>([1, 1500, 1280])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_3372_cast_fp16 = transpose(perm = var_3371, x = var_3370_cast_fp16)[name = string("transpose_100")]; | |
| tensor<fp16, [1, 1500, 1280]> x_371_cast_fp16 = reshape(shape = concat_30, x = var_3372_cast_fp16)[name = string("x_371_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_3376_to_fp16 = const()[name = string("op_3376_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(560789696)))]; | |
| tensor<fp16, [1280]> var_3377_to_fp16 = const()[name = string("op_3377_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(564066560)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_183_cast_fp16 = linear(bias = var_3377_to_fp16, weight = var_3376_to_fp16, x = x_371_cast_fp16)[name = string("linear_183_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_373_cast_fp16 = add(x = x_367_cast_fp16, y = linear_183_cast_fp16)[name = string("x_373_cast_fp16")]; | |
| tensor<int32, [1]> var_3384_axes_0 = const()[name = string("op_3384_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_30_mlp_ln_weight_to_fp16 = const()[name = string("blocks_30_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(564069184)))]; | |
| tensor<fp16, [1280]> blocks_30_mlp_ln_bias_to_fp16 = const()[name = string("blocks_30_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(564071808)))]; | |
| tensor<fp16, [1, 1500, 1280]> var_3384_cast_fp16 = layer_norm(axes = var_3384_axes_0, beta = blocks_30_mlp_ln_bias_to_fp16, epsilon = var_3310_to_fp16, gamma = blocks_30_mlp_ln_weight_to_fp16, x = x_373_cast_fp16)[name = string("op_3384_cast_fp16")]; | |
| tensor<fp16, [5120, 1280]> var_3393_to_fp16 = const()[name = string("op_3393_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(564074432)))]; | |
| tensor<fp16, [5120]> var_3394_to_fp16 = const()[name = string("op_3394_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(577181696)))]; | |
| tensor<fp16, [1, 1500, 5120]> linear_184_cast_fp16 = linear(bias = var_3394_to_fp16, weight = var_3393_to_fp16, x = var_3384_cast_fp16)[name = string("linear_184_cast_fp16")]; | |
| string x_377_mode_0 = const()[name = string("x_377_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 5120]> x_377_cast_fp16 = gelu(mode = x_377_mode_0, x = linear_184_cast_fp16)[name = string("x_377_cast_fp16")]; | |
| tensor<fp16, [1280, 5120]> var_3399_to_fp16 = const()[name = string("op_3399_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(577192000)))]; | |
| tensor<fp16, [1280]> var_3400_to_fp16 = const()[name = string("op_3400_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(590299264)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_185_cast_fp16 = linear(bias = var_3400_to_fp16, weight = var_3399_to_fp16, x = x_377_cast_fp16)[name = string("linear_185_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_379_cast_fp16 = add(x = x_373_cast_fp16, y = linear_185_cast_fp16)[name = string("x_379_cast_fp16")]; | |
| int32 var_3410 = const()[name = string("op_3410"), val = int32(-1)]; | |
| tensor<int32, [1]> var_3426_axes_0 = const()[name = string("op_3426_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_31_attn_ln_weight_to_fp16 = const()[name = string("blocks_31_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(590301888)))]; | |
| tensor<fp16, [1280]> blocks_31_attn_ln_bias_to_fp16 = const()[name = string("blocks_31_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(590304512)))]; | |
| fp16 var_3416_to_fp16 = const()[name = string("op_3416_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1280]> var_3426_cast_fp16 = layer_norm(axes = var_3426_axes_0, beta = blocks_31_attn_ln_bias_to_fp16, epsilon = var_3416_to_fp16, gamma = blocks_31_attn_ln_weight_to_fp16, x = x_379_cast_fp16)[name = string("op_3426_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_3437_to_fp16 = const()[name = string("op_3437_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(590307136)))]; | |
| tensor<fp16, [1280]> var_3438_to_fp16 = const()[name = string("op_3438_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(593584000)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_186_cast_fp16 = linear(bias = var_3438_to_fp16, weight = var_3437_to_fp16, x = var_3426_cast_fp16)[name = string("linear_186_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_3441_to_fp16 = const()[name = string("op_3441_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(593586624)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_187_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = var_3441_to_fp16, x = var_3426_cast_fp16)[name = string("linear_187_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_3445_to_fp16 = const()[name = string("op_3445_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(596863488)))]; | |
| tensor<fp16, [1280]> var_3446_to_fp16 = const()[name = string("op_3446_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(600140352)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_188_cast_fp16 = linear(bias = var_3446_to_fp16, weight = var_3445_to_fp16, x = var_3426_cast_fp16)[name = string("linear_188_cast_fp16")]; | |
| tensor<int32, [4]> var_3454 = const()[name = string("op_3454"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_3455_cast_fp16 = reshape(shape = var_3454, x = linear_186_cast_fp16)[name = string("op_3455_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_286_to_fp16 = const()[name = string("const_286_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> q_cast_fp16 = mul(x = var_3455_cast_fp16, y = const_286_to_fp16)[name = string("q_cast_fp16")]; | |
| tensor<int32, [4]> var_3461 = const()[name = string("op_3461"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_3462_cast_fp16 = reshape(shape = var_3461, x = linear_187_cast_fp16)[name = string("op_3462_cast_fp16")]; | |
| tensor<fp16, [1, 1, 1, 1]> const_287_to_fp16 = const()[name = string("const_287_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 20, 64]> k_cast_fp16 = mul(x = var_3462_cast_fp16, y = const_287_to_fp16)[name = string("k_cast_fp16")]; | |
| tensor<int32, [4]> var_3468 = const()[name = string("op_3468"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_3469_cast_fp16 = reshape(shape = var_3468, x = linear_188_cast_fp16)[name = string("op_3469_cast_fp16")]; | |
| tensor<int32, [4]> var_3470 = const()[name = string("op_3470"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| 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_94_perm_0 = const()[name = string("transpose_94_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])]; | |
| tensor<int32, [4]> transpose_95_perm_0 = const()[name = string("transpose_95_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])]; | |
| tensor<fp16, [1, 20, 64, 1500]> transpose_95 = transpose(perm = transpose_95_perm_0, x = k_cast_fp16)[name = string("transpose_98")]; | |
| tensor<fp16, [1, 20, 1500, 64]> transpose_94 = transpose(perm = transpose_94_perm_0, x = q_cast_fp16)[name = string("transpose_99")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> qk_cast_fp16 = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_94, y = transpose_95)[name = string("qk_cast_fp16")]; | |
| tensor<fp16, [1, 20, 1500, 1500]> var_3474_cast_fp16 = softmax(axis = var_3410, x = qk_cast_fp16)[name = string("op_3474_cast_fp16")]; | |
| bool var_3476_transpose_x_0 = const()[name = string("op_3476_transpose_x_0"), val = bool(false)]; | |
| bool var_3476_transpose_y_0 = const()[name = string("op_3476_transpose_y_0"), val = bool(false)]; | |
| tensor<fp16, [1, 20, 1500, 64]> v_cast_fp16 = transpose(perm = var_3470, x = var_3469_cast_fp16)[name = string("transpose_97")]; | |
| tensor<fp16, [1, 20, 1500, 64]> var_3476_cast_fp16 = matmul(transpose_x = var_3476_transpose_x_0, transpose_y = var_3476_transpose_y_0, x = var_3474_cast_fp16, y = v_cast_fp16)[name = string("op_3476_cast_fp16")]; | |
| tensor<int32, [4]> var_3477 = const()[name = string("op_3477"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [3]> concat_31 = const()[name = string("concat_31"), val = tensor<int32, [3]>([1, 1500, 1280])]; | |
| tensor<fp16, [1, 1500, 20, 64]> var_3478_cast_fp16 = transpose(perm = var_3477, x = var_3476_cast_fp16)[name = string("transpose_96")]; | |
| tensor<fp16, [1, 1500, 1280]> x_383_cast_fp16 = reshape(shape = concat_31, x = var_3478_cast_fp16)[name = string("x_383_cast_fp16")]; | |
| tensor<fp16, [1280, 1280]> var_3482_to_fp16 = const()[name = string("op_3482_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(600142976)))]; | |
| tensor<fp16, [1280]> var_3483_to_fp16 = const()[name = string("op_3483_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(603419840)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_189_cast_fp16 = linear(bias = var_3483_to_fp16, weight = var_3482_to_fp16, x = x_383_cast_fp16)[name = string("linear_189_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_385_cast_fp16 = add(x = x_379_cast_fp16, y = linear_189_cast_fp16)[name = string("x_385_cast_fp16")]; | |
| tensor<int32, [1]> var_3490_axes_0 = const()[name = string("op_3490_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> blocks_31_mlp_ln_weight_to_fp16 = const()[name = string("blocks_31_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(603422464)))]; | |
| tensor<fp16, [1280]> blocks_31_mlp_ln_bias_to_fp16 = const()[name = string("blocks_31_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(603425088)))]; | |
| tensor<fp16, [1, 1500, 1280]> var_3490_cast_fp16 = layer_norm(axes = var_3490_axes_0, beta = blocks_31_mlp_ln_bias_to_fp16, epsilon = var_3416_to_fp16, gamma = blocks_31_mlp_ln_weight_to_fp16, x = x_385_cast_fp16)[name = string("op_3490_cast_fp16")]; | |
| tensor<fp16, [5120, 1280]> var_3499_to_fp16 = const()[name = string("op_3499_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(603427712)))]; | |
| tensor<fp16, [5120]> var_3500_to_fp16 = const()[name = string("op_3500_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(616534976)))]; | |
| tensor<fp16, [1, 1500, 5120]> linear_190_cast_fp16 = linear(bias = var_3500_to_fp16, weight = var_3499_to_fp16, x = var_3490_cast_fp16)[name = string("linear_190_cast_fp16")]; | |
| string x_389_mode_0 = const()[name = string("x_389_mode_0"), val = string("EXACT")]; | |
| tensor<fp16, [1, 1500, 5120]> x_389_cast_fp16 = gelu(mode = x_389_mode_0, x = linear_190_cast_fp16)[name = string("x_389_cast_fp16")]; | |
| tensor<fp16, [1280, 5120]> var_3505_to_fp16 = const()[name = string("op_3505_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(616545280)))]; | |
| tensor<fp16, [1280]> var_3506_to_fp16 = const()[name = string("op_3506_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(629652544)))]; | |
| tensor<fp16, [1, 1500, 1280]> linear_191_cast_fp16 = linear(bias = var_3506_to_fp16, weight = var_3505_to_fp16, x = x_389_cast_fp16)[name = string("linear_191_cast_fp16")]; | |
| tensor<fp16, [1, 1500, 1280]> x_cast_fp16 = add(x = x_385_cast_fp16, y = linear_191_cast_fp16)[name = string("x_cast_fp16")]; | |
| tensor<int32, [1]> var_3519_axes_0 = const()[name = string("op_3519_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [1280]> ln_post_weight_to_fp16 = const()[name = string("ln_post_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(629655168)))]; | |
| tensor<fp16, [1280]> ln_post_bias_to_fp16 = const()[name = string("ln_post_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = string("@model_path/weights/1-weight.bin"), offset = uint64(629657792)))]; | |
| fp16 var_3510_to_fp16 = const()[name = string("op_3510_to_fp16"), val = fp16(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 1280]> output = layer_norm(axes = var_3519_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_3510_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast_fp16)[name = string("op_3519_cast_fp16")]; | |
| } -> (output); | |
| } |