| program(1.3) |
| [buildInfo = dict<string, string>({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}, {"coremltools-component-torch", "2.9.1"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})] |
| { |
| func main<ios18>(tensor<fp16, [1, 80, 3000]> logmel_data) { |
| string var_68_pad_type_0 = const()[name = string("op_68_pad_type_0"), val = string("custom")]; |
| tensor<int32, [2]> var_68_pad_0 = const()[name = string("op_68_pad_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [1]> var_68_strides_0 = const()[name = string("op_68_strides_0"), val = tensor<int32, [1]>([1])]; |
| tensor<int32, [1]> var_68_dilations_0 = const()[name = string("op_68_dilations_0"), val = tensor<int32, [1]>([1])]; |
| int32 var_68_groups_0 = const()[name = string("op_68_groups_0"), val = int32(1)]; |
| tensor<fp16, [1024, 80, 3]> const_0_to_fp16 = const()[name = string("const_0_to_fp16"), val = tensor<fp16, [1024, 80, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))]; |
| tensor<fp16, [1024]> const_1_to_fp16 = const()[name = string("const_1_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(491648)))]; |
| tensor<fp16, [1, 1024, 3000]> var_68_cast_fp16 = conv(bias = const_1_to_fp16, dilations = var_68_dilations_0, groups = var_68_groups_0, pad = var_68_pad_0, pad_type = var_68_pad_type_0, strides = var_68_strides_0, weight = const_0_to_fp16, x = logmel_data)[name = string("op_68_cast_fp16")]; |
| string input_1_mode_0 = const()[name = string("input_1_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 1024, 3000]> input_1_cast_fp16 = gelu(mode = input_1_mode_0, x = var_68_cast_fp16)[name = string("input_1_cast_fp16")]; |
| string var_86_pad_type_0 = const()[name = string("op_86_pad_type_0"), val = string("custom")]; |
| tensor<int32, [2]> var_86_pad_0 = const()[name = string("op_86_pad_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<int32, [1]> var_86_strides_0 = const()[name = string("op_86_strides_0"), val = tensor<int32, [1]>([2])]; |
| tensor<int32, [1]> var_86_dilations_0 = const()[name = string("op_86_dilations_0"), val = tensor<int32, [1]>([1])]; |
| int32 var_86_groups_0 = const()[name = string("op_86_groups_0"), val = int32(1)]; |
| tensor<fp16, [1024, 1024, 3]> const_2_to_fp16 = const()[name = string("const_2_to_fp16"), val = tensor<fp16, [1024, 1024, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(493760)))]; |
| tensor<fp16, [1024]> const_3_to_fp16 = const()[name = string("const_3_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6785280)))]; |
| tensor<fp16, [1, 1024, 1500]> var_86_cast_fp16 = conv(bias = const_3_to_fp16, dilations = var_86_dilations_0, groups = var_86_groups_0, pad = var_86_pad_0, pad_type = var_86_pad_type_0, strides = var_86_strides_0, weight = const_2_to_fp16, x = input_1_cast_fp16)[name = string("op_86_cast_fp16")]; |
| string x_3_mode_0 = const()[name = string("x_3_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 1024, 1500]> x_3_cast_fp16 = gelu(mode = x_3_mode_0, x = var_86_cast_fp16)[name = string("x_3_cast_fp16")]; |
| tensor<int32, [3]> var_92 = const()[name = string("op_92"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<fp16, [1500, 1024]> positional_embedding_to_fp16 = const()[name = string("positional_embedding_to_fp16"), val = tensor<fp16, [1500, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6787392)))]; |
| tensor<fp16, [1, 1500, 1024]> x_5_cast_fp16 = transpose(perm = var_92, x = x_3_cast_fp16)[name = string("transpose_312")]; |
| tensor<fp16, [1, 1500, 1024]> var_95_cast_fp16 = add(x = x_5_cast_fp16, y = positional_embedding_to_fp16)[name = string("op_95_cast_fp16")]; |
| tensor<int32, [1]> var_123_axes_0 = const()[name = string("op_123_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_0_attn_ln_weight_to_fp16 = const()[name = string("blocks_0_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9859456)))]; |
| tensor<fp16, [1024]> blocks_0_attn_ln_bias_to_fp16 = const()[name = string("blocks_0_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9861568)))]; |
| fp16 var_113_to_fp16 = const()[name = string("op_113_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_123_cast_fp16 = layer_norm(axes = var_123_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_113_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_95_cast_fp16)[name = string("op_123_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_4_to_fp16 = const()[name = string("const_4_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9863680)))]; |
| tensor<fp16, [1024]> const_5_to_fp16 = const()[name = string("const_5_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11960896)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_0_cast_fp16 = linear(bias = const_5_to_fp16, weight = const_4_to_fp16, x = var_123_cast_fp16)[name = string("linear_0_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_6_to_fp16 = const()[name = string("const_6_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(11963008)))]; |
| tensor<fp16, [1024]> linear_1_bias_0_to_fp16 = const()[name = string("linear_1_bias_0_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14060224)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_1_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_6_to_fp16, x = var_123_cast_fp16)[name = string("linear_1_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_7_to_fp16 = const()[name = string("const_7_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(14062336)))]; |
| tensor<fp16, [1024]> const_8_to_fp16 = const()[name = string("const_8_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16159552)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_2_cast_fp16 = linear(bias = const_8_to_fp16, weight = const_7_to_fp16, x = var_123_cast_fp16)[name = string("linear_2_cast_fp16")]; |
| tensor<int32, [4]> var_147 = const()[name = string("op_147"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_148_cast_fp16 = reshape(shape = var_147, x = linear_0_cast_fp16)[name = string("op_148_cast_fp16")]; |
| tensor<int32, [4]> var_153 = const()[name = string("op_153"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_154_cast_fp16 = reshape(shape = var_153, x = linear_1_cast_fp16)[name = string("op_154_cast_fp16")]; |
| tensor<int32, [4]> var_159 = const()[name = string("op_159"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_160_cast_fp16 = reshape(shape = var_159, x = linear_2_cast_fp16)[name = string("op_160_cast_fp16")]; |
| tensor<int32, [4]> transpose_144_perm_0 = const()[name = string("transpose_144_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_145_perm_0 = const()[name = string("transpose_145_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_146_perm_0 = const()[name = string("transpose_146_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_146 = transpose(perm = transpose_146_perm_0, x = var_160_cast_fp16)[name = string("transpose_309")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_145 = transpose(perm = transpose_145_perm_0, x = var_154_cast_fp16)[name = string("transpose_310")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_144 = transpose(perm = transpose_144_perm_0, x = var_148_cast_fp16)[name = string("transpose_311")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_1_cast_fp16 = scaled_dot_product_attention(key = transpose_145, query = transpose_144, value = transpose_146)[name = string("a_1_cast_fp16")]; |
| tensor<int32, [4]> var_164 = const()[name = string("op_164"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_0 = const()[name = string("concat_0"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_165_cast_fp16 = transpose(perm = var_164, x = a_1_cast_fp16)[name = string("transpose_308")]; |
| tensor<fp16, [1, 1500, 1024]> x_11_cast_fp16 = reshape(shape = concat_0, x = var_165_cast_fp16)[name = string("x_11_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_15_to_fp16 = const()[name = string("const_15_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(16161664)))]; |
| tensor<fp16, [1024]> const_16_to_fp16 = const()[name = string("const_16_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18258880)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_3_cast_fp16 = linear(bias = const_16_to_fp16, weight = const_15_to_fp16, x = x_11_cast_fp16)[name = string("linear_3_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_13_cast_fp16 = add(x = var_95_cast_fp16, y = linear_3_cast_fp16)[name = string("x_13_cast_fp16")]; |
| tensor<int32, [1]> var_177_axes_0 = const()[name = string("op_177_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_0_mlp_ln_weight_to_fp16 = const()[name = string("blocks_0_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18260992)))]; |
| tensor<fp16, [1024]> blocks_0_mlp_ln_bias_to_fp16 = const()[name = string("blocks_0_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18263104)))]; |
| tensor<fp16, [1, 1500, 1024]> var_177_cast_fp16 = layer_norm(axes = var_177_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_113_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast_fp16)[name = string("op_177_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_17_to_fp16 = const()[name = string("const_17_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(18265216)))]; |
| tensor<fp16, [4096]> const_18_to_fp16 = const()[name = string("const_18_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26653888)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_4_cast_fp16 = linear(bias = const_18_to_fp16, weight = const_17_to_fp16, x = var_177_cast_fp16)[name = string("linear_4_cast_fp16")]; |
| string x_17_mode_0 = const()[name = string("x_17_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 1500, 4096]> x_17_cast_fp16 = gelu(mode = x_17_mode_0, x = linear_4_cast_fp16)[name = string("x_17_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_19_to_fp16 = const()[name = string("const_19_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26662144)))]; |
| tensor<fp16, [1024]> const_20_to_fp16 = const()[name = string("const_20_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35050816)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_5_cast_fp16 = linear(bias = const_20_to_fp16, weight = const_19_to_fp16, x = x_17_cast_fp16)[name = string("linear_5_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_19_cast_fp16 = add(x = x_13_cast_fp16, y = linear_5_cast_fp16)[name = string("x_19_cast_fp16")]; |
| tensor<int32, [1]> var_218_axes_0 = const()[name = string("op_218_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_1_attn_ln_weight_to_fp16 = const()[name = string("blocks_1_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35052928)))]; |
| tensor<fp16, [1024]> blocks_1_attn_ln_bias_to_fp16 = const()[name = string("blocks_1_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35055040)))]; |
| fp16 var_208_to_fp16 = const()[name = string("op_208_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_218_cast_fp16 = layer_norm(axes = var_218_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_208_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast_fp16)[name = string("op_218_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_21_to_fp16 = const()[name = string("const_21_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(35057152)))]; |
| tensor<fp16, [1024]> const_22_to_fp16 = const()[name = string("const_22_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37154368)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_6_cast_fp16 = linear(bias = const_22_to_fp16, weight = const_21_to_fp16, x = var_218_cast_fp16)[name = string("linear_6_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_23_to_fp16 = const()[name = string("const_23_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(37156480)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_7_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_23_to_fp16, x = var_218_cast_fp16)[name = string("linear_7_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_24_to_fp16 = const()[name = string("const_24_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(39253696)))]; |
| tensor<fp16, [1024]> const_25_to_fp16 = const()[name = string("const_25_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41350912)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_8_cast_fp16 = linear(bias = const_25_to_fp16, weight = const_24_to_fp16, x = var_218_cast_fp16)[name = string("linear_8_cast_fp16")]; |
| tensor<int32, [4]> var_242 = const()[name = string("op_242"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_243_cast_fp16 = reshape(shape = var_242, x = linear_6_cast_fp16)[name = string("op_243_cast_fp16")]; |
| tensor<int32, [4]> var_248 = const()[name = string("op_248"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_249_cast_fp16 = reshape(shape = var_248, x = linear_7_cast_fp16)[name = string("op_249_cast_fp16")]; |
| tensor<int32, [4]> var_254 = const()[name = string("op_254"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_255_cast_fp16 = reshape(shape = var_254, x = linear_8_cast_fp16)[name = string("op_255_cast_fp16")]; |
| tensor<int32, [4]> transpose_147_perm_0 = const()[name = string("transpose_147_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_148_perm_0 = const()[name = string("transpose_148_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_149_perm_0 = const()[name = string("transpose_149_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_149 = transpose(perm = transpose_149_perm_0, x = var_255_cast_fp16)[name = string("transpose_305")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_148 = transpose(perm = transpose_148_perm_0, x = var_249_cast_fp16)[name = string("transpose_306")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_147 = transpose(perm = transpose_147_perm_0, x = var_243_cast_fp16)[name = string("transpose_307")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_3_cast_fp16 = scaled_dot_product_attention(key = transpose_148, query = transpose_147, value = transpose_149)[name = string("a_3_cast_fp16")]; |
| tensor<int32, [4]> var_259 = const()[name = string("op_259"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_1 = const()[name = string("concat_1"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_260_cast_fp16 = transpose(perm = var_259, x = a_3_cast_fp16)[name = string("transpose_304")]; |
| tensor<fp16, [1, 1500, 1024]> x_23_cast_fp16 = reshape(shape = concat_1, x = var_260_cast_fp16)[name = string("x_23_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_32_to_fp16 = const()[name = string("const_32_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41353024)))]; |
| tensor<fp16, [1024]> const_33_to_fp16 = const()[name = string("const_33_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43450240)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_9_cast_fp16 = linear(bias = const_33_to_fp16, weight = const_32_to_fp16, x = x_23_cast_fp16)[name = string("linear_9_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_25_cast_fp16 = add(x = x_19_cast_fp16, y = linear_9_cast_fp16)[name = string("x_25_cast_fp16")]; |
| tensor<int32, [1]> var_272_axes_0 = const()[name = string("op_272_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_1_mlp_ln_weight_to_fp16 = const()[name = string("blocks_1_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43452352)))]; |
| tensor<fp16, [1024]> blocks_1_mlp_ln_bias_to_fp16 = const()[name = string("blocks_1_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43454464)))]; |
| tensor<fp16, [1, 1500, 1024]> var_272_cast_fp16 = layer_norm(axes = var_272_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_208_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast_fp16)[name = string("op_272_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_34_to_fp16 = const()[name = string("const_34_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(43456576)))]; |
| tensor<fp16, [4096]> const_35_to_fp16 = const()[name = string("const_35_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51845248)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_10_cast_fp16 = linear(bias = const_35_to_fp16, weight = const_34_to_fp16, x = var_272_cast_fp16)[name = string("linear_10_cast_fp16")]; |
| string x_29_mode_0 = const()[name = string("x_29_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 1500, 4096]> x_29_cast_fp16 = gelu(mode = x_29_mode_0, x = linear_10_cast_fp16)[name = string("x_29_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_36_to_fp16 = const()[name = string("const_36_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(51853504)))]; |
| tensor<fp16, [1024]> const_37_to_fp16 = const()[name = string("const_37_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60242176)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_11_cast_fp16 = linear(bias = const_37_to_fp16, weight = const_36_to_fp16, x = x_29_cast_fp16)[name = string("linear_11_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_31_cast_fp16 = add(x = x_25_cast_fp16, y = linear_11_cast_fp16)[name = string("x_31_cast_fp16")]; |
| tensor<int32, [1]> var_313_axes_0 = const()[name = string("op_313_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_2_attn_ln_weight_to_fp16 = const()[name = string("blocks_2_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60244288)))]; |
| tensor<fp16, [1024]> blocks_2_attn_ln_bias_to_fp16 = const()[name = string("blocks_2_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60246400)))]; |
| fp16 var_303_to_fp16 = const()[name = string("op_303_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_313_cast_fp16 = layer_norm(axes = var_313_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_303_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast_fp16)[name = string("op_313_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_38_to_fp16 = const()[name = string("const_38_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60248512)))]; |
| tensor<fp16, [1024]> const_39_to_fp16 = const()[name = string("const_39_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62345728)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_12_cast_fp16 = linear(bias = const_39_to_fp16, weight = const_38_to_fp16, x = var_313_cast_fp16)[name = string("linear_12_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_40_to_fp16 = const()[name = string("const_40_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62347840)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_13_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_40_to_fp16, x = var_313_cast_fp16)[name = string("linear_13_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_41_to_fp16 = const()[name = string("const_41_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64445056)))]; |
| tensor<fp16, [1024]> const_42_to_fp16 = const()[name = string("const_42_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66542272)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_14_cast_fp16 = linear(bias = const_42_to_fp16, weight = const_41_to_fp16, x = var_313_cast_fp16)[name = string("linear_14_cast_fp16")]; |
| tensor<int32, [4]> var_337 = const()[name = string("op_337"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_338_cast_fp16 = reshape(shape = var_337, x = linear_12_cast_fp16)[name = string("op_338_cast_fp16")]; |
| tensor<int32, [4]> var_343 = const()[name = string("op_343"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_344_cast_fp16 = reshape(shape = var_343, x = linear_13_cast_fp16)[name = string("op_344_cast_fp16")]; |
| tensor<int32, [4]> var_349 = const()[name = string("op_349"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_350_cast_fp16 = reshape(shape = var_349, x = linear_14_cast_fp16)[name = string("op_350_cast_fp16")]; |
| tensor<int32, [4]> transpose_150_perm_0 = const()[name = string("transpose_150_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_151_perm_0 = const()[name = string("transpose_151_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_152_perm_0 = const()[name = string("transpose_152_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_152 = transpose(perm = transpose_152_perm_0, x = var_350_cast_fp16)[name = string("transpose_301")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_151 = transpose(perm = transpose_151_perm_0, x = var_344_cast_fp16)[name = string("transpose_302")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_150 = transpose(perm = transpose_150_perm_0, x = var_338_cast_fp16)[name = string("transpose_303")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_5_cast_fp16 = scaled_dot_product_attention(key = transpose_151, query = transpose_150, value = transpose_152)[name = string("a_5_cast_fp16")]; |
| tensor<int32, [4]> var_354 = const()[name = string("op_354"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_2 = const()[name = string("concat_2"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_355_cast_fp16 = transpose(perm = var_354, x = a_5_cast_fp16)[name = string("transpose_300")]; |
| tensor<fp16, [1, 1500, 1024]> x_35_cast_fp16 = reshape(shape = concat_2, x = var_355_cast_fp16)[name = string("x_35_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_49_to_fp16 = const()[name = string("const_49_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(66544384)))]; |
| tensor<fp16, [1024]> const_50_to_fp16 = const()[name = string("const_50_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68641600)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_15_cast_fp16 = linear(bias = const_50_to_fp16, weight = const_49_to_fp16, x = x_35_cast_fp16)[name = string("linear_15_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_37_cast_fp16 = add(x = x_31_cast_fp16, y = linear_15_cast_fp16)[name = string("x_37_cast_fp16")]; |
| tensor<int32, [1]> var_367_axes_0 = const()[name = string("op_367_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_2_mlp_ln_weight_to_fp16 = const()[name = string("blocks_2_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68643712)))]; |
| tensor<fp16, [1024]> blocks_2_mlp_ln_bias_to_fp16 = const()[name = string("blocks_2_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68645824)))]; |
| tensor<fp16, [1, 1500, 1024]> var_367_cast_fp16 = layer_norm(axes = var_367_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_303_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast_fp16)[name = string("op_367_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_51_to_fp16 = const()[name = string("const_51_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(68647936)))]; |
| tensor<fp16, [4096]> const_52_to_fp16 = const()[name = string("const_52_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77036608)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_16_cast_fp16 = linear(bias = const_52_to_fp16, weight = const_51_to_fp16, x = var_367_cast_fp16)[name = string("linear_16_cast_fp16")]; |
| string x_41_mode_0 = const()[name = string("x_41_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 1500, 4096]> x_41_cast_fp16 = gelu(mode = x_41_mode_0, x = linear_16_cast_fp16)[name = string("x_41_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_53_to_fp16 = const()[name = string("const_53_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(77044864)))]; |
| tensor<fp16, [1024]> const_54_to_fp16 = const()[name = string("const_54_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85433536)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_17_cast_fp16 = linear(bias = const_54_to_fp16, weight = const_53_to_fp16, x = x_41_cast_fp16)[name = string("linear_17_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_43_cast_fp16 = add(x = x_37_cast_fp16, y = linear_17_cast_fp16)[name = string("x_43_cast_fp16")]; |
| tensor<int32, [1]> var_408_axes_0 = const()[name = string("op_408_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_3_attn_ln_weight_to_fp16 = const()[name = string("blocks_3_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85435648)))]; |
| tensor<fp16, [1024]> blocks_3_attn_ln_bias_to_fp16 = const()[name = string("blocks_3_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85437760)))]; |
| fp16 var_398_to_fp16 = const()[name = string("op_398_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_408_cast_fp16 = layer_norm(axes = var_408_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_398_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast_fp16)[name = string("op_408_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_55_to_fp16 = const()[name = string("const_55_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(85439872)))]; |
| tensor<fp16, [1024]> const_56_to_fp16 = const()[name = string("const_56_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87537088)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_18_cast_fp16 = linear(bias = const_56_to_fp16, weight = const_55_to_fp16, x = var_408_cast_fp16)[name = string("linear_18_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_57_to_fp16 = const()[name = string("const_57_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(87539200)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_19_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_57_to_fp16, x = var_408_cast_fp16)[name = string("linear_19_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_58_to_fp16 = const()[name = string("const_58_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(89636416)))]; |
| tensor<fp16, [1024]> const_59_to_fp16 = const()[name = string("const_59_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91733632)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_20_cast_fp16 = linear(bias = const_59_to_fp16, weight = const_58_to_fp16, x = var_408_cast_fp16)[name = string("linear_20_cast_fp16")]; |
| tensor<int32, [4]> var_432 = const()[name = string("op_432"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_433_cast_fp16 = reshape(shape = var_432, x = linear_18_cast_fp16)[name = string("op_433_cast_fp16")]; |
| tensor<int32, [4]> var_438 = const()[name = string("op_438"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_439_cast_fp16 = reshape(shape = var_438, x = linear_19_cast_fp16)[name = string("op_439_cast_fp16")]; |
| tensor<int32, [4]> var_444 = const()[name = string("op_444"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_445_cast_fp16 = reshape(shape = var_444, x = linear_20_cast_fp16)[name = string("op_445_cast_fp16")]; |
| tensor<int32, [4]> transpose_153_perm_0 = const()[name = string("transpose_153_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_154_perm_0 = const()[name = string("transpose_154_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_155_perm_0 = const()[name = string("transpose_155_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_155 = transpose(perm = transpose_155_perm_0, x = var_445_cast_fp16)[name = string("transpose_297")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_154 = transpose(perm = transpose_154_perm_0, x = var_439_cast_fp16)[name = string("transpose_298")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_153 = transpose(perm = transpose_153_perm_0, x = var_433_cast_fp16)[name = string("transpose_299")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_7_cast_fp16 = scaled_dot_product_attention(key = transpose_154, query = transpose_153, value = transpose_155)[name = string("a_7_cast_fp16")]; |
| tensor<int32, [4]> var_449 = const()[name = string("op_449"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_3 = const()[name = string("concat_3"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_450_cast_fp16 = transpose(perm = var_449, x = a_7_cast_fp16)[name = string("transpose_296")]; |
| tensor<fp16, [1, 1500, 1024]> x_47_cast_fp16 = reshape(shape = concat_3, x = var_450_cast_fp16)[name = string("x_47_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_66_to_fp16 = const()[name = string("const_66_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(91735744)))]; |
| tensor<fp16, [1024]> const_67_to_fp16 = const()[name = string("const_67_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93832960)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_21_cast_fp16 = linear(bias = const_67_to_fp16, weight = const_66_to_fp16, x = x_47_cast_fp16)[name = string("linear_21_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_49_cast_fp16 = add(x = x_43_cast_fp16, y = linear_21_cast_fp16)[name = string("x_49_cast_fp16")]; |
| tensor<int32, [1]> var_462_axes_0 = const()[name = string("op_462_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_3_mlp_ln_weight_to_fp16 = const()[name = string("blocks_3_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93835072)))]; |
| tensor<fp16, [1024]> blocks_3_mlp_ln_bias_to_fp16 = const()[name = string("blocks_3_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93837184)))]; |
| tensor<fp16, [1, 1500, 1024]> var_462_cast_fp16 = layer_norm(axes = var_462_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_398_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast_fp16)[name = string("op_462_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_68_to_fp16 = const()[name = string("const_68_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(93839296)))]; |
| tensor<fp16, [4096]> const_69_to_fp16 = const()[name = string("const_69_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102227968)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_22_cast_fp16 = linear(bias = const_69_to_fp16, weight = const_68_to_fp16, x = var_462_cast_fp16)[name = string("linear_22_cast_fp16")]; |
| string x_53_mode_0 = const()[name = string("x_53_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 1500, 4096]> x_53_cast_fp16 = gelu(mode = x_53_mode_0, x = linear_22_cast_fp16)[name = string("x_53_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_70_to_fp16 = const()[name = string("const_70_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102236224)))]; |
| tensor<fp16, [1024]> const_71_to_fp16 = const()[name = string("const_71_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110624896)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_23_cast_fp16 = linear(bias = const_71_to_fp16, weight = const_70_to_fp16, x = x_53_cast_fp16)[name = string("linear_23_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_55_cast_fp16 = add(x = x_49_cast_fp16, y = linear_23_cast_fp16)[name = string("x_55_cast_fp16")]; |
| tensor<int32, [1]> var_503_axes_0 = const()[name = string("op_503_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_4_attn_ln_weight_to_fp16 = const()[name = string("blocks_4_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110627008)))]; |
| tensor<fp16, [1024]> blocks_4_attn_ln_bias_to_fp16 = const()[name = string("blocks_4_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110629120)))]; |
| fp16 var_493_to_fp16 = const()[name = string("op_493_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_503_cast_fp16 = layer_norm(axes = var_503_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_493_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast_fp16)[name = string("op_503_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_72_to_fp16 = const()[name = string("const_72_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(110631232)))]; |
| tensor<fp16, [1024]> const_73_to_fp16 = const()[name = string("const_73_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112728448)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_24_cast_fp16 = linear(bias = const_73_to_fp16, weight = const_72_to_fp16, x = var_503_cast_fp16)[name = string("linear_24_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_74_to_fp16 = const()[name = string("const_74_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112730560)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_25_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_74_to_fp16, x = var_503_cast_fp16)[name = string("linear_25_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_75_to_fp16 = const()[name = string("const_75_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(114827776)))]; |
| tensor<fp16, [1024]> const_76_to_fp16 = const()[name = string("const_76_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116924992)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_26_cast_fp16 = linear(bias = const_76_to_fp16, weight = const_75_to_fp16, x = var_503_cast_fp16)[name = string("linear_26_cast_fp16")]; |
| tensor<int32, [4]> var_527 = const()[name = string("op_527"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_528_cast_fp16 = reshape(shape = var_527, x = linear_24_cast_fp16)[name = string("op_528_cast_fp16")]; |
| tensor<int32, [4]> var_533 = const()[name = string("op_533"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_534_cast_fp16 = reshape(shape = var_533, x = linear_25_cast_fp16)[name = string("op_534_cast_fp16")]; |
| tensor<int32, [4]> var_539 = const()[name = string("op_539"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_540_cast_fp16 = reshape(shape = var_539, x = linear_26_cast_fp16)[name = string("op_540_cast_fp16")]; |
| tensor<int32, [4]> transpose_156_perm_0 = const()[name = string("transpose_156_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_157_perm_0 = const()[name = string("transpose_157_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_158_perm_0 = const()[name = string("transpose_158_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_158 = transpose(perm = transpose_158_perm_0, x = var_540_cast_fp16)[name = string("transpose_293")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_157 = transpose(perm = transpose_157_perm_0, x = var_534_cast_fp16)[name = string("transpose_294")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_156 = transpose(perm = transpose_156_perm_0, x = var_528_cast_fp16)[name = string("transpose_295")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_9_cast_fp16 = scaled_dot_product_attention(key = transpose_157, query = transpose_156, value = transpose_158)[name = string("a_9_cast_fp16")]; |
| tensor<int32, [4]> var_544 = const()[name = string("op_544"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_4 = const()[name = string("concat_4"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_545_cast_fp16 = transpose(perm = var_544, x = a_9_cast_fp16)[name = string("transpose_292")]; |
| tensor<fp16, [1, 1500, 1024]> x_59_cast_fp16 = reshape(shape = concat_4, x = var_545_cast_fp16)[name = string("x_59_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_83_to_fp16 = const()[name = string("const_83_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(116927104)))]; |
| tensor<fp16, [1024]> const_84_to_fp16 = const()[name = string("const_84_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119024320)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_27_cast_fp16 = linear(bias = const_84_to_fp16, weight = const_83_to_fp16, x = x_59_cast_fp16)[name = string("linear_27_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_61_cast_fp16 = add(x = x_55_cast_fp16, y = linear_27_cast_fp16)[name = string("x_61_cast_fp16")]; |
| tensor<int32, [1]> var_557_axes_0 = const()[name = string("op_557_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_4_mlp_ln_weight_to_fp16 = const()[name = string("blocks_4_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119026432)))]; |
| tensor<fp16, [1024]> blocks_4_mlp_ln_bias_to_fp16 = const()[name = string("blocks_4_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119028544)))]; |
| tensor<fp16, [1, 1500, 1024]> var_557_cast_fp16 = layer_norm(axes = var_557_axes_0, beta = blocks_4_mlp_ln_bias_to_fp16, epsilon = var_493_to_fp16, gamma = blocks_4_mlp_ln_weight_to_fp16, x = x_61_cast_fp16)[name = string("op_557_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_85_to_fp16 = const()[name = string("const_85_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(119030656)))]; |
| tensor<fp16, [4096]> const_86_to_fp16 = const()[name = string("const_86_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127419328)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_28_cast_fp16 = linear(bias = const_86_to_fp16, weight = const_85_to_fp16, x = var_557_cast_fp16)[name = string("linear_28_cast_fp16")]; |
| string x_65_mode_0 = const()[name = string("x_65_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 1500, 4096]> x_65_cast_fp16 = gelu(mode = x_65_mode_0, x = linear_28_cast_fp16)[name = string("x_65_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_87_to_fp16 = const()[name = string("const_87_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(127427584)))]; |
| tensor<fp16, [1024]> const_88_to_fp16 = const()[name = string("const_88_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135816256)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_29_cast_fp16 = linear(bias = const_88_to_fp16, weight = const_87_to_fp16, x = x_65_cast_fp16)[name = string("linear_29_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_67_cast_fp16 = add(x = x_61_cast_fp16, y = linear_29_cast_fp16)[name = string("x_67_cast_fp16")]; |
| tensor<int32, [1]> var_598_axes_0 = const()[name = string("op_598_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_5_attn_ln_weight_to_fp16 = const()[name = string("blocks_5_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135818368)))]; |
| tensor<fp16, [1024]> blocks_5_attn_ln_bias_to_fp16 = const()[name = string("blocks_5_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135820480)))]; |
| fp16 var_588_to_fp16 = const()[name = string("op_588_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_598_cast_fp16 = layer_norm(axes = var_598_axes_0, beta = blocks_5_attn_ln_bias_to_fp16, epsilon = var_588_to_fp16, gamma = blocks_5_attn_ln_weight_to_fp16, x = x_67_cast_fp16)[name = string("op_598_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_89_to_fp16 = const()[name = string("const_89_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(135822592)))]; |
| tensor<fp16, [1024]> const_90_to_fp16 = const()[name = string("const_90_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137919808)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_30_cast_fp16 = linear(bias = const_90_to_fp16, weight = const_89_to_fp16, x = var_598_cast_fp16)[name = string("linear_30_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_91_to_fp16 = const()[name = string("const_91_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137921920)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_31_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_91_to_fp16, x = var_598_cast_fp16)[name = string("linear_31_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_92_to_fp16 = const()[name = string("const_92_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(140019136)))]; |
| tensor<fp16, [1024]> const_93_to_fp16 = const()[name = string("const_93_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142116352)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_32_cast_fp16 = linear(bias = const_93_to_fp16, weight = const_92_to_fp16, x = var_598_cast_fp16)[name = string("linear_32_cast_fp16")]; |
| tensor<int32, [4]> var_622 = const()[name = string("op_622"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_623_cast_fp16 = reshape(shape = var_622, x = linear_30_cast_fp16)[name = string("op_623_cast_fp16")]; |
| tensor<int32, [4]> var_628 = const()[name = string("op_628"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_629_cast_fp16 = reshape(shape = var_628, x = linear_31_cast_fp16)[name = string("op_629_cast_fp16")]; |
| tensor<int32, [4]> var_634 = const()[name = string("op_634"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_635_cast_fp16 = reshape(shape = var_634, x = linear_32_cast_fp16)[name = string("op_635_cast_fp16")]; |
| tensor<int32, [4]> transpose_159_perm_0 = const()[name = string("transpose_159_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_160_perm_0 = const()[name = string("transpose_160_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_161_perm_0 = const()[name = string("transpose_161_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_161 = transpose(perm = transpose_161_perm_0, x = var_635_cast_fp16)[name = string("transpose_289")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_160 = transpose(perm = transpose_160_perm_0, x = var_629_cast_fp16)[name = string("transpose_290")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_159 = transpose(perm = transpose_159_perm_0, x = var_623_cast_fp16)[name = string("transpose_291")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_11_cast_fp16 = scaled_dot_product_attention(key = transpose_160, query = transpose_159, value = transpose_161)[name = string("a_11_cast_fp16")]; |
| tensor<int32, [4]> var_639 = const()[name = string("op_639"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_5 = const()[name = string("concat_5"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_640_cast_fp16 = transpose(perm = var_639, x = a_11_cast_fp16)[name = string("transpose_288")]; |
| tensor<fp16, [1, 1500, 1024]> x_71_cast_fp16 = reshape(shape = concat_5, x = var_640_cast_fp16)[name = string("x_71_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_100_to_fp16 = const()[name = string("const_100_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142118464)))]; |
| tensor<fp16, [1024]> const_101_to_fp16 = const()[name = string("const_101_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144215680)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_33_cast_fp16 = linear(bias = const_101_to_fp16, weight = const_100_to_fp16, x = x_71_cast_fp16)[name = string("linear_33_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_73_cast_fp16 = add(x = x_67_cast_fp16, y = linear_33_cast_fp16)[name = string("x_73_cast_fp16")]; |
| tensor<int32, [1]> var_652_axes_0 = const()[name = string("op_652_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_5_mlp_ln_weight_to_fp16 = const()[name = string("blocks_5_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144217792)))]; |
| tensor<fp16, [1024]> blocks_5_mlp_ln_bias_to_fp16 = const()[name = string("blocks_5_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144219904)))]; |
| tensor<fp16, [1, 1500, 1024]> var_652_cast_fp16 = layer_norm(axes = var_652_axes_0, beta = blocks_5_mlp_ln_bias_to_fp16, epsilon = var_588_to_fp16, gamma = blocks_5_mlp_ln_weight_to_fp16, x = x_73_cast_fp16)[name = string("op_652_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_102_to_fp16 = const()[name = string("const_102_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144222016)))]; |
| tensor<fp16, [4096]> const_103_to_fp16 = const()[name = string("const_103_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152610688)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_34_cast_fp16 = linear(bias = const_103_to_fp16, weight = const_102_to_fp16, x = var_652_cast_fp16)[name = string("linear_34_cast_fp16")]; |
| string x_77_mode_0 = const()[name = string("x_77_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 1500, 4096]> x_77_cast_fp16 = gelu(mode = x_77_mode_0, x = linear_34_cast_fp16)[name = string("x_77_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_104_to_fp16 = const()[name = string("const_104_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(152618944)))]; |
| tensor<fp16, [1024]> const_105_to_fp16 = const()[name = string("const_105_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161007616)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_35_cast_fp16 = linear(bias = const_105_to_fp16, weight = const_104_to_fp16, x = x_77_cast_fp16)[name = string("linear_35_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_79_cast_fp16 = add(x = x_73_cast_fp16, y = linear_35_cast_fp16)[name = string("x_79_cast_fp16")]; |
| tensor<int32, [1]> var_693_axes_0 = const()[name = string("op_693_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_6_attn_ln_weight_to_fp16 = const()[name = string("blocks_6_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161009728)))]; |
| tensor<fp16, [1024]> blocks_6_attn_ln_bias_to_fp16 = const()[name = string("blocks_6_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161011840)))]; |
| fp16 var_683_to_fp16 = const()[name = string("op_683_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_693_cast_fp16 = layer_norm(axes = var_693_axes_0, beta = blocks_6_attn_ln_bias_to_fp16, epsilon = var_683_to_fp16, gamma = blocks_6_attn_ln_weight_to_fp16, x = x_79_cast_fp16)[name = string("op_693_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_106_to_fp16 = const()[name = string("const_106_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161013952)))]; |
| tensor<fp16, [1024]> const_107_to_fp16 = const()[name = string("const_107_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163111168)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_36_cast_fp16 = linear(bias = const_107_to_fp16, weight = const_106_to_fp16, x = var_693_cast_fp16)[name = string("linear_36_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_108_to_fp16 = const()[name = string("const_108_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163113280)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_37_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_108_to_fp16, x = var_693_cast_fp16)[name = string("linear_37_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_109_to_fp16 = const()[name = string("const_109_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(165210496)))]; |
| tensor<fp16, [1024]> const_110_to_fp16 = const()[name = string("const_110_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167307712)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_38_cast_fp16 = linear(bias = const_110_to_fp16, weight = const_109_to_fp16, x = var_693_cast_fp16)[name = string("linear_38_cast_fp16")]; |
| tensor<int32, [4]> var_717 = const()[name = string("op_717"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_718_cast_fp16 = reshape(shape = var_717, x = linear_36_cast_fp16)[name = string("op_718_cast_fp16")]; |
| tensor<int32, [4]> var_723 = const()[name = string("op_723"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_724_cast_fp16 = reshape(shape = var_723, x = linear_37_cast_fp16)[name = string("op_724_cast_fp16")]; |
| tensor<int32, [4]> var_729 = const()[name = string("op_729"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_730_cast_fp16 = reshape(shape = var_729, x = linear_38_cast_fp16)[name = string("op_730_cast_fp16")]; |
| tensor<int32, [4]> transpose_162_perm_0 = const()[name = string("transpose_162_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_163_perm_0 = const()[name = string("transpose_163_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_164_perm_0 = const()[name = string("transpose_164_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_164 = transpose(perm = transpose_164_perm_0, x = var_730_cast_fp16)[name = string("transpose_285")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_163 = transpose(perm = transpose_163_perm_0, x = var_724_cast_fp16)[name = string("transpose_286")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_162 = transpose(perm = transpose_162_perm_0, x = var_718_cast_fp16)[name = string("transpose_287")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_13_cast_fp16 = scaled_dot_product_attention(key = transpose_163, query = transpose_162, value = transpose_164)[name = string("a_13_cast_fp16")]; |
| tensor<int32, [4]> var_734 = const()[name = string("op_734"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_6 = const()[name = string("concat_6"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_735_cast_fp16 = transpose(perm = var_734, x = a_13_cast_fp16)[name = string("transpose_284")]; |
| tensor<fp16, [1, 1500, 1024]> x_83_cast_fp16 = reshape(shape = concat_6, x = var_735_cast_fp16)[name = string("x_83_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_117_to_fp16 = const()[name = string("const_117_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(167309824)))]; |
| tensor<fp16, [1024]> const_118_to_fp16 = const()[name = string("const_118_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169407040)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_39_cast_fp16 = linear(bias = const_118_to_fp16, weight = const_117_to_fp16, x = x_83_cast_fp16)[name = string("linear_39_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_85_cast_fp16 = add(x = x_79_cast_fp16, y = linear_39_cast_fp16)[name = string("x_85_cast_fp16")]; |
| tensor<int32, [1]> var_747_axes_0 = const()[name = string("op_747_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_6_mlp_ln_weight_to_fp16 = const()[name = string("blocks_6_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169409152)))]; |
| tensor<fp16, [1024]> blocks_6_mlp_ln_bias_to_fp16 = const()[name = string("blocks_6_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169411264)))]; |
| tensor<fp16, [1, 1500, 1024]> var_747_cast_fp16 = layer_norm(axes = var_747_axes_0, beta = blocks_6_mlp_ln_bias_to_fp16, epsilon = var_683_to_fp16, gamma = blocks_6_mlp_ln_weight_to_fp16, x = x_85_cast_fp16)[name = string("op_747_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_119_to_fp16 = const()[name = string("const_119_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169413376)))]; |
| tensor<fp16, [4096]> const_120_to_fp16 = const()[name = string("const_120_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177802048)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_40_cast_fp16 = linear(bias = const_120_to_fp16, weight = const_119_to_fp16, x = var_747_cast_fp16)[name = string("linear_40_cast_fp16")]; |
| string x_89_mode_0 = const()[name = string("x_89_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 1500, 4096]> x_89_cast_fp16 = gelu(mode = x_89_mode_0, x = linear_40_cast_fp16)[name = string("x_89_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_121_to_fp16 = const()[name = string("const_121_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(177810304)))]; |
| tensor<fp16, [1024]> const_122_to_fp16 = const()[name = string("const_122_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186198976)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_41_cast_fp16 = linear(bias = const_122_to_fp16, weight = const_121_to_fp16, x = x_89_cast_fp16)[name = string("linear_41_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_91_cast_fp16 = add(x = x_85_cast_fp16, y = linear_41_cast_fp16)[name = string("x_91_cast_fp16")]; |
| tensor<int32, [1]> var_788_axes_0 = const()[name = string("op_788_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_7_attn_ln_weight_to_fp16 = const()[name = string("blocks_7_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186201088)))]; |
| tensor<fp16, [1024]> blocks_7_attn_ln_bias_to_fp16 = const()[name = string("blocks_7_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186203200)))]; |
| fp16 var_778_to_fp16 = const()[name = string("op_778_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_788_cast_fp16 = layer_norm(axes = var_788_axes_0, beta = blocks_7_attn_ln_bias_to_fp16, epsilon = var_778_to_fp16, gamma = blocks_7_attn_ln_weight_to_fp16, x = x_91_cast_fp16)[name = string("op_788_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_123_to_fp16 = const()[name = string("const_123_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(186205312)))]; |
| tensor<fp16, [1024]> const_124_to_fp16 = const()[name = string("const_124_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188302528)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_42_cast_fp16 = linear(bias = const_124_to_fp16, weight = const_123_to_fp16, x = var_788_cast_fp16)[name = string("linear_42_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_125_to_fp16 = const()[name = string("const_125_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(188304640)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_43_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_125_to_fp16, x = var_788_cast_fp16)[name = string("linear_43_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_126_to_fp16 = const()[name = string("const_126_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(190401856)))]; |
| tensor<fp16, [1024]> const_127_to_fp16 = const()[name = string("const_127_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192499072)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_44_cast_fp16 = linear(bias = const_127_to_fp16, weight = const_126_to_fp16, x = var_788_cast_fp16)[name = string("linear_44_cast_fp16")]; |
| tensor<int32, [4]> var_812 = const()[name = string("op_812"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_813_cast_fp16 = reshape(shape = var_812, x = linear_42_cast_fp16)[name = string("op_813_cast_fp16")]; |
| tensor<int32, [4]> var_818 = const()[name = string("op_818"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_819_cast_fp16 = reshape(shape = var_818, x = linear_43_cast_fp16)[name = string("op_819_cast_fp16")]; |
| tensor<int32, [4]> var_824 = const()[name = string("op_824"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_825_cast_fp16 = reshape(shape = var_824, x = linear_44_cast_fp16)[name = string("op_825_cast_fp16")]; |
| tensor<int32, [4]> transpose_165_perm_0 = const()[name = string("transpose_165_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_166_perm_0 = const()[name = string("transpose_166_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_167_perm_0 = const()[name = string("transpose_167_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_167 = transpose(perm = transpose_167_perm_0, x = var_825_cast_fp16)[name = string("transpose_281")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_166 = transpose(perm = transpose_166_perm_0, x = var_819_cast_fp16)[name = string("transpose_282")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_165 = transpose(perm = transpose_165_perm_0, x = var_813_cast_fp16)[name = string("transpose_283")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_15_cast_fp16 = scaled_dot_product_attention(key = transpose_166, query = transpose_165, value = transpose_167)[name = string("a_15_cast_fp16")]; |
| tensor<int32, [4]> var_829 = const()[name = string("op_829"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_7 = const()[name = string("concat_7"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_830_cast_fp16 = transpose(perm = var_829, x = a_15_cast_fp16)[name = string("transpose_280")]; |
| tensor<fp16, [1, 1500, 1024]> x_95_cast_fp16 = reshape(shape = concat_7, x = var_830_cast_fp16)[name = string("x_95_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_134_to_fp16 = const()[name = string("const_134_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(192501184)))]; |
| tensor<fp16, [1024]> const_135_to_fp16 = const()[name = string("const_135_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194598400)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_45_cast_fp16 = linear(bias = const_135_to_fp16, weight = const_134_to_fp16, x = x_95_cast_fp16)[name = string("linear_45_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_97_cast_fp16 = add(x = x_91_cast_fp16, y = linear_45_cast_fp16)[name = string("x_97_cast_fp16")]; |
| tensor<int32, [1]> var_842_axes_0 = const()[name = string("op_842_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_7_mlp_ln_weight_to_fp16 = const()[name = string("blocks_7_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194600512)))]; |
| tensor<fp16, [1024]> blocks_7_mlp_ln_bias_to_fp16 = const()[name = string("blocks_7_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194602624)))]; |
| tensor<fp16, [1, 1500, 1024]> var_842_cast_fp16 = layer_norm(axes = var_842_axes_0, beta = blocks_7_mlp_ln_bias_to_fp16, epsilon = var_778_to_fp16, gamma = blocks_7_mlp_ln_weight_to_fp16, x = x_97_cast_fp16)[name = string("op_842_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_136_to_fp16 = const()[name = string("const_136_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(194604736)))]; |
| tensor<fp16, [4096]> const_137_to_fp16 = const()[name = string("const_137_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202993408)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_46_cast_fp16 = linear(bias = const_137_to_fp16, weight = const_136_to_fp16, x = var_842_cast_fp16)[name = string("linear_46_cast_fp16")]; |
| string x_101_mode_0 = const()[name = string("x_101_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 1500, 4096]> x_101_cast_fp16 = gelu(mode = x_101_mode_0, x = linear_46_cast_fp16)[name = string("x_101_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_138_to_fp16 = const()[name = string("const_138_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(203001664)))]; |
| tensor<fp16, [1024]> const_139_to_fp16 = const()[name = string("const_139_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211390336)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_47_cast_fp16 = linear(bias = const_139_to_fp16, weight = const_138_to_fp16, x = x_101_cast_fp16)[name = string("linear_47_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_103_cast_fp16 = add(x = x_97_cast_fp16, y = linear_47_cast_fp16)[name = string("x_103_cast_fp16")]; |
| tensor<int32, [1]> var_883_axes_0 = const()[name = string("op_883_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_8_attn_ln_weight_to_fp16 = const()[name = string("blocks_8_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211392448)))]; |
| tensor<fp16, [1024]> blocks_8_attn_ln_bias_to_fp16 = const()[name = string("blocks_8_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211394560)))]; |
| fp16 var_873_to_fp16 = const()[name = string("op_873_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_883_cast_fp16 = layer_norm(axes = var_883_axes_0, beta = blocks_8_attn_ln_bias_to_fp16, epsilon = var_873_to_fp16, gamma = blocks_8_attn_ln_weight_to_fp16, x = x_103_cast_fp16)[name = string("op_883_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_140_to_fp16 = const()[name = string("const_140_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(211396672)))]; |
| tensor<fp16, [1024]> const_141_to_fp16 = const()[name = string("const_141_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213493888)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_48_cast_fp16 = linear(bias = const_141_to_fp16, weight = const_140_to_fp16, x = var_883_cast_fp16)[name = string("linear_48_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_142_to_fp16 = const()[name = string("const_142_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(213496000)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_49_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_142_to_fp16, x = var_883_cast_fp16)[name = string("linear_49_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_143_to_fp16 = const()[name = string("const_143_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215593216)))]; |
| tensor<fp16, [1024]> const_144_to_fp16 = const()[name = string("const_144_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217690432)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_50_cast_fp16 = linear(bias = const_144_to_fp16, weight = const_143_to_fp16, x = var_883_cast_fp16)[name = string("linear_50_cast_fp16")]; |
| tensor<int32, [4]> var_907 = const()[name = string("op_907"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_908_cast_fp16 = reshape(shape = var_907, x = linear_48_cast_fp16)[name = string("op_908_cast_fp16")]; |
| tensor<int32, [4]> var_913 = const()[name = string("op_913"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_914_cast_fp16 = reshape(shape = var_913, x = linear_49_cast_fp16)[name = string("op_914_cast_fp16")]; |
| tensor<int32, [4]> var_919 = const()[name = string("op_919"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_920_cast_fp16 = reshape(shape = var_919, x = linear_50_cast_fp16)[name = string("op_920_cast_fp16")]; |
| tensor<int32, [4]> transpose_168_perm_0 = const()[name = string("transpose_168_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_169_perm_0 = const()[name = string("transpose_169_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_170_perm_0 = const()[name = string("transpose_170_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_170 = transpose(perm = transpose_170_perm_0, x = var_920_cast_fp16)[name = string("transpose_277")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_169 = transpose(perm = transpose_169_perm_0, x = var_914_cast_fp16)[name = string("transpose_278")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_168 = transpose(perm = transpose_168_perm_0, x = var_908_cast_fp16)[name = string("transpose_279")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_17_cast_fp16 = scaled_dot_product_attention(key = transpose_169, query = transpose_168, value = transpose_170)[name = string("a_17_cast_fp16")]; |
| tensor<int32, [4]> var_924 = const()[name = string("op_924"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_8 = const()[name = string("concat_8"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_925_cast_fp16 = transpose(perm = var_924, x = a_17_cast_fp16)[name = string("transpose_276")]; |
| tensor<fp16, [1, 1500, 1024]> x_107_cast_fp16 = reshape(shape = concat_8, x = var_925_cast_fp16)[name = string("x_107_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_151_to_fp16 = const()[name = string("const_151_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(217692544)))]; |
| tensor<fp16, [1024]> const_152_to_fp16 = const()[name = string("const_152_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219789760)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_51_cast_fp16 = linear(bias = const_152_to_fp16, weight = const_151_to_fp16, x = x_107_cast_fp16)[name = string("linear_51_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_109_cast_fp16 = add(x = x_103_cast_fp16, y = linear_51_cast_fp16)[name = string("x_109_cast_fp16")]; |
| tensor<int32, [1]> var_937_axes_0 = const()[name = string("op_937_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_8_mlp_ln_weight_to_fp16 = const()[name = string("blocks_8_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219791872)))]; |
| tensor<fp16, [1024]> blocks_8_mlp_ln_bias_to_fp16 = const()[name = string("blocks_8_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219793984)))]; |
| tensor<fp16, [1, 1500, 1024]> var_937_cast_fp16 = layer_norm(axes = var_937_axes_0, beta = blocks_8_mlp_ln_bias_to_fp16, epsilon = var_873_to_fp16, gamma = blocks_8_mlp_ln_weight_to_fp16, x = x_109_cast_fp16)[name = string("op_937_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_153_to_fp16 = const()[name = string("const_153_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219796096)))]; |
| tensor<fp16, [4096]> const_154_to_fp16 = const()[name = string("const_154_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228184768)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_52_cast_fp16 = linear(bias = const_154_to_fp16, weight = const_153_to_fp16, x = var_937_cast_fp16)[name = string("linear_52_cast_fp16")]; |
| string x_113_mode_0 = const()[name = string("x_113_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 1500, 4096]> x_113_cast_fp16 = gelu(mode = x_113_mode_0, x = linear_52_cast_fp16)[name = string("x_113_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_155_to_fp16 = const()[name = string("const_155_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(228193024)))]; |
| tensor<fp16, [1024]> const_156_to_fp16 = const()[name = string("const_156_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236581696)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_53_cast_fp16 = linear(bias = const_156_to_fp16, weight = const_155_to_fp16, x = x_113_cast_fp16)[name = string("linear_53_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_115_cast_fp16 = add(x = x_109_cast_fp16, y = linear_53_cast_fp16)[name = string("x_115_cast_fp16")]; |
| tensor<int32, [1]> var_978_axes_0 = const()[name = string("op_978_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_9_attn_ln_weight_to_fp16 = const()[name = string("blocks_9_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236583808)))]; |
| tensor<fp16, [1024]> blocks_9_attn_ln_bias_to_fp16 = const()[name = string("blocks_9_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236585920)))]; |
| fp16 var_968_to_fp16 = const()[name = string("op_968_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_978_cast_fp16 = layer_norm(axes = var_978_axes_0, beta = blocks_9_attn_ln_bias_to_fp16, epsilon = var_968_to_fp16, gamma = blocks_9_attn_ln_weight_to_fp16, x = x_115_cast_fp16)[name = string("op_978_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_157_to_fp16 = const()[name = string("const_157_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(236588032)))]; |
| tensor<fp16, [1024]> const_158_to_fp16 = const()[name = string("const_158_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238685248)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_54_cast_fp16 = linear(bias = const_158_to_fp16, weight = const_157_to_fp16, x = var_978_cast_fp16)[name = string("linear_54_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_159_to_fp16 = const()[name = string("const_159_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(238687360)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_55_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_159_to_fp16, x = var_978_cast_fp16)[name = string("linear_55_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_160_to_fp16 = const()[name = string("const_160_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240784576)))]; |
| tensor<fp16, [1024]> const_161_to_fp16 = const()[name = string("const_161_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242881792)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_56_cast_fp16 = linear(bias = const_161_to_fp16, weight = const_160_to_fp16, x = var_978_cast_fp16)[name = string("linear_56_cast_fp16")]; |
| tensor<int32, [4]> var_1002 = const()[name = string("op_1002"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1003_cast_fp16 = reshape(shape = var_1002, x = linear_54_cast_fp16)[name = string("op_1003_cast_fp16")]; |
| tensor<int32, [4]> var_1008 = const()[name = string("op_1008"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1009_cast_fp16 = reshape(shape = var_1008, x = linear_55_cast_fp16)[name = string("op_1009_cast_fp16")]; |
| tensor<int32, [4]> var_1014 = const()[name = string("op_1014"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1015_cast_fp16 = reshape(shape = var_1014, x = linear_56_cast_fp16)[name = string("op_1015_cast_fp16")]; |
| tensor<int32, [4]> transpose_171_perm_0 = const()[name = string("transpose_171_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_172_perm_0 = const()[name = string("transpose_172_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_173_perm_0 = const()[name = string("transpose_173_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_173 = transpose(perm = transpose_173_perm_0, x = var_1015_cast_fp16)[name = string("transpose_273")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_172 = transpose(perm = transpose_172_perm_0, x = var_1009_cast_fp16)[name = string("transpose_274")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_171 = transpose(perm = transpose_171_perm_0, x = var_1003_cast_fp16)[name = string("transpose_275")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_19_cast_fp16 = scaled_dot_product_attention(key = transpose_172, query = transpose_171, value = transpose_173)[name = string("a_19_cast_fp16")]; |
| tensor<int32, [4]> var_1019 = const()[name = string("op_1019"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_9 = const()[name = string("concat_9"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1020_cast_fp16 = transpose(perm = var_1019, x = a_19_cast_fp16)[name = string("transpose_272")]; |
| tensor<fp16, [1, 1500, 1024]> x_119_cast_fp16 = reshape(shape = concat_9, x = var_1020_cast_fp16)[name = string("x_119_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_168_to_fp16 = const()[name = string("const_168_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(242883904)))]; |
| tensor<fp16, [1024]> const_169_to_fp16 = const()[name = string("const_169_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244981120)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_57_cast_fp16 = linear(bias = const_169_to_fp16, weight = const_168_to_fp16, x = x_119_cast_fp16)[name = string("linear_57_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_121_cast_fp16 = add(x = x_115_cast_fp16, y = linear_57_cast_fp16)[name = string("x_121_cast_fp16")]; |
| tensor<int32, [1]> var_1032_axes_0 = const()[name = string("op_1032_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_9_mlp_ln_weight_to_fp16 = const()[name = string("blocks_9_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244983232)))]; |
| tensor<fp16, [1024]> blocks_9_mlp_ln_bias_to_fp16 = const()[name = string("blocks_9_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244985344)))]; |
| tensor<fp16, [1, 1500, 1024]> var_1032_cast_fp16 = layer_norm(axes = var_1032_axes_0, beta = blocks_9_mlp_ln_bias_to_fp16, epsilon = var_968_to_fp16, gamma = blocks_9_mlp_ln_weight_to_fp16, x = x_121_cast_fp16)[name = string("op_1032_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_170_to_fp16 = const()[name = string("const_170_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(244987456)))]; |
| tensor<fp16, [4096]> const_171_to_fp16 = const()[name = string("const_171_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253376128)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_58_cast_fp16 = linear(bias = const_171_to_fp16, weight = const_170_to_fp16, x = var_1032_cast_fp16)[name = string("linear_58_cast_fp16")]; |
| string x_125_mode_0 = const()[name = string("x_125_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 1500, 4096]> x_125_cast_fp16 = gelu(mode = x_125_mode_0, x = linear_58_cast_fp16)[name = string("x_125_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_172_to_fp16 = const()[name = string("const_172_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253384384)))]; |
| tensor<fp16, [1024]> const_173_to_fp16 = const()[name = string("const_173_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261773056)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_59_cast_fp16 = linear(bias = const_173_to_fp16, weight = const_172_to_fp16, x = x_125_cast_fp16)[name = string("linear_59_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_127_cast_fp16 = add(x = x_121_cast_fp16, y = linear_59_cast_fp16)[name = string("x_127_cast_fp16")]; |
| tensor<int32, [1]> var_1073_axes_0 = const()[name = string("op_1073_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_10_attn_ln_weight_to_fp16 = const()[name = string("blocks_10_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261775168)))]; |
| tensor<fp16, [1024]> blocks_10_attn_ln_bias_to_fp16 = const()[name = string("blocks_10_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261777280)))]; |
| fp16 var_1063_to_fp16 = const()[name = string("op_1063_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_1073_cast_fp16 = layer_norm(axes = var_1073_axes_0, beta = blocks_10_attn_ln_bias_to_fp16, epsilon = var_1063_to_fp16, gamma = blocks_10_attn_ln_weight_to_fp16, x = x_127_cast_fp16)[name = string("op_1073_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_174_to_fp16 = const()[name = string("const_174_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(261779392)))]; |
| tensor<fp16, [1024]> const_175_to_fp16 = const()[name = string("const_175_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263876608)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_60_cast_fp16 = linear(bias = const_175_to_fp16, weight = const_174_to_fp16, x = var_1073_cast_fp16)[name = string("linear_60_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_176_to_fp16 = const()[name = string("const_176_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(263878720)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_61_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_176_to_fp16, x = var_1073_cast_fp16)[name = string("linear_61_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_177_to_fp16 = const()[name = string("const_177_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(265975936)))]; |
| tensor<fp16, [1024]> const_178_to_fp16 = const()[name = string("const_178_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268073152)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_62_cast_fp16 = linear(bias = const_178_to_fp16, weight = const_177_to_fp16, x = var_1073_cast_fp16)[name = string("linear_62_cast_fp16")]; |
| tensor<int32, [4]> var_1097 = const()[name = string("op_1097"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1098_cast_fp16 = reshape(shape = var_1097, x = linear_60_cast_fp16)[name = string("op_1098_cast_fp16")]; |
| tensor<int32, [4]> var_1103 = const()[name = string("op_1103"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1104_cast_fp16 = reshape(shape = var_1103, x = linear_61_cast_fp16)[name = string("op_1104_cast_fp16")]; |
| tensor<int32, [4]> var_1109 = const()[name = string("op_1109"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1110_cast_fp16 = reshape(shape = var_1109, x = linear_62_cast_fp16)[name = string("op_1110_cast_fp16")]; |
| tensor<int32, [4]> transpose_174_perm_0 = const()[name = string("transpose_174_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_175_perm_0 = const()[name = string("transpose_175_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_176_perm_0 = const()[name = string("transpose_176_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_176 = transpose(perm = transpose_176_perm_0, x = var_1110_cast_fp16)[name = string("transpose_269")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_175 = transpose(perm = transpose_175_perm_0, x = var_1104_cast_fp16)[name = string("transpose_270")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_174 = transpose(perm = transpose_174_perm_0, x = var_1098_cast_fp16)[name = string("transpose_271")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_21_cast_fp16 = scaled_dot_product_attention(key = transpose_175, query = transpose_174, value = transpose_176)[name = string("a_21_cast_fp16")]; |
| tensor<int32, [4]> var_1114 = const()[name = string("op_1114"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_10 = const()[name = string("concat_10"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1115_cast_fp16 = transpose(perm = var_1114, x = a_21_cast_fp16)[name = string("transpose_268")]; |
| tensor<fp16, [1, 1500, 1024]> x_131_cast_fp16 = reshape(shape = concat_10, x = var_1115_cast_fp16)[name = string("x_131_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_185_to_fp16 = const()[name = string("const_185_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(268075264)))]; |
| tensor<fp16, [1024]> const_186_to_fp16 = const()[name = string("const_186_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270172480)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_63_cast_fp16 = linear(bias = const_186_to_fp16, weight = const_185_to_fp16, x = x_131_cast_fp16)[name = string("linear_63_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_133_cast_fp16 = add(x = x_127_cast_fp16, y = linear_63_cast_fp16)[name = string("x_133_cast_fp16")]; |
| tensor<int32, [1]> var_1127_axes_0 = const()[name = string("op_1127_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_10_mlp_ln_weight_to_fp16 = const()[name = string("blocks_10_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270174592)))]; |
| tensor<fp16, [1024]> blocks_10_mlp_ln_bias_to_fp16 = const()[name = string("blocks_10_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270176704)))]; |
| tensor<fp16, [1, 1500, 1024]> var_1127_cast_fp16 = layer_norm(axes = var_1127_axes_0, beta = blocks_10_mlp_ln_bias_to_fp16, epsilon = var_1063_to_fp16, gamma = blocks_10_mlp_ln_weight_to_fp16, x = x_133_cast_fp16)[name = string("op_1127_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_187_to_fp16 = const()[name = string("const_187_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(270178816)))]; |
| tensor<fp16, [4096]> const_188_to_fp16 = const()[name = string("const_188_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278567488)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_64_cast_fp16 = linear(bias = const_188_to_fp16, weight = const_187_to_fp16, x = var_1127_cast_fp16)[name = string("linear_64_cast_fp16")]; |
| string x_137_mode_0 = const()[name = string("x_137_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 1500, 4096]> x_137_cast_fp16 = gelu(mode = x_137_mode_0, x = linear_64_cast_fp16)[name = string("x_137_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_189_to_fp16 = const()[name = string("const_189_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(278575744)))]; |
| tensor<fp16, [1024]> const_190_to_fp16 = const()[name = string("const_190_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286964416)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_65_cast_fp16 = linear(bias = const_190_to_fp16, weight = const_189_to_fp16, x = x_137_cast_fp16)[name = string("linear_65_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_139_cast_fp16 = add(x = x_133_cast_fp16, y = linear_65_cast_fp16)[name = string("x_139_cast_fp16")]; |
| tensor<int32, [1]> var_1168_axes_0 = const()[name = string("op_1168_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_11_attn_ln_weight_to_fp16 = const()[name = string("blocks_11_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286966528)))]; |
| tensor<fp16, [1024]> blocks_11_attn_ln_bias_to_fp16 = const()[name = string("blocks_11_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286968640)))]; |
| fp16 var_1158_to_fp16 = const()[name = string("op_1158_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_1168_cast_fp16 = layer_norm(axes = var_1168_axes_0, beta = blocks_11_attn_ln_bias_to_fp16, epsilon = var_1158_to_fp16, gamma = blocks_11_attn_ln_weight_to_fp16, x = x_139_cast_fp16)[name = string("op_1168_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_191_to_fp16 = const()[name = string("const_191_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(286970752)))]; |
| tensor<fp16, [1024]> const_192_to_fp16 = const()[name = string("const_192_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289067968)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_66_cast_fp16 = linear(bias = const_192_to_fp16, weight = const_191_to_fp16, x = var_1168_cast_fp16)[name = string("linear_66_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_193_to_fp16 = const()[name = string("const_193_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(289070080)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_67_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_193_to_fp16, x = var_1168_cast_fp16)[name = string("linear_67_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_194_to_fp16 = const()[name = string("const_194_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(291167296)))]; |
| tensor<fp16, [1024]> const_195_to_fp16 = const()[name = string("const_195_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293264512)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_68_cast_fp16 = linear(bias = const_195_to_fp16, weight = const_194_to_fp16, x = var_1168_cast_fp16)[name = string("linear_68_cast_fp16")]; |
| tensor<int32, [4]> var_1192 = const()[name = string("op_1192"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1193_cast_fp16 = reshape(shape = var_1192, x = linear_66_cast_fp16)[name = string("op_1193_cast_fp16")]; |
| tensor<int32, [4]> var_1198 = const()[name = string("op_1198"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1199_cast_fp16 = reshape(shape = var_1198, x = linear_67_cast_fp16)[name = string("op_1199_cast_fp16")]; |
| tensor<int32, [4]> var_1204 = const()[name = string("op_1204"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1205_cast_fp16 = reshape(shape = var_1204, x = linear_68_cast_fp16)[name = string("op_1205_cast_fp16")]; |
| tensor<int32, [4]> transpose_177_perm_0 = const()[name = string("transpose_177_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_178_perm_0 = const()[name = string("transpose_178_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_179_perm_0 = const()[name = string("transpose_179_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_179 = transpose(perm = transpose_179_perm_0, x = var_1205_cast_fp16)[name = string("transpose_265")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_178 = transpose(perm = transpose_178_perm_0, x = var_1199_cast_fp16)[name = string("transpose_266")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_177 = transpose(perm = transpose_177_perm_0, x = var_1193_cast_fp16)[name = string("transpose_267")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_23_cast_fp16 = scaled_dot_product_attention(key = transpose_178, query = transpose_177, value = transpose_179)[name = string("a_23_cast_fp16")]; |
| tensor<int32, [4]> var_1209 = const()[name = string("op_1209"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_11 = const()[name = string("concat_11"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1210_cast_fp16 = transpose(perm = var_1209, x = a_23_cast_fp16)[name = string("transpose_264")]; |
| tensor<fp16, [1, 1500, 1024]> x_143_cast_fp16 = reshape(shape = concat_11, x = var_1210_cast_fp16)[name = string("x_143_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_202_to_fp16 = const()[name = string("const_202_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293266624)))]; |
| tensor<fp16, [1024]> const_203_to_fp16 = const()[name = string("const_203_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295363840)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_69_cast_fp16 = linear(bias = const_203_to_fp16, weight = const_202_to_fp16, x = x_143_cast_fp16)[name = string("linear_69_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_145_cast_fp16 = add(x = x_139_cast_fp16, y = linear_69_cast_fp16)[name = string("x_145_cast_fp16")]; |
| tensor<int32, [1]> var_1222_axes_0 = const()[name = string("op_1222_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_11_mlp_ln_weight_to_fp16 = const()[name = string("blocks_11_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295365952)))]; |
| tensor<fp16, [1024]> blocks_11_mlp_ln_bias_to_fp16 = const()[name = string("blocks_11_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295368064)))]; |
| tensor<fp16, [1, 1500, 1024]> var_1222_cast_fp16 = layer_norm(axes = var_1222_axes_0, beta = blocks_11_mlp_ln_bias_to_fp16, epsilon = var_1158_to_fp16, gamma = blocks_11_mlp_ln_weight_to_fp16, x = x_145_cast_fp16)[name = string("op_1222_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_204_to_fp16 = const()[name = string("const_204_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295370176)))]; |
| tensor<fp16, [4096]> const_205_to_fp16 = const()[name = string("const_205_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303758848)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_70_cast_fp16 = linear(bias = const_205_to_fp16, weight = const_204_to_fp16, x = var_1222_cast_fp16)[name = string("linear_70_cast_fp16")]; |
| string x_149_mode_0 = const()[name = string("x_149_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 1500, 4096]> x_149_cast_fp16 = gelu(mode = x_149_mode_0, x = linear_70_cast_fp16)[name = string("x_149_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_206_to_fp16 = const()[name = string("const_206_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(303767104)))]; |
| tensor<fp16, [1024]> const_207_to_fp16 = const()[name = string("const_207_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312155776)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_71_cast_fp16 = linear(bias = const_207_to_fp16, weight = const_206_to_fp16, x = x_149_cast_fp16)[name = string("linear_71_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_151_cast_fp16 = add(x = x_145_cast_fp16, y = linear_71_cast_fp16)[name = string("x_151_cast_fp16")]; |
| tensor<int32, [1]> var_1263_axes_0 = const()[name = string("op_1263_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_12_attn_ln_weight_to_fp16 = const()[name = string("blocks_12_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312157888)))]; |
| tensor<fp16, [1024]> blocks_12_attn_ln_bias_to_fp16 = const()[name = string("blocks_12_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312160000)))]; |
| fp16 var_1253_to_fp16 = const()[name = string("op_1253_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_1263_cast_fp16 = layer_norm(axes = var_1263_axes_0, beta = blocks_12_attn_ln_bias_to_fp16, epsilon = var_1253_to_fp16, gamma = blocks_12_attn_ln_weight_to_fp16, x = x_151_cast_fp16)[name = string("op_1263_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_208_to_fp16 = const()[name = string("const_208_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(312162112)))]; |
| tensor<fp16, [1024]> const_209_to_fp16 = const()[name = string("const_209_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314259328)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_72_cast_fp16 = linear(bias = const_209_to_fp16, weight = const_208_to_fp16, x = var_1263_cast_fp16)[name = string("linear_72_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_210_to_fp16 = const()[name = string("const_210_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314261440)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_73_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_210_to_fp16, x = var_1263_cast_fp16)[name = string("linear_73_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_211_to_fp16 = const()[name = string("const_211_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316358656)))]; |
| tensor<fp16, [1024]> const_212_to_fp16 = const()[name = string("const_212_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(318455872)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_74_cast_fp16 = linear(bias = const_212_to_fp16, weight = const_211_to_fp16, x = var_1263_cast_fp16)[name = string("linear_74_cast_fp16")]; |
| tensor<int32, [4]> var_1287 = const()[name = string("op_1287"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1288_cast_fp16 = reshape(shape = var_1287, x = linear_72_cast_fp16)[name = string("op_1288_cast_fp16")]; |
| tensor<int32, [4]> var_1293 = const()[name = string("op_1293"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1294_cast_fp16 = reshape(shape = var_1293, x = linear_73_cast_fp16)[name = string("op_1294_cast_fp16")]; |
| tensor<int32, [4]> var_1299 = const()[name = string("op_1299"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1300_cast_fp16 = reshape(shape = var_1299, x = linear_74_cast_fp16)[name = string("op_1300_cast_fp16")]; |
| tensor<int32, [4]> transpose_180_perm_0 = const()[name = string("transpose_180_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_181_perm_0 = const()[name = string("transpose_181_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_182_perm_0 = const()[name = string("transpose_182_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_182 = transpose(perm = transpose_182_perm_0, x = var_1300_cast_fp16)[name = string("transpose_261")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_181 = transpose(perm = transpose_181_perm_0, x = var_1294_cast_fp16)[name = string("transpose_262")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_180 = transpose(perm = transpose_180_perm_0, x = var_1288_cast_fp16)[name = string("transpose_263")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_25_cast_fp16 = scaled_dot_product_attention(key = transpose_181, query = transpose_180, value = transpose_182)[name = string("a_25_cast_fp16")]; |
| tensor<int32, [4]> var_1304 = const()[name = string("op_1304"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_12 = const()[name = string("concat_12"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1305_cast_fp16 = transpose(perm = var_1304, x = a_25_cast_fp16)[name = string("transpose_260")]; |
| tensor<fp16, [1, 1500, 1024]> x_155_cast_fp16 = reshape(shape = concat_12, x = var_1305_cast_fp16)[name = string("x_155_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_219_to_fp16 = const()[name = string("const_219_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(318457984)))]; |
| tensor<fp16, [1024]> const_220_to_fp16 = const()[name = string("const_220_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320555200)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_75_cast_fp16 = linear(bias = const_220_to_fp16, weight = const_219_to_fp16, x = x_155_cast_fp16)[name = string("linear_75_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_157_cast_fp16 = add(x = x_151_cast_fp16, y = linear_75_cast_fp16)[name = string("x_157_cast_fp16")]; |
| tensor<int32, [1]> var_1317_axes_0 = const()[name = string("op_1317_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_12_mlp_ln_weight_to_fp16 = const()[name = string("blocks_12_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320557312)))]; |
| tensor<fp16, [1024]> blocks_12_mlp_ln_bias_to_fp16 = const()[name = string("blocks_12_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320559424)))]; |
| tensor<fp16, [1, 1500, 1024]> var_1317_cast_fp16 = layer_norm(axes = var_1317_axes_0, beta = blocks_12_mlp_ln_bias_to_fp16, epsilon = var_1253_to_fp16, gamma = blocks_12_mlp_ln_weight_to_fp16, x = x_157_cast_fp16)[name = string("op_1317_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_221_to_fp16 = const()[name = string("const_221_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(320561536)))]; |
| tensor<fp16, [4096]> const_222_to_fp16 = const()[name = string("const_222_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(328950208)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_76_cast_fp16 = linear(bias = const_222_to_fp16, weight = const_221_to_fp16, x = var_1317_cast_fp16)[name = string("linear_76_cast_fp16")]; |
| string x_161_mode_0 = const()[name = string("x_161_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 1500, 4096]> x_161_cast_fp16 = gelu(mode = x_161_mode_0, x = linear_76_cast_fp16)[name = string("x_161_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_223_to_fp16 = const()[name = string("const_223_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(328958464)))]; |
| tensor<fp16, [1024]> const_224_to_fp16 = const()[name = string("const_224_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337347136)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_77_cast_fp16 = linear(bias = const_224_to_fp16, weight = const_223_to_fp16, x = x_161_cast_fp16)[name = string("linear_77_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_163_cast_fp16 = add(x = x_157_cast_fp16, y = linear_77_cast_fp16)[name = string("x_163_cast_fp16")]; |
| tensor<int32, [1]> var_1358_axes_0 = const()[name = string("op_1358_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_13_attn_ln_weight_to_fp16 = const()[name = string("blocks_13_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337349248)))]; |
| tensor<fp16, [1024]> blocks_13_attn_ln_bias_to_fp16 = const()[name = string("blocks_13_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337351360)))]; |
| fp16 var_1348_to_fp16 = const()[name = string("op_1348_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_1358_cast_fp16 = layer_norm(axes = var_1358_axes_0, beta = blocks_13_attn_ln_bias_to_fp16, epsilon = var_1348_to_fp16, gamma = blocks_13_attn_ln_weight_to_fp16, x = x_163_cast_fp16)[name = string("op_1358_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_225_to_fp16 = const()[name = string("const_225_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337353472)))]; |
| tensor<fp16, [1024]> const_226_to_fp16 = const()[name = string("const_226_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339450688)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_78_cast_fp16 = linear(bias = const_226_to_fp16, weight = const_225_to_fp16, x = var_1358_cast_fp16)[name = string("linear_78_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_227_to_fp16 = const()[name = string("const_227_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(339452800)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_79_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_227_to_fp16, x = var_1358_cast_fp16)[name = string("linear_79_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_228_to_fp16 = const()[name = string("const_228_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(341550016)))]; |
| tensor<fp16, [1024]> const_229_to_fp16 = const()[name = string("const_229_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(343647232)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_80_cast_fp16 = linear(bias = const_229_to_fp16, weight = const_228_to_fp16, x = var_1358_cast_fp16)[name = string("linear_80_cast_fp16")]; |
| tensor<int32, [4]> var_1382 = const()[name = string("op_1382"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1383_cast_fp16 = reshape(shape = var_1382, x = linear_78_cast_fp16)[name = string("op_1383_cast_fp16")]; |
| tensor<int32, [4]> var_1388 = const()[name = string("op_1388"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1389_cast_fp16 = reshape(shape = var_1388, x = linear_79_cast_fp16)[name = string("op_1389_cast_fp16")]; |
| tensor<int32, [4]> var_1394 = const()[name = string("op_1394"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1395_cast_fp16 = reshape(shape = var_1394, x = linear_80_cast_fp16)[name = string("op_1395_cast_fp16")]; |
| tensor<int32, [4]> transpose_183_perm_0 = const()[name = string("transpose_183_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_184_perm_0 = const()[name = string("transpose_184_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_185_perm_0 = const()[name = string("transpose_185_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_185 = transpose(perm = transpose_185_perm_0, x = var_1395_cast_fp16)[name = string("transpose_257")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_184 = transpose(perm = transpose_184_perm_0, x = var_1389_cast_fp16)[name = string("transpose_258")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_183 = transpose(perm = transpose_183_perm_0, x = var_1383_cast_fp16)[name = string("transpose_259")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_27_cast_fp16 = scaled_dot_product_attention(key = transpose_184, query = transpose_183, value = transpose_185)[name = string("a_27_cast_fp16")]; |
| tensor<int32, [4]> var_1399 = const()[name = string("op_1399"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_13 = const()[name = string("concat_13"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1400_cast_fp16 = transpose(perm = var_1399, x = a_27_cast_fp16)[name = string("transpose_256")]; |
| tensor<fp16, [1, 1500, 1024]> x_167_cast_fp16 = reshape(shape = concat_13, x = var_1400_cast_fp16)[name = string("x_167_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_236_to_fp16 = const()[name = string("const_236_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(343649344)))]; |
| tensor<fp16, [1024]> const_237_to_fp16 = const()[name = string("const_237_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345746560)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_81_cast_fp16 = linear(bias = const_237_to_fp16, weight = const_236_to_fp16, x = x_167_cast_fp16)[name = string("linear_81_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_169_cast_fp16 = add(x = x_163_cast_fp16, y = linear_81_cast_fp16)[name = string("x_169_cast_fp16")]; |
| tensor<int32, [1]> var_1412_axes_0 = const()[name = string("op_1412_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_13_mlp_ln_weight_to_fp16 = const()[name = string("blocks_13_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345748672)))]; |
| tensor<fp16, [1024]> blocks_13_mlp_ln_bias_to_fp16 = const()[name = string("blocks_13_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345750784)))]; |
| tensor<fp16, [1, 1500, 1024]> var_1412_cast_fp16 = layer_norm(axes = var_1412_axes_0, beta = blocks_13_mlp_ln_bias_to_fp16, epsilon = var_1348_to_fp16, gamma = blocks_13_mlp_ln_weight_to_fp16, x = x_169_cast_fp16)[name = string("op_1412_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_238_to_fp16 = const()[name = string("const_238_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(345752896)))]; |
| tensor<fp16, [4096]> const_239_to_fp16 = const()[name = string("const_239_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354141568)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_82_cast_fp16 = linear(bias = const_239_to_fp16, weight = const_238_to_fp16, x = var_1412_cast_fp16)[name = string("linear_82_cast_fp16")]; |
| string x_173_mode_0 = const()[name = string("x_173_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 1500, 4096]> x_173_cast_fp16 = gelu(mode = x_173_mode_0, x = linear_82_cast_fp16)[name = string("x_173_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_240_to_fp16 = const()[name = string("const_240_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354149824)))]; |
| tensor<fp16, [1024]> const_241_to_fp16 = const()[name = string("const_241_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(362538496)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_83_cast_fp16 = linear(bias = const_241_to_fp16, weight = const_240_to_fp16, x = x_173_cast_fp16)[name = string("linear_83_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_175_cast_fp16 = add(x = x_169_cast_fp16, y = linear_83_cast_fp16)[name = string("x_175_cast_fp16")]; |
| tensor<int32, [1]> var_1453_axes_0 = const()[name = string("op_1453_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_14_attn_ln_weight_to_fp16 = const()[name = string("blocks_14_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(362540608)))]; |
| tensor<fp16, [1024]> blocks_14_attn_ln_bias_to_fp16 = const()[name = string("blocks_14_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(362542720)))]; |
| fp16 var_1443_to_fp16 = const()[name = string("op_1443_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_1453_cast_fp16 = layer_norm(axes = var_1453_axes_0, beta = blocks_14_attn_ln_bias_to_fp16, epsilon = var_1443_to_fp16, gamma = blocks_14_attn_ln_weight_to_fp16, x = x_175_cast_fp16)[name = string("op_1453_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_242_to_fp16 = const()[name = string("const_242_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(362544832)))]; |
| tensor<fp16, [1024]> const_243_to_fp16 = const()[name = string("const_243_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(364642048)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_84_cast_fp16 = linear(bias = const_243_to_fp16, weight = const_242_to_fp16, x = var_1453_cast_fp16)[name = string("linear_84_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_244_to_fp16 = const()[name = string("const_244_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(364644160)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_85_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_244_to_fp16, x = var_1453_cast_fp16)[name = string("linear_85_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_245_to_fp16 = const()[name = string("const_245_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(366741376)))]; |
| tensor<fp16, [1024]> const_246_to_fp16 = const()[name = string("const_246_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368838592)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_86_cast_fp16 = linear(bias = const_246_to_fp16, weight = const_245_to_fp16, x = var_1453_cast_fp16)[name = string("linear_86_cast_fp16")]; |
| tensor<int32, [4]> var_1477 = const()[name = string("op_1477"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1478_cast_fp16 = reshape(shape = var_1477, x = linear_84_cast_fp16)[name = string("op_1478_cast_fp16")]; |
| tensor<int32, [4]> var_1483 = const()[name = string("op_1483"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1484_cast_fp16 = reshape(shape = var_1483, x = linear_85_cast_fp16)[name = string("op_1484_cast_fp16")]; |
| tensor<int32, [4]> var_1489 = const()[name = string("op_1489"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1490_cast_fp16 = reshape(shape = var_1489, x = linear_86_cast_fp16)[name = string("op_1490_cast_fp16")]; |
| tensor<int32, [4]> transpose_186_perm_0 = const()[name = string("transpose_186_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_187_perm_0 = const()[name = string("transpose_187_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_188_perm_0 = const()[name = string("transpose_188_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_188 = transpose(perm = transpose_188_perm_0, x = var_1490_cast_fp16)[name = string("transpose_253")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_187 = transpose(perm = transpose_187_perm_0, x = var_1484_cast_fp16)[name = string("transpose_254")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_186 = transpose(perm = transpose_186_perm_0, x = var_1478_cast_fp16)[name = string("transpose_255")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_29_cast_fp16 = scaled_dot_product_attention(key = transpose_187, query = transpose_186, value = transpose_188)[name = string("a_29_cast_fp16")]; |
| tensor<int32, [4]> var_1494 = const()[name = string("op_1494"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_14 = const()[name = string("concat_14"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1495_cast_fp16 = transpose(perm = var_1494, x = a_29_cast_fp16)[name = string("transpose_252")]; |
| tensor<fp16, [1, 1500, 1024]> x_179_cast_fp16 = reshape(shape = concat_14, x = var_1495_cast_fp16)[name = string("x_179_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_253_to_fp16 = const()[name = string("const_253_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(368840704)))]; |
| tensor<fp16, [1024]> const_254_to_fp16 = const()[name = string("const_254_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370937920)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_87_cast_fp16 = linear(bias = const_254_to_fp16, weight = const_253_to_fp16, x = x_179_cast_fp16)[name = string("linear_87_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_181_cast_fp16 = add(x = x_175_cast_fp16, y = linear_87_cast_fp16)[name = string("x_181_cast_fp16")]; |
| tensor<int32, [1]> var_1507_axes_0 = const()[name = string("op_1507_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_14_mlp_ln_weight_to_fp16 = const()[name = string("blocks_14_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370940032)))]; |
| tensor<fp16, [1024]> blocks_14_mlp_ln_bias_to_fp16 = const()[name = string("blocks_14_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370942144)))]; |
| tensor<fp16, [1, 1500, 1024]> var_1507_cast_fp16 = layer_norm(axes = var_1507_axes_0, beta = blocks_14_mlp_ln_bias_to_fp16, epsilon = var_1443_to_fp16, gamma = blocks_14_mlp_ln_weight_to_fp16, x = x_181_cast_fp16)[name = string("op_1507_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_255_to_fp16 = const()[name = string("const_255_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370944256)))]; |
| tensor<fp16, [4096]> const_256_to_fp16 = const()[name = string("const_256_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(379332928)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_88_cast_fp16 = linear(bias = const_256_to_fp16, weight = const_255_to_fp16, x = var_1507_cast_fp16)[name = string("linear_88_cast_fp16")]; |
| string x_185_mode_0 = const()[name = string("x_185_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 1500, 4096]> x_185_cast_fp16 = gelu(mode = x_185_mode_0, x = linear_88_cast_fp16)[name = string("x_185_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_257_to_fp16 = const()[name = string("const_257_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(379341184)))]; |
| tensor<fp16, [1024]> const_258_to_fp16 = const()[name = string("const_258_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(387729856)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_89_cast_fp16 = linear(bias = const_258_to_fp16, weight = const_257_to_fp16, x = x_185_cast_fp16)[name = string("linear_89_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_187_cast_fp16 = add(x = x_181_cast_fp16, y = linear_89_cast_fp16)[name = string("x_187_cast_fp16")]; |
| tensor<int32, [1]> var_1548_axes_0 = const()[name = string("op_1548_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_15_attn_ln_weight_to_fp16 = const()[name = string("blocks_15_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(387731968)))]; |
| tensor<fp16, [1024]> blocks_15_attn_ln_bias_to_fp16 = const()[name = string("blocks_15_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(387734080)))]; |
| fp16 var_1538_to_fp16 = const()[name = string("op_1538_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_1548_cast_fp16 = layer_norm(axes = var_1548_axes_0, beta = blocks_15_attn_ln_bias_to_fp16, epsilon = var_1538_to_fp16, gamma = blocks_15_attn_ln_weight_to_fp16, x = x_187_cast_fp16)[name = string("op_1548_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_259_to_fp16 = const()[name = string("const_259_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(387736192)))]; |
| tensor<fp16, [1024]> const_260_to_fp16 = const()[name = string("const_260_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(389833408)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_90_cast_fp16 = linear(bias = const_260_to_fp16, weight = const_259_to_fp16, x = var_1548_cast_fp16)[name = string("linear_90_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_261_to_fp16 = const()[name = string("const_261_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(389835520)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_91_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_261_to_fp16, x = var_1548_cast_fp16)[name = string("linear_91_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_262_to_fp16 = const()[name = string("const_262_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(391932736)))]; |
| tensor<fp16, [1024]> const_263_to_fp16 = const()[name = string("const_263_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(394029952)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_92_cast_fp16 = linear(bias = const_263_to_fp16, weight = const_262_to_fp16, x = var_1548_cast_fp16)[name = string("linear_92_cast_fp16")]; |
| tensor<int32, [4]> var_1572 = const()[name = string("op_1572"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1573_cast_fp16 = reshape(shape = var_1572, x = linear_90_cast_fp16)[name = string("op_1573_cast_fp16")]; |
| tensor<int32, [4]> var_1578 = const()[name = string("op_1578"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1579_cast_fp16 = reshape(shape = var_1578, x = linear_91_cast_fp16)[name = string("op_1579_cast_fp16")]; |
| tensor<int32, [4]> var_1584 = const()[name = string("op_1584"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1585_cast_fp16 = reshape(shape = var_1584, x = linear_92_cast_fp16)[name = string("op_1585_cast_fp16")]; |
| tensor<int32, [4]> transpose_189_perm_0 = const()[name = string("transpose_189_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_190_perm_0 = const()[name = string("transpose_190_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_191_perm_0 = const()[name = string("transpose_191_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_191 = transpose(perm = transpose_191_perm_0, x = var_1585_cast_fp16)[name = string("transpose_249")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_190 = transpose(perm = transpose_190_perm_0, x = var_1579_cast_fp16)[name = string("transpose_250")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_189 = transpose(perm = transpose_189_perm_0, x = var_1573_cast_fp16)[name = string("transpose_251")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_31_cast_fp16 = scaled_dot_product_attention(key = transpose_190, query = transpose_189, value = transpose_191)[name = string("a_31_cast_fp16")]; |
| tensor<int32, [4]> var_1589 = const()[name = string("op_1589"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_15 = const()[name = string("concat_15"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1590_cast_fp16 = transpose(perm = var_1589, x = a_31_cast_fp16)[name = string("transpose_248")]; |
| tensor<fp16, [1, 1500, 1024]> x_191_cast_fp16 = reshape(shape = concat_15, x = var_1590_cast_fp16)[name = string("x_191_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_270_to_fp16 = const()[name = string("const_270_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(394032064)))]; |
| tensor<fp16, [1024]> const_271_to_fp16 = const()[name = string("const_271_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396129280)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_93_cast_fp16 = linear(bias = const_271_to_fp16, weight = const_270_to_fp16, x = x_191_cast_fp16)[name = string("linear_93_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_193_cast_fp16 = add(x = x_187_cast_fp16, y = linear_93_cast_fp16)[name = string("x_193_cast_fp16")]; |
| tensor<int32, [1]> var_1602_axes_0 = const()[name = string("op_1602_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_15_mlp_ln_weight_to_fp16 = const()[name = string("blocks_15_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396131392)))]; |
| tensor<fp16, [1024]> blocks_15_mlp_ln_bias_to_fp16 = const()[name = string("blocks_15_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396133504)))]; |
| tensor<fp16, [1, 1500, 1024]> var_1602_cast_fp16 = layer_norm(axes = var_1602_axes_0, beta = blocks_15_mlp_ln_bias_to_fp16, epsilon = var_1538_to_fp16, gamma = blocks_15_mlp_ln_weight_to_fp16, x = x_193_cast_fp16)[name = string("op_1602_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_272_to_fp16 = const()[name = string("const_272_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(396135616)))]; |
| tensor<fp16, [4096]> const_273_to_fp16 = const()[name = string("const_273_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(404524288)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_94_cast_fp16 = linear(bias = const_273_to_fp16, weight = const_272_to_fp16, x = var_1602_cast_fp16)[name = string("linear_94_cast_fp16")]; |
| string x_197_mode_0 = const()[name = string("x_197_mode_0"), val = string("EXACT")]; |
| tensor<fp16, [1, 1500, 4096]> x_197_cast_fp16 = gelu(mode = x_197_mode_0, x = linear_94_cast_fp16)[name = string("x_197_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_274_to_fp16 = const()[name = string("const_274_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(404532544)))]; |
| tensor<fp16, [1024]> const_275_to_fp16 = const()[name = string("const_275_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412921216)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_95_cast_fp16 = linear(bias = const_275_to_fp16, weight = const_274_to_fp16, x = x_197_cast_fp16)[name = string("linear_95_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_199_cast_fp16 = add(x = x_193_cast_fp16, y = linear_95_cast_fp16)[name = string("x_199_cast_fp16")]; |
| tensor<int32, [1]> var_1643_axes_0 = const()[name = string("op_1643_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_16_attn_ln_weight_to_fp16 = const()[name = string("blocks_16_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412923328)))]; |
| tensor<fp16, [1024]> blocks_16_attn_ln_bias_to_fp16 = const()[name = string("blocks_16_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412925440)))]; |
| fp16 var_1633_to_fp16 = const()[name = string("op_1633_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_1643_cast_fp16 = layer_norm(axes = var_1643_axes_0, beta = blocks_16_attn_ln_bias_to_fp16, epsilon = var_1633_to_fp16, gamma = blocks_16_attn_ln_weight_to_fp16, x = x_199_cast_fp16)[name = string("op_1643_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_276_to_fp16 = const()[name = string("const_276_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(412927552)))]; |
| tensor<fp16, [1024]> const_277_to_fp16 = const()[name = string("const_277_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(415024768)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_96_cast_fp16 = linear(bias = const_277_to_fp16, weight = const_276_to_fp16, x = var_1643_cast_fp16)[name = string("linear_96_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_278_to_fp16 = const()[name = string("const_278_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(415026880)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_97_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_278_to_fp16, x = var_1643_cast_fp16)[name = string("linear_97_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_279_to_fp16 = const()[name = string("const_279_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(417124096)))]; |
| tensor<fp16, [1024]> const_280_to_fp16 = const()[name = string("const_280_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(419221312)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_98_cast_fp16 = linear(bias = const_280_to_fp16, weight = const_279_to_fp16, x = var_1643_cast_fp16)[name = string("linear_98_cast_fp16")]; |
| tensor<int32, [4]> var_1667 = const()[name = string("op_1667"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1668_cast_fp16 = reshape(shape = var_1667, x = linear_96_cast_fp16)[name = string("op_1668_cast_fp16")]; |
| tensor<int32, [4]> var_1673 = const()[name = string("op_1673"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1674_cast_fp16 = reshape(shape = var_1673, x = linear_97_cast_fp16)[name = string("op_1674_cast_fp16")]; |
| tensor<int32, [4]> var_1679 = const()[name = string("op_1679"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1680_cast_fp16 = reshape(shape = var_1679, x = linear_98_cast_fp16)[name = string("op_1680_cast_fp16")]; |
| tensor<int32, [4]> transpose_192_perm_0 = const()[name = string("transpose_192_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_193_perm_0 = const()[name = string("transpose_193_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_194_perm_0 = const()[name = string("transpose_194_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_194 = transpose(perm = transpose_194_perm_0, x = var_1680_cast_fp16)[name = string("transpose_245")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_193 = transpose(perm = transpose_193_perm_0, x = var_1674_cast_fp16)[name = string("transpose_246")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_192 = transpose(perm = transpose_192_perm_0, x = var_1668_cast_fp16)[name = string("transpose_247")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_33_cast_fp16 = scaled_dot_product_attention(key = transpose_193, query = transpose_192, value = transpose_194)[name = string("a_33_cast_fp16")]; |
| tensor<int32, [4]> var_1684 = const()[name = string("op_1684"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_16 = const()[name = string("concat_16"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1685_cast_fp16 = transpose(perm = var_1684, x = a_33_cast_fp16)[name = string("transpose_244")]; |
| tensor<fp16, [1, 1500, 1024]> x_203_cast_fp16 = reshape(shape = concat_16, x = var_1685_cast_fp16)[name = string("x_203_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_287_to_fp16 = const()[name = string("const_287_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(419223424)))]; |
| tensor<fp16, [1024]> const_288_to_fp16 = const()[name = string("const_288_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421320640)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_99_cast_fp16 = linear(bias = const_288_to_fp16, weight = const_287_to_fp16, x = x_203_cast_fp16)[name = string("linear_99_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> 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_1697_axes_0 = const()[name = string("op_1697_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_16_mlp_ln_weight_to_fp16 = const()[name = string("blocks_16_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421322752)))]; |
| tensor<fp16, [1024]> blocks_16_mlp_ln_bias_to_fp16 = const()[name = string("blocks_16_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421324864)))]; |
| tensor<fp16, [1, 1500, 1024]> var_1697_cast_fp16 = layer_norm(axes = var_1697_axes_0, beta = blocks_16_mlp_ln_bias_to_fp16, epsilon = var_1633_to_fp16, gamma = blocks_16_mlp_ln_weight_to_fp16, x = x_205_cast_fp16)[name = string("op_1697_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_289_to_fp16 = const()[name = string("const_289_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(421326976)))]; |
| tensor<fp16, [4096]> const_290_to_fp16 = const()[name = string("const_290_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(429715648)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_100_cast_fp16 = linear(bias = const_290_to_fp16, weight = const_289_to_fp16, x = var_1697_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, 4096]> x_209_cast_fp16 = gelu(mode = x_209_mode_0, x = linear_100_cast_fp16)[name = string("x_209_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_291_to_fp16 = const()[name = string("const_291_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(429723904)))]; |
| tensor<fp16, [1024]> const_292_to_fp16 = const()[name = string("const_292_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438112576)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_101_cast_fp16 = linear(bias = const_292_to_fp16, weight = const_291_to_fp16, x = x_209_cast_fp16)[name = string("linear_101_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_211_cast_fp16 = add(x = x_205_cast_fp16, y = linear_101_cast_fp16)[name = string("x_211_cast_fp16")]; |
| tensor<int32, [1]> var_1738_axes_0 = const()[name = string("op_1738_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_17_attn_ln_weight_to_fp16 = const()[name = string("blocks_17_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438114688)))]; |
| tensor<fp16, [1024]> blocks_17_attn_ln_bias_to_fp16 = const()[name = string("blocks_17_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438116800)))]; |
| fp16 var_1728_to_fp16 = const()[name = string("op_1728_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_1738_cast_fp16 = layer_norm(axes = var_1738_axes_0, beta = blocks_17_attn_ln_bias_to_fp16, epsilon = var_1728_to_fp16, gamma = blocks_17_attn_ln_weight_to_fp16, x = x_211_cast_fp16)[name = string("op_1738_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_293_to_fp16 = const()[name = string("const_293_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(438118912)))]; |
| tensor<fp16, [1024]> const_294_to_fp16 = const()[name = string("const_294_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(440216128)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_102_cast_fp16 = linear(bias = const_294_to_fp16, weight = const_293_to_fp16, x = var_1738_cast_fp16)[name = string("linear_102_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_295_to_fp16 = const()[name = string("const_295_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(440218240)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_103_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_295_to_fp16, x = var_1738_cast_fp16)[name = string("linear_103_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_296_to_fp16 = const()[name = string("const_296_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(442315456)))]; |
| tensor<fp16, [1024]> const_297_to_fp16 = const()[name = string("const_297_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(444412672)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_104_cast_fp16 = linear(bias = const_297_to_fp16, weight = const_296_to_fp16, x = var_1738_cast_fp16)[name = string("linear_104_cast_fp16")]; |
| tensor<int32, [4]> var_1762 = const()[name = string("op_1762"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1763_cast_fp16 = reshape(shape = var_1762, x = linear_102_cast_fp16)[name = string("op_1763_cast_fp16")]; |
| tensor<int32, [4]> var_1768 = const()[name = string("op_1768"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1769_cast_fp16 = reshape(shape = var_1768, x = linear_103_cast_fp16)[name = string("op_1769_cast_fp16")]; |
| tensor<int32, [4]> var_1774 = const()[name = string("op_1774"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1775_cast_fp16 = reshape(shape = var_1774, x = linear_104_cast_fp16)[name = string("op_1775_cast_fp16")]; |
| tensor<int32, [4]> transpose_195_perm_0 = const()[name = string("transpose_195_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_196_perm_0 = const()[name = string("transpose_196_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_197_perm_0 = const()[name = string("transpose_197_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_197 = transpose(perm = transpose_197_perm_0, x = var_1775_cast_fp16)[name = string("transpose_241")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_196 = transpose(perm = transpose_196_perm_0, x = var_1769_cast_fp16)[name = string("transpose_242")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_195 = transpose(perm = transpose_195_perm_0, x = var_1763_cast_fp16)[name = string("transpose_243")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_35_cast_fp16 = scaled_dot_product_attention(key = transpose_196, query = transpose_195, value = transpose_197)[name = string("a_35_cast_fp16")]; |
| tensor<int32, [4]> var_1779 = const()[name = string("op_1779"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_17 = const()[name = string("concat_17"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1780_cast_fp16 = transpose(perm = var_1779, x = a_35_cast_fp16)[name = string("transpose_240")]; |
| tensor<fp16, [1, 1500, 1024]> x_215_cast_fp16 = reshape(shape = concat_17, x = var_1780_cast_fp16)[name = string("x_215_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_304_to_fp16 = const()[name = string("const_304_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(444414784)))]; |
| tensor<fp16, [1024]> const_305_to_fp16 = const()[name = string("const_305_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(446512000)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_105_cast_fp16 = linear(bias = const_305_to_fp16, weight = const_304_to_fp16, x = x_215_cast_fp16)[name = string("linear_105_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> 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_1792_axes_0 = const()[name = string("op_1792_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_17_mlp_ln_weight_to_fp16 = const()[name = string("blocks_17_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(446514112)))]; |
| tensor<fp16, [1024]> blocks_17_mlp_ln_bias_to_fp16 = const()[name = string("blocks_17_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(446516224)))]; |
| tensor<fp16, [1, 1500, 1024]> var_1792_cast_fp16 = layer_norm(axes = var_1792_axes_0, beta = blocks_17_mlp_ln_bias_to_fp16, epsilon = var_1728_to_fp16, gamma = blocks_17_mlp_ln_weight_to_fp16, x = x_217_cast_fp16)[name = string("op_1792_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_306_to_fp16 = const()[name = string("const_306_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(446518336)))]; |
| tensor<fp16, [4096]> const_307_to_fp16 = const()[name = string("const_307_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(454907008)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_106_cast_fp16 = linear(bias = const_307_to_fp16, weight = const_306_to_fp16, x = var_1792_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, 4096]> x_221_cast_fp16 = gelu(mode = x_221_mode_0, x = linear_106_cast_fp16)[name = string("x_221_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_308_to_fp16 = const()[name = string("const_308_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(454915264)))]; |
| tensor<fp16, [1024]> const_309_to_fp16 = const()[name = string("const_309_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(463303936)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_107_cast_fp16 = linear(bias = const_309_to_fp16, weight = const_308_to_fp16, x = x_221_cast_fp16)[name = string("linear_107_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_223_cast_fp16 = add(x = x_217_cast_fp16, y = linear_107_cast_fp16)[name = string("x_223_cast_fp16")]; |
| tensor<int32, [1]> var_1833_axes_0 = const()[name = string("op_1833_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_18_attn_ln_weight_to_fp16 = const()[name = string("blocks_18_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(463306048)))]; |
| tensor<fp16, [1024]> blocks_18_attn_ln_bias_to_fp16 = const()[name = string("blocks_18_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(463308160)))]; |
| fp16 var_1823_to_fp16 = const()[name = string("op_1823_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_1833_cast_fp16 = layer_norm(axes = var_1833_axes_0, beta = blocks_18_attn_ln_bias_to_fp16, epsilon = var_1823_to_fp16, gamma = blocks_18_attn_ln_weight_to_fp16, x = x_223_cast_fp16)[name = string("op_1833_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_310_to_fp16 = const()[name = string("const_310_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(463310272)))]; |
| tensor<fp16, [1024]> const_311_to_fp16 = const()[name = string("const_311_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(465407488)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_108_cast_fp16 = linear(bias = const_311_to_fp16, weight = const_310_to_fp16, x = var_1833_cast_fp16)[name = string("linear_108_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_312_to_fp16 = const()[name = string("const_312_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(465409600)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_109_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_312_to_fp16, x = var_1833_cast_fp16)[name = string("linear_109_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_313_to_fp16 = const()[name = string("const_313_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(467506816)))]; |
| tensor<fp16, [1024]> const_314_to_fp16 = const()[name = string("const_314_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(469604032)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_110_cast_fp16 = linear(bias = const_314_to_fp16, weight = const_313_to_fp16, x = var_1833_cast_fp16)[name = string("linear_110_cast_fp16")]; |
| tensor<int32, [4]> var_1857 = const()[name = string("op_1857"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1858_cast_fp16 = reshape(shape = var_1857, x = linear_108_cast_fp16)[name = string("op_1858_cast_fp16")]; |
| tensor<int32, [4]> var_1863 = const()[name = string("op_1863"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1864_cast_fp16 = reshape(shape = var_1863, x = linear_109_cast_fp16)[name = string("op_1864_cast_fp16")]; |
| tensor<int32, [4]> var_1869 = const()[name = string("op_1869"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1870_cast_fp16 = reshape(shape = var_1869, x = linear_110_cast_fp16)[name = string("op_1870_cast_fp16")]; |
| tensor<int32, [4]> transpose_198_perm_0 = const()[name = string("transpose_198_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_199_perm_0 = const()[name = string("transpose_199_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_200_perm_0 = const()[name = string("transpose_200_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_200 = transpose(perm = transpose_200_perm_0, x = var_1870_cast_fp16)[name = string("transpose_237")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_199 = transpose(perm = transpose_199_perm_0, x = var_1864_cast_fp16)[name = string("transpose_238")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_198 = transpose(perm = transpose_198_perm_0, x = var_1858_cast_fp16)[name = string("transpose_239")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_37_cast_fp16 = scaled_dot_product_attention(key = transpose_199, query = transpose_198, value = transpose_200)[name = string("a_37_cast_fp16")]; |
| tensor<int32, [4]> var_1874 = const()[name = string("op_1874"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_18 = const()[name = string("concat_18"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1875_cast_fp16 = transpose(perm = var_1874, x = a_37_cast_fp16)[name = string("transpose_236")]; |
| tensor<fp16, [1, 1500, 1024]> x_227_cast_fp16 = reshape(shape = concat_18, x = var_1875_cast_fp16)[name = string("x_227_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_321_to_fp16 = const()[name = string("const_321_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(469606144)))]; |
| tensor<fp16, [1024]> const_322_to_fp16 = const()[name = string("const_322_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(471703360)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_111_cast_fp16 = linear(bias = const_322_to_fp16, weight = const_321_to_fp16, x = x_227_cast_fp16)[name = string("linear_111_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> 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_1887_axes_0 = const()[name = string("op_1887_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_18_mlp_ln_weight_to_fp16 = const()[name = string("blocks_18_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(471705472)))]; |
| tensor<fp16, [1024]> blocks_18_mlp_ln_bias_to_fp16 = const()[name = string("blocks_18_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(471707584)))]; |
| tensor<fp16, [1, 1500, 1024]> var_1887_cast_fp16 = layer_norm(axes = var_1887_axes_0, beta = blocks_18_mlp_ln_bias_to_fp16, epsilon = var_1823_to_fp16, gamma = blocks_18_mlp_ln_weight_to_fp16, x = x_229_cast_fp16)[name = string("op_1887_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_323_to_fp16 = const()[name = string("const_323_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(471709696)))]; |
| tensor<fp16, [4096]> const_324_to_fp16 = const()[name = string("const_324_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(480098368)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_112_cast_fp16 = linear(bias = const_324_to_fp16, weight = const_323_to_fp16, x = var_1887_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, 4096]> x_233_cast_fp16 = gelu(mode = x_233_mode_0, x = linear_112_cast_fp16)[name = string("x_233_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_325_to_fp16 = const()[name = string("const_325_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(480106624)))]; |
| tensor<fp16, [1024]> const_326_to_fp16 = const()[name = string("const_326_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(488495296)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_113_cast_fp16 = linear(bias = const_326_to_fp16, weight = const_325_to_fp16, x = x_233_cast_fp16)[name = string("linear_113_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_235_cast_fp16 = add(x = x_229_cast_fp16, y = linear_113_cast_fp16)[name = string("x_235_cast_fp16")]; |
| tensor<int32, [1]> var_1928_axes_0 = const()[name = string("op_1928_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_19_attn_ln_weight_to_fp16 = const()[name = string("blocks_19_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(488497408)))]; |
| tensor<fp16, [1024]> blocks_19_attn_ln_bias_to_fp16 = const()[name = string("blocks_19_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(488499520)))]; |
| fp16 var_1918_to_fp16 = const()[name = string("op_1918_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_1928_cast_fp16 = layer_norm(axes = var_1928_axes_0, beta = blocks_19_attn_ln_bias_to_fp16, epsilon = var_1918_to_fp16, gamma = blocks_19_attn_ln_weight_to_fp16, x = x_235_cast_fp16)[name = string("op_1928_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_327_to_fp16 = const()[name = string("const_327_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(488501632)))]; |
| tensor<fp16, [1024]> const_328_to_fp16 = const()[name = string("const_328_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(490598848)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_114_cast_fp16 = linear(bias = const_328_to_fp16, weight = const_327_to_fp16, x = var_1928_cast_fp16)[name = string("linear_114_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_329_to_fp16 = const()[name = string("const_329_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(490600960)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_115_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_329_to_fp16, x = var_1928_cast_fp16)[name = string("linear_115_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_330_to_fp16 = const()[name = string("const_330_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(492698176)))]; |
| tensor<fp16, [1024]> const_331_to_fp16 = const()[name = string("const_331_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(494795392)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_116_cast_fp16 = linear(bias = const_331_to_fp16, weight = const_330_to_fp16, x = var_1928_cast_fp16)[name = string("linear_116_cast_fp16")]; |
| tensor<int32, [4]> var_1952 = const()[name = string("op_1952"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1953_cast_fp16 = reshape(shape = var_1952, x = linear_114_cast_fp16)[name = string("op_1953_cast_fp16")]; |
| tensor<int32, [4]> var_1958 = const()[name = string("op_1958"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1959_cast_fp16 = reshape(shape = var_1958, x = linear_115_cast_fp16)[name = string("op_1959_cast_fp16")]; |
| tensor<int32, [4]> var_1964 = const()[name = string("op_1964"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1965_cast_fp16 = reshape(shape = var_1964, x = linear_116_cast_fp16)[name = string("op_1965_cast_fp16")]; |
| tensor<int32, [4]> transpose_201_perm_0 = const()[name = string("transpose_201_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_202_perm_0 = const()[name = string("transpose_202_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_203_perm_0 = const()[name = string("transpose_203_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_203 = transpose(perm = transpose_203_perm_0, x = var_1965_cast_fp16)[name = string("transpose_233")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_202 = transpose(perm = transpose_202_perm_0, x = var_1959_cast_fp16)[name = string("transpose_234")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_201 = transpose(perm = transpose_201_perm_0, x = var_1953_cast_fp16)[name = string("transpose_235")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_39_cast_fp16 = scaled_dot_product_attention(key = transpose_202, query = transpose_201, value = transpose_203)[name = string("a_39_cast_fp16")]; |
| tensor<int32, [4]> var_1969 = const()[name = string("op_1969"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_19 = const()[name = string("concat_19"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_1970_cast_fp16 = transpose(perm = var_1969, x = a_39_cast_fp16)[name = string("transpose_232")]; |
| tensor<fp16, [1, 1500, 1024]> x_239_cast_fp16 = reshape(shape = concat_19, x = var_1970_cast_fp16)[name = string("x_239_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_338_to_fp16 = const()[name = string("const_338_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(494797504)))]; |
| tensor<fp16, [1024]> const_339_to_fp16 = const()[name = string("const_339_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(496894720)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_117_cast_fp16 = linear(bias = const_339_to_fp16, weight = const_338_to_fp16, x = x_239_cast_fp16)[name = string("linear_117_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> 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_1982_axes_0 = const()[name = string("op_1982_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_19_mlp_ln_weight_to_fp16 = const()[name = string("blocks_19_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(496896832)))]; |
| tensor<fp16, [1024]> blocks_19_mlp_ln_bias_to_fp16 = const()[name = string("blocks_19_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(496898944)))]; |
| tensor<fp16, [1, 1500, 1024]> var_1982_cast_fp16 = layer_norm(axes = var_1982_axes_0, beta = blocks_19_mlp_ln_bias_to_fp16, epsilon = var_1918_to_fp16, gamma = blocks_19_mlp_ln_weight_to_fp16, x = x_241_cast_fp16)[name = string("op_1982_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_340_to_fp16 = const()[name = string("const_340_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(496901056)))]; |
| tensor<fp16, [4096]> const_341_to_fp16 = const()[name = string("const_341_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(505289728)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_118_cast_fp16 = linear(bias = const_341_to_fp16, weight = const_340_to_fp16, x = var_1982_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, 4096]> x_245_cast_fp16 = gelu(mode = x_245_mode_0, x = linear_118_cast_fp16)[name = string("x_245_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_342_to_fp16 = const()[name = string("const_342_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(505297984)))]; |
| tensor<fp16, [1024]> const_343_to_fp16 = const()[name = string("const_343_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(513686656)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_119_cast_fp16 = linear(bias = const_343_to_fp16, weight = const_342_to_fp16, x = x_245_cast_fp16)[name = string("linear_119_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_247_cast_fp16 = add(x = x_241_cast_fp16, y = linear_119_cast_fp16)[name = string("x_247_cast_fp16")]; |
| tensor<int32, [1]> var_2023_axes_0 = const()[name = string("op_2023_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_20_attn_ln_weight_to_fp16 = const()[name = string("blocks_20_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(513688768)))]; |
| tensor<fp16, [1024]> blocks_20_attn_ln_bias_to_fp16 = const()[name = string("blocks_20_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(513690880)))]; |
| fp16 var_2013_to_fp16 = const()[name = string("op_2013_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_2023_cast_fp16 = layer_norm(axes = var_2023_axes_0, beta = blocks_20_attn_ln_bias_to_fp16, epsilon = var_2013_to_fp16, gamma = blocks_20_attn_ln_weight_to_fp16, x = x_247_cast_fp16)[name = string("op_2023_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_344_to_fp16 = const()[name = string("const_344_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(513692992)))]; |
| tensor<fp16, [1024]> const_345_to_fp16 = const()[name = string("const_345_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(515790208)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_120_cast_fp16 = linear(bias = const_345_to_fp16, weight = const_344_to_fp16, x = var_2023_cast_fp16)[name = string("linear_120_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_346_to_fp16 = const()[name = string("const_346_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(515792320)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_121_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_346_to_fp16, x = var_2023_cast_fp16)[name = string("linear_121_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_347_to_fp16 = const()[name = string("const_347_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(517889536)))]; |
| tensor<fp16, [1024]> const_348_to_fp16 = const()[name = string("const_348_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(519986752)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_122_cast_fp16 = linear(bias = const_348_to_fp16, weight = const_347_to_fp16, x = var_2023_cast_fp16)[name = string("linear_122_cast_fp16")]; |
| tensor<int32, [4]> var_2047 = const()[name = string("op_2047"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_2048_cast_fp16 = reshape(shape = var_2047, x = linear_120_cast_fp16)[name = string("op_2048_cast_fp16")]; |
| tensor<int32, [4]> var_2053 = const()[name = string("op_2053"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_2054_cast_fp16 = reshape(shape = var_2053, x = linear_121_cast_fp16)[name = string("op_2054_cast_fp16")]; |
| tensor<int32, [4]> var_2059 = const()[name = string("op_2059"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_2060_cast_fp16 = reshape(shape = var_2059, x = linear_122_cast_fp16)[name = string("op_2060_cast_fp16")]; |
| tensor<int32, [4]> transpose_204_perm_0 = const()[name = string("transpose_204_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_205_perm_0 = const()[name = string("transpose_205_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_206_perm_0 = const()[name = string("transpose_206_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_206 = transpose(perm = transpose_206_perm_0, x = var_2060_cast_fp16)[name = string("transpose_229")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_205 = transpose(perm = transpose_205_perm_0, x = var_2054_cast_fp16)[name = string("transpose_230")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_204 = transpose(perm = transpose_204_perm_0, x = var_2048_cast_fp16)[name = string("transpose_231")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_41_cast_fp16 = scaled_dot_product_attention(key = transpose_205, query = transpose_204, value = transpose_206)[name = string("a_41_cast_fp16")]; |
| tensor<int32, [4]> var_2064 = const()[name = string("op_2064"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_20 = const()[name = string("concat_20"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_2065_cast_fp16 = transpose(perm = var_2064, x = a_41_cast_fp16)[name = string("transpose_228")]; |
| tensor<fp16, [1, 1500, 1024]> x_251_cast_fp16 = reshape(shape = concat_20, x = var_2065_cast_fp16)[name = string("x_251_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_355_to_fp16 = const()[name = string("const_355_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(519988864)))]; |
| tensor<fp16, [1024]> const_356_to_fp16 = const()[name = string("const_356_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(522086080)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_123_cast_fp16 = linear(bias = const_356_to_fp16, weight = const_355_to_fp16, x = x_251_cast_fp16)[name = string("linear_123_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> 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_2077_axes_0 = const()[name = string("op_2077_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_20_mlp_ln_weight_to_fp16 = const()[name = string("blocks_20_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(522088192)))]; |
| tensor<fp16, [1024]> blocks_20_mlp_ln_bias_to_fp16 = const()[name = string("blocks_20_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(522090304)))]; |
| tensor<fp16, [1, 1500, 1024]> var_2077_cast_fp16 = layer_norm(axes = var_2077_axes_0, beta = blocks_20_mlp_ln_bias_to_fp16, epsilon = var_2013_to_fp16, gamma = blocks_20_mlp_ln_weight_to_fp16, x = x_253_cast_fp16)[name = string("op_2077_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_357_to_fp16 = const()[name = string("const_357_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(522092416)))]; |
| tensor<fp16, [4096]> const_358_to_fp16 = const()[name = string("const_358_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(530481088)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_124_cast_fp16 = linear(bias = const_358_to_fp16, weight = const_357_to_fp16, x = var_2077_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, 4096]> x_257_cast_fp16 = gelu(mode = x_257_mode_0, x = linear_124_cast_fp16)[name = string("x_257_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_359_to_fp16 = const()[name = string("const_359_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(530489344)))]; |
| tensor<fp16, [1024]> const_360_to_fp16 = const()[name = string("const_360_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(538878016)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_125_cast_fp16 = linear(bias = const_360_to_fp16, weight = const_359_to_fp16, x = x_257_cast_fp16)[name = string("linear_125_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_259_cast_fp16 = add(x = x_253_cast_fp16, y = linear_125_cast_fp16)[name = string("x_259_cast_fp16")]; |
| tensor<int32, [1]> var_2118_axes_0 = const()[name = string("op_2118_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_21_attn_ln_weight_to_fp16 = const()[name = string("blocks_21_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(538880128)))]; |
| tensor<fp16, [1024]> blocks_21_attn_ln_bias_to_fp16 = const()[name = string("blocks_21_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(538882240)))]; |
| fp16 var_2108_to_fp16 = const()[name = string("op_2108_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_2118_cast_fp16 = layer_norm(axes = var_2118_axes_0, beta = blocks_21_attn_ln_bias_to_fp16, epsilon = var_2108_to_fp16, gamma = blocks_21_attn_ln_weight_to_fp16, x = x_259_cast_fp16)[name = string("op_2118_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_361_to_fp16 = const()[name = string("const_361_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(538884352)))]; |
| tensor<fp16, [1024]> const_362_to_fp16 = const()[name = string("const_362_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(540981568)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_126_cast_fp16 = linear(bias = const_362_to_fp16, weight = const_361_to_fp16, x = var_2118_cast_fp16)[name = string("linear_126_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_363_to_fp16 = const()[name = string("const_363_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(540983680)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_127_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_363_to_fp16, x = var_2118_cast_fp16)[name = string("linear_127_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_364_to_fp16 = const()[name = string("const_364_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(543080896)))]; |
| tensor<fp16, [1024]> const_365_to_fp16 = const()[name = string("const_365_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(545178112)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_128_cast_fp16 = linear(bias = const_365_to_fp16, weight = const_364_to_fp16, x = var_2118_cast_fp16)[name = string("linear_128_cast_fp16")]; |
| tensor<int32, [4]> var_2142 = const()[name = string("op_2142"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_2143_cast_fp16 = reshape(shape = var_2142, x = linear_126_cast_fp16)[name = string("op_2143_cast_fp16")]; |
| tensor<int32, [4]> var_2148 = const()[name = string("op_2148"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_2149_cast_fp16 = reshape(shape = var_2148, x = linear_127_cast_fp16)[name = string("op_2149_cast_fp16")]; |
| tensor<int32, [4]> var_2154 = const()[name = string("op_2154"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_2155_cast_fp16 = reshape(shape = var_2154, x = linear_128_cast_fp16)[name = string("op_2155_cast_fp16")]; |
| tensor<int32, [4]> transpose_207_perm_0 = const()[name = string("transpose_207_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_208_perm_0 = const()[name = string("transpose_208_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_209_perm_0 = const()[name = string("transpose_209_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_209 = transpose(perm = transpose_209_perm_0, x = var_2155_cast_fp16)[name = string("transpose_225")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_208 = transpose(perm = transpose_208_perm_0, x = var_2149_cast_fp16)[name = string("transpose_226")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_207 = transpose(perm = transpose_207_perm_0, x = var_2143_cast_fp16)[name = string("transpose_227")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_43_cast_fp16 = scaled_dot_product_attention(key = transpose_208, query = transpose_207, value = transpose_209)[name = string("a_43_cast_fp16")]; |
| tensor<int32, [4]> var_2159 = const()[name = string("op_2159"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_21 = const()[name = string("concat_21"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_2160_cast_fp16 = transpose(perm = var_2159, x = a_43_cast_fp16)[name = string("transpose_224")]; |
| tensor<fp16, [1, 1500, 1024]> x_263_cast_fp16 = reshape(shape = concat_21, x = var_2160_cast_fp16)[name = string("x_263_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_372_to_fp16 = const()[name = string("const_372_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(545180224)))]; |
| tensor<fp16, [1024]> const_373_to_fp16 = const()[name = string("const_373_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(547277440)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_129_cast_fp16 = linear(bias = const_373_to_fp16, weight = const_372_to_fp16, x = x_263_cast_fp16)[name = string("linear_129_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> 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_2172_axes_0 = const()[name = string("op_2172_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_21_mlp_ln_weight_to_fp16 = const()[name = string("blocks_21_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(547279552)))]; |
| tensor<fp16, [1024]> blocks_21_mlp_ln_bias_to_fp16 = const()[name = string("blocks_21_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(547281664)))]; |
| tensor<fp16, [1, 1500, 1024]> var_2172_cast_fp16 = layer_norm(axes = var_2172_axes_0, beta = blocks_21_mlp_ln_bias_to_fp16, epsilon = var_2108_to_fp16, gamma = blocks_21_mlp_ln_weight_to_fp16, x = x_265_cast_fp16)[name = string("op_2172_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_374_to_fp16 = const()[name = string("const_374_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(547283776)))]; |
| tensor<fp16, [4096]> const_375_to_fp16 = const()[name = string("const_375_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(555672448)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_130_cast_fp16 = linear(bias = const_375_to_fp16, weight = const_374_to_fp16, x = var_2172_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, 4096]> x_269_cast_fp16 = gelu(mode = x_269_mode_0, x = linear_130_cast_fp16)[name = string("x_269_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_376_to_fp16 = const()[name = string("const_376_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(555680704)))]; |
| tensor<fp16, [1024]> const_377_to_fp16 = const()[name = string("const_377_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(564069376)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_131_cast_fp16 = linear(bias = const_377_to_fp16, weight = const_376_to_fp16, x = x_269_cast_fp16)[name = string("linear_131_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_271_cast_fp16 = add(x = x_265_cast_fp16, y = linear_131_cast_fp16)[name = string("x_271_cast_fp16")]; |
| tensor<int32, [1]> var_2213_axes_0 = const()[name = string("op_2213_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_22_attn_ln_weight_to_fp16 = const()[name = string("blocks_22_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(564071488)))]; |
| tensor<fp16, [1024]> blocks_22_attn_ln_bias_to_fp16 = const()[name = string("blocks_22_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(564073600)))]; |
| fp16 var_2203_to_fp16 = const()[name = string("op_2203_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_2213_cast_fp16 = layer_norm(axes = var_2213_axes_0, beta = blocks_22_attn_ln_bias_to_fp16, epsilon = var_2203_to_fp16, gamma = blocks_22_attn_ln_weight_to_fp16, x = x_271_cast_fp16)[name = string("op_2213_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_378_to_fp16 = const()[name = string("const_378_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(564075712)))]; |
| tensor<fp16, [1024]> const_379_to_fp16 = const()[name = string("const_379_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(566172928)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_132_cast_fp16 = linear(bias = const_379_to_fp16, weight = const_378_to_fp16, x = var_2213_cast_fp16)[name = string("linear_132_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_380_to_fp16 = const()[name = string("const_380_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(566175040)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_133_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_380_to_fp16, x = var_2213_cast_fp16)[name = string("linear_133_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_381_to_fp16 = const()[name = string("const_381_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(568272256)))]; |
| tensor<fp16, [1024]> const_382_to_fp16 = const()[name = string("const_382_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(570369472)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_134_cast_fp16 = linear(bias = const_382_to_fp16, weight = const_381_to_fp16, x = var_2213_cast_fp16)[name = string("linear_134_cast_fp16")]; |
| tensor<int32, [4]> var_2237 = const()[name = string("op_2237"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_2238_cast_fp16 = reshape(shape = var_2237, x = linear_132_cast_fp16)[name = string("op_2238_cast_fp16")]; |
| tensor<int32, [4]> var_2243 = const()[name = string("op_2243"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_2244_cast_fp16 = reshape(shape = var_2243, x = linear_133_cast_fp16)[name = string("op_2244_cast_fp16")]; |
| tensor<int32, [4]> var_2249 = const()[name = string("op_2249"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_2250_cast_fp16 = reshape(shape = var_2249, x = linear_134_cast_fp16)[name = string("op_2250_cast_fp16")]; |
| tensor<int32, [4]> transpose_210_perm_0 = const()[name = string("transpose_210_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_211_perm_0 = const()[name = string("transpose_211_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_212_perm_0 = const()[name = string("transpose_212_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_212 = transpose(perm = transpose_212_perm_0, x = var_2250_cast_fp16)[name = string("transpose_221")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_211 = transpose(perm = transpose_211_perm_0, x = var_2244_cast_fp16)[name = string("transpose_222")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_210 = transpose(perm = transpose_210_perm_0, x = var_2238_cast_fp16)[name = string("transpose_223")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_45_cast_fp16 = scaled_dot_product_attention(key = transpose_211, query = transpose_210, value = transpose_212)[name = string("a_45_cast_fp16")]; |
| tensor<int32, [4]> var_2254 = const()[name = string("op_2254"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_22 = const()[name = string("concat_22"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_2255_cast_fp16 = transpose(perm = var_2254, x = a_45_cast_fp16)[name = string("transpose_220")]; |
| tensor<fp16, [1, 1500, 1024]> x_275_cast_fp16 = reshape(shape = concat_22, x = var_2255_cast_fp16)[name = string("x_275_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_389_to_fp16 = const()[name = string("const_389_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(570371584)))]; |
| tensor<fp16, [1024]> const_390_to_fp16 = const()[name = string("const_390_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(572468800)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_135_cast_fp16 = linear(bias = const_390_to_fp16, weight = const_389_to_fp16, x = x_275_cast_fp16)[name = string("linear_135_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> 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_2267_axes_0 = const()[name = string("op_2267_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_22_mlp_ln_weight_to_fp16 = const()[name = string("blocks_22_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(572470912)))]; |
| tensor<fp16, [1024]> blocks_22_mlp_ln_bias_to_fp16 = const()[name = string("blocks_22_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(572473024)))]; |
| tensor<fp16, [1, 1500, 1024]> var_2267_cast_fp16 = layer_norm(axes = var_2267_axes_0, beta = blocks_22_mlp_ln_bias_to_fp16, epsilon = var_2203_to_fp16, gamma = blocks_22_mlp_ln_weight_to_fp16, x = x_277_cast_fp16)[name = string("op_2267_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_391_to_fp16 = const()[name = string("const_391_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(572475136)))]; |
| tensor<fp16, [4096]> const_392_to_fp16 = const()[name = string("const_392_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(580863808)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_136_cast_fp16 = linear(bias = const_392_to_fp16, weight = const_391_to_fp16, x = var_2267_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, 4096]> x_281_cast_fp16 = gelu(mode = x_281_mode_0, x = linear_136_cast_fp16)[name = string("x_281_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_393_to_fp16 = const()[name = string("const_393_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(580872064)))]; |
| tensor<fp16, [1024]> const_394_to_fp16 = const()[name = string("const_394_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(589260736)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_137_cast_fp16 = linear(bias = const_394_to_fp16, weight = const_393_to_fp16, x = x_281_cast_fp16)[name = string("linear_137_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_283_cast_fp16 = add(x = x_277_cast_fp16, y = linear_137_cast_fp16)[name = string("x_283_cast_fp16")]; |
| tensor<int32, [1]> var_2308_axes_0 = const()[name = string("op_2308_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_23_attn_ln_weight_to_fp16 = const()[name = string("blocks_23_attn_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(589262848)))]; |
| tensor<fp16, [1024]> blocks_23_attn_ln_bias_to_fp16 = const()[name = string("blocks_23_attn_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(589264960)))]; |
| fp16 var_2298_to_fp16 = const()[name = string("op_2298_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> var_2308_cast_fp16 = layer_norm(axes = var_2308_axes_0, beta = blocks_23_attn_ln_bias_to_fp16, epsilon = var_2298_to_fp16, gamma = blocks_23_attn_ln_weight_to_fp16, x = x_283_cast_fp16)[name = string("op_2308_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_395_to_fp16 = const()[name = string("const_395_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(589267072)))]; |
| tensor<fp16, [1024]> const_396_to_fp16 = const()[name = string("const_396_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(591364288)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_138_cast_fp16 = linear(bias = const_396_to_fp16, weight = const_395_to_fp16, x = var_2308_cast_fp16)[name = string("linear_138_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_397_to_fp16 = const()[name = string("const_397_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(591366400)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_139_cast_fp16 = linear(bias = linear_1_bias_0_to_fp16, weight = const_397_to_fp16, x = var_2308_cast_fp16)[name = string("linear_139_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_398_to_fp16 = const()[name = string("const_398_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(593463616)))]; |
| tensor<fp16, [1024]> const_399_to_fp16 = const()[name = string("const_399_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(595560832)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_140_cast_fp16 = linear(bias = const_399_to_fp16, weight = const_398_to_fp16, x = var_2308_cast_fp16)[name = string("linear_140_cast_fp16")]; |
| tensor<int32, [4]> var_2332 = const()[name = string("op_2332"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_2333_cast_fp16 = reshape(shape = var_2332, x = linear_138_cast_fp16)[name = string("op_2333_cast_fp16")]; |
| tensor<int32, [4]> var_2338 = const()[name = string("op_2338"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_2339_cast_fp16 = reshape(shape = var_2338, x = linear_139_cast_fp16)[name = string("op_2339_cast_fp16")]; |
| tensor<int32, [4]> var_2344 = const()[name = string("op_2344"), val = tensor<int32, [4]>([1, 1500, 16, -1])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_2345_cast_fp16 = reshape(shape = var_2344, x = linear_140_cast_fp16)[name = string("op_2345_cast_fp16")]; |
| tensor<int32, [4]> transpose_213_perm_0 = const()[name = string("transpose_213_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_214_perm_0 = const()[name = string("transpose_214_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_215_perm_0 = const()[name = string("transpose_215_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_215 = transpose(perm = transpose_215_perm_0, x = var_2345_cast_fp16)[name = string("transpose_217")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_214 = transpose(perm = transpose_214_perm_0, x = var_2339_cast_fp16)[name = string("transpose_218")]; |
| tensor<fp16, [1, 16, 1500, 64]> transpose_213 = transpose(perm = transpose_213_perm_0, x = var_2333_cast_fp16)[name = string("transpose_219")]; |
| tensor<fp16, [1, 16, 1500, 64]> a_cast_fp16 = scaled_dot_product_attention(key = transpose_214, query = transpose_213, value = transpose_215)[name = string("a_cast_fp16")]; |
| tensor<int32, [4]> var_2349 = const()[name = string("op_2349"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_23 = const()[name = string("concat_23"), val = tensor<int32, [3]>([1, 1500, 1024])]; |
| tensor<fp16, [1, 1500, 16, 64]> var_2350_cast_fp16 = transpose(perm = var_2349, x = a_cast_fp16)[name = string("transpose_216")]; |
| tensor<fp16, [1, 1500, 1024]> x_287_cast_fp16 = reshape(shape = concat_23, x = var_2350_cast_fp16)[name = string("x_287_cast_fp16")]; |
| tensor<fp16, [1024, 1024]> const_406_to_fp16 = const()[name = string("const_406_to_fp16"), val = tensor<fp16, [1024, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(595562944)))]; |
| tensor<fp16, [1024]> const_407_to_fp16 = const()[name = string("const_407_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(597660160)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_141_cast_fp16 = linear(bias = const_407_to_fp16, weight = const_406_to_fp16, x = x_287_cast_fp16)[name = string("linear_141_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> 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_2362_axes_0 = const()[name = string("op_2362_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> blocks_23_mlp_ln_weight_to_fp16 = const()[name = string("blocks_23_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(597662272)))]; |
| tensor<fp16, [1024]> blocks_23_mlp_ln_bias_to_fp16 = const()[name = string("blocks_23_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(597664384)))]; |
| tensor<fp16, [1, 1500, 1024]> var_2362_cast_fp16 = layer_norm(axes = var_2362_axes_0, beta = blocks_23_mlp_ln_bias_to_fp16, epsilon = var_2298_to_fp16, gamma = blocks_23_mlp_ln_weight_to_fp16, x = x_289_cast_fp16)[name = string("op_2362_cast_fp16")]; |
| tensor<fp16, [4096, 1024]> const_408_to_fp16 = const()[name = string("const_408_to_fp16"), val = tensor<fp16, [4096, 1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(597666496)))]; |
| tensor<fp16, [4096]> const_409_to_fp16 = const()[name = string("const_409_to_fp16"), val = tensor<fp16, [4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(606055168)))]; |
| tensor<fp16, [1, 1500, 4096]> linear_142_cast_fp16 = linear(bias = const_409_to_fp16, weight = const_408_to_fp16, x = var_2362_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, 4096]> x_293_cast_fp16 = gelu(mode = x_293_mode_0, x = linear_142_cast_fp16)[name = string("x_293_cast_fp16")]; |
| tensor<fp16, [1024, 4096]> const_410_to_fp16 = const()[name = string("const_410_to_fp16"), val = tensor<fp16, [1024, 4096]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(606063424)))]; |
| tensor<fp16, [1024]> const_411_to_fp16 = const()[name = string("const_411_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(614452096)))]; |
| tensor<fp16, [1, 1500, 1024]> linear_143_cast_fp16 = linear(bias = const_411_to_fp16, weight = const_410_to_fp16, x = x_293_cast_fp16)[name = string("linear_143_cast_fp16")]; |
| tensor<fp16, [1, 1500, 1024]> x_cast_fp16 = add(x = x_289_cast_fp16, y = linear_143_cast_fp16)[name = string("x_cast_fp16")]; |
| tensor<int32, [1]> var_2391_axes_0 = const()[name = string("op_2391_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1024]> ln_post_weight_to_fp16 = const()[name = string("ln_post_weight_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(614454208)))]; |
| tensor<fp16, [1024]> ln_post_bias_to_fp16 = const()[name = string("ln_post_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(614456320)))]; |
| fp16 var_2382_to_fp16 = const()[name = string("op_2382_to_fp16"), val = fp16(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1024]> output = layer_norm(axes = var_2391_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_2382_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast_fp16)[name = string("op_2391_cast_fp16")]; |
| } -> (output); |
| } |
