| program(1.0) |
| [buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "4.28.4"}, {"coremlc-version", "1436.100.10"}})] |
| { |
| func main<ios15>(tensor<fp32, [1, 80, 3000]> logmel_data) { |
| tensor<int32, []> var_72 = const()[name = tensor<string, []>("op_72"), val = tensor<int32, []>(1)]; |
| tensor<int32, [1]> var_80 = const()[name = tensor<string, []>("op_80"), val = tensor<int32, [1]>([1])]; |
| tensor<int32, [1]> var_82 = const()[name = tensor<string, []>("op_82"), val = tensor<int32, [1]>([1])]; |
| tensor<string, []> var_84_pad_type_0 = const()[name = tensor<string, []>("op_84_pad_type_0"), val = tensor<string, []>("custom")]; |
| tensor<int32, [2]> var_84_pad_0 = const()[name = tensor<string, []>("op_84_pad_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<string, []> logmel_data_to_fp16_dtype_0 = const()[name = tensor<string, []>("logmel_data_to_fp16_dtype_0"), val = tensor<string, []>("fp16")]; |
| tensor<fp16, [1280, 80, 3]> weight_3_to_fp16 = const()[name = tensor<string, []>("weight_3_to_fp16"), val = tensor<fp16, [1280, 80, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))]; |
| tensor<fp16, [1280]> bias_3_to_fp16 = const()[name = tensor<string, []>("bias_3_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(614528)))]; |
| tensor<fp16, [1, 80, 3000]> cast_967 = cast(dtype = logmel_data_to_fp16_dtype_0, x = logmel_data); |
| tensor<fp16, [1, 1280, 3000]> var_84_cast = conv(bias = bias_3_to_fp16, dilations = var_82, groups = var_72, pad = var_84_pad_0, pad_type = var_84_pad_type_0, strides = var_80, weight = weight_3_to_fp16, x = cast_967); |
| tensor<string, []> input_1_mode_0 = const()[name = tensor<string, []>("input_1_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1280, 3000]> input_1_cast = gelu(mode = input_1_mode_0, x = var_84_cast); |
| tensor<int32, []> var_88 = const()[name = tensor<string, []>("op_88"), val = tensor<int32, []>(1)]; |
| tensor<int32, [1]> var_97 = const()[name = tensor<string, []>("op_97"), val = tensor<int32, [1]>([2])]; |
| tensor<int32, [1]> var_99 = const()[name = tensor<string, []>("op_99"), val = tensor<int32, [1]>([1])]; |
| tensor<string, []> var_101_pad_type_0 = const()[name = tensor<string, []>("op_101_pad_type_0"), val = tensor<string, []>("custom")]; |
| tensor<int32, [2]> var_101_pad_0 = const()[name = tensor<string, []>("op_101_pad_0"), val = tensor<int32, [2]>([1, 1])]; |
| tensor<fp16, [1280, 1280, 3]> weight_7_to_fp16 = const()[name = tensor<string, []>("weight_7_to_fp16"), val = tensor<fp16, [1280, 1280, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(617152)))]; |
| tensor<fp16, [1280]> bias_7_to_fp16 = const()[name = tensor<string, []>("bias_7_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10447616)))]; |
| tensor<fp16, [1, 1280, 1500]> var_101_cast = conv(bias = bias_7_to_fp16, dilations = var_99, groups = var_88, pad = var_101_pad_0, pad_type = var_101_pad_type_0, strides = var_97, weight = weight_7_to_fp16, x = input_1_cast); |
| tensor<string, []> x_3_mode_0 = const()[name = tensor<string, []>("x_3_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1280, 1500]> x_3_cast = gelu(mode = x_3_mode_0, x = var_101_cast); |
| tensor<int32, [3]> var_106 = const()[name = tensor<string, []>("op_106"), val = tensor<int32, [3]>([0, 2, 1])]; |
| tensor<fp16, [1500, 1280]> positional_embedding_to_fp16 = const()[name = tensor<string, []>("positional_embedding_to_fp16"), val = tensor<fp16, [1500, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10450240)))]; |
| tensor<fp16, [1, 1500, 1280]> transpose_256 = transpose(perm = var_106, x = x_3_cast); |
| tensor<fp16, [1, 1500, 1280]> var_109_cast = add(x = transpose_256, y = positional_embedding_to_fp16); |
| tensor<int32, []> var_122 = const()[name = tensor<string, []>("op_122"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_139_axes_0 = const()[name = tensor<string, []>("op_139_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_0_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_0_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14290304)))]; |
| tensor<fp16, [1280]> blocks_0_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_0_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14292928)))]; |
| tensor<fp16, []> var_128_to_fp16 = const()[name = tensor<string, []>("op_128_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_139_cast = layer_norm(axes = var_139_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_128_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_109_cast); |
| tensor<fp16, [1280, 1280]> var_150_to_fp16 = const()[name = tensor<string, []>("op_150_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14295552)))]; |
| tensor<fp16, [1280]> var_151_to_fp16 = const()[name = tensor<string, []>("op_151_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17572416)))]; |
| tensor<fp16, [1, 1500, 1280]> q_1_cast = linear(bias = var_151_to_fp16, weight = var_150_to_fp16, x = var_139_cast); |
| tensor<fp16, [1280, 1280]> var_154_to_fp16 = const()[name = tensor<string, []>("op_154_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17575040)))]; |
| tensor<fp16, [1280]> k_1_bias_0_to_fp16 = const()[name = tensor<string, []>("k_1_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20851904)))]; |
| tensor<fp16, [1, 1500, 1280]> k_1_cast = linear(bias = k_1_bias_0_to_fp16, weight = var_154_to_fp16, x = var_139_cast); |
| tensor<fp16, [1280, 1280]> var_158_to_fp16 = const()[name = tensor<string, []>("op_158_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20854528)))]; |
| tensor<fp16, [1280]> var_159_to_fp16 = const()[name = tensor<string, []>("op_159_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24131392)))]; |
| tensor<fp16, [1, 1500, 1280]> v_1_cast = linear(bias = var_159_to_fp16, weight = var_158_to_fp16, x = var_139_cast); |
| tensor<int32, [4]> var_167 = const()[name = tensor<string, []>("op_167"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_168_cast = reshape(shape = var_167, x = q_1_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_224_to_fp16 = const()[name = tensor<string, []>("const_224_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_3_cast = mul(x = var_168_cast, y = const_224_to_fp16); |
| tensor<int32, [4]> var_174 = const()[name = tensor<string, []>("op_174"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_175_cast = reshape(shape = var_174, x = k_1_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_225_to_fp16 = const()[name = tensor<string, []>("const_225_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_3_cast = mul(x = var_175_cast, y = const_225_to_fp16); |
| tensor<int32, [4]> var_181 = const()[name = tensor<string, []>("op_181"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_182_cast = reshape(shape = var_181, x = v_1_cast); |
| tensor<int32, [4]> var_183 = const()[name = tensor<string, []>("op_183"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_1_transpose_x_0 = const()[name = tensor<string, []>("qk_1_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_1_transpose_y_0 = const()[name = tensor<string, []>("qk_1_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_64_perm_0 = const()[name = tensor<string, []>("transpose_64_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_65_perm_0 = const()[name = tensor<string, []>("transpose_65_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_253 = transpose(perm = transpose_65_perm_0, x = k_3_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_254 = transpose(perm = transpose_64_perm_0, x = q_3_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_1_cast = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_254, y = transpose_253); |
| tensor<fp16, [1, 20, 1500, 1500]> var_187_cast = softmax(axis = var_122, x = qk_1_cast); |
| tensor<bool, []> var_189_transpose_x_0 = const()[name = tensor<string, []>("op_189_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_189_transpose_y_0 = const()[name = tensor<string, []>("op_189_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_255 = transpose(perm = var_183, x = var_182_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_189_cast = matmul(transpose_x = var_189_transpose_x_0, transpose_y = var_189_transpose_y_0, x = var_187_cast, y = transpose_255); |
| tensor<int32, [4]> var_190 = const()[name = tensor<string, []>("op_190"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_0 = const()[name = tensor<string, []>("concat_0"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_252 = transpose(perm = var_190, x = var_189_cast); |
| tensor<fp16, [1, 1500, 1280]> x_11_cast = reshape(shape = concat_0, x = transpose_252); |
| tensor<fp16, [1280, 1280]> var_195_to_fp16 = const()[name = tensor<string, []>("op_195_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24134016)))]; |
| tensor<fp16, [1280]> var_196_to_fp16 = const()[name = tensor<string, []>("op_196_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27410880)))]; |
| tensor<fp16, [1, 1500, 1280]> var_197_cast = linear(bias = var_196_to_fp16, weight = var_195_to_fp16, x = x_11_cast); |
| tensor<fp16, [1, 1500, 1280]> x_13_cast = add(x = var_109_cast, y = var_197_cast); |
| tensor<int32, [1]> var_203_axes_0 = const()[name = tensor<string, []>("op_203_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_0_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_0_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27413504)))]; |
| tensor<fp16, [1280]> blocks_0_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_0_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27416128)))]; |
| tensor<fp16, [1, 1500, 1280]> var_203_cast = layer_norm(axes = var_203_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_128_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast); |
| tensor<fp16, [5120, 1280]> var_212_to_fp16 = const()[name = tensor<string, []>("op_212_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27418752)))]; |
| tensor<fp16, [5120]> var_213_to_fp16 = const()[name = tensor<string, []>("op_213_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40526016)))]; |
| tensor<fp16, [1, 1500, 5120]> input_9_cast = linear(bias = var_213_to_fp16, weight = var_212_to_fp16, x = var_203_cast); |
| tensor<string, []> x_17_mode_0 = const()[name = tensor<string, []>("x_17_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_17_cast = gelu(mode = x_17_mode_0, x = input_9_cast); |
| tensor<fp16, [1280, 5120]> var_218_to_fp16 = const()[name = tensor<string, []>("op_218_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40536320)))]; |
| tensor<fp16, [1280]> var_219_to_fp16 = const()[name = tensor<string, []>("op_219_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53643584)))]; |
| tensor<fp16, [1, 1500, 1280]> var_220_cast = linear(bias = var_219_to_fp16, weight = var_218_to_fp16, x = x_17_cast); |
| tensor<fp16, [1, 1500, 1280]> x_19_cast = add(x = x_13_cast, y = var_220_cast); |
| tensor<int32, []> var_229 = const()[name = tensor<string, []>("op_229"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_246_axes_0 = const()[name = tensor<string, []>("op_246_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_1_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_1_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53646208)))]; |
| tensor<fp16, [1280]> blocks_1_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_1_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53648832)))]; |
| tensor<fp16, []> var_235_to_fp16 = const()[name = tensor<string, []>("op_235_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_246_cast = layer_norm(axes = var_246_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_235_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast); |
| tensor<fp16, [1280, 1280]> var_257_to_fp16 = const()[name = tensor<string, []>("op_257_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53651456)))]; |
| tensor<fp16, [1280]> var_258_to_fp16 = const()[name = tensor<string, []>("op_258_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56928320)))]; |
| tensor<fp16, [1, 1500, 1280]> q_5_cast = linear(bias = var_258_to_fp16, weight = var_257_to_fp16, x = var_246_cast); |
| tensor<fp16, [1280, 1280]> var_261_to_fp16 = const()[name = tensor<string, []>("op_261_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56930944)))]; |
| tensor<fp16, [1280]> k_5_bias_0_to_fp16 = const()[name = tensor<string, []>("k_5_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60207808)))]; |
| tensor<fp16, [1, 1500, 1280]> k_5_cast = linear(bias = k_5_bias_0_to_fp16, weight = var_261_to_fp16, x = var_246_cast); |
| tensor<fp16, [1280, 1280]> var_265_to_fp16 = const()[name = tensor<string, []>("op_265_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60210432)))]; |
| tensor<fp16, [1280]> var_266_to_fp16 = const()[name = tensor<string, []>("op_266_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63487296)))]; |
| tensor<fp16, [1, 1500, 1280]> v_5_cast = linear(bias = var_266_to_fp16, weight = var_265_to_fp16, x = var_246_cast); |
| tensor<int32, [4]> var_274 = const()[name = tensor<string, []>("op_274"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_275_cast = reshape(shape = var_274, x = q_5_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_226_to_fp16 = const()[name = tensor<string, []>("const_226_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_7_cast = mul(x = var_275_cast, y = const_226_to_fp16); |
| tensor<int32, [4]> var_281 = const()[name = tensor<string, []>("op_281"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_282_cast = reshape(shape = var_281, x = k_5_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_227_to_fp16 = const()[name = tensor<string, []>("const_227_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_7_cast = mul(x = var_282_cast, y = const_227_to_fp16); |
| tensor<int32, [4]> var_288 = const()[name = tensor<string, []>("op_288"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_289_cast = reshape(shape = var_288, x = v_5_cast); |
| tensor<int32, [4]> var_290 = const()[name = tensor<string, []>("op_290"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_3_transpose_x_0 = const()[name = tensor<string, []>("qk_3_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_3_transpose_y_0 = const()[name = tensor<string, []>("qk_3_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_66_perm_0 = const()[name = tensor<string, []>("transpose_66_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_67_perm_0 = const()[name = tensor<string, []>("transpose_67_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_249 = transpose(perm = transpose_67_perm_0, x = k_7_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_250 = transpose(perm = transpose_66_perm_0, x = q_7_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_3_cast = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_250, y = transpose_249); |
| tensor<fp16, [1, 20, 1500, 1500]> var_294_cast = softmax(axis = var_229, x = qk_3_cast); |
| tensor<bool, []> var_296_transpose_x_0 = const()[name = tensor<string, []>("op_296_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_296_transpose_y_0 = const()[name = tensor<string, []>("op_296_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_251 = transpose(perm = var_290, x = var_289_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_296_cast = matmul(transpose_x = var_296_transpose_x_0, transpose_y = var_296_transpose_y_0, x = var_294_cast, y = transpose_251); |
| tensor<int32, [4]> var_297 = const()[name = tensor<string, []>("op_297"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_1 = const()[name = tensor<string, []>("concat_1"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_248 = transpose(perm = var_297, x = var_296_cast); |
| tensor<fp16, [1, 1500, 1280]> x_23_cast = reshape(shape = concat_1, x = transpose_248); |
| tensor<fp16, [1280, 1280]> var_302_to_fp16 = const()[name = tensor<string, []>("op_302_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63489920)))]; |
| tensor<fp16, [1280]> var_303_to_fp16 = const()[name = tensor<string, []>("op_303_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66766784)))]; |
| tensor<fp16, [1, 1500, 1280]> var_304_cast = linear(bias = var_303_to_fp16, weight = var_302_to_fp16, x = x_23_cast); |
| tensor<fp16, [1, 1500, 1280]> x_25_cast = add(x = x_19_cast, y = var_304_cast); |
| tensor<int32, [1]> var_310_axes_0 = const()[name = tensor<string, []>("op_310_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_1_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_1_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66769408)))]; |
| tensor<fp16, [1280]> blocks_1_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_1_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66772032)))]; |
| tensor<fp16, [1, 1500, 1280]> var_310_cast = layer_norm(axes = var_310_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_235_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast); |
| tensor<fp16, [5120, 1280]> var_319_to_fp16 = const()[name = tensor<string, []>("op_319_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66774656)))]; |
| tensor<fp16, [5120]> var_320_to_fp16 = const()[name = tensor<string, []>("op_320_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79881920)))]; |
| tensor<fp16, [1, 1500, 5120]> input_17_cast = linear(bias = var_320_to_fp16, weight = var_319_to_fp16, x = var_310_cast); |
| tensor<string, []> x_29_mode_0 = const()[name = tensor<string, []>("x_29_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_29_cast = gelu(mode = x_29_mode_0, x = input_17_cast); |
| tensor<fp16, [1280, 5120]> var_325_to_fp16 = const()[name = tensor<string, []>("op_325_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79892224)))]; |
| tensor<fp16, [1280]> var_326_to_fp16 = const()[name = tensor<string, []>("op_326_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92999488)))]; |
| tensor<fp16, [1, 1500, 1280]> var_327_cast = linear(bias = var_326_to_fp16, weight = var_325_to_fp16, x = x_29_cast); |
| tensor<fp16, [1, 1500, 1280]> x_31_cast = add(x = x_25_cast, y = var_327_cast); |
| tensor<int32, []> var_336 = const()[name = tensor<string, []>("op_336"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_353_axes_0 = const()[name = tensor<string, []>("op_353_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_2_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_2_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93002112)))]; |
| tensor<fp16, [1280]> blocks_2_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_2_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93004736)))]; |
| tensor<fp16, []> var_342_to_fp16 = const()[name = tensor<string, []>("op_342_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_353_cast = layer_norm(axes = var_353_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_342_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast); |
| tensor<fp16, [1280, 1280]> var_364_to_fp16 = const()[name = tensor<string, []>("op_364_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93007360)))]; |
| tensor<fp16, [1280]> var_365_to_fp16 = const()[name = tensor<string, []>("op_365_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96284224)))]; |
| tensor<fp16, [1, 1500, 1280]> q_9_cast = linear(bias = var_365_to_fp16, weight = var_364_to_fp16, x = var_353_cast); |
| tensor<fp16, [1280, 1280]> var_368_to_fp16 = const()[name = tensor<string, []>("op_368_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96286848)))]; |
| tensor<fp16, [1280]> k_9_bias_0_to_fp16 = const()[name = tensor<string, []>("k_9_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99563712)))]; |
| tensor<fp16, [1, 1500, 1280]> k_9_cast = linear(bias = k_9_bias_0_to_fp16, weight = var_368_to_fp16, x = var_353_cast); |
| tensor<fp16, [1280, 1280]> var_372_to_fp16 = const()[name = tensor<string, []>("op_372_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99566336)))]; |
| tensor<fp16, [1280]> var_373_to_fp16 = const()[name = tensor<string, []>("op_373_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102843200)))]; |
| tensor<fp16, [1, 1500, 1280]> v_9_cast = linear(bias = var_373_to_fp16, weight = var_372_to_fp16, x = var_353_cast); |
| tensor<int32, [4]> var_381 = const()[name = tensor<string, []>("op_381"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_382_cast = reshape(shape = var_381, x = q_9_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_228_to_fp16 = const()[name = tensor<string, []>("const_228_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_11_cast = mul(x = var_382_cast, y = const_228_to_fp16); |
| tensor<int32, [4]> var_388 = const()[name = tensor<string, []>("op_388"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_389_cast = reshape(shape = var_388, x = k_9_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_229_to_fp16 = const()[name = tensor<string, []>("const_229_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_11_cast = mul(x = var_389_cast, y = const_229_to_fp16); |
| tensor<int32, [4]> var_395 = const()[name = tensor<string, []>("op_395"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_396_cast = reshape(shape = var_395, x = v_9_cast); |
| tensor<int32, [4]> var_397 = const()[name = tensor<string, []>("op_397"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_5_transpose_x_0 = const()[name = tensor<string, []>("qk_5_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_5_transpose_y_0 = const()[name = tensor<string, []>("qk_5_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_68_perm_0 = const()[name = tensor<string, []>("transpose_68_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_69_perm_0 = const()[name = tensor<string, []>("transpose_69_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_245 = transpose(perm = transpose_69_perm_0, x = k_11_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_246 = transpose(perm = transpose_68_perm_0, x = q_11_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_5_cast = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_246, y = transpose_245); |
| tensor<fp16, [1, 20, 1500, 1500]> var_401_cast = softmax(axis = var_336, x = qk_5_cast); |
| tensor<bool, []> var_403_transpose_x_0 = const()[name = tensor<string, []>("op_403_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_403_transpose_y_0 = const()[name = tensor<string, []>("op_403_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_247 = transpose(perm = var_397, x = var_396_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_403_cast = matmul(transpose_x = var_403_transpose_x_0, transpose_y = var_403_transpose_y_0, x = var_401_cast, y = transpose_247); |
| tensor<int32, [4]> var_404 = const()[name = tensor<string, []>("op_404"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_2 = const()[name = tensor<string, []>("concat_2"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_244 = transpose(perm = var_404, x = var_403_cast); |
| tensor<fp16, [1, 1500, 1280]> x_35_cast = reshape(shape = concat_2, x = transpose_244); |
| tensor<fp16, [1280, 1280]> var_409_to_fp16 = const()[name = tensor<string, []>("op_409_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102845824)))]; |
| tensor<fp16, [1280]> var_410_to_fp16 = const()[name = tensor<string, []>("op_410_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106122688)))]; |
| tensor<fp16, [1, 1500, 1280]> var_411_cast = linear(bias = var_410_to_fp16, weight = var_409_to_fp16, x = x_35_cast); |
| tensor<fp16, [1, 1500, 1280]> x_37_cast = add(x = x_31_cast, y = var_411_cast); |
| tensor<int32, [1]> var_417_axes_0 = const()[name = tensor<string, []>("op_417_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_2_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_2_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106125312)))]; |
| tensor<fp16, [1280]> blocks_2_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_2_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106127936)))]; |
| tensor<fp16, [1, 1500, 1280]> var_417_cast = layer_norm(axes = var_417_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_342_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast); |
| tensor<fp16, [5120, 1280]> var_426_to_fp16 = const()[name = tensor<string, []>("op_426_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106130560)))]; |
| tensor<fp16, [5120]> var_427_to_fp16 = const()[name = tensor<string, []>("op_427_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119237824)))]; |
| tensor<fp16, [1, 1500, 5120]> input_25_cast = linear(bias = var_427_to_fp16, weight = var_426_to_fp16, x = var_417_cast); |
| tensor<string, []> x_41_mode_0 = const()[name = tensor<string, []>("x_41_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_41_cast = gelu(mode = x_41_mode_0, x = input_25_cast); |
| tensor<fp16, [1280, 5120]> var_432_to_fp16 = const()[name = tensor<string, []>("op_432_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119248128)))]; |
| tensor<fp16, [1280]> var_433_to_fp16 = const()[name = tensor<string, []>("op_433_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132355392)))]; |
| tensor<fp16, [1, 1500, 1280]> var_434_cast = linear(bias = var_433_to_fp16, weight = var_432_to_fp16, x = x_41_cast); |
| tensor<fp16, [1, 1500, 1280]> x_43_cast = add(x = x_37_cast, y = var_434_cast); |
| tensor<int32, []> var_443 = const()[name = tensor<string, []>("op_443"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_460_axes_0 = const()[name = tensor<string, []>("op_460_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_3_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_3_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132358016)))]; |
| tensor<fp16, [1280]> blocks_3_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_3_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132360640)))]; |
| tensor<fp16, []> var_449_to_fp16 = const()[name = tensor<string, []>("op_449_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_460_cast = layer_norm(axes = var_460_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_449_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast); |
| tensor<fp16, [1280, 1280]> var_471_to_fp16 = const()[name = tensor<string, []>("op_471_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(132363264)))]; |
| tensor<fp16, [1280]> var_472_to_fp16 = const()[name = tensor<string, []>("op_472_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135640128)))]; |
| tensor<fp16, [1, 1500, 1280]> q_13_cast = linear(bias = var_472_to_fp16, weight = var_471_to_fp16, x = var_460_cast); |
| tensor<fp16, [1280, 1280]> var_475_to_fp16 = const()[name = tensor<string, []>("op_475_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(135642752)))]; |
| tensor<fp16, [1280]> k_13_bias_0_to_fp16 = const()[name = tensor<string, []>("k_13_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138919616)))]; |
| tensor<fp16, [1, 1500, 1280]> k_13_cast = linear(bias = k_13_bias_0_to_fp16, weight = var_475_to_fp16, x = var_460_cast); |
| tensor<fp16, [1280, 1280]> var_479_to_fp16 = const()[name = tensor<string, []>("op_479_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138922240)))]; |
| tensor<fp16, [1280]> var_480_to_fp16 = const()[name = tensor<string, []>("op_480_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142199104)))]; |
| tensor<fp16, [1, 1500, 1280]> v_13_cast = linear(bias = var_480_to_fp16, weight = var_479_to_fp16, x = var_460_cast); |
| tensor<int32, [4]> var_488 = const()[name = tensor<string, []>("op_488"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_489_cast = reshape(shape = var_488, x = q_13_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_230_to_fp16 = const()[name = tensor<string, []>("const_230_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_15_cast = mul(x = var_489_cast, y = const_230_to_fp16); |
| tensor<int32, [4]> var_495 = const()[name = tensor<string, []>("op_495"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_496_cast = reshape(shape = var_495, x = k_13_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_231_to_fp16 = const()[name = tensor<string, []>("const_231_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_15_cast = mul(x = var_496_cast, y = const_231_to_fp16); |
| tensor<int32, [4]> var_502 = const()[name = tensor<string, []>("op_502"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_503_cast = reshape(shape = var_502, x = v_13_cast); |
| tensor<int32, [4]> var_504 = const()[name = tensor<string, []>("op_504"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_7_transpose_x_0 = const()[name = tensor<string, []>("qk_7_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_7_transpose_y_0 = const()[name = tensor<string, []>("qk_7_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_70_perm_0 = const()[name = tensor<string, []>("transpose_70_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_71_perm_0 = const()[name = tensor<string, []>("transpose_71_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_241 = transpose(perm = transpose_71_perm_0, x = k_15_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_242 = transpose(perm = transpose_70_perm_0, x = q_15_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_7_cast = matmul(transpose_x = qk_7_transpose_x_0, transpose_y = qk_7_transpose_y_0, x = transpose_242, y = transpose_241); |
| tensor<fp16, [1, 20, 1500, 1500]> var_508_cast = softmax(axis = var_443, x = qk_7_cast); |
| tensor<bool, []> var_510_transpose_x_0 = const()[name = tensor<string, []>("op_510_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_510_transpose_y_0 = const()[name = tensor<string, []>("op_510_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_243 = transpose(perm = var_504, x = var_503_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_510_cast = matmul(transpose_x = var_510_transpose_x_0, transpose_y = var_510_transpose_y_0, x = var_508_cast, y = transpose_243); |
| tensor<int32, [4]> var_511 = const()[name = tensor<string, []>("op_511"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_3 = const()[name = tensor<string, []>("concat_3"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_240 = transpose(perm = var_511, x = var_510_cast); |
| tensor<fp16, [1, 1500, 1280]> x_47_cast = reshape(shape = concat_3, x = transpose_240); |
| tensor<fp16, [1280, 1280]> var_516_to_fp16 = const()[name = tensor<string, []>("op_516_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(142201728)))]; |
| tensor<fp16, [1280]> var_517_to_fp16 = const()[name = tensor<string, []>("op_517_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145478592)))]; |
| tensor<fp16, [1, 1500, 1280]> var_518_cast = linear(bias = var_517_to_fp16, weight = var_516_to_fp16, x = x_47_cast); |
| tensor<fp16, [1, 1500, 1280]> x_49_cast = add(x = x_43_cast, y = var_518_cast); |
| tensor<int32, [1]> var_524_axes_0 = const()[name = tensor<string, []>("op_524_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_3_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_3_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145481216)))]; |
| tensor<fp16, [1280]> blocks_3_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_3_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145483840)))]; |
| tensor<fp16, [1, 1500, 1280]> var_524_cast = layer_norm(axes = var_524_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_449_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast); |
| tensor<fp16, [5120, 1280]> var_533_to_fp16 = const()[name = tensor<string, []>("op_533_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(145486464)))]; |
| tensor<fp16, [5120]> var_534_to_fp16 = const()[name = tensor<string, []>("op_534_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158593728)))]; |
| tensor<fp16, [1, 1500, 5120]> input_33_cast = linear(bias = var_534_to_fp16, weight = var_533_to_fp16, x = var_524_cast); |
| tensor<string, []> x_53_mode_0 = const()[name = tensor<string, []>("x_53_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_53_cast = gelu(mode = x_53_mode_0, x = input_33_cast); |
| tensor<fp16, [1280, 5120]> var_539_to_fp16 = const()[name = tensor<string, []>("op_539_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(158604032)))]; |
| tensor<fp16, [1280]> var_540_to_fp16 = const()[name = tensor<string, []>("op_540_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(171711296)))]; |
| tensor<fp16, [1, 1500, 1280]> var_541_cast = linear(bias = var_540_to_fp16, weight = var_539_to_fp16, x = x_53_cast); |
| tensor<fp16, [1, 1500, 1280]> x_55_cast = add(x = x_49_cast, y = var_541_cast); |
| tensor<int32, []> var_550 = const()[name = tensor<string, []>("op_550"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_567_axes_0 = const()[name = tensor<string, []>("op_567_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_4_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_4_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(171713920)))]; |
| tensor<fp16, [1280]> blocks_4_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_4_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(171716544)))]; |
| tensor<fp16, []> var_556_to_fp16 = const()[name = tensor<string, []>("op_556_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_567_cast = layer_norm(axes = var_567_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_556_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast); |
| tensor<fp16, [1280, 1280]> var_578_to_fp16 = const()[name = tensor<string, []>("op_578_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(171719168)))]; |
| tensor<fp16, [1280]> var_579_to_fp16 = const()[name = tensor<string, []>("op_579_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(174996032)))]; |
| tensor<fp16, [1, 1500, 1280]> q_17_cast = linear(bias = var_579_to_fp16, weight = var_578_to_fp16, x = var_567_cast); |
| tensor<fp16, [1280, 1280]> var_582_to_fp16 = const()[name = tensor<string, []>("op_582_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(174998656)))]; |
| tensor<fp16, [1280]> k_17_bias_0_to_fp16 = const()[name = tensor<string, []>("k_17_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178275520)))]; |
| tensor<fp16, [1, 1500, 1280]> k_17_cast = linear(bias = k_17_bias_0_to_fp16, weight = var_582_to_fp16, x = var_567_cast); |
| tensor<fp16, [1280, 1280]> var_586_to_fp16 = const()[name = tensor<string, []>("op_586_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(178278144)))]; |
| tensor<fp16, [1280]> var_587_to_fp16 = const()[name = tensor<string, []>("op_587_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(181555008)))]; |
| tensor<fp16, [1, 1500, 1280]> v_17_cast = linear(bias = var_587_to_fp16, weight = var_586_to_fp16, x = var_567_cast); |
| tensor<int32, [4]> var_595 = const()[name = tensor<string, []>("op_595"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_596_cast = reshape(shape = var_595, x = q_17_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_232_to_fp16 = const()[name = tensor<string, []>("const_232_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_19_cast = mul(x = var_596_cast, y = const_232_to_fp16); |
| tensor<int32, [4]> var_602 = const()[name = tensor<string, []>("op_602"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_603_cast = reshape(shape = var_602, x = k_17_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_233_to_fp16 = const()[name = tensor<string, []>("const_233_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_19_cast = mul(x = var_603_cast, y = const_233_to_fp16); |
| tensor<int32, [4]> var_609 = const()[name = tensor<string, []>("op_609"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_610_cast = reshape(shape = var_609, x = v_17_cast); |
| tensor<int32, [4]> var_611 = const()[name = tensor<string, []>("op_611"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_9_transpose_x_0 = const()[name = tensor<string, []>("qk_9_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_9_transpose_y_0 = const()[name = tensor<string, []>("qk_9_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_72_perm_0 = const()[name = tensor<string, []>("transpose_72_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_73_perm_0 = const()[name = tensor<string, []>("transpose_73_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_237 = transpose(perm = transpose_73_perm_0, x = k_19_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_238 = transpose(perm = transpose_72_perm_0, x = q_19_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_9_cast = matmul(transpose_x = qk_9_transpose_x_0, transpose_y = qk_9_transpose_y_0, x = transpose_238, y = transpose_237); |
| tensor<fp16, [1, 20, 1500, 1500]> var_615_cast = softmax(axis = var_550, x = qk_9_cast); |
| tensor<bool, []> var_617_transpose_x_0 = const()[name = tensor<string, []>("op_617_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_617_transpose_y_0 = const()[name = tensor<string, []>("op_617_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_239 = transpose(perm = var_611, x = var_610_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_617_cast = matmul(transpose_x = var_617_transpose_x_0, transpose_y = var_617_transpose_y_0, x = var_615_cast, y = transpose_239); |
| tensor<int32, [4]> var_618 = const()[name = tensor<string, []>("op_618"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_4 = const()[name = tensor<string, []>("concat_4"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_236 = transpose(perm = var_618, x = var_617_cast); |
| tensor<fp16, [1, 1500, 1280]> x_59_cast = reshape(shape = concat_4, x = transpose_236); |
| tensor<fp16, [1280, 1280]> var_623_to_fp16 = const()[name = tensor<string, []>("op_623_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(181557632)))]; |
| tensor<fp16, [1280]> var_624_to_fp16 = const()[name = tensor<string, []>("op_624_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(184834496)))]; |
| tensor<fp16, [1, 1500, 1280]> var_625_cast = linear(bias = var_624_to_fp16, weight = var_623_to_fp16, x = x_59_cast); |
| tensor<fp16, [1, 1500, 1280]> x_61_cast = add(x = x_55_cast, y = var_625_cast); |
| tensor<int32, [1]> var_631_axes_0 = const()[name = tensor<string, []>("op_631_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_4_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_4_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(184837120)))]; |
| tensor<fp16, [1280]> blocks_4_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_4_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(184839744)))]; |
| tensor<fp16, [1, 1500, 1280]> var_631_cast = layer_norm(axes = var_631_axes_0, beta = blocks_4_mlp_ln_bias_to_fp16, epsilon = var_556_to_fp16, gamma = blocks_4_mlp_ln_weight_to_fp16, x = x_61_cast); |
| tensor<fp16, [5120, 1280]> var_640_to_fp16 = const()[name = tensor<string, []>("op_640_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(184842368)))]; |
| tensor<fp16, [5120]> var_641_to_fp16 = const()[name = tensor<string, []>("op_641_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(197949632)))]; |
| tensor<fp16, [1, 1500, 5120]> input_41_cast = linear(bias = var_641_to_fp16, weight = var_640_to_fp16, x = var_631_cast); |
| tensor<string, []> x_65_mode_0 = const()[name = tensor<string, []>("x_65_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_65_cast = gelu(mode = x_65_mode_0, x = input_41_cast); |
| tensor<fp16, [1280, 5120]> var_646_to_fp16 = const()[name = tensor<string, []>("op_646_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(197959936)))]; |
| tensor<fp16, [1280]> var_647_to_fp16 = const()[name = tensor<string, []>("op_647_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211067200)))]; |
| tensor<fp16, [1, 1500, 1280]> var_648_cast = linear(bias = var_647_to_fp16, weight = var_646_to_fp16, x = x_65_cast); |
| tensor<fp16, [1, 1500, 1280]> x_67_cast = add(x = x_61_cast, y = var_648_cast); |
| tensor<int32, []> var_657 = const()[name = tensor<string, []>("op_657"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_674_axes_0 = const()[name = tensor<string, []>("op_674_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_5_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_5_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211069824)))]; |
| tensor<fp16, [1280]> blocks_5_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_5_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211072448)))]; |
| tensor<fp16, []> var_663_to_fp16 = const()[name = tensor<string, []>("op_663_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_674_cast = layer_norm(axes = var_674_axes_0, beta = blocks_5_attn_ln_bias_to_fp16, epsilon = var_663_to_fp16, gamma = blocks_5_attn_ln_weight_to_fp16, x = x_67_cast); |
| tensor<fp16, [1280, 1280]> var_685_to_fp16 = const()[name = tensor<string, []>("op_685_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(211075072)))]; |
| tensor<fp16, [1280]> var_686_to_fp16 = const()[name = tensor<string, []>("op_686_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214351936)))]; |
| tensor<fp16, [1, 1500, 1280]> q_21_cast = linear(bias = var_686_to_fp16, weight = var_685_to_fp16, x = var_674_cast); |
| tensor<fp16, [1280, 1280]> var_689_to_fp16 = const()[name = tensor<string, []>("op_689_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(214354560)))]; |
| tensor<fp16, [1280]> k_21_bias_0_to_fp16 = const()[name = tensor<string, []>("k_21_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217631424)))]; |
| tensor<fp16, [1, 1500, 1280]> k_21_cast = linear(bias = k_21_bias_0_to_fp16, weight = var_689_to_fp16, x = var_674_cast); |
| tensor<fp16, [1280, 1280]> var_693_to_fp16 = const()[name = tensor<string, []>("op_693_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(217634048)))]; |
| tensor<fp16, [1280]> var_694_to_fp16 = const()[name = tensor<string, []>("op_694_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220910912)))]; |
| tensor<fp16, [1, 1500, 1280]> v_21_cast = linear(bias = var_694_to_fp16, weight = var_693_to_fp16, x = var_674_cast); |
| tensor<int32, [4]> var_702 = const()[name = tensor<string, []>("op_702"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_703_cast = reshape(shape = var_702, x = q_21_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_234_to_fp16 = const()[name = tensor<string, []>("const_234_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_23_cast = mul(x = var_703_cast, y = const_234_to_fp16); |
| tensor<int32, [4]> var_709 = const()[name = tensor<string, []>("op_709"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_710_cast = reshape(shape = var_709, x = k_21_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_235_to_fp16 = const()[name = tensor<string, []>("const_235_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_23_cast = mul(x = var_710_cast, y = const_235_to_fp16); |
| tensor<int32, [4]> var_716 = const()[name = tensor<string, []>("op_716"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_717_cast = reshape(shape = var_716, x = v_21_cast); |
| tensor<int32, [4]> var_718 = const()[name = tensor<string, []>("op_718"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_11_transpose_x_0 = const()[name = tensor<string, []>("qk_11_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_11_transpose_y_0 = const()[name = tensor<string, []>("qk_11_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_74_perm_0 = const()[name = tensor<string, []>("transpose_74_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_75_perm_0 = const()[name = tensor<string, []>("transpose_75_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_233 = transpose(perm = transpose_75_perm_0, x = k_23_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_234 = transpose(perm = transpose_74_perm_0, x = q_23_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_11_cast = matmul(transpose_x = qk_11_transpose_x_0, transpose_y = qk_11_transpose_y_0, x = transpose_234, y = transpose_233); |
| tensor<fp16, [1, 20, 1500, 1500]> var_722_cast = softmax(axis = var_657, x = qk_11_cast); |
| tensor<bool, []> var_724_transpose_x_0 = const()[name = tensor<string, []>("op_724_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_724_transpose_y_0 = const()[name = tensor<string, []>("op_724_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_235 = transpose(perm = var_718, x = var_717_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_724_cast = matmul(transpose_x = var_724_transpose_x_0, transpose_y = var_724_transpose_y_0, x = var_722_cast, y = transpose_235); |
| tensor<int32, [4]> var_725 = const()[name = tensor<string, []>("op_725"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_5 = const()[name = tensor<string, []>("concat_5"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_232 = transpose(perm = var_725, x = var_724_cast); |
| tensor<fp16, [1, 1500, 1280]> x_71_cast = reshape(shape = concat_5, x = transpose_232); |
| tensor<fp16, [1280, 1280]> var_730_to_fp16 = const()[name = tensor<string, []>("op_730_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(220913536)))]; |
| tensor<fp16, [1280]> var_731_to_fp16 = const()[name = tensor<string, []>("op_731_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(224190400)))]; |
| tensor<fp16, [1, 1500, 1280]> var_732_cast = linear(bias = var_731_to_fp16, weight = var_730_to_fp16, x = x_71_cast); |
| tensor<fp16, [1, 1500, 1280]> x_73_cast = add(x = x_67_cast, y = var_732_cast); |
| tensor<int32, [1]> var_738_axes_0 = const()[name = tensor<string, []>("op_738_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_5_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_5_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(224193024)))]; |
| tensor<fp16, [1280]> blocks_5_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_5_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(224195648)))]; |
| tensor<fp16, [1, 1500, 1280]> var_738_cast = layer_norm(axes = var_738_axes_0, beta = blocks_5_mlp_ln_bias_to_fp16, epsilon = var_663_to_fp16, gamma = blocks_5_mlp_ln_weight_to_fp16, x = x_73_cast); |
| tensor<fp16, [5120, 1280]> var_747_to_fp16 = const()[name = tensor<string, []>("op_747_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(224198272)))]; |
| tensor<fp16, [5120]> var_748_to_fp16 = const()[name = tensor<string, []>("op_748_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(237305536)))]; |
| tensor<fp16, [1, 1500, 5120]> input_49_cast = linear(bias = var_748_to_fp16, weight = var_747_to_fp16, x = var_738_cast); |
| tensor<string, []> x_77_mode_0 = const()[name = tensor<string, []>("x_77_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_77_cast = gelu(mode = x_77_mode_0, x = input_49_cast); |
| tensor<fp16, [1280, 5120]> var_753_to_fp16 = const()[name = tensor<string, []>("op_753_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(237315840)))]; |
| tensor<fp16, [1280]> var_754_to_fp16 = const()[name = tensor<string, []>("op_754_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250423104)))]; |
| tensor<fp16, [1, 1500, 1280]> var_755_cast = linear(bias = var_754_to_fp16, weight = var_753_to_fp16, x = x_77_cast); |
| tensor<fp16, [1, 1500, 1280]> x_79_cast = add(x = x_73_cast, y = var_755_cast); |
| tensor<int32, []> var_764 = const()[name = tensor<string, []>("op_764"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_781_axes_0 = const()[name = tensor<string, []>("op_781_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_6_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_6_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250425728)))]; |
| tensor<fp16, [1280]> blocks_6_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_6_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250428352)))]; |
| tensor<fp16, []> var_770_to_fp16 = const()[name = tensor<string, []>("op_770_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_781_cast = layer_norm(axes = var_781_axes_0, beta = blocks_6_attn_ln_bias_to_fp16, epsilon = var_770_to_fp16, gamma = blocks_6_attn_ln_weight_to_fp16, x = x_79_cast); |
| tensor<fp16, [1280, 1280]> var_792_to_fp16 = const()[name = tensor<string, []>("op_792_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(250430976)))]; |
| tensor<fp16, [1280]> var_793_to_fp16 = const()[name = tensor<string, []>("op_793_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(253707840)))]; |
| tensor<fp16, [1, 1500, 1280]> q_25_cast = linear(bias = var_793_to_fp16, weight = var_792_to_fp16, x = var_781_cast); |
| tensor<fp16, [1280, 1280]> var_796_to_fp16 = const()[name = tensor<string, []>("op_796_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(253710464)))]; |
| tensor<fp16, [1280]> k_25_bias_0_to_fp16 = const()[name = tensor<string, []>("k_25_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(256987328)))]; |
| tensor<fp16, [1, 1500, 1280]> k_25_cast = linear(bias = k_25_bias_0_to_fp16, weight = var_796_to_fp16, x = var_781_cast); |
| tensor<fp16, [1280, 1280]> var_800_to_fp16 = const()[name = tensor<string, []>("op_800_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(256989952)))]; |
| tensor<fp16, [1280]> var_801_to_fp16 = const()[name = tensor<string, []>("op_801_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(260266816)))]; |
| tensor<fp16, [1, 1500, 1280]> v_25_cast = linear(bias = var_801_to_fp16, weight = var_800_to_fp16, x = var_781_cast); |
| tensor<int32, [4]> var_809 = const()[name = tensor<string, []>("op_809"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_810_cast = reshape(shape = var_809, x = q_25_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_236_to_fp16 = const()[name = tensor<string, []>("const_236_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_27_cast = mul(x = var_810_cast, y = const_236_to_fp16); |
| tensor<int32, [4]> var_816 = const()[name = tensor<string, []>("op_816"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_817_cast = reshape(shape = var_816, x = k_25_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_237_to_fp16 = const()[name = tensor<string, []>("const_237_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_27_cast = mul(x = var_817_cast, y = const_237_to_fp16); |
| tensor<int32, [4]> var_823 = const()[name = tensor<string, []>("op_823"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_824_cast = reshape(shape = var_823, x = v_25_cast); |
| tensor<int32, [4]> var_825 = const()[name = tensor<string, []>("op_825"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_13_transpose_x_0 = const()[name = tensor<string, []>("qk_13_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_13_transpose_y_0 = const()[name = tensor<string, []>("qk_13_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_76_perm_0 = const()[name = tensor<string, []>("transpose_76_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_77_perm_0 = const()[name = tensor<string, []>("transpose_77_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_229 = transpose(perm = transpose_77_perm_0, x = k_27_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_230 = transpose(perm = transpose_76_perm_0, x = q_27_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_13_cast = matmul(transpose_x = qk_13_transpose_x_0, transpose_y = qk_13_transpose_y_0, x = transpose_230, y = transpose_229); |
| tensor<fp16, [1, 20, 1500, 1500]> var_829_cast = softmax(axis = var_764, x = qk_13_cast); |
| tensor<bool, []> var_831_transpose_x_0 = const()[name = tensor<string, []>("op_831_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_831_transpose_y_0 = const()[name = tensor<string, []>("op_831_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_231 = transpose(perm = var_825, x = var_824_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_831_cast = matmul(transpose_x = var_831_transpose_x_0, transpose_y = var_831_transpose_y_0, x = var_829_cast, y = transpose_231); |
| tensor<int32, [4]> var_832 = const()[name = tensor<string, []>("op_832"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_6 = const()[name = tensor<string, []>("concat_6"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_228 = transpose(perm = var_832, x = var_831_cast); |
| tensor<fp16, [1, 1500, 1280]> x_83_cast = reshape(shape = concat_6, x = transpose_228); |
| tensor<fp16, [1280, 1280]> var_837_to_fp16 = const()[name = tensor<string, []>("op_837_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(260269440)))]; |
| tensor<fp16, [1280]> var_838_to_fp16 = const()[name = tensor<string, []>("op_838_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(263546304)))]; |
| tensor<fp16, [1, 1500, 1280]> var_839_cast = linear(bias = var_838_to_fp16, weight = var_837_to_fp16, x = x_83_cast); |
| tensor<fp16, [1, 1500, 1280]> x_85_cast = add(x = x_79_cast, y = var_839_cast); |
| tensor<int32, [1]> var_845_axes_0 = const()[name = tensor<string, []>("op_845_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_6_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_6_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(263548928)))]; |
| tensor<fp16, [1280]> blocks_6_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_6_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(263551552)))]; |
| tensor<fp16, [1, 1500, 1280]> var_845_cast = layer_norm(axes = var_845_axes_0, beta = blocks_6_mlp_ln_bias_to_fp16, epsilon = var_770_to_fp16, gamma = blocks_6_mlp_ln_weight_to_fp16, x = x_85_cast); |
| tensor<fp16, [5120, 1280]> var_854_to_fp16 = const()[name = tensor<string, []>("op_854_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(263554176)))]; |
| tensor<fp16, [5120]> var_855_to_fp16 = const()[name = tensor<string, []>("op_855_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(276661440)))]; |
| tensor<fp16, [1, 1500, 5120]> input_57_cast = linear(bias = var_855_to_fp16, weight = var_854_to_fp16, x = var_845_cast); |
| tensor<string, []> x_89_mode_0 = const()[name = tensor<string, []>("x_89_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_89_cast = gelu(mode = x_89_mode_0, x = input_57_cast); |
| tensor<fp16, [1280, 5120]> var_860_to_fp16 = const()[name = tensor<string, []>("op_860_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(276671744)))]; |
| tensor<fp16, [1280]> var_861_to_fp16 = const()[name = tensor<string, []>("op_861_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(289779008)))]; |
| tensor<fp16, [1, 1500, 1280]> var_862_cast = linear(bias = var_861_to_fp16, weight = var_860_to_fp16, x = x_89_cast); |
| tensor<fp16, [1, 1500, 1280]> x_91_cast = add(x = x_85_cast, y = var_862_cast); |
| tensor<int32, []> var_871 = const()[name = tensor<string, []>("op_871"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_888_axes_0 = const()[name = tensor<string, []>("op_888_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_7_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_7_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(289781632)))]; |
| tensor<fp16, [1280]> blocks_7_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_7_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(289784256)))]; |
| tensor<fp16, []> var_877_to_fp16 = const()[name = tensor<string, []>("op_877_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_888_cast = layer_norm(axes = var_888_axes_0, beta = blocks_7_attn_ln_bias_to_fp16, epsilon = var_877_to_fp16, gamma = blocks_7_attn_ln_weight_to_fp16, x = x_91_cast); |
| tensor<fp16, [1280, 1280]> var_899_to_fp16 = const()[name = tensor<string, []>("op_899_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(289786880)))]; |
| tensor<fp16, [1280]> var_900_to_fp16 = const()[name = tensor<string, []>("op_900_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(293063744)))]; |
| tensor<fp16, [1, 1500, 1280]> q_29_cast = linear(bias = var_900_to_fp16, weight = var_899_to_fp16, x = var_888_cast); |
| tensor<fp16, [1280, 1280]> var_903_to_fp16 = const()[name = tensor<string, []>("op_903_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(293066368)))]; |
| tensor<fp16, [1280]> k_29_bias_0_to_fp16 = const()[name = tensor<string, []>("k_29_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(296343232)))]; |
| tensor<fp16, [1, 1500, 1280]> k_29_cast = linear(bias = k_29_bias_0_to_fp16, weight = var_903_to_fp16, x = var_888_cast); |
| tensor<fp16, [1280, 1280]> var_907_to_fp16 = const()[name = tensor<string, []>("op_907_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(296345856)))]; |
| tensor<fp16, [1280]> var_908_to_fp16 = const()[name = tensor<string, []>("op_908_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(299622720)))]; |
| tensor<fp16, [1, 1500, 1280]> v_29_cast = linear(bias = var_908_to_fp16, weight = var_907_to_fp16, x = var_888_cast); |
| tensor<int32, [4]> var_916 = const()[name = tensor<string, []>("op_916"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_917_cast = reshape(shape = var_916, x = q_29_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_238_to_fp16 = const()[name = tensor<string, []>("const_238_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_31_cast = mul(x = var_917_cast, y = const_238_to_fp16); |
| tensor<int32, [4]> var_923 = const()[name = tensor<string, []>("op_923"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_924_cast = reshape(shape = var_923, x = k_29_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_239_to_fp16 = const()[name = tensor<string, []>("const_239_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_31_cast = mul(x = var_924_cast, y = const_239_to_fp16); |
| tensor<int32, [4]> var_930 = const()[name = tensor<string, []>("op_930"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_931_cast = reshape(shape = var_930, x = v_29_cast); |
| tensor<int32, [4]> var_932 = const()[name = tensor<string, []>("op_932"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_15_transpose_x_0 = const()[name = tensor<string, []>("qk_15_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_15_transpose_y_0 = const()[name = tensor<string, []>("qk_15_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_78_perm_0 = const()[name = tensor<string, []>("transpose_78_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_79_perm_0 = const()[name = tensor<string, []>("transpose_79_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_225 = transpose(perm = transpose_79_perm_0, x = k_31_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_226 = transpose(perm = transpose_78_perm_0, x = q_31_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_15_cast = matmul(transpose_x = qk_15_transpose_x_0, transpose_y = qk_15_transpose_y_0, x = transpose_226, y = transpose_225); |
| tensor<fp16, [1, 20, 1500, 1500]> var_936_cast = softmax(axis = var_871, x = qk_15_cast); |
| tensor<bool, []> var_938_transpose_x_0 = const()[name = tensor<string, []>("op_938_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_938_transpose_y_0 = const()[name = tensor<string, []>("op_938_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_227 = transpose(perm = var_932, x = var_931_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_938_cast = matmul(transpose_x = var_938_transpose_x_0, transpose_y = var_938_transpose_y_0, x = var_936_cast, y = transpose_227); |
| tensor<int32, [4]> var_939 = const()[name = tensor<string, []>("op_939"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_7 = const()[name = tensor<string, []>("concat_7"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_224 = transpose(perm = var_939, x = var_938_cast); |
| tensor<fp16, [1, 1500, 1280]> x_95_cast = reshape(shape = concat_7, x = transpose_224); |
| tensor<fp16, [1280, 1280]> var_944_to_fp16 = const()[name = tensor<string, []>("op_944_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(299625344)))]; |
| tensor<fp16, [1280]> var_945_to_fp16 = const()[name = tensor<string, []>("op_945_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(302902208)))]; |
| tensor<fp16, [1, 1500, 1280]> var_946_cast = linear(bias = var_945_to_fp16, weight = var_944_to_fp16, x = x_95_cast); |
| tensor<fp16, [1, 1500, 1280]> x_97_cast = add(x = x_91_cast, y = var_946_cast); |
| tensor<int32, [1]> var_952_axes_0 = const()[name = tensor<string, []>("op_952_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_7_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_7_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(302904832)))]; |
| tensor<fp16, [1280]> blocks_7_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_7_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(302907456)))]; |
| tensor<fp16, [1, 1500, 1280]> var_952_cast = layer_norm(axes = var_952_axes_0, beta = blocks_7_mlp_ln_bias_to_fp16, epsilon = var_877_to_fp16, gamma = blocks_7_mlp_ln_weight_to_fp16, x = x_97_cast); |
| tensor<fp16, [5120, 1280]> var_961_to_fp16 = const()[name = tensor<string, []>("op_961_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(302910080)))]; |
| tensor<fp16, [5120]> var_962_to_fp16 = const()[name = tensor<string, []>("op_962_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(316017344)))]; |
| tensor<fp16, [1, 1500, 5120]> input_65_cast = linear(bias = var_962_to_fp16, weight = var_961_to_fp16, x = var_952_cast); |
| tensor<string, []> x_101_mode_0 = const()[name = tensor<string, []>("x_101_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_101_cast = gelu(mode = x_101_mode_0, x = input_65_cast); |
| tensor<fp16, [1280, 5120]> var_967_to_fp16 = const()[name = tensor<string, []>("op_967_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(316027648)))]; |
| tensor<fp16, [1280]> var_968_to_fp16 = const()[name = tensor<string, []>("op_968_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(329134912)))]; |
| tensor<fp16, [1, 1500, 1280]> var_969_cast = linear(bias = var_968_to_fp16, weight = var_967_to_fp16, x = x_101_cast); |
| tensor<fp16, [1, 1500, 1280]> x_103_cast = add(x = x_97_cast, y = var_969_cast); |
| tensor<int32, []> var_978 = const()[name = tensor<string, []>("op_978"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_995_axes_0 = const()[name = tensor<string, []>("op_995_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_8_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_8_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(329137536)))]; |
| tensor<fp16, [1280]> blocks_8_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_8_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(329140160)))]; |
| tensor<fp16, []> var_984_to_fp16 = const()[name = tensor<string, []>("op_984_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_995_cast = layer_norm(axes = var_995_axes_0, beta = blocks_8_attn_ln_bias_to_fp16, epsilon = var_984_to_fp16, gamma = blocks_8_attn_ln_weight_to_fp16, x = x_103_cast); |
| tensor<fp16, [1280, 1280]> var_1006_to_fp16 = const()[name = tensor<string, []>("op_1006_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(329142784)))]; |
| tensor<fp16, [1280]> var_1007_to_fp16 = const()[name = tensor<string, []>("op_1007_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(332419648)))]; |
| tensor<fp16, [1, 1500, 1280]> q_33_cast = linear(bias = var_1007_to_fp16, weight = var_1006_to_fp16, x = var_995_cast); |
| tensor<fp16, [1280, 1280]> var_1010_to_fp16 = const()[name = tensor<string, []>("op_1010_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(332422272)))]; |
| tensor<fp16, [1280]> k_33_bias_0_to_fp16 = const()[name = tensor<string, []>("k_33_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335699136)))]; |
| tensor<fp16, [1, 1500, 1280]> k_33_cast = linear(bias = k_33_bias_0_to_fp16, weight = var_1010_to_fp16, x = var_995_cast); |
| tensor<fp16, [1280, 1280]> var_1014_to_fp16 = const()[name = tensor<string, []>("op_1014_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(335701760)))]; |
| tensor<fp16, [1280]> var_1015_to_fp16 = const()[name = tensor<string, []>("op_1015_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(338978624)))]; |
| tensor<fp16, [1, 1500, 1280]> v_33_cast = linear(bias = var_1015_to_fp16, weight = var_1014_to_fp16, x = var_995_cast); |
| tensor<int32, [4]> var_1023 = const()[name = tensor<string, []>("op_1023"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1024_cast = reshape(shape = var_1023, x = q_33_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_240_to_fp16 = const()[name = tensor<string, []>("const_240_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_35_cast = mul(x = var_1024_cast, y = const_240_to_fp16); |
| tensor<int32, [4]> var_1030 = const()[name = tensor<string, []>("op_1030"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1031_cast = reshape(shape = var_1030, x = k_33_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_241_to_fp16 = const()[name = tensor<string, []>("const_241_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_35_cast = mul(x = var_1031_cast, y = const_241_to_fp16); |
| tensor<int32, [4]> var_1037 = const()[name = tensor<string, []>("op_1037"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1038_cast = reshape(shape = var_1037, x = v_33_cast); |
| tensor<int32, [4]> var_1039 = const()[name = tensor<string, []>("op_1039"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_17_transpose_x_0 = const()[name = tensor<string, []>("qk_17_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_17_transpose_y_0 = const()[name = tensor<string, []>("qk_17_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_80_perm_0 = const()[name = tensor<string, []>("transpose_80_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_81_perm_0 = const()[name = tensor<string, []>("transpose_81_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_221 = transpose(perm = transpose_81_perm_0, x = k_35_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_222 = transpose(perm = transpose_80_perm_0, x = q_35_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_17_cast = matmul(transpose_x = qk_17_transpose_x_0, transpose_y = qk_17_transpose_y_0, x = transpose_222, y = transpose_221); |
| tensor<fp16, [1, 20, 1500, 1500]> var_1043_cast = softmax(axis = var_978, x = qk_17_cast); |
| tensor<bool, []> var_1045_transpose_x_0 = const()[name = tensor<string, []>("op_1045_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_1045_transpose_y_0 = const()[name = tensor<string, []>("op_1045_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_223 = transpose(perm = var_1039, x = var_1038_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_1045_cast = matmul(transpose_x = var_1045_transpose_x_0, transpose_y = var_1045_transpose_y_0, x = var_1043_cast, y = transpose_223); |
| tensor<int32, [4]> var_1046 = const()[name = tensor<string, []>("op_1046"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_8 = const()[name = tensor<string, []>("concat_8"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_220 = transpose(perm = var_1046, x = var_1045_cast); |
| tensor<fp16, [1, 1500, 1280]> x_107_cast = reshape(shape = concat_8, x = transpose_220); |
| tensor<fp16, [1280, 1280]> var_1051_to_fp16 = const()[name = tensor<string, []>("op_1051_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(338981248)))]; |
| tensor<fp16, [1280]> var_1052_to_fp16 = const()[name = tensor<string, []>("op_1052_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(342258112)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1053_cast = linear(bias = var_1052_to_fp16, weight = var_1051_to_fp16, x = x_107_cast); |
| tensor<fp16, [1, 1500, 1280]> x_109_cast = add(x = x_103_cast, y = var_1053_cast); |
| tensor<int32, [1]> var_1059_axes_0 = const()[name = tensor<string, []>("op_1059_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_8_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_8_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(342260736)))]; |
| tensor<fp16, [1280]> blocks_8_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_8_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(342263360)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1059_cast = layer_norm(axes = var_1059_axes_0, beta = blocks_8_mlp_ln_bias_to_fp16, epsilon = var_984_to_fp16, gamma = blocks_8_mlp_ln_weight_to_fp16, x = x_109_cast); |
| tensor<fp16, [5120, 1280]> var_1068_to_fp16 = const()[name = tensor<string, []>("op_1068_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(342265984)))]; |
| tensor<fp16, [5120]> var_1069_to_fp16 = const()[name = tensor<string, []>("op_1069_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(355373248)))]; |
| tensor<fp16, [1, 1500, 5120]> input_73_cast = linear(bias = var_1069_to_fp16, weight = var_1068_to_fp16, x = var_1059_cast); |
| tensor<string, []> x_113_mode_0 = const()[name = tensor<string, []>("x_113_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_113_cast = gelu(mode = x_113_mode_0, x = input_73_cast); |
| tensor<fp16, [1280, 5120]> var_1074_to_fp16 = const()[name = tensor<string, []>("op_1074_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(355383552)))]; |
| tensor<fp16, [1280]> var_1075_to_fp16 = const()[name = tensor<string, []>("op_1075_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(368490816)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1076_cast = linear(bias = var_1075_to_fp16, weight = var_1074_to_fp16, x = x_113_cast); |
| tensor<fp16, [1, 1500, 1280]> x_115_cast = add(x = x_109_cast, y = var_1076_cast); |
| tensor<int32, []> var_1085 = const()[name = tensor<string, []>("op_1085"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_1102_axes_0 = const()[name = tensor<string, []>("op_1102_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_9_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_9_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(368493440)))]; |
| tensor<fp16, [1280]> blocks_9_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_9_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(368496064)))]; |
| tensor<fp16, []> var_1091_to_fp16 = const()[name = tensor<string, []>("op_1091_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_1102_cast = layer_norm(axes = var_1102_axes_0, beta = blocks_9_attn_ln_bias_to_fp16, epsilon = var_1091_to_fp16, gamma = blocks_9_attn_ln_weight_to_fp16, x = x_115_cast); |
| tensor<fp16, [1280, 1280]> var_1113_to_fp16 = const()[name = tensor<string, []>("op_1113_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(368498688)))]; |
| tensor<fp16, [1280]> var_1114_to_fp16 = const()[name = tensor<string, []>("op_1114_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(371775552)))]; |
| tensor<fp16, [1, 1500, 1280]> q_37_cast = linear(bias = var_1114_to_fp16, weight = var_1113_to_fp16, x = var_1102_cast); |
| tensor<fp16, [1280, 1280]> var_1117_to_fp16 = const()[name = tensor<string, []>("op_1117_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(371778176)))]; |
| tensor<fp16, [1280]> k_37_bias_0_to_fp16 = const()[name = tensor<string, []>("k_37_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(375055040)))]; |
| tensor<fp16, [1, 1500, 1280]> k_37_cast = linear(bias = k_37_bias_0_to_fp16, weight = var_1117_to_fp16, x = var_1102_cast); |
| tensor<fp16, [1280, 1280]> var_1121_to_fp16 = const()[name = tensor<string, []>("op_1121_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(375057664)))]; |
| tensor<fp16, [1280]> var_1122_to_fp16 = const()[name = tensor<string, []>("op_1122_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(378334528)))]; |
| tensor<fp16, [1, 1500, 1280]> v_37_cast = linear(bias = var_1122_to_fp16, weight = var_1121_to_fp16, x = var_1102_cast); |
| tensor<int32, [4]> var_1130 = const()[name = tensor<string, []>("op_1130"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1131_cast = reshape(shape = var_1130, x = q_37_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_242_to_fp16 = const()[name = tensor<string, []>("const_242_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_39_cast = mul(x = var_1131_cast, y = const_242_to_fp16); |
| tensor<int32, [4]> var_1137 = const()[name = tensor<string, []>("op_1137"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1138_cast = reshape(shape = var_1137, x = k_37_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_243_to_fp16 = const()[name = tensor<string, []>("const_243_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_39_cast = mul(x = var_1138_cast, y = const_243_to_fp16); |
| tensor<int32, [4]> var_1144 = const()[name = tensor<string, []>("op_1144"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1145_cast = reshape(shape = var_1144, x = v_37_cast); |
| tensor<int32, [4]> var_1146 = const()[name = tensor<string, []>("op_1146"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_19_transpose_x_0 = const()[name = tensor<string, []>("qk_19_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_19_transpose_y_0 = const()[name = tensor<string, []>("qk_19_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_82_perm_0 = const()[name = tensor<string, []>("transpose_82_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_83_perm_0 = const()[name = tensor<string, []>("transpose_83_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_217 = transpose(perm = transpose_83_perm_0, x = k_39_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_218 = transpose(perm = transpose_82_perm_0, x = q_39_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_19_cast = matmul(transpose_x = qk_19_transpose_x_0, transpose_y = qk_19_transpose_y_0, x = transpose_218, y = transpose_217); |
| tensor<fp16, [1, 20, 1500, 1500]> var_1150_cast = softmax(axis = var_1085, x = qk_19_cast); |
| tensor<bool, []> var_1152_transpose_x_0 = const()[name = tensor<string, []>("op_1152_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_1152_transpose_y_0 = const()[name = tensor<string, []>("op_1152_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_219 = transpose(perm = var_1146, x = var_1145_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_1152_cast = matmul(transpose_x = var_1152_transpose_x_0, transpose_y = var_1152_transpose_y_0, x = var_1150_cast, y = transpose_219); |
| tensor<int32, [4]> var_1153 = const()[name = tensor<string, []>("op_1153"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_9 = const()[name = tensor<string, []>("concat_9"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_216 = transpose(perm = var_1153, x = var_1152_cast); |
| tensor<fp16, [1, 1500, 1280]> x_119_cast = reshape(shape = concat_9, x = transpose_216); |
| tensor<fp16, [1280, 1280]> var_1158_to_fp16 = const()[name = tensor<string, []>("op_1158_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(378337152)))]; |
| tensor<fp16, [1280]> var_1159_to_fp16 = const()[name = tensor<string, []>("op_1159_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(381614016)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1160_cast = linear(bias = var_1159_to_fp16, weight = var_1158_to_fp16, x = x_119_cast); |
| tensor<fp16, [1, 1500, 1280]> x_121_cast = add(x = x_115_cast, y = var_1160_cast); |
| tensor<int32, [1]> var_1166_axes_0 = const()[name = tensor<string, []>("op_1166_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_9_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_9_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(381616640)))]; |
| tensor<fp16, [1280]> blocks_9_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_9_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(381619264)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1166_cast = layer_norm(axes = var_1166_axes_0, beta = blocks_9_mlp_ln_bias_to_fp16, epsilon = var_1091_to_fp16, gamma = blocks_9_mlp_ln_weight_to_fp16, x = x_121_cast); |
| tensor<fp16, [5120, 1280]> var_1175_to_fp16 = const()[name = tensor<string, []>("op_1175_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(381621888)))]; |
| tensor<fp16, [5120]> var_1176_to_fp16 = const()[name = tensor<string, []>("op_1176_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(394729152)))]; |
| tensor<fp16, [1, 1500, 5120]> input_81_cast = linear(bias = var_1176_to_fp16, weight = var_1175_to_fp16, x = var_1166_cast); |
| tensor<string, []> x_125_mode_0 = const()[name = tensor<string, []>("x_125_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_125_cast = gelu(mode = x_125_mode_0, x = input_81_cast); |
| tensor<fp16, [1280, 5120]> var_1181_to_fp16 = const()[name = tensor<string, []>("op_1181_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(394739456)))]; |
| tensor<fp16, [1280]> var_1182_to_fp16 = const()[name = tensor<string, []>("op_1182_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(407846720)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1183_cast = linear(bias = var_1182_to_fp16, weight = var_1181_to_fp16, x = x_125_cast); |
| tensor<fp16, [1, 1500, 1280]> x_127_cast = add(x = x_121_cast, y = var_1183_cast); |
| tensor<int32, []> var_1192 = const()[name = tensor<string, []>("op_1192"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_1209_axes_0 = const()[name = tensor<string, []>("op_1209_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_10_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_10_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(407849344)))]; |
| tensor<fp16, [1280]> blocks_10_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_10_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(407851968)))]; |
| tensor<fp16, []> var_1198_to_fp16 = const()[name = tensor<string, []>("op_1198_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_1209_cast = layer_norm(axes = var_1209_axes_0, beta = blocks_10_attn_ln_bias_to_fp16, epsilon = var_1198_to_fp16, gamma = blocks_10_attn_ln_weight_to_fp16, x = x_127_cast); |
| tensor<fp16, [1280, 1280]> var_1220_to_fp16 = const()[name = tensor<string, []>("op_1220_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(407854592)))]; |
| tensor<fp16, [1280]> var_1221_to_fp16 = const()[name = tensor<string, []>("op_1221_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(411131456)))]; |
| tensor<fp16, [1, 1500, 1280]> q_41_cast = linear(bias = var_1221_to_fp16, weight = var_1220_to_fp16, x = var_1209_cast); |
| tensor<fp16, [1280, 1280]> var_1224_to_fp16 = const()[name = tensor<string, []>("op_1224_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(411134080)))]; |
| tensor<fp16, [1280]> k_41_bias_0_to_fp16 = const()[name = tensor<string, []>("k_41_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(414410944)))]; |
| tensor<fp16, [1, 1500, 1280]> k_41_cast = linear(bias = k_41_bias_0_to_fp16, weight = var_1224_to_fp16, x = var_1209_cast); |
| tensor<fp16, [1280, 1280]> var_1228_to_fp16 = const()[name = tensor<string, []>("op_1228_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(414413568)))]; |
| tensor<fp16, [1280]> var_1229_to_fp16 = const()[name = tensor<string, []>("op_1229_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(417690432)))]; |
| tensor<fp16, [1, 1500, 1280]> v_41_cast = linear(bias = var_1229_to_fp16, weight = var_1228_to_fp16, x = var_1209_cast); |
| tensor<int32, [4]> var_1237 = const()[name = tensor<string, []>("op_1237"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1238_cast = reshape(shape = var_1237, x = q_41_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_244_to_fp16 = const()[name = tensor<string, []>("const_244_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_43_cast = mul(x = var_1238_cast, y = const_244_to_fp16); |
| tensor<int32, [4]> var_1244 = const()[name = tensor<string, []>("op_1244"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1245_cast = reshape(shape = var_1244, x = k_41_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_245_to_fp16 = const()[name = tensor<string, []>("const_245_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_43_cast = mul(x = var_1245_cast, y = const_245_to_fp16); |
| tensor<int32, [4]> var_1251 = const()[name = tensor<string, []>("op_1251"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1252_cast = reshape(shape = var_1251, x = v_41_cast); |
| tensor<int32, [4]> var_1253 = const()[name = tensor<string, []>("op_1253"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_21_transpose_x_0 = const()[name = tensor<string, []>("qk_21_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_21_transpose_y_0 = const()[name = tensor<string, []>("qk_21_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_84_perm_0 = const()[name = tensor<string, []>("transpose_84_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_85_perm_0 = const()[name = tensor<string, []>("transpose_85_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_213 = transpose(perm = transpose_85_perm_0, x = k_43_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_214 = transpose(perm = transpose_84_perm_0, x = q_43_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_21_cast = matmul(transpose_x = qk_21_transpose_x_0, transpose_y = qk_21_transpose_y_0, x = transpose_214, y = transpose_213); |
| tensor<fp16, [1, 20, 1500, 1500]> var_1257_cast = softmax(axis = var_1192, x = qk_21_cast); |
| tensor<bool, []> var_1259_transpose_x_0 = const()[name = tensor<string, []>("op_1259_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_1259_transpose_y_0 = const()[name = tensor<string, []>("op_1259_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_215 = transpose(perm = var_1253, x = var_1252_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_1259_cast = matmul(transpose_x = var_1259_transpose_x_0, transpose_y = var_1259_transpose_y_0, x = var_1257_cast, y = transpose_215); |
| tensor<int32, [4]> var_1260 = const()[name = tensor<string, []>("op_1260"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_10 = const()[name = tensor<string, []>("concat_10"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_212 = transpose(perm = var_1260, x = var_1259_cast); |
| tensor<fp16, [1, 1500, 1280]> x_131_cast = reshape(shape = concat_10, x = transpose_212); |
| tensor<fp16, [1280, 1280]> var_1265_to_fp16 = const()[name = tensor<string, []>("op_1265_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(417693056)))]; |
| tensor<fp16, [1280]> var_1266_to_fp16 = const()[name = tensor<string, []>("op_1266_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(420969920)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1267_cast = linear(bias = var_1266_to_fp16, weight = var_1265_to_fp16, x = x_131_cast); |
| tensor<fp16, [1, 1500, 1280]> x_133_cast = add(x = x_127_cast, y = var_1267_cast); |
| tensor<int32, [1]> var_1273_axes_0 = const()[name = tensor<string, []>("op_1273_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_10_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_10_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(420972544)))]; |
| tensor<fp16, [1280]> blocks_10_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_10_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(420975168)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1273_cast = layer_norm(axes = var_1273_axes_0, beta = blocks_10_mlp_ln_bias_to_fp16, epsilon = var_1198_to_fp16, gamma = blocks_10_mlp_ln_weight_to_fp16, x = x_133_cast); |
| tensor<fp16, [5120, 1280]> var_1282_to_fp16 = const()[name = tensor<string, []>("op_1282_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(420977792)))]; |
| tensor<fp16, [5120]> var_1283_to_fp16 = const()[name = tensor<string, []>("op_1283_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(434085056)))]; |
| tensor<fp16, [1, 1500, 5120]> input_89_cast = linear(bias = var_1283_to_fp16, weight = var_1282_to_fp16, x = var_1273_cast); |
| tensor<string, []> x_137_mode_0 = const()[name = tensor<string, []>("x_137_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_137_cast = gelu(mode = x_137_mode_0, x = input_89_cast); |
| tensor<fp16, [1280, 5120]> var_1288_to_fp16 = const()[name = tensor<string, []>("op_1288_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(434095360)))]; |
| tensor<fp16, [1280]> var_1289_to_fp16 = const()[name = tensor<string, []>("op_1289_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(447202624)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1290_cast = linear(bias = var_1289_to_fp16, weight = var_1288_to_fp16, x = x_137_cast); |
| tensor<fp16, [1, 1500, 1280]> x_139_cast = add(x = x_133_cast, y = var_1290_cast); |
| tensor<int32, []> var_1299 = const()[name = tensor<string, []>("op_1299"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_1316_axes_0 = const()[name = tensor<string, []>("op_1316_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_11_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_11_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(447205248)))]; |
| tensor<fp16, [1280]> blocks_11_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_11_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(447207872)))]; |
| tensor<fp16, []> var_1305_to_fp16 = const()[name = tensor<string, []>("op_1305_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_1316_cast = layer_norm(axes = var_1316_axes_0, beta = blocks_11_attn_ln_bias_to_fp16, epsilon = var_1305_to_fp16, gamma = blocks_11_attn_ln_weight_to_fp16, x = x_139_cast); |
| tensor<fp16, [1280, 1280]> var_1327_to_fp16 = const()[name = tensor<string, []>("op_1327_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(447210496)))]; |
| tensor<fp16, [1280]> var_1328_to_fp16 = const()[name = tensor<string, []>("op_1328_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(450487360)))]; |
| tensor<fp16, [1, 1500, 1280]> q_45_cast = linear(bias = var_1328_to_fp16, weight = var_1327_to_fp16, x = var_1316_cast); |
| tensor<fp16, [1280, 1280]> var_1331_to_fp16 = const()[name = tensor<string, []>("op_1331_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(450489984)))]; |
| tensor<fp16, [1280]> k_45_bias_0_to_fp16 = const()[name = tensor<string, []>("k_45_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(453766848)))]; |
| tensor<fp16, [1, 1500, 1280]> k_45_cast = linear(bias = k_45_bias_0_to_fp16, weight = var_1331_to_fp16, x = var_1316_cast); |
| tensor<fp16, [1280, 1280]> var_1335_to_fp16 = const()[name = tensor<string, []>("op_1335_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(453769472)))]; |
| tensor<fp16, [1280]> var_1336_to_fp16 = const()[name = tensor<string, []>("op_1336_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(457046336)))]; |
| tensor<fp16, [1, 1500, 1280]> v_45_cast = linear(bias = var_1336_to_fp16, weight = var_1335_to_fp16, x = var_1316_cast); |
| tensor<int32, [4]> var_1344 = const()[name = tensor<string, []>("op_1344"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1345_cast = reshape(shape = var_1344, x = q_45_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_246_to_fp16 = const()[name = tensor<string, []>("const_246_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_47_cast = mul(x = var_1345_cast, y = const_246_to_fp16); |
| tensor<int32, [4]> var_1351 = const()[name = tensor<string, []>("op_1351"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1352_cast = reshape(shape = var_1351, x = k_45_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_247_to_fp16 = const()[name = tensor<string, []>("const_247_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_47_cast = mul(x = var_1352_cast, y = const_247_to_fp16); |
| tensor<int32, [4]> var_1358 = const()[name = tensor<string, []>("op_1358"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1359_cast = reshape(shape = var_1358, x = v_45_cast); |
| tensor<int32, [4]> var_1360 = const()[name = tensor<string, []>("op_1360"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_23_transpose_x_0 = const()[name = tensor<string, []>("qk_23_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_23_transpose_y_0 = const()[name = tensor<string, []>("qk_23_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_86_perm_0 = const()[name = tensor<string, []>("transpose_86_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_87_perm_0 = const()[name = tensor<string, []>("transpose_87_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_209 = transpose(perm = transpose_87_perm_0, x = k_47_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_210 = transpose(perm = transpose_86_perm_0, x = q_47_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_23_cast = matmul(transpose_x = qk_23_transpose_x_0, transpose_y = qk_23_transpose_y_0, x = transpose_210, y = transpose_209); |
| tensor<fp16, [1, 20, 1500, 1500]> var_1364_cast = softmax(axis = var_1299, x = qk_23_cast); |
| tensor<bool, []> var_1366_transpose_x_0 = const()[name = tensor<string, []>("op_1366_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_1366_transpose_y_0 = const()[name = tensor<string, []>("op_1366_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_211 = transpose(perm = var_1360, x = var_1359_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_1366_cast = matmul(transpose_x = var_1366_transpose_x_0, transpose_y = var_1366_transpose_y_0, x = var_1364_cast, y = transpose_211); |
| tensor<int32, [4]> var_1367 = const()[name = tensor<string, []>("op_1367"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_11 = const()[name = tensor<string, []>("concat_11"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_208 = transpose(perm = var_1367, x = var_1366_cast); |
| tensor<fp16, [1, 1500, 1280]> x_143_cast = reshape(shape = concat_11, x = transpose_208); |
| tensor<fp16, [1280, 1280]> var_1372_to_fp16 = const()[name = tensor<string, []>("op_1372_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(457048960)))]; |
| tensor<fp16, [1280]> var_1373_to_fp16 = const()[name = tensor<string, []>("op_1373_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(460325824)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1374_cast = linear(bias = var_1373_to_fp16, weight = var_1372_to_fp16, x = x_143_cast); |
| tensor<fp16, [1, 1500, 1280]> x_145_cast = add(x = x_139_cast, y = var_1374_cast); |
| tensor<int32, [1]> var_1380_axes_0 = const()[name = tensor<string, []>("op_1380_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_11_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_11_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(460328448)))]; |
| tensor<fp16, [1280]> blocks_11_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_11_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(460331072)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1380_cast = layer_norm(axes = var_1380_axes_0, beta = blocks_11_mlp_ln_bias_to_fp16, epsilon = var_1305_to_fp16, gamma = blocks_11_mlp_ln_weight_to_fp16, x = x_145_cast); |
| tensor<fp16, [5120, 1280]> var_1389_to_fp16 = const()[name = tensor<string, []>("op_1389_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(460333696)))]; |
| tensor<fp16, [5120]> var_1390_to_fp16 = const()[name = tensor<string, []>("op_1390_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(473440960)))]; |
| tensor<fp16, [1, 1500, 5120]> input_97_cast = linear(bias = var_1390_to_fp16, weight = var_1389_to_fp16, x = var_1380_cast); |
| tensor<string, []> x_149_mode_0 = const()[name = tensor<string, []>("x_149_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_149_cast = gelu(mode = x_149_mode_0, x = input_97_cast); |
| tensor<fp16, [1280, 5120]> var_1395_to_fp16 = const()[name = tensor<string, []>("op_1395_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(473451264)))]; |
| tensor<fp16, [1280]> var_1396_to_fp16 = const()[name = tensor<string, []>("op_1396_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(486558528)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1397_cast = linear(bias = var_1396_to_fp16, weight = var_1395_to_fp16, x = x_149_cast); |
| tensor<fp16, [1, 1500, 1280]> x_151_cast = add(x = x_145_cast, y = var_1397_cast); |
| tensor<int32, []> var_1406 = const()[name = tensor<string, []>("op_1406"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_1423_axes_0 = const()[name = tensor<string, []>("op_1423_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_12_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_12_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(486561152)))]; |
| tensor<fp16, [1280]> blocks_12_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_12_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(486563776)))]; |
| tensor<fp16, []> var_1412_to_fp16 = const()[name = tensor<string, []>("op_1412_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_1423_cast = layer_norm(axes = var_1423_axes_0, beta = blocks_12_attn_ln_bias_to_fp16, epsilon = var_1412_to_fp16, gamma = blocks_12_attn_ln_weight_to_fp16, x = x_151_cast); |
| tensor<fp16, [1280, 1280]> var_1434_to_fp16 = const()[name = tensor<string, []>("op_1434_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(486566400)))]; |
| tensor<fp16, [1280]> var_1435_to_fp16 = const()[name = tensor<string, []>("op_1435_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(489843264)))]; |
| tensor<fp16, [1, 1500, 1280]> q_49_cast = linear(bias = var_1435_to_fp16, weight = var_1434_to_fp16, x = var_1423_cast); |
| tensor<fp16, [1280, 1280]> var_1438_to_fp16 = const()[name = tensor<string, []>("op_1438_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(489845888)))]; |
| tensor<fp16, [1280]> k_49_bias_0_to_fp16 = const()[name = tensor<string, []>("k_49_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(493122752)))]; |
| tensor<fp16, [1, 1500, 1280]> k_49_cast = linear(bias = k_49_bias_0_to_fp16, weight = var_1438_to_fp16, x = var_1423_cast); |
| tensor<fp16, [1280, 1280]> var_1442_to_fp16 = const()[name = tensor<string, []>("op_1442_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(493125376)))]; |
| tensor<fp16, [1280]> var_1443_to_fp16 = const()[name = tensor<string, []>("op_1443_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496402240)))]; |
| tensor<fp16, [1, 1500, 1280]> v_49_cast = linear(bias = var_1443_to_fp16, weight = var_1442_to_fp16, x = var_1423_cast); |
| tensor<int32, [4]> var_1451 = const()[name = tensor<string, []>("op_1451"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1452_cast = reshape(shape = var_1451, x = q_49_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_248_to_fp16 = const()[name = tensor<string, []>("const_248_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_51_cast = mul(x = var_1452_cast, y = const_248_to_fp16); |
| tensor<int32, [4]> var_1458 = const()[name = tensor<string, []>("op_1458"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1459_cast = reshape(shape = var_1458, x = k_49_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_249_to_fp16 = const()[name = tensor<string, []>("const_249_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_51_cast = mul(x = var_1459_cast, y = const_249_to_fp16); |
| tensor<int32, [4]> var_1465 = const()[name = tensor<string, []>("op_1465"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1466_cast = reshape(shape = var_1465, x = v_49_cast); |
| tensor<int32, [4]> var_1467 = const()[name = tensor<string, []>("op_1467"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_25_transpose_x_0 = const()[name = tensor<string, []>("qk_25_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_25_transpose_y_0 = const()[name = tensor<string, []>("qk_25_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_88_perm_0 = const()[name = tensor<string, []>("transpose_88_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_89_perm_0 = const()[name = tensor<string, []>("transpose_89_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_205 = transpose(perm = transpose_89_perm_0, x = k_51_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_206 = transpose(perm = transpose_88_perm_0, x = q_51_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_25_cast = matmul(transpose_x = qk_25_transpose_x_0, transpose_y = qk_25_transpose_y_0, x = transpose_206, y = transpose_205); |
| tensor<fp16, [1, 20, 1500, 1500]> var_1471_cast = softmax(axis = var_1406, x = qk_25_cast); |
| tensor<bool, []> var_1473_transpose_x_0 = const()[name = tensor<string, []>("op_1473_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_1473_transpose_y_0 = const()[name = tensor<string, []>("op_1473_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_207 = transpose(perm = var_1467, x = var_1466_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_1473_cast = matmul(transpose_x = var_1473_transpose_x_0, transpose_y = var_1473_transpose_y_0, x = var_1471_cast, y = transpose_207); |
| tensor<int32, [4]> var_1474 = const()[name = tensor<string, []>("op_1474"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_12 = const()[name = tensor<string, []>("concat_12"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_204 = transpose(perm = var_1474, x = var_1473_cast); |
| tensor<fp16, [1, 1500, 1280]> x_155_cast = reshape(shape = concat_12, x = transpose_204); |
| tensor<fp16, [1280, 1280]> var_1479_to_fp16 = const()[name = tensor<string, []>("op_1479_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(496404864)))]; |
| tensor<fp16, [1280]> var_1480_to_fp16 = const()[name = tensor<string, []>("op_1480_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(499681728)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1481_cast = linear(bias = var_1480_to_fp16, weight = var_1479_to_fp16, x = x_155_cast); |
| tensor<fp16, [1, 1500, 1280]> x_157_cast = add(x = x_151_cast, y = var_1481_cast); |
| tensor<int32, [1]> var_1487_axes_0 = const()[name = tensor<string, []>("op_1487_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_12_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_12_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(499684352)))]; |
| tensor<fp16, [1280]> blocks_12_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_12_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(499686976)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1487_cast = layer_norm(axes = var_1487_axes_0, beta = blocks_12_mlp_ln_bias_to_fp16, epsilon = var_1412_to_fp16, gamma = blocks_12_mlp_ln_weight_to_fp16, x = x_157_cast); |
| tensor<fp16, [5120, 1280]> var_1496_to_fp16 = const()[name = tensor<string, []>("op_1496_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(499689600)))]; |
| tensor<fp16, [5120]> var_1497_to_fp16 = const()[name = tensor<string, []>("op_1497_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(512796864)))]; |
| tensor<fp16, [1, 1500, 5120]> input_105_cast = linear(bias = var_1497_to_fp16, weight = var_1496_to_fp16, x = var_1487_cast); |
| tensor<string, []> x_161_mode_0 = const()[name = tensor<string, []>("x_161_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_161_cast = gelu(mode = x_161_mode_0, x = input_105_cast); |
| tensor<fp16, [1280, 5120]> var_1502_to_fp16 = const()[name = tensor<string, []>("op_1502_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(512807168)))]; |
| tensor<fp16, [1280]> var_1503_to_fp16 = const()[name = tensor<string, []>("op_1503_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(525914432)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1504_cast = linear(bias = var_1503_to_fp16, weight = var_1502_to_fp16, x = x_161_cast); |
| tensor<fp16, [1, 1500, 1280]> x_163_cast = add(x = x_157_cast, y = var_1504_cast); |
| tensor<int32, []> var_1513 = const()[name = tensor<string, []>("op_1513"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_1530_axes_0 = const()[name = tensor<string, []>("op_1530_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_13_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_13_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(525917056)))]; |
| tensor<fp16, [1280]> blocks_13_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_13_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(525919680)))]; |
| tensor<fp16, []> var_1519_to_fp16 = const()[name = tensor<string, []>("op_1519_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_1530_cast = layer_norm(axes = var_1530_axes_0, beta = blocks_13_attn_ln_bias_to_fp16, epsilon = var_1519_to_fp16, gamma = blocks_13_attn_ln_weight_to_fp16, x = x_163_cast); |
| tensor<fp16, [1280, 1280]> var_1541_to_fp16 = const()[name = tensor<string, []>("op_1541_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(525922304)))]; |
| tensor<fp16, [1280]> var_1542_to_fp16 = const()[name = tensor<string, []>("op_1542_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(529199168)))]; |
| tensor<fp16, [1, 1500, 1280]> q_53_cast = linear(bias = var_1542_to_fp16, weight = var_1541_to_fp16, x = var_1530_cast); |
| tensor<fp16, [1280, 1280]> var_1545_to_fp16 = const()[name = tensor<string, []>("op_1545_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(529201792)))]; |
| tensor<fp16, [1280]> k_53_bias_0_to_fp16 = const()[name = tensor<string, []>("k_53_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(532478656)))]; |
| tensor<fp16, [1, 1500, 1280]> k_53_cast = linear(bias = k_53_bias_0_to_fp16, weight = var_1545_to_fp16, x = var_1530_cast); |
| tensor<fp16, [1280, 1280]> var_1549_to_fp16 = const()[name = tensor<string, []>("op_1549_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(532481280)))]; |
| tensor<fp16, [1280]> var_1550_to_fp16 = const()[name = tensor<string, []>("op_1550_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(535758144)))]; |
| tensor<fp16, [1, 1500, 1280]> v_53_cast = linear(bias = var_1550_to_fp16, weight = var_1549_to_fp16, x = var_1530_cast); |
| tensor<int32, [4]> var_1558 = const()[name = tensor<string, []>("op_1558"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1559_cast = reshape(shape = var_1558, x = q_53_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_250_to_fp16 = const()[name = tensor<string, []>("const_250_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_55_cast = mul(x = var_1559_cast, y = const_250_to_fp16); |
| tensor<int32, [4]> var_1565 = const()[name = tensor<string, []>("op_1565"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1566_cast = reshape(shape = var_1565, x = k_53_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_251_to_fp16 = const()[name = tensor<string, []>("const_251_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_55_cast = mul(x = var_1566_cast, y = const_251_to_fp16); |
| tensor<int32, [4]> var_1572 = const()[name = tensor<string, []>("op_1572"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1573_cast = reshape(shape = var_1572, x = v_53_cast); |
| tensor<int32, [4]> var_1574 = const()[name = tensor<string, []>("op_1574"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_27_transpose_x_0 = const()[name = tensor<string, []>("qk_27_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_27_transpose_y_0 = const()[name = tensor<string, []>("qk_27_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_90_perm_0 = const()[name = tensor<string, []>("transpose_90_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_91_perm_0 = const()[name = tensor<string, []>("transpose_91_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_201 = transpose(perm = transpose_91_perm_0, x = k_55_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_202 = transpose(perm = transpose_90_perm_0, x = q_55_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_27_cast = matmul(transpose_x = qk_27_transpose_x_0, transpose_y = qk_27_transpose_y_0, x = transpose_202, y = transpose_201); |
| tensor<fp16, [1, 20, 1500, 1500]> var_1578_cast = softmax(axis = var_1513, x = qk_27_cast); |
| tensor<bool, []> var_1580_transpose_x_0 = const()[name = tensor<string, []>("op_1580_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_1580_transpose_y_0 = const()[name = tensor<string, []>("op_1580_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_203 = transpose(perm = var_1574, x = var_1573_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_1580_cast = matmul(transpose_x = var_1580_transpose_x_0, transpose_y = var_1580_transpose_y_0, x = var_1578_cast, y = transpose_203); |
| tensor<int32, [4]> var_1581 = const()[name = tensor<string, []>("op_1581"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_13 = const()[name = tensor<string, []>("concat_13"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_200 = transpose(perm = var_1581, x = var_1580_cast); |
| tensor<fp16, [1, 1500, 1280]> x_167_cast = reshape(shape = concat_13, x = transpose_200); |
| tensor<fp16, [1280, 1280]> var_1586_to_fp16 = const()[name = tensor<string, []>("op_1586_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(535760768)))]; |
| tensor<fp16, [1280]> var_1587_to_fp16 = const()[name = tensor<string, []>("op_1587_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(539037632)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1588_cast = linear(bias = var_1587_to_fp16, weight = var_1586_to_fp16, x = x_167_cast); |
| tensor<fp16, [1, 1500, 1280]> x_169_cast = add(x = x_163_cast, y = var_1588_cast); |
| tensor<int32, [1]> var_1594_axes_0 = const()[name = tensor<string, []>("op_1594_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_13_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_13_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(539040256)))]; |
| tensor<fp16, [1280]> blocks_13_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_13_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(539042880)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1594_cast = layer_norm(axes = var_1594_axes_0, beta = blocks_13_mlp_ln_bias_to_fp16, epsilon = var_1519_to_fp16, gamma = blocks_13_mlp_ln_weight_to_fp16, x = x_169_cast); |
| tensor<fp16, [5120, 1280]> var_1603_to_fp16 = const()[name = tensor<string, []>("op_1603_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(539045504)))]; |
| tensor<fp16, [5120]> var_1604_to_fp16 = const()[name = tensor<string, []>("op_1604_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(552152768)))]; |
| tensor<fp16, [1, 1500, 5120]> input_113_cast = linear(bias = var_1604_to_fp16, weight = var_1603_to_fp16, x = var_1594_cast); |
| tensor<string, []> x_173_mode_0 = const()[name = tensor<string, []>("x_173_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_173_cast = gelu(mode = x_173_mode_0, x = input_113_cast); |
| tensor<fp16, [1280, 5120]> var_1609_to_fp16 = const()[name = tensor<string, []>("op_1609_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(552163072)))]; |
| tensor<fp16, [1280]> var_1610_to_fp16 = const()[name = tensor<string, []>("op_1610_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(565270336)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1611_cast = linear(bias = var_1610_to_fp16, weight = var_1609_to_fp16, x = x_173_cast); |
| tensor<fp16, [1, 1500, 1280]> x_175_cast = add(x = x_169_cast, y = var_1611_cast); |
| tensor<int32, []> var_1620 = const()[name = tensor<string, []>("op_1620"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_1637_axes_0 = const()[name = tensor<string, []>("op_1637_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_14_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_14_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(565272960)))]; |
| tensor<fp16, [1280]> blocks_14_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_14_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(565275584)))]; |
| tensor<fp16, []> var_1626_to_fp16 = const()[name = tensor<string, []>("op_1626_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_1637_cast = layer_norm(axes = var_1637_axes_0, beta = blocks_14_attn_ln_bias_to_fp16, epsilon = var_1626_to_fp16, gamma = blocks_14_attn_ln_weight_to_fp16, x = x_175_cast); |
| tensor<fp16, [1280, 1280]> var_1648_to_fp16 = const()[name = tensor<string, []>("op_1648_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(565278208)))]; |
| tensor<fp16, [1280]> var_1649_to_fp16 = const()[name = tensor<string, []>("op_1649_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(568555072)))]; |
| tensor<fp16, [1, 1500, 1280]> q_57_cast = linear(bias = var_1649_to_fp16, weight = var_1648_to_fp16, x = var_1637_cast); |
| tensor<fp16, [1280, 1280]> var_1652_to_fp16 = const()[name = tensor<string, []>("op_1652_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(568557696)))]; |
| tensor<fp16, [1280]> k_57_bias_0_to_fp16 = const()[name = tensor<string, []>("k_57_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(571834560)))]; |
| tensor<fp16, [1, 1500, 1280]> k_57_cast = linear(bias = k_57_bias_0_to_fp16, weight = var_1652_to_fp16, x = var_1637_cast); |
| tensor<fp16, [1280, 1280]> var_1656_to_fp16 = const()[name = tensor<string, []>("op_1656_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(571837184)))]; |
| tensor<fp16, [1280]> var_1657_to_fp16 = const()[name = tensor<string, []>("op_1657_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(575114048)))]; |
| tensor<fp16, [1, 1500, 1280]> v_57_cast = linear(bias = var_1657_to_fp16, weight = var_1656_to_fp16, x = var_1637_cast); |
| tensor<int32, [4]> var_1665 = const()[name = tensor<string, []>("op_1665"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1666_cast = reshape(shape = var_1665, x = q_57_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_252_to_fp16 = const()[name = tensor<string, []>("const_252_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_59_cast = mul(x = var_1666_cast, y = const_252_to_fp16); |
| tensor<int32, [4]> var_1672 = const()[name = tensor<string, []>("op_1672"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1673_cast = reshape(shape = var_1672, x = k_57_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_253_to_fp16 = const()[name = tensor<string, []>("const_253_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_59_cast = mul(x = var_1673_cast, y = const_253_to_fp16); |
| tensor<int32, [4]> var_1679 = const()[name = tensor<string, []>("op_1679"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1680_cast = reshape(shape = var_1679, x = v_57_cast); |
| tensor<int32, [4]> var_1681 = const()[name = tensor<string, []>("op_1681"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_29_transpose_x_0 = const()[name = tensor<string, []>("qk_29_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_29_transpose_y_0 = const()[name = tensor<string, []>("qk_29_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_92_perm_0 = const()[name = tensor<string, []>("transpose_92_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_93_perm_0 = const()[name = tensor<string, []>("transpose_93_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_197 = transpose(perm = transpose_93_perm_0, x = k_59_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_198 = transpose(perm = transpose_92_perm_0, x = q_59_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_29_cast = matmul(transpose_x = qk_29_transpose_x_0, transpose_y = qk_29_transpose_y_0, x = transpose_198, y = transpose_197); |
| tensor<fp16, [1, 20, 1500, 1500]> var_1685_cast = softmax(axis = var_1620, x = qk_29_cast); |
| tensor<bool, []> var_1687_transpose_x_0 = const()[name = tensor<string, []>("op_1687_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_1687_transpose_y_0 = const()[name = tensor<string, []>("op_1687_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_199 = transpose(perm = var_1681, x = var_1680_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_1687_cast = matmul(transpose_x = var_1687_transpose_x_0, transpose_y = var_1687_transpose_y_0, x = var_1685_cast, y = transpose_199); |
| tensor<int32, [4]> var_1688 = const()[name = tensor<string, []>("op_1688"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_14 = const()[name = tensor<string, []>("concat_14"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_196 = transpose(perm = var_1688, x = var_1687_cast); |
| tensor<fp16, [1, 1500, 1280]> x_179_cast = reshape(shape = concat_14, x = transpose_196); |
| tensor<fp16, [1280, 1280]> var_1693_to_fp16 = const()[name = tensor<string, []>("op_1693_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(575116672)))]; |
| tensor<fp16, [1280]> var_1694_to_fp16 = const()[name = tensor<string, []>("op_1694_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(578393536)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1695_cast = linear(bias = var_1694_to_fp16, weight = var_1693_to_fp16, x = x_179_cast); |
| tensor<fp16, [1, 1500, 1280]> x_181_cast = add(x = x_175_cast, y = var_1695_cast); |
| tensor<int32, [1]> var_1701_axes_0 = const()[name = tensor<string, []>("op_1701_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_14_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_14_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(578396160)))]; |
| tensor<fp16, [1280]> blocks_14_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_14_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(578398784)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1701_cast = layer_norm(axes = var_1701_axes_0, beta = blocks_14_mlp_ln_bias_to_fp16, epsilon = var_1626_to_fp16, gamma = blocks_14_mlp_ln_weight_to_fp16, x = x_181_cast); |
| tensor<fp16, [5120, 1280]> var_1710_to_fp16 = const()[name = tensor<string, []>("op_1710_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(578401408)))]; |
| tensor<fp16, [5120]> var_1711_to_fp16 = const()[name = tensor<string, []>("op_1711_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(591508672)))]; |
| tensor<fp16, [1, 1500, 5120]> input_121_cast = linear(bias = var_1711_to_fp16, weight = var_1710_to_fp16, x = var_1701_cast); |
| tensor<string, []> x_185_mode_0 = const()[name = tensor<string, []>("x_185_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_185_cast = gelu(mode = x_185_mode_0, x = input_121_cast); |
| tensor<fp16, [1280, 5120]> var_1716_to_fp16 = const()[name = tensor<string, []>("op_1716_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(591518976)))]; |
| tensor<fp16, [1280]> var_1717_to_fp16 = const()[name = tensor<string, []>("op_1717_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(604626240)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1718_cast = linear(bias = var_1717_to_fp16, weight = var_1716_to_fp16, x = x_185_cast); |
| tensor<fp16, [1, 1500, 1280]> x_187_cast = add(x = x_181_cast, y = var_1718_cast); |
| tensor<int32, []> var_1727 = const()[name = tensor<string, []>("op_1727"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_1744_axes_0 = const()[name = tensor<string, []>("op_1744_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_15_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_15_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(604628864)))]; |
| tensor<fp16, [1280]> blocks_15_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_15_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(604631488)))]; |
| tensor<fp16, []> var_1733_to_fp16 = const()[name = tensor<string, []>("op_1733_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_1744_cast = layer_norm(axes = var_1744_axes_0, beta = blocks_15_attn_ln_bias_to_fp16, epsilon = var_1733_to_fp16, gamma = blocks_15_attn_ln_weight_to_fp16, x = x_187_cast); |
| tensor<fp16, [1280, 1280]> var_1755_to_fp16 = const()[name = tensor<string, []>("op_1755_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(604634112)))]; |
| tensor<fp16, [1280]> var_1756_to_fp16 = const()[name = tensor<string, []>("op_1756_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(607910976)))]; |
| tensor<fp16, [1, 1500, 1280]> q_61_cast = linear(bias = var_1756_to_fp16, weight = var_1755_to_fp16, x = var_1744_cast); |
| tensor<fp16, [1280, 1280]> var_1759_to_fp16 = const()[name = tensor<string, []>("op_1759_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(607913600)))]; |
| tensor<fp16, [1280]> k_61_bias_0_to_fp16 = const()[name = tensor<string, []>("k_61_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(611190464)))]; |
| tensor<fp16, [1, 1500, 1280]> k_61_cast = linear(bias = k_61_bias_0_to_fp16, weight = var_1759_to_fp16, x = var_1744_cast); |
| tensor<fp16, [1280, 1280]> var_1763_to_fp16 = const()[name = tensor<string, []>("op_1763_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(611193088)))]; |
| tensor<fp16, [1280]> var_1764_to_fp16 = const()[name = tensor<string, []>("op_1764_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(614469952)))]; |
| tensor<fp16, [1, 1500, 1280]> v_61_cast = linear(bias = var_1764_to_fp16, weight = var_1763_to_fp16, x = var_1744_cast); |
| tensor<int32, [4]> var_1772 = const()[name = tensor<string, []>("op_1772"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1773_cast = reshape(shape = var_1772, x = q_61_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_254_to_fp16 = const()[name = tensor<string, []>("const_254_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_63_cast = mul(x = var_1773_cast, y = const_254_to_fp16); |
| tensor<int32, [4]> var_1779 = const()[name = tensor<string, []>("op_1779"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1780_cast = reshape(shape = var_1779, x = k_61_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_255_to_fp16 = const()[name = tensor<string, []>("const_255_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_63_cast = mul(x = var_1780_cast, y = const_255_to_fp16); |
| tensor<int32, [4]> var_1786 = const()[name = tensor<string, []>("op_1786"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1787_cast = reshape(shape = var_1786, x = v_61_cast); |
| tensor<int32, [4]> var_1788 = const()[name = tensor<string, []>("op_1788"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_31_transpose_x_0 = const()[name = tensor<string, []>("qk_31_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_31_transpose_y_0 = const()[name = tensor<string, []>("qk_31_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_94_perm_0 = const()[name = tensor<string, []>("transpose_94_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_95_perm_0 = const()[name = tensor<string, []>("transpose_95_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_193 = transpose(perm = transpose_95_perm_0, x = k_63_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_194 = transpose(perm = transpose_94_perm_0, x = q_63_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_31_cast = matmul(transpose_x = qk_31_transpose_x_0, transpose_y = qk_31_transpose_y_0, x = transpose_194, y = transpose_193); |
| tensor<fp16, [1, 20, 1500, 1500]> var_1792_cast = softmax(axis = var_1727, x = qk_31_cast); |
| tensor<bool, []> var_1794_transpose_x_0 = const()[name = tensor<string, []>("op_1794_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_1794_transpose_y_0 = const()[name = tensor<string, []>("op_1794_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_195 = transpose(perm = var_1788, x = var_1787_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_1794_cast = matmul(transpose_x = var_1794_transpose_x_0, transpose_y = var_1794_transpose_y_0, x = var_1792_cast, y = transpose_195); |
| tensor<int32, [4]> var_1795 = const()[name = tensor<string, []>("op_1795"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_15 = const()[name = tensor<string, []>("concat_15"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_192 = transpose(perm = var_1795, x = var_1794_cast); |
| tensor<fp16, [1, 1500, 1280]> x_191_cast = reshape(shape = concat_15, x = transpose_192); |
| tensor<fp16, [1280, 1280]> var_1800_to_fp16 = const()[name = tensor<string, []>("op_1800_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(614472576)))]; |
| tensor<fp16, [1280]> var_1801_to_fp16 = const()[name = tensor<string, []>("op_1801_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(617749440)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1802_cast = linear(bias = var_1801_to_fp16, weight = var_1800_to_fp16, x = x_191_cast); |
| tensor<fp16, [1, 1500, 1280]> x_193_cast = add(x = x_187_cast, y = var_1802_cast); |
| tensor<int32, [1]> var_1808_axes_0 = const()[name = tensor<string, []>("op_1808_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_15_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_15_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(617752064)))]; |
| tensor<fp16, [1280]> blocks_15_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_15_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(617754688)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1808_cast = layer_norm(axes = var_1808_axes_0, beta = blocks_15_mlp_ln_bias_to_fp16, epsilon = var_1733_to_fp16, gamma = blocks_15_mlp_ln_weight_to_fp16, x = x_193_cast); |
| tensor<fp16, [5120, 1280]> var_1817_to_fp16 = const()[name = tensor<string, []>("op_1817_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(617757312)))]; |
| tensor<fp16, [5120]> var_1818_to_fp16 = const()[name = tensor<string, []>("op_1818_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(630864576)))]; |
| tensor<fp16, [1, 1500, 5120]> input_129_cast = linear(bias = var_1818_to_fp16, weight = var_1817_to_fp16, x = var_1808_cast); |
| tensor<string, []> x_197_mode_0 = const()[name = tensor<string, []>("x_197_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_197_cast = gelu(mode = x_197_mode_0, x = input_129_cast); |
| tensor<fp16, [1280, 5120]> var_1823_to_fp16 = const()[name = tensor<string, []>("op_1823_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(630874880)))]; |
| tensor<fp16, [1280]> var_1824_to_fp16 = const()[name = tensor<string, []>("op_1824_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643982144)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1825_cast = linear(bias = var_1824_to_fp16, weight = var_1823_to_fp16, x = x_197_cast); |
| tensor<fp16, [1, 1500, 1280]> x_199_cast = add(x = x_193_cast, y = var_1825_cast); |
| tensor<int32, []> var_1834 = const()[name = tensor<string, []>("op_1834"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_1851_axes_0 = const()[name = tensor<string, []>("op_1851_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_16_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_16_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643984768)))]; |
| tensor<fp16, [1280]> blocks_16_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_16_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643987392)))]; |
| tensor<fp16, []> var_1840_to_fp16 = const()[name = tensor<string, []>("op_1840_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_1851_cast = layer_norm(axes = var_1851_axes_0, beta = blocks_16_attn_ln_bias_to_fp16, epsilon = var_1840_to_fp16, gamma = blocks_16_attn_ln_weight_to_fp16, x = x_199_cast); |
| tensor<fp16, [1280, 1280]> var_1862_to_fp16 = const()[name = tensor<string, []>("op_1862_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(643990016)))]; |
| tensor<fp16, [1280]> var_1863_to_fp16 = const()[name = tensor<string, []>("op_1863_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(647266880)))]; |
| tensor<fp16, [1, 1500, 1280]> q_65_cast = linear(bias = var_1863_to_fp16, weight = var_1862_to_fp16, x = var_1851_cast); |
| tensor<fp16, [1280, 1280]> var_1866_to_fp16 = const()[name = tensor<string, []>("op_1866_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(647269504)))]; |
| tensor<fp16, [1280]> k_65_bias_0_to_fp16 = const()[name = tensor<string, []>("k_65_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(650546368)))]; |
| tensor<fp16, [1, 1500, 1280]> k_65_cast = linear(bias = k_65_bias_0_to_fp16, weight = var_1866_to_fp16, x = var_1851_cast); |
| tensor<fp16, [1280, 1280]> var_1870_to_fp16 = const()[name = tensor<string, []>("op_1870_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(650548992)))]; |
| tensor<fp16, [1280]> var_1871_to_fp16 = const()[name = tensor<string, []>("op_1871_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(653825856)))]; |
| tensor<fp16, [1, 1500, 1280]> v_65_cast = linear(bias = var_1871_to_fp16, weight = var_1870_to_fp16, x = var_1851_cast); |
| tensor<int32, [4]> var_1879 = const()[name = tensor<string, []>("op_1879"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1880_cast = reshape(shape = var_1879, x = q_65_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_256_to_fp16 = const()[name = tensor<string, []>("const_256_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_67_cast = mul(x = var_1880_cast, y = const_256_to_fp16); |
| tensor<int32, [4]> var_1886 = const()[name = tensor<string, []>("op_1886"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1887_cast = reshape(shape = var_1886, x = k_65_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_257_to_fp16 = const()[name = tensor<string, []>("const_257_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_67_cast = mul(x = var_1887_cast, y = const_257_to_fp16); |
| tensor<int32, [4]> var_1893 = const()[name = tensor<string, []>("op_1893"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1894_cast = reshape(shape = var_1893, x = v_65_cast); |
| tensor<int32, [4]> var_1895 = const()[name = tensor<string, []>("op_1895"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_33_transpose_x_0 = const()[name = tensor<string, []>("qk_33_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_33_transpose_y_0 = const()[name = tensor<string, []>("qk_33_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_96_perm_0 = const()[name = tensor<string, []>("transpose_96_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_97_perm_0 = const()[name = tensor<string, []>("transpose_97_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_189 = transpose(perm = transpose_97_perm_0, x = k_67_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_190 = transpose(perm = transpose_96_perm_0, x = q_67_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_33_cast = matmul(transpose_x = qk_33_transpose_x_0, transpose_y = qk_33_transpose_y_0, x = transpose_190, y = transpose_189); |
| tensor<fp16, [1, 20, 1500, 1500]> var_1899_cast = softmax(axis = var_1834, x = qk_33_cast); |
| tensor<bool, []> var_1901_transpose_x_0 = const()[name = tensor<string, []>("op_1901_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_1901_transpose_y_0 = const()[name = tensor<string, []>("op_1901_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_191 = transpose(perm = var_1895, x = var_1894_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_1901_cast = matmul(transpose_x = var_1901_transpose_x_0, transpose_y = var_1901_transpose_y_0, x = var_1899_cast, y = transpose_191); |
| tensor<int32, [4]> var_1902 = const()[name = tensor<string, []>("op_1902"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_16 = const()[name = tensor<string, []>("concat_16"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_188 = transpose(perm = var_1902, x = var_1901_cast); |
| tensor<fp16, [1, 1500, 1280]> x_203_cast = reshape(shape = concat_16, x = transpose_188); |
| tensor<fp16, [1280, 1280]> var_1907_to_fp16 = const()[name = tensor<string, []>("op_1907_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(653828480)))]; |
| tensor<fp16, [1280]> var_1908_to_fp16 = const()[name = tensor<string, []>("op_1908_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(657105344)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1909_cast = linear(bias = var_1908_to_fp16, weight = var_1907_to_fp16, x = x_203_cast); |
| tensor<fp16, [1, 1500, 1280]> x_205_cast = add(x = x_199_cast, y = var_1909_cast); |
| tensor<int32, [1]> var_1915_axes_0 = const()[name = tensor<string, []>("op_1915_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_16_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_16_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(657107968)))]; |
| tensor<fp16, [1280]> blocks_16_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_16_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(657110592)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1915_cast = layer_norm(axes = var_1915_axes_0, beta = blocks_16_mlp_ln_bias_to_fp16, epsilon = var_1840_to_fp16, gamma = blocks_16_mlp_ln_weight_to_fp16, x = x_205_cast); |
| tensor<fp16, [5120, 1280]> var_1924_to_fp16 = const()[name = tensor<string, []>("op_1924_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(657113216)))]; |
| tensor<fp16, [5120]> var_1925_to_fp16 = const()[name = tensor<string, []>("op_1925_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(670220480)))]; |
| tensor<fp16, [1, 1500, 5120]> input_137_cast = linear(bias = var_1925_to_fp16, weight = var_1924_to_fp16, x = var_1915_cast); |
| tensor<string, []> x_209_mode_0 = const()[name = tensor<string, []>("x_209_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_209_cast = gelu(mode = x_209_mode_0, x = input_137_cast); |
| tensor<fp16, [1280, 5120]> var_1930_to_fp16 = const()[name = tensor<string, []>("op_1930_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(670230784)))]; |
| tensor<fp16, [1280]> var_1931_to_fp16 = const()[name = tensor<string, []>("op_1931_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(683338048)))]; |
| tensor<fp16, [1, 1500, 1280]> var_1932_cast = linear(bias = var_1931_to_fp16, weight = var_1930_to_fp16, x = x_209_cast); |
| tensor<fp16, [1, 1500, 1280]> x_211_cast = add(x = x_205_cast, y = var_1932_cast); |
| tensor<int32, []> var_1941 = const()[name = tensor<string, []>("op_1941"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_1958_axes_0 = const()[name = tensor<string, []>("op_1958_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_17_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_17_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(683340672)))]; |
| tensor<fp16, [1280]> blocks_17_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_17_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(683343296)))]; |
| tensor<fp16, []> var_1947_to_fp16 = const()[name = tensor<string, []>("op_1947_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_1958_cast = layer_norm(axes = var_1958_axes_0, beta = blocks_17_attn_ln_bias_to_fp16, epsilon = var_1947_to_fp16, gamma = blocks_17_attn_ln_weight_to_fp16, x = x_211_cast); |
| tensor<fp16, [1280, 1280]> var_1969_to_fp16 = const()[name = tensor<string, []>("op_1969_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(683345920)))]; |
| tensor<fp16, [1280]> var_1970_to_fp16 = const()[name = tensor<string, []>("op_1970_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(686622784)))]; |
| tensor<fp16, [1, 1500, 1280]> q_69_cast = linear(bias = var_1970_to_fp16, weight = var_1969_to_fp16, x = var_1958_cast); |
| tensor<fp16, [1280, 1280]> var_1973_to_fp16 = const()[name = tensor<string, []>("op_1973_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(686625408)))]; |
| tensor<fp16, [1280]> k_69_bias_0_to_fp16 = const()[name = tensor<string, []>("k_69_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(689902272)))]; |
| tensor<fp16, [1, 1500, 1280]> k_69_cast = linear(bias = k_69_bias_0_to_fp16, weight = var_1973_to_fp16, x = var_1958_cast); |
| tensor<fp16, [1280, 1280]> var_1977_to_fp16 = const()[name = tensor<string, []>("op_1977_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(689904896)))]; |
| tensor<fp16, [1280]> var_1978_to_fp16 = const()[name = tensor<string, []>("op_1978_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(693181760)))]; |
| tensor<fp16, [1, 1500, 1280]> v_69_cast = linear(bias = var_1978_to_fp16, weight = var_1977_to_fp16, x = var_1958_cast); |
| tensor<int32, [4]> var_1986 = const()[name = tensor<string, []>("op_1986"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1987_cast = reshape(shape = var_1986, x = q_69_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_258_to_fp16 = const()[name = tensor<string, []>("const_258_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_71_cast = mul(x = var_1987_cast, y = const_258_to_fp16); |
| tensor<int32, [4]> var_1993 = const()[name = tensor<string, []>("op_1993"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_1994_cast = reshape(shape = var_1993, x = k_69_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_259_to_fp16 = const()[name = tensor<string, []>("const_259_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_71_cast = mul(x = var_1994_cast, y = const_259_to_fp16); |
| tensor<int32, [4]> var_2000 = const()[name = tensor<string, []>("op_2000"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2001_cast = reshape(shape = var_2000, x = v_69_cast); |
| tensor<int32, [4]> var_2002 = const()[name = tensor<string, []>("op_2002"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_35_transpose_x_0 = const()[name = tensor<string, []>("qk_35_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_35_transpose_y_0 = const()[name = tensor<string, []>("qk_35_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_98_perm_0 = const()[name = tensor<string, []>("transpose_98_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_99_perm_0 = const()[name = tensor<string, []>("transpose_99_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_185 = transpose(perm = transpose_99_perm_0, x = k_71_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_186 = transpose(perm = transpose_98_perm_0, x = q_71_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_35_cast = matmul(transpose_x = qk_35_transpose_x_0, transpose_y = qk_35_transpose_y_0, x = transpose_186, y = transpose_185); |
| tensor<fp16, [1, 20, 1500, 1500]> var_2006_cast = softmax(axis = var_1941, x = qk_35_cast); |
| tensor<bool, []> var_2008_transpose_x_0 = const()[name = tensor<string, []>("op_2008_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_2008_transpose_y_0 = const()[name = tensor<string, []>("op_2008_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_187 = transpose(perm = var_2002, x = var_2001_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_2008_cast = matmul(transpose_x = var_2008_transpose_x_0, transpose_y = var_2008_transpose_y_0, x = var_2006_cast, y = transpose_187); |
| tensor<int32, [4]> var_2009 = const()[name = tensor<string, []>("op_2009"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_17 = const()[name = tensor<string, []>("concat_17"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_184 = transpose(perm = var_2009, x = var_2008_cast); |
| tensor<fp16, [1, 1500, 1280]> x_215_cast = reshape(shape = concat_17, x = transpose_184); |
| tensor<fp16, [1280, 1280]> var_2014_to_fp16 = const()[name = tensor<string, []>("op_2014_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(693184384)))]; |
| tensor<fp16, [1280]> var_2015_to_fp16 = const()[name = tensor<string, []>("op_2015_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(696461248)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2016_cast = linear(bias = var_2015_to_fp16, weight = var_2014_to_fp16, x = x_215_cast); |
| tensor<fp16, [1, 1500, 1280]> x_217_cast = add(x = x_211_cast, y = var_2016_cast); |
| tensor<int32, [1]> var_2022_axes_0 = const()[name = tensor<string, []>("op_2022_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_17_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_17_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(696463872)))]; |
| tensor<fp16, [1280]> blocks_17_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_17_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(696466496)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2022_cast = layer_norm(axes = var_2022_axes_0, beta = blocks_17_mlp_ln_bias_to_fp16, epsilon = var_1947_to_fp16, gamma = blocks_17_mlp_ln_weight_to_fp16, x = x_217_cast); |
| tensor<fp16, [5120, 1280]> var_2031_to_fp16 = const()[name = tensor<string, []>("op_2031_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(696469120)))]; |
| tensor<fp16, [5120]> var_2032_to_fp16 = const()[name = tensor<string, []>("op_2032_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(709576384)))]; |
| tensor<fp16, [1, 1500, 5120]> input_145_cast = linear(bias = var_2032_to_fp16, weight = var_2031_to_fp16, x = var_2022_cast); |
| tensor<string, []> x_221_mode_0 = const()[name = tensor<string, []>("x_221_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_221_cast = gelu(mode = x_221_mode_0, x = input_145_cast); |
| tensor<fp16, [1280, 5120]> var_2037_to_fp16 = const()[name = tensor<string, []>("op_2037_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(709586688)))]; |
| tensor<fp16, [1280]> var_2038_to_fp16 = const()[name = tensor<string, []>("op_2038_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(722693952)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2039_cast = linear(bias = var_2038_to_fp16, weight = var_2037_to_fp16, x = x_221_cast); |
| tensor<fp16, [1, 1500, 1280]> x_223_cast = add(x = x_217_cast, y = var_2039_cast); |
| tensor<int32, []> var_2048 = const()[name = tensor<string, []>("op_2048"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_2065_axes_0 = const()[name = tensor<string, []>("op_2065_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_18_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_18_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(722696576)))]; |
| tensor<fp16, [1280]> blocks_18_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_18_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(722699200)))]; |
| tensor<fp16, []> var_2054_to_fp16 = const()[name = tensor<string, []>("op_2054_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_2065_cast = layer_norm(axes = var_2065_axes_0, beta = blocks_18_attn_ln_bias_to_fp16, epsilon = var_2054_to_fp16, gamma = blocks_18_attn_ln_weight_to_fp16, x = x_223_cast); |
| tensor<fp16, [1280, 1280]> var_2076_to_fp16 = const()[name = tensor<string, []>("op_2076_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(722701824)))]; |
| tensor<fp16, [1280]> var_2077_to_fp16 = const()[name = tensor<string, []>("op_2077_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(725978688)))]; |
| tensor<fp16, [1, 1500, 1280]> q_73_cast = linear(bias = var_2077_to_fp16, weight = var_2076_to_fp16, x = var_2065_cast); |
| tensor<fp16, [1280, 1280]> var_2080_to_fp16 = const()[name = tensor<string, []>("op_2080_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(725981312)))]; |
| tensor<fp16, [1280]> k_73_bias_0_to_fp16 = const()[name = tensor<string, []>("k_73_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(729258176)))]; |
| tensor<fp16, [1, 1500, 1280]> k_73_cast = linear(bias = k_73_bias_0_to_fp16, weight = var_2080_to_fp16, x = var_2065_cast); |
| tensor<fp16, [1280, 1280]> var_2084_to_fp16 = const()[name = tensor<string, []>("op_2084_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(729260800)))]; |
| tensor<fp16, [1280]> var_2085_to_fp16 = const()[name = tensor<string, []>("op_2085_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(732537664)))]; |
| tensor<fp16, [1, 1500, 1280]> v_73_cast = linear(bias = var_2085_to_fp16, weight = var_2084_to_fp16, x = var_2065_cast); |
| tensor<int32, [4]> var_2093 = const()[name = tensor<string, []>("op_2093"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2094_cast = reshape(shape = var_2093, x = q_73_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_260_to_fp16 = const()[name = tensor<string, []>("const_260_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_75_cast = mul(x = var_2094_cast, y = const_260_to_fp16); |
| tensor<int32, [4]> var_2100 = const()[name = tensor<string, []>("op_2100"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2101_cast = reshape(shape = var_2100, x = k_73_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_261_to_fp16 = const()[name = tensor<string, []>("const_261_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_75_cast = mul(x = var_2101_cast, y = const_261_to_fp16); |
| tensor<int32, [4]> var_2107 = const()[name = tensor<string, []>("op_2107"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2108_cast = reshape(shape = var_2107, x = v_73_cast); |
| tensor<int32, [4]> var_2109 = const()[name = tensor<string, []>("op_2109"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_37_transpose_x_0 = const()[name = tensor<string, []>("qk_37_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_37_transpose_y_0 = const()[name = tensor<string, []>("qk_37_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_100_perm_0 = const()[name = tensor<string, []>("transpose_100_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_101_perm_0 = const()[name = tensor<string, []>("transpose_101_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_181 = transpose(perm = transpose_101_perm_0, x = k_75_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_182 = transpose(perm = transpose_100_perm_0, x = q_75_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_37_cast = matmul(transpose_x = qk_37_transpose_x_0, transpose_y = qk_37_transpose_y_0, x = transpose_182, y = transpose_181); |
| tensor<fp16, [1, 20, 1500, 1500]> var_2113_cast = softmax(axis = var_2048, x = qk_37_cast); |
| tensor<bool, []> var_2115_transpose_x_0 = const()[name = tensor<string, []>("op_2115_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_2115_transpose_y_0 = const()[name = tensor<string, []>("op_2115_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_183 = transpose(perm = var_2109, x = var_2108_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_2115_cast = matmul(transpose_x = var_2115_transpose_x_0, transpose_y = var_2115_transpose_y_0, x = var_2113_cast, y = transpose_183); |
| tensor<int32, [4]> var_2116 = const()[name = tensor<string, []>("op_2116"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_18 = const()[name = tensor<string, []>("concat_18"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_180 = transpose(perm = var_2116, x = var_2115_cast); |
| tensor<fp16, [1, 1500, 1280]> x_227_cast = reshape(shape = concat_18, x = transpose_180); |
| tensor<fp16, [1280, 1280]> var_2121_to_fp16 = const()[name = tensor<string, []>("op_2121_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(732540288)))]; |
| tensor<fp16, [1280]> var_2122_to_fp16 = const()[name = tensor<string, []>("op_2122_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(735817152)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2123_cast = linear(bias = var_2122_to_fp16, weight = var_2121_to_fp16, x = x_227_cast); |
| tensor<fp16, [1, 1500, 1280]> x_229_cast = add(x = x_223_cast, y = var_2123_cast); |
| tensor<int32, [1]> var_2129_axes_0 = const()[name = tensor<string, []>("op_2129_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_18_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_18_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(735819776)))]; |
| tensor<fp16, [1280]> blocks_18_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_18_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(735822400)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2129_cast = layer_norm(axes = var_2129_axes_0, beta = blocks_18_mlp_ln_bias_to_fp16, epsilon = var_2054_to_fp16, gamma = blocks_18_mlp_ln_weight_to_fp16, x = x_229_cast); |
| tensor<fp16, [5120, 1280]> var_2138_to_fp16 = const()[name = tensor<string, []>("op_2138_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(735825024)))]; |
| tensor<fp16, [5120]> var_2139_to_fp16 = const()[name = tensor<string, []>("op_2139_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(748932288)))]; |
| tensor<fp16, [1, 1500, 5120]> input_153_cast = linear(bias = var_2139_to_fp16, weight = var_2138_to_fp16, x = var_2129_cast); |
| tensor<string, []> x_233_mode_0 = const()[name = tensor<string, []>("x_233_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_233_cast = gelu(mode = x_233_mode_0, x = input_153_cast); |
| tensor<fp16, [1280, 5120]> var_2144_to_fp16 = const()[name = tensor<string, []>("op_2144_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(748942592)))]; |
| tensor<fp16, [1280]> var_2145_to_fp16 = const()[name = tensor<string, []>("op_2145_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(762049856)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2146_cast = linear(bias = var_2145_to_fp16, weight = var_2144_to_fp16, x = x_233_cast); |
| tensor<fp16, [1, 1500, 1280]> x_235_cast = add(x = x_229_cast, y = var_2146_cast); |
| tensor<int32, []> var_2155 = const()[name = tensor<string, []>("op_2155"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_2172_axes_0 = const()[name = tensor<string, []>("op_2172_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_19_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_19_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(762052480)))]; |
| tensor<fp16, [1280]> blocks_19_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_19_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(762055104)))]; |
| tensor<fp16, []> var_2161_to_fp16 = const()[name = tensor<string, []>("op_2161_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_2172_cast = layer_norm(axes = var_2172_axes_0, beta = blocks_19_attn_ln_bias_to_fp16, epsilon = var_2161_to_fp16, gamma = blocks_19_attn_ln_weight_to_fp16, x = x_235_cast); |
| tensor<fp16, [1280, 1280]> var_2183_to_fp16 = const()[name = tensor<string, []>("op_2183_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(762057728)))]; |
| tensor<fp16, [1280]> var_2184_to_fp16 = const()[name = tensor<string, []>("op_2184_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(765334592)))]; |
| tensor<fp16, [1, 1500, 1280]> q_77_cast = linear(bias = var_2184_to_fp16, weight = var_2183_to_fp16, x = var_2172_cast); |
| tensor<fp16, [1280, 1280]> var_2187_to_fp16 = const()[name = tensor<string, []>("op_2187_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(765337216)))]; |
| tensor<fp16, [1280]> k_77_bias_0_to_fp16 = const()[name = tensor<string, []>("k_77_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(768614080)))]; |
| tensor<fp16, [1, 1500, 1280]> k_77_cast = linear(bias = k_77_bias_0_to_fp16, weight = var_2187_to_fp16, x = var_2172_cast); |
| tensor<fp16, [1280, 1280]> var_2191_to_fp16 = const()[name = tensor<string, []>("op_2191_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(768616704)))]; |
| tensor<fp16, [1280]> var_2192_to_fp16 = const()[name = tensor<string, []>("op_2192_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(771893568)))]; |
| tensor<fp16, [1, 1500, 1280]> v_77_cast = linear(bias = var_2192_to_fp16, weight = var_2191_to_fp16, x = var_2172_cast); |
| tensor<int32, [4]> var_2200 = const()[name = tensor<string, []>("op_2200"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2201_cast = reshape(shape = var_2200, x = q_77_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_262_to_fp16 = const()[name = tensor<string, []>("const_262_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_79_cast = mul(x = var_2201_cast, y = const_262_to_fp16); |
| tensor<int32, [4]> var_2207 = const()[name = tensor<string, []>("op_2207"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2208_cast = reshape(shape = var_2207, x = k_77_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_263_to_fp16 = const()[name = tensor<string, []>("const_263_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_79_cast = mul(x = var_2208_cast, y = const_263_to_fp16); |
| tensor<int32, [4]> var_2214 = const()[name = tensor<string, []>("op_2214"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2215_cast = reshape(shape = var_2214, x = v_77_cast); |
| tensor<int32, [4]> var_2216 = const()[name = tensor<string, []>("op_2216"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_39_transpose_x_0 = const()[name = tensor<string, []>("qk_39_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_39_transpose_y_0 = const()[name = tensor<string, []>("qk_39_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_102_perm_0 = const()[name = tensor<string, []>("transpose_102_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_103_perm_0 = const()[name = tensor<string, []>("transpose_103_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_177 = transpose(perm = transpose_103_perm_0, x = k_79_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_178 = transpose(perm = transpose_102_perm_0, x = q_79_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_39_cast = matmul(transpose_x = qk_39_transpose_x_0, transpose_y = qk_39_transpose_y_0, x = transpose_178, y = transpose_177); |
| tensor<fp16, [1, 20, 1500, 1500]> var_2220_cast = softmax(axis = var_2155, x = qk_39_cast); |
| tensor<bool, []> var_2222_transpose_x_0 = const()[name = tensor<string, []>("op_2222_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_2222_transpose_y_0 = const()[name = tensor<string, []>("op_2222_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_179 = transpose(perm = var_2216, x = var_2215_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_2222_cast = matmul(transpose_x = var_2222_transpose_x_0, transpose_y = var_2222_transpose_y_0, x = var_2220_cast, y = transpose_179); |
| tensor<int32, [4]> var_2223 = const()[name = tensor<string, []>("op_2223"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_19 = const()[name = tensor<string, []>("concat_19"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_176 = transpose(perm = var_2223, x = var_2222_cast); |
| tensor<fp16, [1, 1500, 1280]> x_239_cast = reshape(shape = concat_19, x = transpose_176); |
| tensor<fp16, [1280, 1280]> var_2228_to_fp16 = const()[name = tensor<string, []>("op_2228_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(771896192)))]; |
| tensor<fp16, [1280]> var_2229_to_fp16 = const()[name = tensor<string, []>("op_2229_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(775173056)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2230_cast = linear(bias = var_2229_to_fp16, weight = var_2228_to_fp16, x = x_239_cast); |
| tensor<fp16, [1, 1500, 1280]> x_241_cast = add(x = x_235_cast, y = var_2230_cast); |
| tensor<int32, [1]> var_2236_axes_0 = const()[name = tensor<string, []>("op_2236_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_19_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_19_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(775175680)))]; |
| tensor<fp16, [1280]> blocks_19_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_19_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(775178304)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2236_cast = layer_norm(axes = var_2236_axes_0, beta = blocks_19_mlp_ln_bias_to_fp16, epsilon = var_2161_to_fp16, gamma = blocks_19_mlp_ln_weight_to_fp16, x = x_241_cast); |
| tensor<fp16, [5120, 1280]> var_2245_to_fp16 = const()[name = tensor<string, []>("op_2245_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(775180928)))]; |
| tensor<fp16, [5120]> var_2246_to_fp16 = const()[name = tensor<string, []>("op_2246_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(788288192)))]; |
| tensor<fp16, [1, 1500, 5120]> input_161_cast = linear(bias = var_2246_to_fp16, weight = var_2245_to_fp16, x = var_2236_cast); |
| tensor<string, []> x_245_mode_0 = const()[name = tensor<string, []>("x_245_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_245_cast = gelu(mode = x_245_mode_0, x = input_161_cast); |
| tensor<fp16, [1280, 5120]> var_2251_to_fp16 = const()[name = tensor<string, []>("op_2251_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(788298496)))]; |
| tensor<fp16, [1280]> var_2252_to_fp16 = const()[name = tensor<string, []>("op_2252_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(801405760)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2253_cast = linear(bias = var_2252_to_fp16, weight = var_2251_to_fp16, x = x_245_cast); |
| tensor<fp16, [1, 1500, 1280]> x_247_cast = add(x = x_241_cast, y = var_2253_cast); |
| tensor<int32, []> var_2262 = const()[name = tensor<string, []>("op_2262"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_2279_axes_0 = const()[name = tensor<string, []>("op_2279_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_20_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_20_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(801408384)))]; |
| tensor<fp16, [1280]> blocks_20_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_20_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(801411008)))]; |
| tensor<fp16, []> var_2268_to_fp16 = const()[name = tensor<string, []>("op_2268_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_2279_cast = layer_norm(axes = var_2279_axes_0, beta = blocks_20_attn_ln_bias_to_fp16, epsilon = var_2268_to_fp16, gamma = blocks_20_attn_ln_weight_to_fp16, x = x_247_cast); |
| tensor<fp16, [1280, 1280]> var_2290_to_fp16 = const()[name = tensor<string, []>("op_2290_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(801413632)))]; |
| tensor<fp16, [1280]> var_2291_to_fp16 = const()[name = tensor<string, []>("op_2291_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(804690496)))]; |
| tensor<fp16, [1, 1500, 1280]> q_81_cast = linear(bias = var_2291_to_fp16, weight = var_2290_to_fp16, x = var_2279_cast); |
| tensor<fp16, [1280, 1280]> var_2294_to_fp16 = const()[name = tensor<string, []>("op_2294_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(804693120)))]; |
| tensor<fp16, [1280]> k_81_bias_0_to_fp16 = const()[name = tensor<string, []>("k_81_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(807969984)))]; |
| tensor<fp16, [1, 1500, 1280]> k_81_cast = linear(bias = k_81_bias_0_to_fp16, weight = var_2294_to_fp16, x = var_2279_cast); |
| tensor<fp16, [1280, 1280]> var_2298_to_fp16 = const()[name = tensor<string, []>("op_2298_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(807972608)))]; |
| tensor<fp16, [1280]> var_2299_to_fp16 = const()[name = tensor<string, []>("op_2299_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(811249472)))]; |
| tensor<fp16, [1, 1500, 1280]> v_81_cast = linear(bias = var_2299_to_fp16, weight = var_2298_to_fp16, x = var_2279_cast); |
| tensor<int32, [4]> var_2307 = const()[name = tensor<string, []>("op_2307"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2308_cast = reshape(shape = var_2307, x = q_81_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_264_to_fp16 = const()[name = tensor<string, []>("const_264_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_83_cast = mul(x = var_2308_cast, y = const_264_to_fp16); |
| tensor<int32, [4]> var_2314 = const()[name = tensor<string, []>("op_2314"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2315_cast = reshape(shape = var_2314, x = k_81_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_265_to_fp16 = const()[name = tensor<string, []>("const_265_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_83_cast = mul(x = var_2315_cast, y = const_265_to_fp16); |
| tensor<int32, [4]> var_2321 = const()[name = tensor<string, []>("op_2321"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2322_cast = reshape(shape = var_2321, x = v_81_cast); |
| tensor<int32, [4]> var_2323 = const()[name = tensor<string, []>("op_2323"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_41_transpose_x_0 = const()[name = tensor<string, []>("qk_41_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_41_transpose_y_0 = const()[name = tensor<string, []>("qk_41_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_104_perm_0 = const()[name = tensor<string, []>("transpose_104_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_105_perm_0 = const()[name = tensor<string, []>("transpose_105_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_173 = transpose(perm = transpose_105_perm_0, x = k_83_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_174 = transpose(perm = transpose_104_perm_0, x = q_83_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_41_cast = matmul(transpose_x = qk_41_transpose_x_0, transpose_y = qk_41_transpose_y_0, x = transpose_174, y = transpose_173); |
| tensor<fp16, [1, 20, 1500, 1500]> var_2327_cast = softmax(axis = var_2262, x = qk_41_cast); |
| tensor<bool, []> var_2329_transpose_x_0 = const()[name = tensor<string, []>("op_2329_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_2329_transpose_y_0 = const()[name = tensor<string, []>("op_2329_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_175 = transpose(perm = var_2323, x = var_2322_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_2329_cast = matmul(transpose_x = var_2329_transpose_x_0, transpose_y = var_2329_transpose_y_0, x = var_2327_cast, y = transpose_175); |
| tensor<int32, [4]> var_2330 = const()[name = tensor<string, []>("op_2330"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_20 = const()[name = tensor<string, []>("concat_20"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_172 = transpose(perm = var_2330, x = var_2329_cast); |
| tensor<fp16, [1, 1500, 1280]> x_251_cast = reshape(shape = concat_20, x = transpose_172); |
| tensor<fp16, [1280, 1280]> var_2335_to_fp16 = const()[name = tensor<string, []>("op_2335_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(811252096)))]; |
| tensor<fp16, [1280]> var_2336_to_fp16 = const()[name = tensor<string, []>("op_2336_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(814528960)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2337_cast = linear(bias = var_2336_to_fp16, weight = var_2335_to_fp16, x = x_251_cast); |
| tensor<fp16, [1, 1500, 1280]> x_253_cast = add(x = x_247_cast, y = var_2337_cast); |
| tensor<int32, [1]> var_2343_axes_0 = const()[name = tensor<string, []>("op_2343_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_20_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_20_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(814531584)))]; |
| tensor<fp16, [1280]> blocks_20_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_20_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(814534208)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2343_cast = layer_norm(axes = var_2343_axes_0, beta = blocks_20_mlp_ln_bias_to_fp16, epsilon = var_2268_to_fp16, gamma = blocks_20_mlp_ln_weight_to_fp16, x = x_253_cast); |
| tensor<fp16, [5120, 1280]> var_2352_to_fp16 = const()[name = tensor<string, []>("op_2352_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(814536832)))]; |
| tensor<fp16, [5120]> var_2353_to_fp16 = const()[name = tensor<string, []>("op_2353_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(827644096)))]; |
| tensor<fp16, [1, 1500, 5120]> input_169_cast = linear(bias = var_2353_to_fp16, weight = var_2352_to_fp16, x = var_2343_cast); |
| tensor<string, []> x_257_mode_0 = const()[name = tensor<string, []>("x_257_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_257_cast = gelu(mode = x_257_mode_0, x = input_169_cast); |
| tensor<fp16, [1280, 5120]> var_2358_to_fp16 = const()[name = tensor<string, []>("op_2358_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(827654400)))]; |
| tensor<fp16, [1280]> var_2359_to_fp16 = const()[name = tensor<string, []>("op_2359_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(840761664)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2360_cast = linear(bias = var_2359_to_fp16, weight = var_2358_to_fp16, x = x_257_cast); |
| tensor<fp16, [1, 1500, 1280]> x_259_cast = add(x = x_253_cast, y = var_2360_cast); |
| tensor<int32, []> var_2369 = const()[name = tensor<string, []>("op_2369"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_2386_axes_0 = const()[name = tensor<string, []>("op_2386_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_21_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_21_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(840764288)))]; |
| tensor<fp16, [1280]> blocks_21_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_21_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(840766912)))]; |
| tensor<fp16, []> var_2375_to_fp16 = const()[name = tensor<string, []>("op_2375_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_2386_cast = layer_norm(axes = var_2386_axes_0, beta = blocks_21_attn_ln_bias_to_fp16, epsilon = var_2375_to_fp16, gamma = blocks_21_attn_ln_weight_to_fp16, x = x_259_cast); |
| tensor<fp16, [1280, 1280]> var_2397_to_fp16 = const()[name = tensor<string, []>("op_2397_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(840769536)))]; |
| tensor<fp16, [1280]> var_2398_to_fp16 = const()[name = tensor<string, []>("op_2398_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(844046400)))]; |
| tensor<fp16, [1, 1500, 1280]> q_85_cast = linear(bias = var_2398_to_fp16, weight = var_2397_to_fp16, x = var_2386_cast); |
| tensor<fp16, [1280, 1280]> var_2401_to_fp16 = const()[name = tensor<string, []>("op_2401_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(844049024)))]; |
| tensor<fp16, [1280]> k_85_bias_0_to_fp16 = const()[name = tensor<string, []>("k_85_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(847325888)))]; |
| tensor<fp16, [1, 1500, 1280]> k_85_cast = linear(bias = k_85_bias_0_to_fp16, weight = var_2401_to_fp16, x = var_2386_cast); |
| tensor<fp16, [1280, 1280]> var_2405_to_fp16 = const()[name = tensor<string, []>("op_2405_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(847328512)))]; |
| tensor<fp16, [1280]> var_2406_to_fp16 = const()[name = tensor<string, []>("op_2406_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(850605376)))]; |
| tensor<fp16, [1, 1500, 1280]> v_85_cast = linear(bias = var_2406_to_fp16, weight = var_2405_to_fp16, x = var_2386_cast); |
| tensor<int32, [4]> var_2414 = const()[name = tensor<string, []>("op_2414"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2415_cast = reshape(shape = var_2414, x = q_85_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_266_to_fp16 = const()[name = tensor<string, []>("const_266_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_87_cast = mul(x = var_2415_cast, y = const_266_to_fp16); |
| tensor<int32, [4]> var_2421 = const()[name = tensor<string, []>("op_2421"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2422_cast = reshape(shape = var_2421, x = k_85_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_267_to_fp16 = const()[name = tensor<string, []>("const_267_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_87_cast = mul(x = var_2422_cast, y = const_267_to_fp16); |
| tensor<int32, [4]> var_2428 = const()[name = tensor<string, []>("op_2428"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2429_cast = reshape(shape = var_2428, x = v_85_cast); |
| tensor<int32, [4]> var_2430 = const()[name = tensor<string, []>("op_2430"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_43_transpose_x_0 = const()[name = tensor<string, []>("qk_43_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_43_transpose_y_0 = const()[name = tensor<string, []>("qk_43_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_106_perm_0 = const()[name = tensor<string, []>("transpose_106_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_107_perm_0 = const()[name = tensor<string, []>("transpose_107_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_169 = transpose(perm = transpose_107_perm_0, x = k_87_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_170 = transpose(perm = transpose_106_perm_0, x = q_87_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_43_cast = matmul(transpose_x = qk_43_transpose_x_0, transpose_y = qk_43_transpose_y_0, x = transpose_170, y = transpose_169); |
| tensor<fp16, [1, 20, 1500, 1500]> var_2434_cast = softmax(axis = var_2369, x = qk_43_cast); |
| tensor<bool, []> var_2436_transpose_x_0 = const()[name = tensor<string, []>("op_2436_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_2436_transpose_y_0 = const()[name = tensor<string, []>("op_2436_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_171 = transpose(perm = var_2430, x = var_2429_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_2436_cast = matmul(transpose_x = var_2436_transpose_x_0, transpose_y = var_2436_transpose_y_0, x = var_2434_cast, y = transpose_171); |
| tensor<int32, [4]> var_2437 = const()[name = tensor<string, []>("op_2437"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_21 = const()[name = tensor<string, []>("concat_21"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_168 = transpose(perm = var_2437, x = var_2436_cast); |
| tensor<fp16, [1, 1500, 1280]> x_263_cast = reshape(shape = concat_21, x = transpose_168); |
| tensor<fp16, [1280, 1280]> var_2442_to_fp16 = const()[name = tensor<string, []>("op_2442_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(850608000)))]; |
| tensor<fp16, [1280]> var_2443_to_fp16 = const()[name = tensor<string, []>("op_2443_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(853884864)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2444_cast = linear(bias = var_2443_to_fp16, weight = var_2442_to_fp16, x = x_263_cast); |
| tensor<fp16, [1, 1500, 1280]> x_265_cast = add(x = x_259_cast, y = var_2444_cast); |
| tensor<int32, [1]> var_2450_axes_0 = const()[name = tensor<string, []>("op_2450_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_21_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_21_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(853887488)))]; |
| tensor<fp16, [1280]> blocks_21_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_21_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(853890112)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2450_cast = layer_norm(axes = var_2450_axes_0, beta = blocks_21_mlp_ln_bias_to_fp16, epsilon = var_2375_to_fp16, gamma = blocks_21_mlp_ln_weight_to_fp16, x = x_265_cast); |
| tensor<fp16, [5120, 1280]> var_2459_to_fp16 = const()[name = tensor<string, []>("op_2459_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(853892736)))]; |
| tensor<fp16, [5120]> var_2460_to_fp16 = const()[name = tensor<string, []>("op_2460_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(867000000)))]; |
| tensor<fp16, [1, 1500, 5120]> input_177_cast = linear(bias = var_2460_to_fp16, weight = var_2459_to_fp16, x = var_2450_cast); |
| tensor<string, []> x_269_mode_0 = const()[name = tensor<string, []>("x_269_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_269_cast = gelu(mode = x_269_mode_0, x = input_177_cast); |
| tensor<fp16, [1280, 5120]> var_2465_to_fp16 = const()[name = tensor<string, []>("op_2465_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(867010304)))]; |
| tensor<fp16, [1280]> var_2466_to_fp16 = const()[name = tensor<string, []>("op_2466_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(880117568)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2467_cast = linear(bias = var_2466_to_fp16, weight = var_2465_to_fp16, x = x_269_cast); |
| tensor<fp16, [1, 1500, 1280]> x_271_cast = add(x = x_265_cast, y = var_2467_cast); |
| tensor<int32, []> var_2476 = const()[name = tensor<string, []>("op_2476"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_2493_axes_0 = const()[name = tensor<string, []>("op_2493_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_22_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_22_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(880120192)))]; |
| tensor<fp16, [1280]> blocks_22_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_22_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(880122816)))]; |
| tensor<fp16, []> var_2482_to_fp16 = const()[name = tensor<string, []>("op_2482_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_2493_cast = layer_norm(axes = var_2493_axes_0, beta = blocks_22_attn_ln_bias_to_fp16, epsilon = var_2482_to_fp16, gamma = blocks_22_attn_ln_weight_to_fp16, x = x_271_cast); |
| tensor<fp16, [1280, 1280]> var_2504_to_fp16 = const()[name = tensor<string, []>("op_2504_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(880125440)))]; |
| tensor<fp16, [1280]> var_2505_to_fp16 = const()[name = tensor<string, []>("op_2505_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(883402304)))]; |
| tensor<fp16, [1, 1500, 1280]> q_89_cast = linear(bias = var_2505_to_fp16, weight = var_2504_to_fp16, x = var_2493_cast); |
| tensor<fp16, [1280, 1280]> var_2508_to_fp16 = const()[name = tensor<string, []>("op_2508_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(883404928)))]; |
| tensor<fp16, [1280]> k_89_bias_0_to_fp16 = const()[name = tensor<string, []>("k_89_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(886681792)))]; |
| tensor<fp16, [1, 1500, 1280]> k_89_cast = linear(bias = k_89_bias_0_to_fp16, weight = var_2508_to_fp16, x = var_2493_cast); |
| tensor<fp16, [1280, 1280]> var_2512_to_fp16 = const()[name = tensor<string, []>("op_2512_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(886684416)))]; |
| tensor<fp16, [1280]> var_2513_to_fp16 = const()[name = tensor<string, []>("op_2513_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(889961280)))]; |
| tensor<fp16, [1, 1500, 1280]> v_89_cast = linear(bias = var_2513_to_fp16, weight = var_2512_to_fp16, x = var_2493_cast); |
| tensor<int32, [4]> var_2521 = const()[name = tensor<string, []>("op_2521"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2522_cast = reshape(shape = var_2521, x = q_89_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_268_to_fp16 = const()[name = tensor<string, []>("const_268_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_91_cast = mul(x = var_2522_cast, y = const_268_to_fp16); |
| tensor<int32, [4]> var_2528 = const()[name = tensor<string, []>("op_2528"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2529_cast = reshape(shape = var_2528, x = k_89_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_269_to_fp16 = const()[name = tensor<string, []>("const_269_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_91_cast = mul(x = var_2529_cast, y = const_269_to_fp16); |
| tensor<int32, [4]> var_2535 = const()[name = tensor<string, []>("op_2535"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2536_cast = reshape(shape = var_2535, x = v_89_cast); |
| tensor<int32, [4]> var_2537 = const()[name = tensor<string, []>("op_2537"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_45_transpose_x_0 = const()[name = tensor<string, []>("qk_45_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_45_transpose_y_0 = const()[name = tensor<string, []>("qk_45_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_108_perm_0 = const()[name = tensor<string, []>("transpose_108_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_109_perm_0 = const()[name = tensor<string, []>("transpose_109_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_165 = transpose(perm = transpose_109_perm_0, x = k_91_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_166 = transpose(perm = transpose_108_perm_0, x = q_91_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_45_cast = matmul(transpose_x = qk_45_transpose_x_0, transpose_y = qk_45_transpose_y_0, x = transpose_166, y = transpose_165); |
| tensor<fp16, [1, 20, 1500, 1500]> var_2541_cast = softmax(axis = var_2476, x = qk_45_cast); |
| tensor<bool, []> var_2543_transpose_x_0 = const()[name = tensor<string, []>("op_2543_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_2543_transpose_y_0 = const()[name = tensor<string, []>("op_2543_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_167 = transpose(perm = var_2537, x = var_2536_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_2543_cast = matmul(transpose_x = var_2543_transpose_x_0, transpose_y = var_2543_transpose_y_0, x = var_2541_cast, y = transpose_167); |
| tensor<int32, [4]> var_2544 = const()[name = tensor<string, []>("op_2544"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_22 = const()[name = tensor<string, []>("concat_22"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_164 = transpose(perm = var_2544, x = var_2543_cast); |
| tensor<fp16, [1, 1500, 1280]> x_275_cast = reshape(shape = concat_22, x = transpose_164); |
| tensor<fp16, [1280, 1280]> var_2549_to_fp16 = const()[name = tensor<string, []>("op_2549_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(889963904)))]; |
| tensor<fp16, [1280]> var_2550_to_fp16 = const()[name = tensor<string, []>("op_2550_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(893240768)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2551_cast = linear(bias = var_2550_to_fp16, weight = var_2549_to_fp16, x = x_275_cast); |
| tensor<fp16, [1, 1500, 1280]> x_277_cast = add(x = x_271_cast, y = var_2551_cast); |
| tensor<int32, [1]> var_2557_axes_0 = const()[name = tensor<string, []>("op_2557_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_22_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_22_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(893243392)))]; |
| tensor<fp16, [1280]> blocks_22_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_22_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(893246016)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2557_cast = layer_norm(axes = var_2557_axes_0, beta = blocks_22_mlp_ln_bias_to_fp16, epsilon = var_2482_to_fp16, gamma = blocks_22_mlp_ln_weight_to_fp16, x = x_277_cast); |
| tensor<fp16, [5120, 1280]> var_2566_to_fp16 = const()[name = tensor<string, []>("op_2566_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(893248640)))]; |
| tensor<fp16, [5120]> var_2567_to_fp16 = const()[name = tensor<string, []>("op_2567_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(906355904)))]; |
| tensor<fp16, [1, 1500, 5120]> input_185_cast = linear(bias = var_2567_to_fp16, weight = var_2566_to_fp16, x = var_2557_cast); |
| tensor<string, []> x_281_mode_0 = const()[name = tensor<string, []>("x_281_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_281_cast = gelu(mode = x_281_mode_0, x = input_185_cast); |
| tensor<fp16, [1280, 5120]> var_2572_to_fp16 = const()[name = tensor<string, []>("op_2572_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(906366208)))]; |
| tensor<fp16, [1280]> var_2573_to_fp16 = const()[name = tensor<string, []>("op_2573_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(919473472)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2574_cast = linear(bias = var_2573_to_fp16, weight = var_2572_to_fp16, x = x_281_cast); |
| tensor<fp16, [1, 1500, 1280]> x_283_cast = add(x = x_277_cast, y = var_2574_cast); |
| tensor<int32, []> var_2583 = const()[name = tensor<string, []>("op_2583"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_2600_axes_0 = const()[name = tensor<string, []>("op_2600_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_23_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_23_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(919476096)))]; |
| tensor<fp16, [1280]> blocks_23_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_23_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(919478720)))]; |
| tensor<fp16, []> var_2589_to_fp16 = const()[name = tensor<string, []>("op_2589_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_2600_cast = layer_norm(axes = var_2600_axes_0, beta = blocks_23_attn_ln_bias_to_fp16, epsilon = var_2589_to_fp16, gamma = blocks_23_attn_ln_weight_to_fp16, x = x_283_cast); |
| tensor<fp16, [1280, 1280]> var_2611_to_fp16 = const()[name = tensor<string, []>("op_2611_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(919481344)))]; |
| tensor<fp16, [1280]> var_2612_to_fp16 = const()[name = tensor<string, []>("op_2612_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(922758208)))]; |
| tensor<fp16, [1, 1500, 1280]> q_93_cast = linear(bias = var_2612_to_fp16, weight = var_2611_to_fp16, x = var_2600_cast); |
| tensor<fp16, [1280, 1280]> var_2615_to_fp16 = const()[name = tensor<string, []>("op_2615_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(922760832)))]; |
| tensor<fp16, [1280]> k_93_bias_0_to_fp16 = const()[name = tensor<string, []>("k_93_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(926037696)))]; |
| tensor<fp16, [1, 1500, 1280]> k_93_cast = linear(bias = k_93_bias_0_to_fp16, weight = var_2615_to_fp16, x = var_2600_cast); |
| tensor<fp16, [1280, 1280]> var_2619_to_fp16 = const()[name = tensor<string, []>("op_2619_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(926040320)))]; |
| tensor<fp16, [1280]> var_2620_to_fp16 = const()[name = tensor<string, []>("op_2620_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(929317184)))]; |
| tensor<fp16, [1, 1500, 1280]> v_93_cast = linear(bias = var_2620_to_fp16, weight = var_2619_to_fp16, x = var_2600_cast); |
| tensor<int32, [4]> var_2628 = const()[name = tensor<string, []>("op_2628"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2629_cast = reshape(shape = var_2628, x = q_93_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_270_to_fp16 = const()[name = tensor<string, []>("const_270_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_95_cast = mul(x = var_2629_cast, y = const_270_to_fp16); |
| tensor<int32, [4]> var_2635 = const()[name = tensor<string, []>("op_2635"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2636_cast = reshape(shape = var_2635, x = k_93_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_271_to_fp16 = const()[name = tensor<string, []>("const_271_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_95_cast = mul(x = var_2636_cast, y = const_271_to_fp16); |
| tensor<int32, [4]> var_2642 = const()[name = tensor<string, []>("op_2642"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2643_cast = reshape(shape = var_2642, x = v_93_cast); |
| tensor<int32, [4]> var_2644 = const()[name = tensor<string, []>("op_2644"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_47_transpose_x_0 = const()[name = tensor<string, []>("qk_47_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_47_transpose_y_0 = const()[name = tensor<string, []>("qk_47_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_110_perm_0 = const()[name = tensor<string, []>("transpose_110_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_111_perm_0 = const()[name = tensor<string, []>("transpose_111_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_161 = transpose(perm = transpose_111_perm_0, x = k_95_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_162 = transpose(perm = transpose_110_perm_0, x = q_95_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_47_cast = matmul(transpose_x = qk_47_transpose_x_0, transpose_y = qk_47_transpose_y_0, x = transpose_162, y = transpose_161); |
| tensor<fp16, [1, 20, 1500, 1500]> var_2648_cast = softmax(axis = var_2583, x = qk_47_cast); |
| tensor<bool, []> var_2650_transpose_x_0 = const()[name = tensor<string, []>("op_2650_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_2650_transpose_y_0 = const()[name = tensor<string, []>("op_2650_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_163 = transpose(perm = var_2644, x = var_2643_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_2650_cast = matmul(transpose_x = var_2650_transpose_x_0, transpose_y = var_2650_transpose_y_0, x = var_2648_cast, y = transpose_163); |
| tensor<int32, [4]> var_2651 = const()[name = tensor<string, []>("op_2651"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_23 = const()[name = tensor<string, []>("concat_23"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_160 = transpose(perm = var_2651, x = var_2650_cast); |
| tensor<fp16, [1, 1500, 1280]> x_287_cast = reshape(shape = concat_23, x = transpose_160); |
| tensor<fp16, [1280, 1280]> var_2656_to_fp16 = const()[name = tensor<string, []>("op_2656_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(929319808)))]; |
| tensor<fp16, [1280]> var_2657_to_fp16 = const()[name = tensor<string, []>("op_2657_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(932596672)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2658_cast = linear(bias = var_2657_to_fp16, weight = var_2656_to_fp16, x = x_287_cast); |
| tensor<fp16, [1, 1500, 1280]> x_289_cast = add(x = x_283_cast, y = var_2658_cast); |
| tensor<int32, [1]> var_2664_axes_0 = const()[name = tensor<string, []>("op_2664_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_23_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_23_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(932599296)))]; |
| tensor<fp16, [1280]> blocks_23_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_23_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(932601920)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2664_cast = layer_norm(axes = var_2664_axes_0, beta = blocks_23_mlp_ln_bias_to_fp16, epsilon = var_2589_to_fp16, gamma = blocks_23_mlp_ln_weight_to_fp16, x = x_289_cast); |
| tensor<fp16, [5120, 1280]> var_2673_to_fp16 = const()[name = tensor<string, []>("op_2673_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(932604544)))]; |
| tensor<fp16, [5120]> var_2674_to_fp16 = const()[name = tensor<string, []>("op_2674_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(945711808)))]; |
| tensor<fp16, [1, 1500, 5120]> input_193_cast = linear(bias = var_2674_to_fp16, weight = var_2673_to_fp16, x = var_2664_cast); |
| tensor<string, []> x_293_mode_0 = const()[name = tensor<string, []>("x_293_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_293_cast = gelu(mode = x_293_mode_0, x = input_193_cast); |
| tensor<fp16, [1280, 5120]> var_2679_to_fp16 = const()[name = tensor<string, []>("op_2679_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(945722112)))]; |
| tensor<fp16, [1280]> var_2680_to_fp16 = const()[name = tensor<string, []>("op_2680_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(958829376)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2681_cast = linear(bias = var_2680_to_fp16, weight = var_2679_to_fp16, x = x_293_cast); |
| tensor<fp16, [1, 1500, 1280]> x_295_cast = add(x = x_289_cast, y = var_2681_cast); |
| tensor<int32, []> var_2690 = const()[name = tensor<string, []>("op_2690"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_2707_axes_0 = const()[name = tensor<string, []>("op_2707_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_24_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_24_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(958832000)))]; |
| tensor<fp16, [1280]> blocks_24_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_24_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(958834624)))]; |
| tensor<fp16, []> var_2696_to_fp16 = const()[name = tensor<string, []>("op_2696_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_2707_cast = layer_norm(axes = var_2707_axes_0, beta = blocks_24_attn_ln_bias_to_fp16, epsilon = var_2696_to_fp16, gamma = blocks_24_attn_ln_weight_to_fp16, x = x_295_cast); |
| tensor<fp16, [1280, 1280]> var_2718_to_fp16 = const()[name = tensor<string, []>("op_2718_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(958837248)))]; |
| tensor<fp16, [1280]> var_2719_to_fp16 = const()[name = tensor<string, []>("op_2719_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(962114112)))]; |
| tensor<fp16, [1, 1500, 1280]> q_97_cast = linear(bias = var_2719_to_fp16, weight = var_2718_to_fp16, x = var_2707_cast); |
| tensor<fp16, [1280, 1280]> var_2722_to_fp16 = const()[name = tensor<string, []>("op_2722_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(962116736)))]; |
| tensor<fp16, [1280]> k_97_bias_0_to_fp16 = const()[name = tensor<string, []>("k_97_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(965393600)))]; |
| tensor<fp16, [1, 1500, 1280]> k_97_cast = linear(bias = k_97_bias_0_to_fp16, weight = var_2722_to_fp16, x = var_2707_cast); |
| tensor<fp16, [1280, 1280]> var_2726_to_fp16 = const()[name = tensor<string, []>("op_2726_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(965396224)))]; |
| tensor<fp16, [1280]> var_2727_to_fp16 = const()[name = tensor<string, []>("op_2727_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(968673088)))]; |
| tensor<fp16, [1, 1500, 1280]> v_97_cast = linear(bias = var_2727_to_fp16, weight = var_2726_to_fp16, x = var_2707_cast); |
| tensor<int32, [4]> var_2735 = const()[name = tensor<string, []>("op_2735"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2736_cast = reshape(shape = var_2735, x = q_97_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_272_to_fp16 = const()[name = tensor<string, []>("const_272_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_99_cast = mul(x = var_2736_cast, y = const_272_to_fp16); |
| tensor<int32, [4]> var_2742 = const()[name = tensor<string, []>("op_2742"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2743_cast = reshape(shape = var_2742, x = k_97_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_273_to_fp16 = const()[name = tensor<string, []>("const_273_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_99_cast = mul(x = var_2743_cast, y = const_273_to_fp16); |
| tensor<int32, [4]> var_2749 = const()[name = tensor<string, []>("op_2749"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2750_cast = reshape(shape = var_2749, x = v_97_cast); |
| tensor<int32, [4]> var_2751 = const()[name = tensor<string, []>("op_2751"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_49_transpose_x_0 = const()[name = tensor<string, []>("qk_49_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_49_transpose_y_0 = const()[name = tensor<string, []>("qk_49_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_112_perm_0 = const()[name = tensor<string, []>("transpose_112_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_113_perm_0 = const()[name = tensor<string, []>("transpose_113_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_157 = transpose(perm = transpose_113_perm_0, x = k_99_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_158 = transpose(perm = transpose_112_perm_0, x = q_99_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_49_cast = matmul(transpose_x = qk_49_transpose_x_0, transpose_y = qk_49_transpose_y_0, x = transpose_158, y = transpose_157); |
| tensor<fp16, [1, 20, 1500, 1500]> var_2755_cast = softmax(axis = var_2690, x = qk_49_cast); |
| tensor<bool, []> var_2757_transpose_x_0 = const()[name = tensor<string, []>("op_2757_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_2757_transpose_y_0 = const()[name = tensor<string, []>("op_2757_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_159 = transpose(perm = var_2751, x = var_2750_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_2757_cast = matmul(transpose_x = var_2757_transpose_x_0, transpose_y = var_2757_transpose_y_0, x = var_2755_cast, y = transpose_159); |
| tensor<int32, [4]> var_2758 = const()[name = tensor<string, []>("op_2758"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_24 = const()[name = tensor<string, []>("concat_24"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_156 = transpose(perm = var_2758, x = var_2757_cast); |
| tensor<fp16, [1, 1500, 1280]> x_299_cast = reshape(shape = concat_24, x = transpose_156); |
| tensor<fp16, [1280, 1280]> var_2763_to_fp16 = const()[name = tensor<string, []>("op_2763_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(968675712)))]; |
| tensor<fp16, [1280]> var_2764_to_fp16 = const()[name = tensor<string, []>("op_2764_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(971952576)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2765_cast = linear(bias = var_2764_to_fp16, weight = var_2763_to_fp16, x = x_299_cast); |
| tensor<fp16, [1, 1500, 1280]> x_301_cast = add(x = x_295_cast, y = var_2765_cast); |
| tensor<int32, [1]> var_2771_axes_0 = const()[name = tensor<string, []>("op_2771_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_24_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_24_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(971955200)))]; |
| tensor<fp16, [1280]> blocks_24_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_24_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(971957824)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2771_cast = layer_norm(axes = var_2771_axes_0, beta = blocks_24_mlp_ln_bias_to_fp16, epsilon = var_2696_to_fp16, gamma = blocks_24_mlp_ln_weight_to_fp16, x = x_301_cast); |
| tensor<fp16, [5120, 1280]> var_2780_to_fp16 = const()[name = tensor<string, []>("op_2780_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(971960448)))]; |
| tensor<fp16, [5120]> var_2781_to_fp16 = const()[name = tensor<string, []>("op_2781_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(985067712)))]; |
| tensor<fp16, [1, 1500, 5120]> input_201_cast = linear(bias = var_2781_to_fp16, weight = var_2780_to_fp16, x = var_2771_cast); |
| tensor<string, []> x_305_mode_0 = const()[name = tensor<string, []>("x_305_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_305_cast = gelu(mode = x_305_mode_0, x = input_201_cast); |
| tensor<fp16, [1280, 5120]> var_2786_to_fp16 = const()[name = tensor<string, []>("op_2786_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(985078016)))]; |
| tensor<fp16, [1280]> var_2787_to_fp16 = const()[name = tensor<string, []>("op_2787_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(998185280)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2788_cast = linear(bias = var_2787_to_fp16, weight = var_2786_to_fp16, x = x_305_cast); |
| tensor<fp16, [1, 1500, 1280]> x_307_cast = add(x = x_301_cast, y = var_2788_cast); |
| tensor<int32, []> var_2797 = const()[name = tensor<string, []>("op_2797"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_2814_axes_0 = const()[name = tensor<string, []>("op_2814_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_25_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_25_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(998187904)))]; |
| tensor<fp16, [1280]> blocks_25_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_25_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(998190528)))]; |
| tensor<fp16, []> var_2803_to_fp16 = const()[name = tensor<string, []>("op_2803_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_2814_cast = layer_norm(axes = var_2814_axes_0, beta = blocks_25_attn_ln_bias_to_fp16, epsilon = var_2803_to_fp16, gamma = blocks_25_attn_ln_weight_to_fp16, x = x_307_cast); |
| tensor<fp16, [1280, 1280]> var_2825_to_fp16 = const()[name = tensor<string, []>("op_2825_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(998193152)))]; |
| tensor<fp16, [1280]> var_2826_to_fp16 = const()[name = tensor<string, []>("op_2826_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1001470016)))]; |
| tensor<fp16, [1, 1500, 1280]> q_101_cast = linear(bias = var_2826_to_fp16, weight = var_2825_to_fp16, x = var_2814_cast); |
| tensor<fp16, [1280, 1280]> var_2829_to_fp16 = const()[name = tensor<string, []>("op_2829_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1001472640)))]; |
| tensor<fp16, [1280]> k_101_bias_0_to_fp16 = const()[name = tensor<string, []>("k_101_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1004749504)))]; |
| tensor<fp16, [1, 1500, 1280]> k_101_cast = linear(bias = k_101_bias_0_to_fp16, weight = var_2829_to_fp16, x = var_2814_cast); |
| tensor<fp16, [1280, 1280]> var_2833_to_fp16 = const()[name = tensor<string, []>("op_2833_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1004752128)))]; |
| tensor<fp16, [1280]> var_2834_to_fp16 = const()[name = tensor<string, []>("op_2834_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1008028992)))]; |
| tensor<fp16, [1, 1500, 1280]> v_101_cast = linear(bias = var_2834_to_fp16, weight = var_2833_to_fp16, x = var_2814_cast); |
| tensor<int32, [4]> var_2842 = const()[name = tensor<string, []>("op_2842"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2843_cast = reshape(shape = var_2842, x = q_101_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_274_to_fp16 = const()[name = tensor<string, []>("const_274_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_103_cast = mul(x = var_2843_cast, y = const_274_to_fp16); |
| tensor<int32, [4]> var_2849 = const()[name = tensor<string, []>("op_2849"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2850_cast = reshape(shape = var_2849, x = k_101_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_275_to_fp16 = const()[name = tensor<string, []>("const_275_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_103_cast = mul(x = var_2850_cast, y = const_275_to_fp16); |
| tensor<int32, [4]> var_2856 = const()[name = tensor<string, []>("op_2856"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2857_cast = reshape(shape = var_2856, x = v_101_cast); |
| tensor<int32, [4]> var_2858 = const()[name = tensor<string, []>("op_2858"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_51_transpose_x_0 = const()[name = tensor<string, []>("qk_51_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_51_transpose_y_0 = const()[name = tensor<string, []>("qk_51_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_114_perm_0 = const()[name = tensor<string, []>("transpose_114_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_115_perm_0 = const()[name = tensor<string, []>("transpose_115_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_153 = transpose(perm = transpose_115_perm_0, x = k_103_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_154 = transpose(perm = transpose_114_perm_0, x = q_103_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_51_cast = matmul(transpose_x = qk_51_transpose_x_0, transpose_y = qk_51_transpose_y_0, x = transpose_154, y = transpose_153); |
| tensor<fp16, [1, 20, 1500, 1500]> var_2862_cast = softmax(axis = var_2797, x = qk_51_cast); |
| tensor<bool, []> var_2864_transpose_x_0 = const()[name = tensor<string, []>("op_2864_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_2864_transpose_y_0 = const()[name = tensor<string, []>("op_2864_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_155 = transpose(perm = var_2858, x = var_2857_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_2864_cast = matmul(transpose_x = var_2864_transpose_x_0, transpose_y = var_2864_transpose_y_0, x = var_2862_cast, y = transpose_155); |
| tensor<int32, [4]> var_2865 = const()[name = tensor<string, []>("op_2865"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_25 = const()[name = tensor<string, []>("concat_25"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_152 = transpose(perm = var_2865, x = var_2864_cast); |
| tensor<fp16, [1, 1500, 1280]> x_311_cast = reshape(shape = concat_25, x = transpose_152); |
| tensor<fp16, [1280, 1280]> var_2870_to_fp16 = const()[name = tensor<string, []>("op_2870_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1008031616)))]; |
| tensor<fp16, [1280]> var_2871_to_fp16 = const()[name = tensor<string, []>("op_2871_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1011308480)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2872_cast = linear(bias = var_2871_to_fp16, weight = var_2870_to_fp16, x = x_311_cast); |
| tensor<fp16, [1, 1500, 1280]> x_313_cast = add(x = x_307_cast, y = var_2872_cast); |
| tensor<int32, [1]> var_2878_axes_0 = const()[name = tensor<string, []>("op_2878_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_25_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_25_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1011311104)))]; |
| tensor<fp16, [1280]> blocks_25_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_25_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1011313728)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2878_cast = layer_norm(axes = var_2878_axes_0, beta = blocks_25_mlp_ln_bias_to_fp16, epsilon = var_2803_to_fp16, gamma = blocks_25_mlp_ln_weight_to_fp16, x = x_313_cast); |
| tensor<fp16, [5120, 1280]> var_2887_to_fp16 = const()[name = tensor<string, []>("op_2887_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1011316352)))]; |
| tensor<fp16, [5120]> var_2888_to_fp16 = const()[name = tensor<string, []>("op_2888_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1024423616)))]; |
| tensor<fp16, [1, 1500, 5120]> input_209_cast = linear(bias = var_2888_to_fp16, weight = var_2887_to_fp16, x = var_2878_cast); |
| tensor<string, []> x_317_mode_0 = const()[name = tensor<string, []>("x_317_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_317_cast = gelu(mode = x_317_mode_0, x = input_209_cast); |
| tensor<fp16, [1280, 5120]> var_2893_to_fp16 = const()[name = tensor<string, []>("op_2893_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1024433920)))]; |
| tensor<fp16, [1280]> var_2894_to_fp16 = const()[name = tensor<string, []>("op_2894_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1037541184)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2895_cast = linear(bias = var_2894_to_fp16, weight = var_2893_to_fp16, x = x_317_cast); |
| tensor<fp16, [1, 1500, 1280]> x_319_cast = add(x = x_313_cast, y = var_2895_cast); |
| tensor<int32, []> var_2904 = const()[name = tensor<string, []>("op_2904"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_2921_axes_0 = const()[name = tensor<string, []>("op_2921_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_26_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_26_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1037543808)))]; |
| tensor<fp16, [1280]> blocks_26_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_26_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1037546432)))]; |
| tensor<fp16, []> var_2910_to_fp16 = const()[name = tensor<string, []>("op_2910_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_2921_cast = layer_norm(axes = var_2921_axes_0, beta = blocks_26_attn_ln_bias_to_fp16, epsilon = var_2910_to_fp16, gamma = blocks_26_attn_ln_weight_to_fp16, x = x_319_cast); |
| tensor<fp16, [1280, 1280]> var_2932_to_fp16 = const()[name = tensor<string, []>("op_2932_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1037549056)))]; |
| tensor<fp16, [1280]> var_2933_to_fp16 = const()[name = tensor<string, []>("op_2933_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1040825920)))]; |
| tensor<fp16, [1, 1500, 1280]> q_105_cast = linear(bias = var_2933_to_fp16, weight = var_2932_to_fp16, x = var_2921_cast); |
| tensor<fp16, [1280, 1280]> var_2936_to_fp16 = const()[name = tensor<string, []>("op_2936_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1040828544)))]; |
| tensor<fp16, [1280]> k_105_bias_0_to_fp16 = const()[name = tensor<string, []>("k_105_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1044105408)))]; |
| tensor<fp16, [1, 1500, 1280]> k_105_cast = linear(bias = k_105_bias_0_to_fp16, weight = var_2936_to_fp16, x = var_2921_cast); |
| tensor<fp16, [1280, 1280]> var_2940_to_fp16 = const()[name = tensor<string, []>("op_2940_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1044108032)))]; |
| tensor<fp16, [1280]> var_2941_to_fp16 = const()[name = tensor<string, []>("op_2941_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1047384896)))]; |
| tensor<fp16, [1, 1500, 1280]> v_105_cast = linear(bias = var_2941_to_fp16, weight = var_2940_to_fp16, x = var_2921_cast); |
| tensor<int32, [4]> var_2949 = const()[name = tensor<string, []>("op_2949"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2950_cast = reshape(shape = var_2949, x = q_105_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_276_to_fp16 = const()[name = tensor<string, []>("const_276_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_107_cast = mul(x = var_2950_cast, y = const_276_to_fp16); |
| tensor<int32, [4]> var_2956 = const()[name = tensor<string, []>("op_2956"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2957_cast = reshape(shape = var_2956, x = k_105_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_277_to_fp16 = const()[name = tensor<string, []>("const_277_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_107_cast = mul(x = var_2957_cast, y = const_277_to_fp16); |
| tensor<int32, [4]> var_2963 = const()[name = tensor<string, []>("op_2963"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_2964_cast = reshape(shape = var_2963, x = v_105_cast); |
| tensor<int32, [4]> var_2965 = const()[name = tensor<string, []>("op_2965"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_53_transpose_x_0 = const()[name = tensor<string, []>("qk_53_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_53_transpose_y_0 = const()[name = tensor<string, []>("qk_53_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_116_perm_0 = const()[name = tensor<string, []>("transpose_116_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_117_perm_0 = const()[name = tensor<string, []>("transpose_117_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_149 = transpose(perm = transpose_117_perm_0, x = k_107_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_150 = transpose(perm = transpose_116_perm_0, x = q_107_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_53_cast = matmul(transpose_x = qk_53_transpose_x_0, transpose_y = qk_53_transpose_y_0, x = transpose_150, y = transpose_149); |
| tensor<fp16, [1, 20, 1500, 1500]> var_2969_cast = softmax(axis = var_2904, x = qk_53_cast); |
| tensor<bool, []> var_2971_transpose_x_0 = const()[name = tensor<string, []>("op_2971_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_2971_transpose_y_0 = const()[name = tensor<string, []>("op_2971_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_151 = transpose(perm = var_2965, x = var_2964_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_2971_cast = matmul(transpose_x = var_2971_transpose_x_0, transpose_y = var_2971_transpose_y_0, x = var_2969_cast, y = transpose_151); |
| tensor<int32, [4]> var_2972 = const()[name = tensor<string, []>("op_2972"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_26 = const()[name = tensor<string, []>("concat_26"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_148 = transpose(perm = var_2972, x = var_2971_cast); |
| tensor<fp16, [1, 1500, 1280]> x_323_cast = reshape(shape = concat_26, x = transpose_148); |
| tensor<fp16, [1280, 1280]> var_2977_to_fp16 = const()[name = tensor<string, []>("op_2977_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1047387520)))]; |
| tensor<fp16, [1280]> var_2978_to_fp16 = const()[name = tensor<string, []>("op_2978_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1050664384)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2979_cast = linear(bias = var_2978_to_fp16, weight = var_2977_to_fp16, x = x_323_cast); |
| tensor<fp16, [1, 1500, 1280]> x_325_cast = add(x = x_319_cast, y = var_2979_cast); |
| tensor<int32, [1]> var_2985_axes_0 = const()[name = tensor<string, []>("op_2985_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_26_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_26_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1050667008)))]; |
| tensor<fp16, [1280]> blocks_26_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_26_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1050669632)))]; |
| tensor<fp16, [1, 1500, 1280]> var_2985_cast = layer_norm(axes = var_2985_axes_0, beta = blocks_26_mlp_ln_bias_to_fp16, epsilon = var_2910_to_fp16, gamma = blocks_26_mlp_ln_weight_to_fp16, x = x_325_cast); |
| tensor<fp16, [5120, 1280]> var_2994_to_fp16 = const()[name = tensor<string, []>("op_2994_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1050672256)))]; |
| tensor<fp16, [5120]> var_2995_to_fp16 = const()[name = tensor<string, []>("op_2995_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1063779520)))]; |
| tensor<fp16, [1, 1500, 5120]> input_217_cast = linear(bias = var_2995_to_fp16, weight = var_2994_to_fp16, x = var_2985_cast); |
| tensor<string, []> x_329_mode_0 = const()[name = tensor<string, []>("x_329_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_329_cast = gelu(mode = x_329_mode_0, x = input_217_cast); |
| tensor<fp16, [1280, 5120]> var_3000_to_fp16 = const()[name = tensor<string, []>("op_3000_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1063789824)))]; |
| tensor<fp16, [1280]> var_3001_to_fp16 = const()[name = tensor<string, []>("op_3001_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1076897088)))]; |
| tensor<fp16, [1, 1500, 1280]> var_3002_cast = linear(bias = var_3001_to_fp16, weight = var_3000_to_fp16, x = x_329_cast); |
| tensor<fp16, [1, 1500, 1280]> x_331_cast = add(x = x_325_cast, y = var_3002_cast); |
| tensor<int32, []> var_3011 = const()[name = tensor<string, []>("op_3011"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_3028_axes_0 = const()[name = tensor<string, []>("op_3028_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_27_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_27_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1076899712)))]; |
| tensor<fp16, [1280]> blocks_27_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_27_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1076902336)))]; |
| tensor<fp16, []> var_3017_to_fp16 = const()[name = tensor<string, []>("op_3017_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_3028_cast = layer_norm(axes = var_3028_axes_0, beta = blocks_27_attn_ln_bias_to_fp16, epsilon = var_3017_to_fp16, gamma = blocks_27_attn_ln_weight_to_fp16, x = x_331_cast); |
| tensor<fp16, [1280, 1280]> var_3039_to_fp16 = const()[name = tensor<string, []>("op_3039_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1076904960)))]; |
| tensor<fp16, [1280]> var_3040_to_fp16 = const()[name = tensor<string, []>("op_3040_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1080181824)))]; |
| tensor<fp16, [1, 1500, 1280]> q_109_cast = linear(bias = var_3040_to_fp16, weight = var_3039_to_fp16, x = var_3028_cast); |
| tensor<fp16, [1280, 1280]> var_3043_to_fp16 = const()[name = tensor<string, []>("op_3043_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1080184448)))]; |
| tensor<fp16, [1280]> k_109_bias_0_to_fp16 = const()[name = tensor<string, []>("k_109_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1083461312)))]; |
| tensor<fp16, [1, 1500, 1280]> k_109_cast = linear(bias = k_109_bias_0_to_fp16, weight = var_3043_to_fp16, x = var_3028_cast); |
| tensor<fp16, [1280, 1280]> var_3047_to_fp16 = const()[name = tensor<string, []>("op_3047_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1083463936)))]; |
| tensor<fp16, [1280]> var_3048_to_fp16 = const()[name = tensor<string, []>("op_3048_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1086740800)))]; |
| tensor<fp16, [1, 1500, 1280]> v_109_cast = linear(bias = var_3048_to_fp16, weight = var_3047_to_fp16, x = var_3028_cast); |
| tensor<int32, [4]> var_3056 = const()[name = tensor<string, []>("op_3056"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_3057_cast = reshape(shape = var_3056, x = q_109_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_278_to_fp16 = const()[name = tensor<string, []>("const_278_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_111_cast = mul(x = var_3057_cast, y = const_278_to_fp16); |
| tensor<int32, [4]> var_3063 = const()[name = tensor<string, []>("op_3063"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_3064_cast = reshape(shape = var_3063, x = k_109_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_279_to_fp16 = const()[name = tensor<string, []>("const_279_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_111_cast = mul(x = var_3064_cast, y = const_279_to_fp16); |
| tensor<int32, [4]> var_3070 = const()[name = tensor<string, []>("op_3070"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_3071_cast = reshape(shape = var_3070, x = v_109_cast); |
| tensor<int32, [4]> var_3072 = const()[name = tensor<string, []>("op_3072"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_55_transpose_x_0 = const()[name = tensor<string, []>("qk_55_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_55_transpose_y_0 = const()[name = tensor<string, []>("qk_55_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_118_perm_0 = const()[name = tensor<string, []>("transpose_118_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_119_perm_0 = const()[name = tensor<string, []>("transpose_119_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_145 = transpose(perm = transpose_119_perm_0, x = k_111_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_146 = transpose(perm = transpose_118_perm_0, x = q_111_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_55_cast = matmul(transpose_x = qk_55_transpose_x_0, transpose_y = qk_55_transpose_y_0, x = transpose_146, y = transpose_145); |
| tensor<fp16, [1, 20, 1500, 1500]> var_3076_cast = softmax(axis = var_3011, x = qk_55_cast); |
| tensor<bool, []> var_3078_transpose_x_0 = const()[name = tensor<string, []>("op_3078_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_3078_transpose_y_0 = const()[name = tensor<string, []>("op_3078_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_147 = transpose(perm = var_3072, x = var_3071_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_3078_cast = matmul(transpose_x = var_3078_transpose_x_0, transpose_y = var_3078_transpose_y_0, x = var_3076_cast, y = transpose_147); |
| tensor<int32, [4]> var_3079 = const()[name = tensor<string, []>("op_3079"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_27 = const()[name = tensor<string, []>("concat_27"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_144 = transpose(perm = var_3079, x = var_3078_cast); |
| tensor<fp16, [1, 1500, 1280]> x_335_cast = reshape(shape = concat_27, x = transpose_144); |
| tensor<fp16, [1280, 1280]> var_3084_to_fp16 = const()[name = tensor<string, []>("op_3084_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1086743424)))]; |
| tensor<fp16, [1280]> var_3085_to_fp16 = const()[name = tensor<string, []>("op_3085_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1090020288)))]; |
| tensor<fp16, [1, 1500, 1280]> var_3086_cast = linear(bias = var_3085_to_fp16, weight = var_3084_to_fp16, x = x_335_cast); |
| tensor<fp16, [1, 1500, 1280]> x_337_cast = add(x = x_331_cast, y = var_3086_cast); |
| tensor<int32, [1]> var_3092_axes_0 = const()[name = tensor<string, []>("op_3092_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_27_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_27_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1090022912)))]; |
| tensor<fp16, [1280]> blocks_27_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_27_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1090025536)))]; |
| tensor<fp16, [1, 1500, 1280]> var_3092_cast = layer_norm(axes = var_3092_axes_0, beta = blocks_27_mlp_ln_bias_to_fp16, epsilon = var_3017_to_fp16, gamma = blocks_27_mlp_ln_weight_to_fp16, x = x_337_cast); |
| tensor<fp16, [5120, 1280]> var_3101_to_fp16 = const()[name = tensor<string, []>("op_3101_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1090028160)))]; |
| tensor<fp16, [5120]> var_3102_to_fp16 = const()[name = tensor<string, []>("op_3102_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1103135424)))]; |
| tensor<fp16, [1, 1500, 5120]> input_225_cast = linear(bias = var_3102_to_fp16, weight = var_3101_to_fp16, x = var_3092_cast); |
| tensor<string, []> x_341_mode_0 = const()[name = tensor<string, []>("x_341_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_341_cast = gelu(mode = x_341_mode_0, x = input_225_cast); |
| tensor<fp16, [1280, 5120]> var_3107_to_fp16 = const()[name = tensor<string, []>("op_3107_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1103145728)))]; |
| tensor<fp16, [1280]> var_3108_to_fp16 = const()[name = tensor<string, []>("op_3108_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1116252992)))]; |
| tensor<fp16, [1, 1500, 1280]> var_3109_cast = linear(bias = var_3108_to_fp16, weight = var_3107_to_fp16, x = x_341_cast); |
| tensor<fp16, [1, 1500, 1280]> x_343_cast = add(x = x_337_cast, y = var_3109_cast); |
| tensor<int32, []> var_3118 = const()[name = tensor<string, []>("op_3118"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_3135_axes_0 = const()[name = tensor<string, []>("op_3135_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_28_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_28_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1116255616)))]; |
| tensor<fp16, [1280]> blocks_28_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_28_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1116258240)))]; |
| tensor<fp16, []> var_3124_to_fp16 = const()[name = tensor<string, []>("op_3124_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_3135_cast = layer_norm(axes = var_3135_axes_0, beta = blocks_28_attn_ln_bias_to_fp16, epsilon = var_3124_to_fp16, gamma = blocks_28_attn_ln_weight_to_fp16, x = x_343_cast); |
| tensor<fp16, [1280, 1280]> var_3146_to_fp16 = const()[name = tensor<string, []>("op_3146_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1116260864)))]; |
| tensor<fp16, [1280]> var_3147_to_fp16 = const()[name = tensor<string, []>("op_3147_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1119537728)))]; |
| tensor<fp16, [1, 1500, 1280]> q_113_cast = linear(bias = var_3147_to_fp16, weight = var_3146_to_fp16, x = var_3135_cast); |
| tensor<fp16, [1280, 1280]> var_3150_to_fp16 = const()[name = tensor<string, []>("op_3150_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1119540352)))]; |
| tensor<fp16, [1280]> k_113_bias_0_to_fp16 = const()[name = tensor<string, []>("k_113_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1122817216)))]; |
| tensor<fp16, [1, 1500, 1280]> k_113_cast = linear(bias = k_113_bias_0_to_fp16, weight = var_3150_to_fp16, x = var_3135_cast); |
| tensor<fp16, [1280, 1280]> var_3154_to_fp16 = const()[name = tensor<string, []>("op_3154_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1122819840)))]; |
| tensor<fp16, [1280]> var_3155_to_fp16 = const()[name = tensor<string, []>("op_3155_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1126096704)))]; |
| tensor<fp16, [1, 1500, 1280]> v_113_cast = linear(bias = var_3155_to_fp16, weight = var_3154_to_fp16, x = var_3135_cast); |
| tensor<int32, [4]> var_3163 = const()[name = tensor<string, []>("op_3163"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_3164_cast = reshape(shape = var_3163, x = q_113_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_280_to_fp16 = const()[name = tensor<string, []>("const_280_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_115_cast = mul(x = var_3164_cast, y = const_280_to_fp16); |
| tensor<int32, [4]> var_3170 = const()[name = tensor<string, []>("op_3170"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_3171_cast = reshape(shape = var_3170, x = k_113_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_281_to_fp16 = const()[name = tensor<string, []>("const_281_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_115_cast = mul(x = var_3171_cast, y = const_281_to_fp16); |
| tensor<int32, [4]> var_3177 = const()[name = tensor<string, []>("op_3177"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_3178_cast = reshape(shape = var_3177, x = v_113_cast); |
| tensor<int32, [4]> var_3179 = const()[name = tensor<string, []>("op_3179"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_57_transpose_x_0 = const()[name = tensor<string, []>("qk_57_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_57_transpose_y_0 = const()[name = tensor<string, []>("qk_57_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_120_perm_0 = const()[name = tensor<string, []>("transpose_120_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_121_perm_0 = const()[name = tensor<string, []>("transpose_121_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_141 = transpose(perm = transpose_121_perm_0, x = k_115_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_142 = transpose(perm = transpose_120_perm_0, x = q_115_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_57_cast = matmul(transpose_x = qk_57_transpose_x_0, transpose_y = qk_57_transpose_y_0, x = transpose_142, y = transpose_141); |
| tensor<fp16, [1, 20, 1500, 1500]> var_3183_cast = softmax(axis = var_3118, x = qk_57_cast); |
| tensor<bool, []> var_3185_transpose_x_0 = const()[name = tensor<string, []>("op_3185_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_3185_transpose_y_0 = const()[name = tensor<string, []>("op_3185_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_143 = transpose(perm = var_3179, x = var_3178_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_3185_cast = matmul(transpose_x = var_3185_transpose_x_0, transpose_y = var_3185_transpose_y_0, x = var_3183_cast, y = transpose_143); |
| tensor<int32, [4]> var_3186 = const()[name = tensor<string, []>("op_3186"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_28 = const()[name = tensor<string, []>("concat_28"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_140 = transpose(perm = var_3186, x = var_3185_cast); |
| tensor<fp16, [1, 1500, 1280]> x_347_cast = reshape(shape = concat_28, x = transpose_140); |
| tensor<fp16, [1280, 1280]> var_3191_to_fp16 = const()[name = tensor<string, []>("op_3191_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1126099328)))]; |
| tensor<fp16, [1280]> var_3192_to_fp16 = const()[name = tensor<string, []>("op_3192_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1129376192)))]; |
| tensor<fp16, [1, 1500, 1280]> var_3193_cast = linear(bias = var_3192_to_fp16, weight = var_3191_to_fp16, x = x_347_cast); |
| tensor<fp16, [1, 1500, 1280]> x_349_cast = add(x = x_343_cast, y = var_3193_cast); |
| tensor<int32, [1]> var_3199_axes_0 = const()[name = tensor<string, []>("op_3199_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_28_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_28_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1129378816)))]; |
| tensor<fp16, [1280]> blocks_28_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_28_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1129381440)))]; |
| tensor<fp16, [1, 1500, 1280]> var_3199_cast = layer_norm(axes = var_3199_axes_0, beta = blocks_28_mlp_ln_bias_to_fp16, epsilon = var_3124_to_fp16, gamma = blocks_28_mlp_ln_weight_to_fp16, x = x_349_cast); |
| tensor<fp16, [5120, 1280]> var_3208_to_fp16 = const()[name = tensor<string, []>("op_3208_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1129384064)))]; |
| tensor<fp16, [5120]> var_3209_to_fp16 = const()[name = tensor<string, []>("op_3209_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1142491328)))]; |
| tensor<fp16, [1, 1500, 5120]> input_233_cast = linear(bias = var_3209_to_fp16, weight = var_3208_to_fp16, x = var_3199_cast); |
| tensor<string, []> x_353_mode_0 = const()[name = tensor<string, []>("x_353_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_353_cast = gelu(mode = x_353_mode_0, x = input_233_cast); |
| tensor<fp16, [1280, 5120]> var_3214_to_fp16 = const()[name = tensor<string, []>("op_3214_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1142501632)))]; |
| tensor<fp16, [1280]> var_3215_to_fp16 = const()[name = tensor<string, []>("op_3215_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1155608896)))]; |
| tensor<fp16, [1, 1500, 1280]> var_3216_cast = linear(bias = var_3215_to_fp16, weight = var_3214_to_fp16, x = x_353_cast); |
| tensor<fp16, [1, 1500, 1280]> x_355_cast = add(x = x_349_cast, y = var_3216_cast); |
| tensor<int32, []> var_3225 = const()[name = tensor<string, []>("op_3225"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_3242_axes_0 = const()[name = tensor<string, []>("op_3242_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_29_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_29_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1155611520)))]; |
| tensor<fp16, [1280]> blocks_29_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_29_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1155614144)))]; |
| tensor<fp16, []> var_3231_to_fp16 = const()[name = tensor<string, []>("op_3231_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_3242_cast = layer_norm(axes = var_3242_axes_0, beta = blocks_29_attn_ln_bias_to_fp16, epsilon = var_3231_to_fp16, gamma = blocks_29_attn_ln_weight_to_fp16, x = x_355_cast); |
| tensor<fp16, [1280, 1280]> var_3253_to_fp16 = const()[name = tensor<string, []>("op_3253_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1155616768)))]; |
| tensor<fp16, [1280]> var_3254_to_fp16 = const()[name = tensor<string, []>("op_3254_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1158893632)))]; |
| tensor<fp16, [1, 1500, 1280]> q_117_cast = linear(bias = var_3254_to_fp16, weight = var_3253_to_fp16, x = var_3242_cast); |
| tensor<fp16, [1280, 1280]> var_3257_to_fp16 = const()[name = tensor<string, []>("op_3257_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1158896256)))]; |
| tensor<fp16, [1280]> k_117_bias_0_to_fp16 = const()[name = tensor<string, []>("k_117_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1162173120)))]; |
| tensor<fp16, [1, 1500, 1280]> k_117_cast = linear(bias = k_117_bias_0_to_fp16, weight = var_3257_to_fp16, x = var_3242_cast); |
| tensor<fp16, [1280, 1280]> var_3261_to_fp16 = const()[name = tensor<string, []>("op_3261_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1162175744)))]; |
| tensor<fp16, [1280]> var_3262_to_fp16 = const()[name = tensor<string, []>("op_3262_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1165452608)))]; |
| tensor<fp16, [1, 1500, 1280]> v_117_cast = linear(bias = var_3262_to_fp16, weight = var_3261_to_fp16, x = var_3242_cast); |
| tensor<int32, [4]> var_3270 = const()[name = tensor<string, []>("op_3270"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_3271_cast = reshape(shape = var_3270, x = q_117_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_282_to_fp16 = const()[name = tensor<string, []>("const_282_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_119_cast = mul(x = var_3271_cast, y = const_282_to_fp16); |
| tensor<int32, [4]> var_3277 = const()[name = tensor<string, []>("op_3277"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_3278_cast = reshape(shape = var_3277, x = k_117_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_283_to_fp16 = const()[name = tensor<string, []>("const_283_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_119_cast = mul(x = var_3278_cast, y = const_283_to_fp16); |
| tensor<int32, [4]> var_3284 = const()[name = tensor<string, []>("op_3284"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_3285_cast = reshape(shape = var_3284, x = v_117_cast); |
| tensor<int32, [4]> var_3286 = const()[name = tensor<string, []>("op_3286"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_59_transpose_x_0 = const()[name = tensor<string, []>("qk_59_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_59_transpose_y_0 = const()[name = tensor<string, []>("qk_59_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_122_perm_0 = const()[name = tensor<string, []>("transpose_122_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_123_perm_0 = const()[name = tensor<string, []>("transpose_123_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_137 = transpose(perm = transpose_123_perm_0, x = k_119_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_138 = transpose(perm = transpose_122_perm_0, x = q_119_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_59_cast = matmul(transpose_x = qk_59_transpose_x_0, transpose_y = qk_59_transpose_y_0, x = transpose_138, y = transpose_137); |
| tensor<fp16, [1, 20, 1500, 1500]> var_3290_cast = softmax(axis = var_3225, x = qk_59_cast); |
| tensor<bool, []> var_3292_transpose_x_0 = const()[name = tensor<string, []>("op_3292_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_3292_transpose_y_0 = const()[name = tensor<string, []>("op_3292_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_139 = transpose(perm = var_3286, x = var_3285_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_3292_cast = matmul(transpose_x = var_3292_transpose_x_0, transpose_y = var_3292_transpose_y_0, x = var_3290_cast, y = transpose_139); |
| tensor<int32, [4]> var_3293 = const()[name = tensor<string, []>("op_3293"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_29 = const()[name = tensor<string, []>("concat_29"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_136 = transpose(perm = var_3293, x = var_3292_cast); |
| tensor<fp16, [1, 1500, 1280]> x_359_cast = reshape(shape = concat_29, x = transpose_136); |
| tensor<fp16, [1280, 1280]> var_3298_to_fp16 = const()[name = tensor<string, []>("op_3298_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1165455232)))]; |
| tensor<fp16, [1280]> var_3299_to_fp16 = const()[name = tensor<string, []>("op_3299_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1168732096)))]; |
| tensor<fp16, [1, 1500, 1280]> var_3300_cast = linear(bias = var_3299_to_fp16, weight = var_3298_to_fp16, x = x_359_cast); |
| tensor<fp16, [1, 1500, 1280]> x_361_cast = add(x = x_355_cast, y = var_3300_cast); |
| tensor<int32, [1]> var_3306_axes_0 = const()[name = tensor<string, []>("op_3306_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_29_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_29_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1168734720)))]; |
| tensor<fp16, [1280]> blocks_29_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_29_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1168737344)))]; |
| tensor<fp16, [1, 1500, 1280]> var_3306_cast = layer_norm(axes = var_3306_axes_0, beta = blocks_29_mlp_ln_bias_to_fp16, epsilon = var_3231_to_fp16, gamma = blocks_29_mlp_ln_weight_to_fp16, x = x_361_cast); |
| tensor<fp16, [5120, 1280]> var_3315_to_fp16 = const()[name = tensor<string, []>("op_3315_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1168739968)))]; |
| tensor<fp16, [5120]> var_3316_to_fp16 = const()[name = tensor<string, []>("op_3316_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1181847232)))]; |
| tensor<fp16, [1, 1500, 5120]> input_241_cast = linear(bias = var_3316_to_fp16, weight = var_3315_to_fp16, x = var_3306_cast); |
| tensor<string, []> x_365_mode_0 = const()[name = tensor<string, []>("x_365_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_365_cast = gelu(mode = x_365_mode_0, x = input_241_cast); |
| tensor<fp16, [1280, 5120]> var_3321_to_fp16 = const()[name = tensor<string, []>("op_3321_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1181857536)))]; |
| tensor<fp16, [1280]> var_3322_to_fp16 = const()[name = tensor<string, []>("op_3322_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1194964800)))]; |
| tensor<fp16, [1, 1500, 1280]> var_3323_cast = linear(bias = var_3322_to_fp16, weight = var_3321_to_fp16, x = x_365_cast); |
| tensor<fp16, [1, 1500, 1280]> x_367_cast = add(x = x_361_cast, y = var_3323_cast); |
| tensor<int32, []> var_3332 = const()[name = tensor<string, []>("op_3332"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_3349_axes_0 = const()[name = tensor<string, []>("op_3349_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_30_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_30_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1194967424)))]; |
| tensor<fp16, [1280]> blocks_30_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_30_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1194970048)))]; |
| tensor<fp16, []> var_3338_to_fp16 = const()[name = tensor<string, []>("op_3338_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_3349_cast = layer_norm(axes = var_3349_axes_0, beta = blocks_30_attn_ln_bias_to_fp16, epsilon = var_3338_to_fp16, gamma = blocks_30_attn_ln_weight_to_fp16, x = x_367_cast); |
| tensor<fp16, [1280, 1280]> var_3360_to_fp16 = const()[name = tensor<string, []>("op_3360_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1194972672)))]; |
| tensor<fp16, [1280]> var_3361_to_fp16 = const()[name = tensor<string, []>("op_3361_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1198249536)))]; |
| tensor<fp16, [1, 1500, 1280]> q_121_cast = linear(bias = var_3361_to_fp16, weight = var_3360_to_fp16, x = var_3349_cast); |
| tensor<fp16, [1280, 1280]> var_3364_to_fp16 = const()[name = tensor<string, []>("op_3364_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1198252160)))]; |
| tensor<fp16, [1280]> k_121_bias_0_to_fp16 = const()[name = tensor<string, []>("k_121_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1201529024)))]; |
| tensor<fp16, [1, 1500, 1280]> k_121_cast = linear(bias = k_121_bias_0_to_fp16, weight = var_3364_to_fp16, x = var_3349_cast); |
| tensor<fp16, [1280, 1280]> var_3368_to_fp16 = const()[name = tensor<string, []>("op_3368_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1201531648)))]; |
| tensor<fp16, [1280]> var_3369_to_fp16 = const()[name = tensor<string, []>("op_3369_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1204808512)))]; |
| tensor<fp16, [1, 1500, 1280]> v_121_cast = linear(bias = var_3369_to_fp16, weight = var_3368_to_fp16, x = var_3349_cast); |
| tensor<int32, [4]> var_3377 = const()[name = tensor<string, []>("op_3377"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_3378_cast = reshape(shape = var_3377, x = q_121_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_284_to_fp16 = const()[name = tensor<string, []>("const_284_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_123_cast = mul(x = var_3378_cast, y = const_284_to_fp16); |
| tensor<int32, [4]> var_3384 = const()[name = tensor<string, []>("op_3384"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_3385_cast = reshape(shape = var_3384, x = k_121_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_285_to_fp16 = const()[name = tensor<string, []>("const_285_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_123_cast = mul(x = var_3385_cast, y = const_285_to_fp16); |
| tensor<int32, [4]> var_3391 = const()[name = tensor<string, []>("op_3391"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_3392_cast = reshape(shape = var_3391, x = v_121_cast); |
| tensor<int32, [4]> var_3393 = const()[name = tensor<string, []>("op_3393"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_61_transpose_x_0 = const()[name = tensor<string, []>("qk_61_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_61_transpose_y_0 = const()[name = tensor<string, []>("qk_61_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_124_perm_0 = const()[name = tensor<string, []>("transpose_124_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_125_perm_0 = const()[name = tensor<string, []>("transpose_125_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_133 = transpose(perm = transpose_125_perm_0, x = k_123_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_134 = transpose(perm = transpose_124_perm_0, x = q_123_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_61_cast = matmul(transpose_x = qk_61_transpose_x_0, transpose_y = qk_61_transpose_y_0, x = transpose_134, y = transpose_133); |
| tensor<fp16, [1, 20, 1500, 1500]> var_3397_cast = softmax(axis = var_3332, x = qk_61_cast); |
| tensor<bool, []> var_3399_transpose_x_0 = const()[name = tensor<string, []>("op_3399_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_3399_transpose_y_0 = const()[name = tensor<string, []>("op_3399_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_135 = transpose(perm = var_3393, x = var_3392_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_3399_cast = matmul(transpose_x = var_3399_transpose_x_0, transpose_y = var_3399_transpose_y_0, x = var_3397_cast, y = transpose_135); |
| tensor<int32, [4]> var_3400 = const()[name = tensor<string, []>("op_3400"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_30 = const()[name = tensor<string, []>("concat_30"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_132 = transpose(perm = var_3400, x = var_3399_cast); |
| tensor<fp16, [1, 1500, 1280]> x_371_cast = reshape(shape = concat_30, x = transpose_132); |
| tensor<fp16, [1280, 1280]> var_3405_to_fp16 = const()[name = tensor<string, []>("op_3405_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1204811136)))]; |
| tensor<fp16, [1280]> var_3406_to_fp16 = const()[name = tensor<string, []>("op_3406_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1208088000)))]; |
| tensor<fp16, [1, 1500, 1280]> var_3407_cast = linear(bias = var_3406_to_fp16, weight = var_3405_to_fp16, x = x_371_cast); |
| tensor<fp16, [1, 1500, 1280]> x_373_cast = add(x = x_367_cast, y = var_3407_cast); |
| tensor<int32, [1]> var_3413_axes_0 = const()[name = tensor<string, []>("op_3413_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_30_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_30_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1208090624)))]; |
| tensor<fp16, [1280]> blocks_30_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_30_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1208093248)))]; |
| tensor<fp16, [1, 1500, 1280]> var_3413_cast = layer_norm(axes = var_3413_axes_0, beta = blocks_30_mlp_ln_bias_to_fp16, epsilon = var_3338_to_fp16, gamma = blocks_30_mlp_ln_weight_to_fp16, x = x_373_cast); |
| tensor<fp16, [5120, 1280]> var_3422_to_fp16 = const()[name = tensor<string, []>("op_3422_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1208095872)))]; |
| tensor<fp16, [5120]> var_3423_to_fp16 = const()[name = tensor<string, []>("op_3423_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1221203136)))]; |
| tensor<fp16, [1, 1500, 5120]> input_249_cast = linear(bias = var_3423_to_fp16, weight = var_3422_to_fp16, x = var_3413_cast); |
| tensor<string, []> x_377_mode_0 = const()[name = tensor<string, []>("x_377_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_377_cast = gelu(mode = x_377_mode_0, x = input_249_cast); |
| tensor<fp16, [1280, 5120]> var_3428_to_fp16 = const()[name = tensor<string, []>("op_3428_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1221213440)))]; |
| tensor<fp16, [1280]> var_3429_to_fp16 = const()[name = tensor<string, []>("op_3429_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1234320704)))]; |
| tensor<fp16, [1, 1500, 1280]> var_3430_cast = linear(bias = var_3429_to_fp16, weight = var_3428_to_fp16, x = x_377_cast); |
| tensor<fp16, [1, 1500, 1280]> x_379_cast = add(x = x_373_cast, y = var_3430_cast); |
| tensor<int32, []> var_3439 = const()[name = tensor<string, []>("op_3439"), val = tensor<int32, []>(-1)]; |
| tensor<int32, [1]> var_3456_axes_0 = const()[name = tensor<string, []>("op_3456_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_31_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_31_attn_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1234323328)))]; |
| tensor<fp16, [1280]> blocks_31_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_31_attn_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1234325952)))]; |
| tensor<fp16, []> var_3445_to_fp16 = const()[name = tensor<string, []>("op_3445_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_3456_cast = layer_norm(axes = var_3456_axes_0, beta = blocks_31_attn_ln_bias_to_fp16, epsilon = var_3445_to_fp16, gamma = blocks_31_attn_ln_weight_to_fp16, x = x_379_cast); |
| tensor<fp16, [1280, 1280]> var_3467_to_fp16 = const()[name = tensor<string, []>("op_3467_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1234328576)))]; |
| tensor<fp16, [1280]> var_3468_to_fp16 = const()[name = tensor<string, []>("op_3468_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1237605440)))]; |
| tensor<fp16, [1, 1500, 1280]> q_125_cast = linear(bias = var_3468_to_fp16, weight = var_3467_to_fp16, x = var_3456_cast); |
| tensor<fp16, [1280, 1280]> var_3471_to_fp16 = const()[name = tensor<string, []>("op_3471_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1237608064)))]; |
| tensor<fp16, [1280]> k_125_bias_0_to_fp16 = const()[name = tensor<string, []>("k_125_bias_0_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1240884928)))]; |
| tensor<fp16, [1, 1500, 1280]> k_125_cast = linear(bias = k_125_bias_0_to_fp16, weight = var_3471_to_fp16, x = var_3456_cast); |
| tensor<fp16, [1280, 1280]> var_3475_to_fp16 = const()[name = tensor<string, []>("op_3475_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1240887552)))]; |
| tensor<fp16, [1280]> var_3476_to_fp16 = const()[name = tensor<string, []>("op_3476_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1244164416)))]; |
| tensor<fp16, [1, 1500, 1280]> v_125_cast = linear(bias = var_3476_to_fp16, weight = var_3475_to_fp16, x = var_3456_cast); |
| tensor<int32, [4]> var_3484 = const()[name = tensor<string, []>("op_3484"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_3485_cast = reshape(shape = var_3484, x = q_125_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_286_to_fp16 = const()[name = tensor<string, []>("const_286_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> q_cast = mul(x = var_3485_cast, y = const_286_to_fp16); |
| tensor<int32, [4]> var_3491 = const()[name = tensor<string, []>("op_3491"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_3492_cast = reshape(shape = var_3491, x = k_125_cast); |
| tensor<fp16, [1, 1, 1, 1]> const_287_to_fp16 = const()[name = tensor<string, []>("const_287_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; |
| tensor<fp16, [1, 1500, 20, 64]> k_cast = mul(x = var_3492_cast, y = const_287_to_fp16); |
| tensor<int32, [4]> var_3498 = const()[name = tensor<string, []>("op_3498"), val = tensor<int32, [4]>([1, 1500, 20, -1])]; |
| tensor<fp16, [1, 1500, 20, 64]> var_3499_cast = reshape(shape = var_3498, x = v_125_cast); |
| tensor<int32, [4]> var_3500 = const()[name = tensor<string, []>("op_3500"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<bool, []> qk_transpose_x_0 = const()[name = tensor<string, []>("qk_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> qk_transpose_y_0 = const()[name = tensor<string, []>("qk_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<int32, [4]> transpose_126_perm_0 = const()[name = tensor<string, []>("transpose_126_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [4]> transpose_127_perm_0 = const()[name = tensor<string, []>("transpose_127_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; |
| tensor<fp16, [1, 20, 64, 1500]> transpose_129 = transpose(perm = transpose_127_perm_0, x = k_cast); |
| tensor<fp16, [1, 20, 1500, 64]> transpose_130 = transpose(perm = transpose_126_perm_0, x = q_cast); |
| tensor<fp16, [1, 20, 1500, 1500]> qk_cast = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_130, y = transpose_129); |
| tensor<fp16, [1, 20, 1500, 1500]> var_3504_cast = softmax(axis = var_3439, x = qk_cast); |
| tensor<bool, []> var_3506_transpose_x_0 = const()[name = tensor<string, []>("op_3506_transpose_x_0"), val = tensor<bool, []>(false)]; |
| tensor<bool, []> var_3506_transpose_y_0 = const()[name = tensor<string, []>("op_3506_transpose_y_0"), val = tensor<bool, []>(false)]; |
| tensor<fp16, [1, 20, 1500, 64]> transpose_131 = transpose(perm = var_3500, x = var_3499_cast); |
| tensor<fp16, [1, 20, 1500, 64]> var_3506_cast = matmul(transpose_x = var_3506_transpose_x_0, transpose_y = var_3506_transpose_y_0, x = var_3504_cast, y = transpose_131); |
| tensor<int32, [4]> var_3507 = const()[name = tensor<string, []>("op_3507"), val = tensor<int32, [4]>([0, 2, 1, 3])]; |
| tensor<int32, [3]> concat_31 = const()[name = tensor<string, []>("concat_31"), val = tensor<int32, [3]>([1, 1500, 1280])]; |
| tensor<fp16, [1, 1500, 20, 64]> transpose_128 = transpose(perm = var_3507, x = var_3506_cast); |
| tensor<fp16, [1, 1500, 1280]> x_383_cast = reshape(shape = concat_31, x = transpose_128); |
| tensor<fp16, [1280, 1280]> var_3512_to_fp16 = const()[name = tensor<string, []>("op_3512_to_fp16"), val = tensor<fp16, [1280, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1244167040)))]; |
| tensor<fp16, [1280]> var_3513_to_fp16 = const()[name = tensor<string, []>("op_3513_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1247443904)))]; |
| tensor<fp16, [1, 1500, 1280]> var_3514_cast = linear(bias = var_3513_to_fp16, weight = var_3512_to_fp16, x = x_383_cast); |
| tensor<fp16, [1, 1500, 1280]> x_385_cast = add(x = x_379_cast, y = var_3514_cast); |
| tensor<int32, [1]> var_3520_axes_0 = const()[name = tensor<string, []>("op_3520_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> blocks_31_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_31_mlp_ln_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1247446528)))]; |
| tensor<fp16, [1280]> blocks_31_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_31_mlp_ln_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1247449152)))]; |
| tensor<fp16, [1, 1500, 1280]> var_3520_cast = layer_norm(axes = var_3520_axes_0, beta = blocks_31_mlp_ln_bias_to_fp16, epsilon = var_3445_to_fp16, gamma = blocks_31_mlp_ln_weight_to_fp16, x = x_385_cast); |
| tensor<fp16, [5120, 1280]> var_3529_to_fp16 = const()[name = tensor<string, []>("op_3529_to_fp16"), val = tensor<fp16, [5120, 1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1247451776)))]; |
| tensor<fp16, [5120]> var_3530_to_fp16 = const()[name = tensor<string, []>("op_3530_to_fp16"), val = tensor<fp16, [5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1260559040)))]; |
| tensor<fp16, [1, 1500, 5120]> input_257_cast = linear(bias = var_3530_to_fp16, weight = var_3529_to_fp16, x = var_3520_cast); |
| tensor<string, []> x_389_mode_0 = const()[name = tensor<string, []>("x_389_mode_0"), val = tensor<string, []>("EXACT")]; |
| tensor<fp16, [1, 1500, 5120]> x_389_cast = gelu(mode = x_389_mode_0, x = input_257_cast); |
| tensor<fp16, [1280, 5120]> var_3535_to_fp16 = const()[name = tensor<string, []>("op_3535_to_fp16"), val = tensor<fp16, [1280, 5120]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1260569344)))]; |
| tensor<fp16, [1280]> var_3536_to_fp16 = const()[name = tensor<string, []>("op_3536_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1273676608)))]; |
| tensor<fp16, [1, 1500, 1280]> var_3537_cast = linear(bias = var_3536_to_fp16, weight = var_3535_to_fp16, x = x_389_cast); |
| tensor<fp16, [1, 1500, 1280]> x_cast = add(x = x_385_cast, y = var_3537_cast); |
| tensor<int32, [1]> var_3550_axes_0 = const()[name = tensor<string, []>("op_3550_axes_0"), val = tensor<int32, [1]>([-1])]; |
| tensor<fp16, [1280]> ln_post_weight_to_fp16 = const()[name = tensor<string, []>("ln_post_weight_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1273679232)))]; |
| tensor<fp16, [1280]> ln_post_bias_to_fp16 = const()[name = tensor<string, []>("ln_post_bias_to_fp16"), val = tensor<fp16, [1280]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(1273681856)))]; |
| tensor<fp16, []> var_3541_to_fp16 = const()[name = tensor<string, []>("op_3541_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; |
| tensor<fp16, [1, 1500, 1280]> var_3550_cast = layer_norm(axes = var_3550_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_3541_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast); |
| tensor<string, []> var_3550_cast_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_3550_cast_to_fp32_dtype_0"), val = tensor<string, []>("fp32")]; |
| tensor<fp32, [1, 1500, 1280]> output = cast(dtype = var_3550_cast_to_fp32_dtype_0, x = var_3550_cast); |
| } -> (output); |
| } |
