| 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_32 = const()[name = tensor<string, []>("op_32"), val = tensor<int32, []>(1)]; | |
| tensor<int32, [1]> var_40 = const()[name = tensor<string, []>("op_40"), val = tensor<int32, [1]>([1])]; | |
| tensor<int32, [1]> var_42 = const()[name = tensor<string, []>("op_42"), val = tensor<int32, [1]>([1])]; | |
| tensor<string, []> var_44_pad_type_0 = const()[name = tensor<string, []>("op_44_pad_type_0"), val = tensor<string, []>("custom")]; | |
| tensor<int32, [2]> var_44_pad_0 = const()[name = tensor<string, []>("op_44_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, [768, 80, 3]> weight_3_to_fp16 = const()[name = tensor<string, []>("weight_3_to_fp16"), val = tensor<fp16, [768, 80, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64)))]; | |
| tensor<fp16, [768]> bias_3_to_fp16 = const()[name = tensor<string, []>("bias_3_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(368768)))]; | |
| tensor<fp16, [1, 80, 3000]> cast_367 = cast(dtype = logmel_data_to_fp16_dtype_0, x = logmel_data); | |
| tensor<fp16, [1, 768, 3000]> var_44_cast = conv(bias = bias_3_to_fp16, dilations = var_42, groups = var_32, pad = var_44_pad_0, pad_type = var_44_pad_type_0, strides = var_40, weight = weight_3_to_fp16, x = cast_367); | |
| tensor<string, []> input_1_mode_0 = const()[name = tensor<string, []>("input_1_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 768, 3000]> input_1_cast = gelu(mode = input_1_mode_0, x = var_44_cast); | |
| tensor<int32, []> var_48 = const()[name = tensor<string, []>("op_48"), val = tensor<int32, []>(1)]; | |
| tensor<int32, [1]> var_57 = const()[name = tensor<string, []>("op_57"), val = tensor<int32, [1]>([2])]; | |
| tensor<int32, [1]> var_59 = const()[name = tensor<string, []>("op_59"), val = tensor<int32, [1]>([1])]; | |
| tensor<string, []> var_61_pad_type_0 = const()[name = tensor<string, []>("op_61_pad_type_0"), val = tensor<string, []>("custom")]; | |
| tensor<int32, [2]> var_61_pad_0 = const()[name = tensor<string, []>("op_61_pad_0"), val = tensor<int32, [2]>([1, 1])]; | |
| tensor<fp16, [768, 768, 3]> weight_7_to_fp16 = const()[name = tensor<string, []>("weight_7_to_fp16"), val = tensor<fp16, [768, 768, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(370368)))]; | |
| tensor<fp16, [768]> bias_7_to_fp16 = const()[name = tensor<string, []>("bias_7_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3909376)))]; | |
| tensor<fp16, [1, 768, 1500]> var_61_cast = conv(bias = bias_7_to_fp16, dilations = var_59, groups = var_48, pad = var_61_pad_0, pad_type = var_61_pad_type_0, strides = var_57, 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, 768, 1500]> x_3_cast = gelu(mode = x_3_mode_0, x = var_61_cast); | |
| tensor<int32, [3]> var_66 = const()[name = tensor<string, []>("op_66"), val = tensor<int32, [3]>([0, 2, 1])]; | |
| tensor<fp16, [1500, 768]> positional_embedding_to_fp16 = const()[name = tensor<string, []>("positional_embedding_to_fp16"), val = tensor<fp16, [1500, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3910976)))]; | |
| tensor<fp16, [1, 1500, 768]> transpose_96 = transpose(perm = var_66, x = x_3_cast); | |
| tensor<fp16, [1, 1500, 768]> var_69_cast = add(x = transpose_96, y = positional_embedding_to_fp16); | |
| tensor<int32, []> var_82 = const()[name = tensor<string, []>("op_82"), val = tensor<int32, []>(-1)]; | |
| tensor<int32, [1]> var_99_axes_0 = const()[name = tensor<string, []>("op_99_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_0_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_0_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6215040)))]; | |
| tensor<fp16, [768]> blocks_0_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_0_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6216640)))]; | |
| tensor<fp16, []> var_88_to_fp16 = const()[name = tensor<string, []>("op_88_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_99_cast = layer_norm(axes = var_99_axes_0, beta = blocks_0_attn_ln_bias_to_fp16, epsilon = var_88_to_fp16, gamma = blocks_0_attn_ln_weight_to_fp16, x = var_69_cast); | |
| tensor<fp16, [768, 768]> var_110_to_fp16 = const()[name = tensor<string, []>("op_110_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6218240)))]; | |
| tensor<fp16, [768]> var_111_to_fp16 = const()[name = tensor<string, []>("op_111_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7397952)))]; | |
| tensor<fp16, [1, 1500, 768]> q_1_cast = linear(bias = var_111_to_fp16, weight = var_110_to_fp16, x = var_99_cast); | |
| tensor<fp16, [768, 768]> var_114_to_fp16 = const()[name = tensor<string, []>("op_114_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7399552)))]; | |
| tensor<fp16, [768]> k_1_bias_0_to_fp16 = const()[name = tensor<string, []>("k_1_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8579264)))]; | |
| tensor<fp16, [1, 1500, 768]> k_1_cast = linear(bias = k_1_bias_0_to_fp16, weight = var_114_to_fp16, x = var_99_cast); | |
| tensor<fp16, [768, 768]> var_118_to_fp16 = const()[name = tensor<string, []>("op_118_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8580864)))]; | |
| tensor<fp16, [768]> var_119_to_fp16 = const()[name = tensor<string, []>("op_119_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9760576)))]; | |
| tensor<fp16, [1, 1500, 768]> v_1_cast = linear(bias = var_119_to_fp16, weight = var_118_to_fp16, x = var_99_cast); | |
| tensor<int32, [4]> var_127 = const()[name = tensor<string, []>("op_127"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_128_cast = reshape(shape = var_127, x = q_1_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_84_to_fp16 = const()[name = tensor<string, []>("const_84_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_3_cast = mul(x = var_128_cast, y = const_84_to_fp16); | |
| tensor<int32, [4]> var_134 = const()[name = tensor<string, []>("op_134"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_135_cast = reshape(shape = var_134, x = k_1_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_85_to_fp16 = const()[name = tensor<string, []>("const_85_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_3_cast = mul(x = var_135_cast, y = const_85_to_fp16); | |
| tensor<int32, [4]> var_141 = const()[name = tensor<string, []>("op_141"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_142_cast = reshape(shape = var_141, x = v_1_cast); | |
| tensor<int32, [4]> var_143 = const()[name = tensor<string, []>("op_143"), 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_24_perm_0 = const()[name = tensor<string, []>("transpose_24_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> transpose_25_perm_0 = const()[name = tensor<string, []>("transpose_25_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_93 = transpose(perm = transpose_25_perm_0, x = k_3_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_94 = transpose(perm = transpose_24_perm_0, x = q_3_cast); | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_1_cast = matmul(transpose_x = qk_1_transpose_x_0, transpose_y = qk_1_transpose_y_0, x = transpose_94, y = transpose_93); | |
| tensor<fp16, [1, 12, 1500, 1500]> var_147_cast = softmax(axis = var_82, x = qk_1_cast); | |
| tensor<bool, []> var_149_transpose_x_0 = const()[name = tensor<string, []>("op_149_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> var_149_transpose_y_0 = const()[name = tensor<string, []>("op_149_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_95 = transpose(perm = var_143, x = var_142_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> var_149_cast = matmul(transpose_x = var_149_transpose_x_0, transpose_y = var_149_transpose_y_0, x = var_147_cast, y = transpose_95); | |
| tensor<int32, [4]> var_150 = const()[name = tensor<string, []>("op_150"), 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, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> transpose_92 = transpose(perm = var_150, x = var_149_cast); | |
| tensor<fp16, [1, 1500, 768]> x_11_cast = reshape(shape = concat_0, x = transpose_92); | |
| tensor<fp16, [768, 768]> var_155_to_fp16 = const()[name = tensor<string, []>("op_155_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9762176)))]; | |
| tensor<fp16, [768]> var_156_to_fp16 = const()[name = tensor<string, []>("op_156_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10941888)))]; | |
| tensor<fp16, [1, 1500, 768]> var_157_cast = linear(bias = var_156_to_fp16, weight = var_155_to_fp16, x = x_11_cast); | |
| tensor<fp16, [1, 1500, 768]> x_13_cast = add(x = var_69_cast, y = var_157_cast); | |
| tensor<int32, [1]> var_163_axes_0 = const()[name = tensor<string, []>("op_163_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_0_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_0_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10943488)))]; | |
| tensor<fp16, [768]> blocks_0_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_0_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10945088)))]; | |
| tensor<fp16, [1, 1500, 768]> var_163_cast = layer_norm(axes = var_163_axes_0, beta = blocks_0_mlp_ln_bias_to_fp16, epsilon = var_88_to_fp16, gamma = blocks_0_mlp_ln_weight_to_fp16, x = x_13_cast); | |
| tensor<fp16, [3072, 768]> var_172_to_fp16 = const()[name = tensor<string, []>("op_172_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10946688)))]; | |
| tensor<fp16, [3072]> var_173_to_fp16 = const()[name = tensor<string, []>("op_173_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15665344)))]; | |
| tensor<fp16, [1, 1500, 3072]> input_9_cast = linear(bias = var_173_to_fp16, weight = var_172_to_fp16, x = var_163_cast); | |
| tensor<string, []> x_17_mode_0 = const()[name = tensor<string, []>("x_17_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_17_cast = gelu(mode = x_17_mode_0, x = input_9_cast); | |
| tensor<fp16, [768, 3072]> var_178_to_fp16 = const()[name = tensor<string, []>("op_178_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15671552)))]; | |
| tensor<fp16, [768]> var_179_to_fp16 = const()[name = tensor<string, []>("op_179_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20390208)))]; | |
| tensor<fp16, [1, 1500, 768]> var_180_cast = linear(bias = var_179_to_fp16, weight = var_178_to_fp16, x = x_17_cast); | |
| tensor<fp16, [1, 1500, 768]> x_19_cast = add(x = x_13_cast, y = var_180_cast); | |
| tensor<int32, []> var_189 = const()[name = tensor<string, []>("op_189"), val = tensor<int32, []>(-1)]; | |
| tensor<int32, [1]> var_206_axes_0 = const()[name = tensor<string, []>("op_206_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_1_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_1_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20391808)))]; | |
| tensor<fp16, [768]> blocks_1_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_1_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20393408)))]; | |
| tensor<fp16, []> var_195_to_fp16 = const()[name = tensor<string, []>("op_195_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_206_cast = layer_norm(axes = var_206_axes_0, beta = blocks_1_attn_ln_bias_to_fp16, epsilon = var_195_to_fp16, gamma = blocks_1_attn_ln_weight_to_fp16, x = x_19_cast); | |
| tensor<fp16, [768, 768]> var_217_to_fp16 = const()[name = tensor<string, []>("op_217_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20395008)))]; | |
| tensor<fp16, [768]> var_218_to_fp16 = const()[name = tensor<string, []>("op_218_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21574720)))]; | |
| tensor<fp16, [1, 1500, 768]> q_5_cast = linear(bias = var_218_to_fp16, weight = var_217_to_fp16, x = var_206_cast); | |
| tensor<fp16, [768, 768]> var_221_to_fp16 = const()[name = tensor<string, []>("op_221_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21576320)))]; | |
| tensor<fp16, [768]> k_5_bias_0_to_fp16 = const()[name = tensor<string, []>("k_5_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22756032)))]; | |
| tensor<fp16, [1, 1500, 768]> k_5_cast = linear(bias = k_5_bias_0_to_fp16, weight = var_221_to_fp16, x = var_206_cast); | |
| tensor<fp16, [768, 768]> var_225_to_fp16 = const()[name = tensor<string, []>("op_225_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22757632)))]; | |
| tensor<fp16, [768]> var_226_to_fp16 = const()[name = tensor<string, []>("op_226_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23937344)))]; | |
| tensor<fp16, [1, 1500, 768]> v_5_cast = linear(bias = var_226_to_fp16, weight = var_225_to_fp16, x = var_206_cast); | |
| tensor<int32, [4]> var_234 = const()[name = tensor<string, []>("op_234"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_235_cast = reshape(shape = var_234, x = q_5_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_86_to_fp16 = const()[name = tensor<string, []>("const_86_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_7_cast = mul(x = var_235_cast, y = const_86_to_fp16); | |
| tensor<int32, [4]> var_241 = const()[name = tensor<string, []>("op_241"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_242_cast = reshape(shape = var_241, x = k_5_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_87_to_fp16 = const()[name = tensor<string, []>("const_87_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_7_cast = mul(x = var_242_cast, y = const_87_to_fp16); | |
| tensor<int32, [4]> var_248 = const()[name = tensor<string, []>("op_248"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_249_cast = reshape(shape = var_248, x = v_5_cast); | |
| tensor<int32, [4]> var_250 = const()[name = tensor<string, []>("op_250"), 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_26_perm_0 = const()[name = tensor<string, []>("transpose_26_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> transpose_27_perm_0 = const()[name = tensor<string, []>("transpose_27_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_89 = transpose(perm = transpose_27_perm_0, x = k_7_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_90 = transpose(perm = transpose_26_perm_0, x = q_7_cast); | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_3_cast = matmul(transpose_x = qk_3_transpose_x_0, transpose_y = qk_3_transpose_y_0, x = transpose_90, y = transpose_89); | |
| tensor<fp16, [1, 12, 1500, 1500]> var_254_cast = softmax(axis = var_189, x = qk_3_cast); | |
| tensor<bool, []> var_256_transpose_x_0 = const()[name = tensor<string, []>("op_256_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> var_256_transpose_y_0 = const()[name = tensor<string, []>("op_256_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_91 = transpose(perm = var_250, x = var_249_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> var_256_cast = matmul(transpose_x = var_256_transpose_x_0, transpose_y = var_256_transpose_y_0, x = var_254_cast, y = transpose_91); | |
| tensor<int32, [4]> var_257 = const()[name = tensor<string, []>("op_257"), 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, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> transpose_88 = transpose(perm = var_257, x = var_256_cast); | |
| tensor<fp16, [1, 1500, 768]> x_23_cast = reshape(shape = concat_1, x = transpose_88); | |
| tensor<fp16, [768, 768]> var_262_to_fp16 = const()[name = tensor<string, []>("op_262_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23938944)))]; | |
| tensor<fp16, [768]> var_263_to_fp16 = const()[name = tensor<string, []>("op_263_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25118656)))]; | |
| tensor<fp16, [1, 1500, 768]> var_264_cast = linear(bias = var_263_to_fp16, weight = var_262_to_fp16, x = x_23_cast); | |
| tensor<fp16, [1, 1500, 768]> x_25_cast = add(x = x_19_cast, y = var_264_cast); | |
| tensor<int32, [1]> var_270_axes_0 = const()[name = tensor<string, []>("op_270_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_1_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_1_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25120256)))]; | |
| tensor<fp16, [768]> blocks_1_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_1_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25121856)))]; | |
| tensor<fp16, [1, 1500, 768]> var_270_cast = layer_norm(axes = var_270_axes_0, beta = blocks_1_mlp_ln_bias_to_fp16, epsilon = var_195_to_fp16, gamma = blocks_1_mlp_ln_weight_to_fp16, x = x_25_cast); | |
| tensor<fp16, [3072, 768]> var_279_to_fp16 = const()[name = tensor<string, []>("op_279_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(25123456)))]; | |
| tensor<fp16, [3072]> var_280_to_fp16 = const()[name = tensor<string, []>("op_280_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29842112)))]; | |
| tensor<fp16, [1, 1500, 3072]> input_17_cast = linear(bias = var_280_to_fp16, weight = var_279_to_fp16, x = var_270_cast); | |
| tensor<string, []> x_29_mode_0 = const()[name = tensor<string, []>("x_29_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_29_cast = gelu(mode = x_29_mode_0, x = input_17_cast); | |
| tensor<fp16, [768, 3072]> var_285_to_fp16 = const()[name = tensor<string, []>("op_285_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29848320)))]; | |
| tensor<fp16, [768]> var_286_to_fp16 = const()[name = tensor<string, []>("op_286_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34566976)))]; | |
| tensor<fp16, [1, 1500, 768]> var_287_cast = linear(bias = var_286_to_fp16, weight = var_285_to_fp16, x = x_29_cast); | |
| tensor<fp16, [1, 1500, 768]> x_31_cast = add(x = x_25_cast, y = var_287_cast); | |
| tensor<int32, []> var_296 = const()[name = tensor<string, []>("op_296"), val = tensor<int32, []>(-1)]; | |
| tensor<int32, [1]> var_313_axes_0 = const()[name = tensor<string, []>("op_313_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_2_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_2_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34568576)))]; | |
| tensor<fp16, [768]> blocks_2_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_2_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34570176)))]; | |
| tensor<fp16, []> var_302_to_fp16 = const()[name = tensor<string, []>("op_302_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_313_cast = layer_norm(axes = var_313_axes_0, beta = blocks_2_attn_ln_bias_to_fp16, epsilon = var_302_to_fp16, gamma = blocks_2_attn_ln_weight_to_fp16, x = x_31_cast); | |
| tensor<fp16, [768, 768]> var_324_to_fp16 = const()[name = tensor<string, []>("op_324_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34571776)))]; | |
| tensor<fp16, [768]> var_325_to_fp16 = const()[name = tensor<string, []>("op_325_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35751488)))]; | |
| tensor<fp16, [1, 1500, 768]> q_9_cast = linear(bias = var_325_to_fp16, weight = var_324_to_fp16, x = var_313_cast); | |
| tensor<fp16, [768, 768]> var_328_to_fp16 = const()[name = tensor<string, []>("op_328_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35753088)))]; | |
| tensor<fp16, [768]> k_9_bias_0_to_fp16 = const()[name = tensor<string, []>("k_9_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36932800)))]; | |
| tensor<fp16, [1, 1500, 768]> k_9_cast = linear(bias = k_9_bias_0_to_fp16, weight = var_328_to_fp16, x = var_313_cast); | |
| tensor<fp16, [768, 768]> var_332_to_fp16 = const()[name = tensor<string, []>("op_332_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36934400)))]; | |
| tensor<fp16, [768]> var_333_to_fp16 = const()[name = tensor<string, []>("op_333_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38114112)))]; | |
| tensor<fp16, [1, 1500, 768]> v_9_cast = linear(bias = var_333_to_fp16, weight = var_332_to_fp16, x = var_313_cast); | |
| tensor<int32, [4]> var_341 = const()[name = tensor<string, []>("op_341"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_342_cast = reshape(shape = var_341, x = q_9_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_88_to_fp16 = const()[name = tensor<string, []>("const_88_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_11_cast = mul(x = var_342_cast, y = const_88_to_fp16); | |
| tensor<int32, [4]> var_348 = const()[name = tensor<string, []>("op_348"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_349_cast = reshape(shape = var_348, x = k_9_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_89_to_fp16 = const()[name = tensor<string, []>("const_89_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_11_cast = mul(x = var_349_cast, y = const_89_to_fp16); | |
| tensor<int32, [4]> var_355 = const()[name = tensor<string, []>("op_355"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_356_cast = reshape(shape = var_355, x = v_9_cast); | |
| tensor<int32, [4]> var_357 = const()[name = tensor<string, []>("op_357"), 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_28_perm_0 = const()[name = tensor<string, []>("transpose_28_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> transpose_29_perm_0 = const()[name = tensor<string, []>("transpose_29_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_85 = transpose(perm = transpose_29_perm_0, x = k_11_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_86 = transpose(perm = transpose_28_perm_0, x = q_11_cast); | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_5_cast = matmul(transpose_x = qk_5_transpose_x_0, transpose_y = qk_5_transpose_y_0, x = transpose_86, y = transpose_85); | |
| tensor<fp16, [1, 12, 1500, 1500]> var_361_cast = softmax(axis = var_296, x = qk_5_cast); | |
| tensor<bool, []> var_363_transpose_x_0 = const()[name = tensor<string, []>("op_363_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> var_363_transpose_y_0 = const()[name = tensor<string, []>("op_363_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_87 = transpose(perm = var_357, x = var_356_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> var_363_cast = matmul(transpose_x = var_363_transpose_x_0, transpose_y = var_363_transpose_y_0, x = var_361_cast, y = transpose_87); | |
| tensor<int32, [4]> var_364 = const()[name = tensor<string, []>("op_364"), 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, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> transpose_84 = transpose(perm = var_364, x = var_363_cast); | |
| tensor<fp16, [1, 1500, 768]> x_35_cast = reshape(shape = concat_2, x = transpose_84); | |
| tensor<fp16, [768, 768]> var_369_to_fp16 = const()[name = tensor<string, []>("op_369_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38115712)))]; | |
| tensor<fp16, [768]> var_370_to_fp16 = const()[name = tensor<string, []>("op_370_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39295424)))]; | |
| tensor<fp16, [1, 1500, 768]> var_371_cast = linear(bias = var_370_to_fp16, weight = var_369_to_fp16, x = x_35_cast); | |
| tensor<fp16, [1, 1500, 768]> x_37_cast = add(x = x_31_cast, y = var_371_cast); | |
| tensor<int32, [1]> var_377_axes_0 = const()[name = tensor<string, []>("op_377_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_2_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_2_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39297024)))]; | |
| tensor<fp16, [768]> blocks_2_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_2_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39298624)))]; | |
| tensor<fp16, [1, 1500, 768]> var_377_cast = layer_norm(axes = var_377_axes_0, beta = blocks_2_mlp_ln_bias_to_fp16, epsilon = var_302_to_fp16, gamma = blocks_2_mlp_ln_weight_to_fp16, x = x_37_cast); | |
| tensor<fp16, [3072, 768]> var_386_to_fp16 = const()[name = tensor<string, []>("op_386_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39300224)))]; | |
| tensor<fp16, [3072]> var_387_to_fp16 = const()[name = tensor<string, []>("op_387_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44018880)))]; | |
| tensor<fp16, [1, 1500, 3072]> input_25_cast = linear(bias = var_387_to_fp16, weight = var_386_to_fp16, x = var_377_cast); | |
| tensor<string, []> x_41_mode_0 = const()[name = tensor<string, []>("x_41_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_41_cast = gelu(mode = x_41_mode_0, x = input_25_cast); | |
| tensor<fp16, [768, 3072]> var_392_to_fp16 = const()[name = tensor<string, []>("op_392_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44025088)))]; | |
| tensor<fp16, [768]> var_393_to_fp16 = const()[name = tensor<string, []>("op_393_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48743744)))]; | |
| tensor<fp16, [1, 1500, 768]> var_394_cast = linear(bias = var_393_to_fp16, weight = var_392_to_fp16, x = x_41_cast); | |
| tensor<fp16, [1, 1500, 768]> x_43_cast = add(x = x_37_cast, y = var_394_cast); | |
| tensor<int32, []> var_403 = const()[name = tensor<string, []>("op_403"), val = tensor<int32, []>(-1)]; | |
| tensor<int32, [1]> var_420_axes_0 = const()[name = tensor<string, []>("op_420_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_3_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_3_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48745344)))]; | |
| tensor<fp16, [768]> blocks_3_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_3_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48746944)))]; | |
| tensor<fp16, []> var_409_to_fp16 = const()[name = tensor<string, []>("op_409_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_420_cast = layer_norm(axes = var_420_axes_0, beta = blocks_3_attn_ln_bias_to_fp16, epsilon = var_409_to_fp16, gamma = blocks_3_attn_ln_weight_to_fp16, x = x_43_cast); | |
| tensor<fp16, [768, 768]> var_431_to_fp16 = const()[name = tensor<string, []>("op_431_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48748544)))]; | |
| tensor<fp16, [768]> var_432_to_fp16 = const()[name = tensor<string, []>("op_432_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49928256)))]; | |
| tensor<fp16, [1, 1500, 768]> q_13_cast = linear(bias = var_432_to_fp16, weight = var_431_to_fp16, x = var_420_cast); | |
| tensor<fp16, [768, 768]> var_435_to_fp16 = const()[name = tensor<string, []>("op_435_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49929856)))]; | |
| tensor<fp16, [768]> k_13_bias_0_to_fp16 = const()[name = tensor<string, []>("k_13_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51109568)))]; | |
| tensor<fp16, [1, 1500, 768]> k_13_cast = linear(bias = k_13_bias_0_to_fp16, weight = var_435_to_fp16, x = var_420_cast); | |
| tensor<fp16, [768, 768]> var_439_to_fp16 = const()[name = tensor<string, []>("op_439_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51111168)))]; | |
| tensor<fp16, [768]> var_440_to_fp16 = const()[name = tensor<string, []>("op_440_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52290880)))]; | |
| tensor<fp16, [1, 1500, 768]> v_13_cast = linear(bias = var_440_to_fp16, weight = var_439_to_fp16, x = var_420_cast); | |
| tensor<int32, [4]> var_448 = const()[name = tensor<string, []>("op_448"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_449_cast = reshape(shape = var_448, x = q_13_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_90_to_fp16 = const()[name = tensor<string, []>("const_90_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_15_cast = mul(x = var_449_cast, y = const_90_to_fp16); | |
| tensor<int32, [4]> var_455 = const()[name = tensor<string, []>("op_455"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_456_cast = reshape(shape = var_455, x = k_13_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_91_to_fp16 = const()[name = tensor<string, []>("const_91_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_15_cast = mul(x = var_456_cast, y = const_91_to_fp16); | |
| tensor<int32, [4]> var_462 = const()[name = tensor<string, []>("op_462"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_463_cast = reshape(shape = var_462, x = v_13_cast); | |
| tensor<int32, [4]> var_464 = const()[name = tensor<string, []>("op_464"), 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_30_perm_0 = const()[name = tensor<string, []>("transpose_30_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> transpose_31_perm_0 = const()[name = tensor<string, []>("transpose_31_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_81 = transpose(perm = transpose_31_perm_0, x = k_15_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_82 = transpose(perm = transpose_30_perm_0, x = q_15_cast); | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_7_cast = matmul(transpose_x = qk_7_transpose_x_0, transpose_y = qk_7_transpose_y_0, x = transpose_82, y = transpose_81); | |
| tensor<fp16, [1, 12, 1500, 1500]> var_468_cast = softmax(axis = var_403, x = qk_7_cast); | |
| tensor<bool, []> var_470_transpose_x_0 = const()[name = tensor<string, []>("op_470_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> var_470_transpose_y_0 = const()[name = tensor<string, []>("op_470_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_83 = transpose(perm = var_464, x = var_463_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> var_470_cast = matmul(transpose_x = var_470_transpose_x_0, transpose_y = var_470_transpose_y_0, x = var_468_cast, y = transpose_83); | |
| tensor<int32, [4]> var_471 = const()[name = tensor<string, []>("op_471"), 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, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> transpose_80 = transpose(perm = var_471, x = var_470_cast); | |
| tensor<fp16, [1, 1500, 768]> x_47_cast = reshape(shape = concat_3, x = transpose_80); | |
| tensor<fp16, [768, 768]> var_476_to_fp16 = const()[name = tensor<string, []>("op_476_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(52292480)))]; | |
| tensor<fp16, [768]> var_477_to_fp16 = const()[name = tensor<string, []>("op_477_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53472192)))]; | |
| tensor<fp16, [1, 1500, 768]> var_478_cast = linear(bias = var_477_to_fp16, weight = var_476_to_fp16, x = x_47_cast); | |
| tensor<fp16, [1, 1500, 768]> x_49_cast = add(x = x_43_cast, y = var_478_cast); | |
| tensor<int32, [1]> var_484_axes_0 = const()[name = tensor<string, []>("op_484_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_3_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_3_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53473792)))]; | |
| tensor<fp16, [768]> blocks_3_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_3_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53475392)))]; | |
| tensor<fp16, [1, 1500, 768]> var_484_cast = layer_norm(axes = var_484_axes_0, beta = blocks_3_mlp_ln_bias_to_fp16, epsilon = var_409_to_fp16, gamma = blocks_3_mlp_ln_weight_to_fp16, x = x_49_cast); | |
| tensor<fp16, [3072, 768]> var_493_to_fp16 = const()[name = tensor<string, []>("op_493_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53476992)))]; | |
| tensor<fp16, [3072]> var_494_to_fp16 = const()[name = tensor<string, []>("op_494_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58195648)))]; | |
| tensor<fp16, [1, 1500, 3072]> input_33_cast = linear(bias = var_494_to_fp16, weight = var_493_to_fp16, x = var_484_cast); | |
| tensor<string, []> x_53_mode_0 = const()[name = tensor<string, []>("x_53_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_53_cast = gelu(mode = x_53_mode_0, x = input_33_cast); | |
| tensor<fp16, [768, 3072]> var_499_to_fp16 = const()[name = tensor<string, []>("op_499_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58201856)))]; | |
| tensor<fp16, [768]> var_500_to_fp16 = const()[name = tensor<string, []>("op_500_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62920512)))]; | |
| tensor<fp16, [1, 1500, 768]> var_501_cast = linear(bias = var_500_to_fp16, weight = var_499_to_fp16, x = x_53_cast); | |
| tensor<fp16, [1, 1500, 768]> x_55_cast = add(x = x_49_cast, y = var_501_cast); | |
| tensor<int32, []> var_510 = const()[name = tensor<string, []>("op_510"), val = tensor<int32, []>(-1)]; | |
| tensor<int32, [1]> var_527_axes_0 = const()[name = tensor<string, []>("op_527_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_4_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_4_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62922112)))]; | |
| tensor<fp16, [768]> blocks_4_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_4_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62923712)))]; | |
| tensor<fp16, []> var_516_to_fp16 = const()[name = tensor<string, []>("op_516_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_527_cast = layer_norm(axes = var_527_axes_0, beta = blocks_4_attn_ln_bias_to_fp16, epsilon = var_516_to_fp16, gamma = blocks_4_attn_ln_weight_to_fp16, x = x_55_cast); | |
| tensor<fp16, [768, 768]> var_538_to_fp16 = const()[name = tensor<string, []>("op_538_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62925312)))]; | |
| tensor<fp16, [768]> var_539_to_fp16 = const()[name = tensor<string, []>("op_539_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64105024)))]; | |
| tensor<fp16, [1, 1500, 768]> q_17_cast = linear(bias = var_539_to_fp16, weight = var_538_to_fp16, x = var_527_cast); | |
| tensor<fp16, [768, 768]> var_542_to_fp16 = const()[name = tensor<string, []>("op_542_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64106624)))]; | |
| tensor<fp16, [768]> k_17_bias_0_to_fp16 = const()[name = tensor<string, []>("k_17_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65286336)))]; | |
| tensor<fp16, [1, 1500, 768]> k_17_cast = linear(bias = k_17_bias_0_to_fp16, weight = var_542_to_fp16, x = var_527_cast); | |
| tensor<fp16, [768, 768]> var_546_to_fp16 = const()[name = tensor<string, []>("op_546_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65287936)))]; | |
| tensor<fp16, [768]> var_547_to_fp16 = const()[name = tensor<string, []>("op_547_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66467648)))]; | |
| tensor<fp16, [1, 1500, 768]> v_17_cast = linear(bias = var_547_to_fp16, weight = var_546_to_fp16, x = var_527_cast); | |
| tensor<int32, [4]> var_555 = const()[name = tensor<string, []>("op_555"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_556_cast = reshape(shape = var_555, x = q_17_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_92_to_fp16 = const()[name = tensor<string, []>("const_92_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_19_cast = mul(x = var_556_cast, y = const_92_to_fp16); | |
| tensor<int32, [4]> var_562 = const()[name = tensor<string, []>("op_562"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_563_cast = reshape(shape = var_562, x = k_17_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_93_to_fp16 = const()[name = tensor<string, []>("const_93_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_19_cast = mul(x = var_563_cast, y = const_93_to_fp16); | |
| tensor<int32, [4]> var_569 = const()[name = tensor<string, []>("op_569"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_570_cast = reshape(shape = var_569, x = v_17_cast); | |
| tensor<int32, [4]> var_571 = const()[name = tensor<string, []>("op_571"), 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_32_perm_0 = const()[name = tensor<string, []>("transpose_32_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> transpose_33_perm_0 = const()[name = tensor<string, []>("transpose_33_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_77 = transpose(perm = transpose_33_perm_0, x = k_19_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_78 = transpose(perm = transpose_32_perm_0, x = q_19_cast); | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_9_cast = matmul(transpose_x = qk_9_transpose_x_0, transpose_y = qk_9_transpose_y_0, x = transpose_78, y = transpose_77); | |
| tensor<fp16, [1, 12, 1500, 1500]> var_575_cast = softmax(axis = var_510, x = qk_9_cast); | |
| tensor<bool, []> var_577_transpose_x_0 = const()[name = tensor<string, []>("op_577_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> var_577_transpose_y_0 = const()[name = tensor<string, []>("op_577_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_79 = transpose(perm = var_571, x = var_570_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> var_577_cast = matmul(transpose_x = var_577_transpose_x_0, transpose_y = var_577_transpose_y_0, x = var_575_cast, y = transpose_79); | |
| tensor<int32, [4]> var_578 = const()[name = tensor<string, []>("op_578"), 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, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> transpose_76 = transpose(perm = var_578, x = var_577_cast); | |
| tensor<fp16, [1, 1500, 768]> x_59_cast = reshape(shape = concat_4, x = transpose_76); | |
| tensor<fp16, [768, 768]> var_583_to_fp16 = const()[name = tensor<string, []>("op_583_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66469248)))]; | |
| tensor<fp16, [768]> var_584_to_fp16 = const()[name = tensor<string, []>("op_584_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67648960)))]; | |
| tensor<fp16, [1, 1500, 768]> var_585_cast = linear(bias = var_584_to_fp16, weight = var_583_to_fp16, x = x_59_cast); | |
| tensor<fp16, [1, 1500, 768]> x_61_cast = add(x = x_55_cast, y = var_585_cast); | |
| tensor<int32, [1]> var_591_axes_0 = const()[name = tensor<string, []>("op_591_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_4_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_4_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67650560)))]; | |
| tensor<fp16, [768]> blocks_4_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_4_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67652160)))]; | |
| tensor<fp16, [1, 1500, 768]> var_591_cast = layer_norm(axes = var_591_axes_0, beta = blocks_4_mlp_ln_bias_to_fp16, epsilon = var_516_to_fp16, gamma = blocks_4_mlp_ln_weight_to_fp16, x = x_61_cast); | |
| tensor<fp16, [3072, 768]> var_600_to_fp16 = const()[name = tensor<string, []>("op_600_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67653760)))]; | |
| tensor<fp16, [3072]> var_601_to_fp16 = const()[name = tensor<string, []>("op_601_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72372416)))]; | |
| tensor<fp16, [1, 1500, 3072]> input_41_cast = linear(bias = var_601_to_fp16, weight = var_600_to_fp16, x = var_591_cast); | |
| tensor<string, []> x_65_mode_0 = const()[name = tensor<string, []>("x_65_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_65_cast = gelu(mode = x_65_mode_0, x = input_41_cast); | |
| tensor<fp16, [768, 3072]> var_606_to_fp16 = const()[name = tensor<string, []>("op_606_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72378624)))]; | |
| tensor<fp16, [768]> var_607_to_fp16 = const()[name = tensor<string, []>("op_607_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77097280)))]; | |
| tensor<fp16, [1, 1500, 768]> var_608_cast = linear(bias = var_607_to_fp16, weight = var_606_to_fp16, x = x_65_cast); | |
| tensor<fp16, [1, 1500, 768]> x_67_cast = add(x = x_61_cast, y = var_608_cast); | |
| tensor<int32, []> var_617 = const()[name = tensor<string, []>("op_617"), val = tensor<int32, []>(-1)]; | |
| tensor<int32, [1]> var_634_axes_0 = const()[name = tensor<string, []>("op_634_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_5_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_5_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77098880)))]; | |
| tensor<fp16, [768]> blocks_5_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_5_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77100480)))]; | |
| tensor<fp16, []> var_623_to_fp16 = const()[name = tensor<string, []>("op_623_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_634_cast = layer_norm(axes = var_634_axes_0, beta = blocks_5_attn_ln_bias_to_fp16, epsilon = var_623_to_fp16, gamma = blocks_5_attn_ln_weight_to_fp16, x = x_67_cast); | |
| tensor<fp16, [768, 768]> var_645_to_fp16 = const()[name = tensor<string, []>("op_645_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77102080)))]; | |
| tensor<fp16, [768]> var_646_to_fp16 = const()[name = tensor<string, []>("op_646_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78281792)))]; | |
| tensor<fp16, [1, 1500, 768]> q_21_cast = linear(bias = var_646_to_fp16, weight = var_645_to_fp16, x = var_634_cast); | |
| tensor<fp16, [768, 768]> var_649_to_fp16 = const()[name = tensor<string, []>("op_649_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78283392)))]; | |
| tensor<fp16, [768]> k_21_bias_0_to_fp16 = const()[name = tensor<string, []>("k_21_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79463104)))]; | |
| tensor<fp16, [1, 1500, 768]> k_21_cast = linear(bias = k_21_bias_0_to_fp16, weight = var_649_to_fp16, x = var_634_cast); | |
| tensor<fp16, [768, 768]> var_653_to_fp16 = const()[name = tensor<string, []>("op_653_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79464704)))]; | |
| tensor<fp16, [768]> var_654_to_fp16 = const()[name = tensor<string, []>("op_654_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80644416)))]; | |
| tensor<fp16, [1, 1500, 768]> v_21_cast = linear(bias = var_654_to_fp16, weight = var_653_to_fp16, x = var_634_cast); | |
| tensor<int32, [4]> var_662 = const()[name = tensor<string, []>("op_662"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_663_cast = reshape(shape = var_662, x = q_21_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_94_to_fp16 = const()[name = tensor<string, []>("const_94_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_23_cast = mul(x = var_663_cast, y = const_94_to_fp16); | |
| tensor<int32, [4]> var_669 = const()[name = tensor<string, []>("op_669"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_670_cast = reshape(shape = var_669, x = k_21_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_95_to_fp16 = const()[name = tensor<string, []>("const_95_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_23_cast = mul(x = var_670_cast, y = const_95_to_fp16); | |
| tensor<int32, [4]> var_676 = const()[name = tensor<string, []>("op_676"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_677_cast = reshape(shape = var_676, x = v_21_cast); | |
| tensor<int32, [4]> var_678 = const()[name = tensor<string, []>("op_678"), 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_34_perm_0 = const()[name = tensor<string, []>("transpose_34_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> transpose_35_perm_0 = const()[name = tensor<string, []>("transpose_35_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_73 = transpose(perm = transpose_35_perm_0, x = k_23_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_74 = transpose(perm = transpose_34_perm_0, x = q_23_cast); | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_11_cast = matmul(transpose_x = qk_11_transpose_x_0, transpose_y = qk_11_transpose_y_0, x = transpose_74, y = transpose_73); | |
| tensor<fp16, [1, 12, 1500, 1500]> var_682_cast = softmax(axis = var_617, x = qk_11_cast); | |
| tensor<bool, []> var_684_transpose_x_0 = const()[name = tensor<string, []>("op_684_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> var_684_transpose_y_0 = const()[name = tensor<string, []>("op_684_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_75 = transpose(perm = var_678, x = var_677_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> var_684_cast = matmul(transpose_x = var_684_transpose_x_0, transpose_y = var_684_transpose_y_0, x = var_682_cast, y = transpose_75); | |
| tensor<int32, [4]> var_685 = const()[name = tensor<string, []>("op_685"), 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, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> transpose_72 = transpose(perm = var_685, x = var_684_cast); | |
| tensor<fp16, [1, 1500, 768]> x_71_cast = reshape(shape = concat_5, x = transpose_72); | |
| tensor<fp16, [768, 768]> var_690_to_fp16 = const()[name = tensor<string, []>("op_690_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80646016)))]; | |
| tensor<fp16, [768]> var_691_to_fp16 = const()[name = tensor<string, []>("op_691_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81825728)))]; | |
| tensor<fp16, [1, 1500, 768]> var_692_cast = linear(bias = var_691_to_fp16, weight = var_690_to_fp16, x = x_71_cast); | |
| tensor<fp16, [1, 1500, 768]> x_73_cast = add(x = x_67_cast, y = var_692_cast); | |
| tensor<int32, [1]> var_698_axes_0 = const()[name = tensor<string, []>("op_698_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_5_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_5_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81827328)))]; | |
| tensor<fp16, [768]> blocks_5_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_5_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81828928)))]; | |
| tensor<fp16, [1, 1500, 768]> var_698_cast = layer_norm(axes = var_698_axes_0, beta = blocks_5_mlp_ln_bias_to_fp16, epsilon = var_623_to_fp16, gamma = blocks_5_mlp_ln_weight_to_fp16, x = x_73_cast); | |
| tensor<fp16, [3072, 768]> var_707_to_fp16 = const()[name = tensor<string, []>("op_707_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81830528)))]; | |
| tensor<fp16, [3072]> var_708_to_fp16 = const()[name = tensor<string, []>("op_708_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86549184)))]; | |
| tensor<fp16, [1, 1500, 3072]> input_49_cast = linear(bias = var_708_to_fp16, weight = var_707_to_fp16, x = var_698_cast); | |
| tensor<string, []> x_77_mode_0 = const()[name = tensor<string, []>("x_77_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_77_cast = gelu(mode = x_77_mode_0, x = input_49_cast); | |
| tensor<fp16, [768, 3072]> var_713_to_fp16 = const()[name = tensor<string, []>("op_713_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86555392)))]; | |
| tensor<fp16, [768]> var_714_to_fp16 = const()[name = tensor<string, []>("op_714_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91274048)))]; | |
| tensor<fp16, [1, 1500, 768]> var_715_cast = linear(bias = var_714_to_fp16, weight = var_713_to_fp16, x = x_77_cast); | |
| tensor<fp16, [1, 1500, 768]> x_79_cast = add(x = x_73_cast, y = var_715_cast); | |
| tensor<int32, []> var_724 = const()[name = tensor<string, []>("op_724"), val = tensor<int32, []>(-1)]; | |
| tensor<int32, [1]> var_741_axes_0 = const()[name = tensor<string, []>("op_741_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_6_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_6_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91275648)))]; | |
| tensor<fp16, [768]> blocks_6_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_6_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91277248)))]; | |
| tensor<fp16, []> var_730_to_fp16 = const()[name = tensor<string, []>("op_730_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_741_cast = layer_norm(axes = var_741_axes_0, beta = blocks_6_attn_ln_bias_to_fp16, epsilon = var_730_to_fp16, gamma = blocks_6_attn_ln_weight_to_fp16, x = x_79_cast); | |
| tensor<fp16, [768, 768]> var_752_to_fp16 = const()[name = tensor<string, []>("op_752_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91278848)))]; | |
| tensor<fp16, [768]> var_753_to_fp16 = const()[name = tensor<string, []>("op_753_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92458560)))]; | |
| tensor<fp16, [1, 1500, 768]> q_25_cast = linear(bias = var_753_to_fp16, weight = var_752_to_fp16, x = var_741_cast); | |
| tensor<fp16, [768, 768]> var_756_to_fp16 = const()[name = tensor<string, []>("op_756_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92460160)))]; | |
| tensor<fp16, [768]> k_25_bias_0_to_fp16 = const()[name = tensor<string, []>("k_25_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93639872)))]; | |
| tensor<fp16, [1, 1500, 768]> k_25_cast = linear(bias = k_25_bias_0_to_fp16, weight = var_756_to_fp16, x = var_741_cast); | |
| tensor<fp16, [768, 768]> var_760_to_fp16 = const()[name = tensor<string, []>("op_760_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93641472)))]; | |
| tensor<fp16, [768]> var_761_to_fp16 = const()[name = tensor<string, []>("op_761_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94821184)))]; | |
| tensor<fp16, [1, 1500, 768]> v_25_cast = linear(bias = var_761_to_fp16, weight = var_760_to_fp16, x = var_741_cast); | |
| tensor<int32, [4]> var_769 = const()[name = tensor<string, []>("op_769"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_770_cast = reshape(shape = var_769, x = q_25_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_96_to_fp16 = const()[name = tensor<string, []>("const_96_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_27_cast = mul(x = var_770_cast, y = const_96_to_fp16); | |
| tensor<int32, [4]> var_776 = const()[name = tensor<string, []>("op_776"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_777_cast = reshape(shape = var_776, x = k_25_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_97_to_fp16 = const()[name = tensor<string, []>("const_97_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_27_cast = mul(x = var_777_cast, y = const_97_to_fp16); | |
| tensor<int32, [4]> var_783 = const()[name = tensor<string, []>("op_783"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_784_cast = reshape(shape = var_783, x = v_25_cast); | |
| tensor<int32, [4]> var_785 = const()[name = tensor<string, []>("op_785"), 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_36_perm_0 = const()[name = tensor<string, []>("transpose_36_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> transpose_37_perm_0 = const()[name = tensor<string, []>("transpose_37_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_69 = transpose(perm = transpose_37_perm_0, x = k_27_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_70 = transpose(perm = transpose_36_perm_0, x = q_27_cast); | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_13_cast = matmul(transpose_x = qk_13_transpose_x_0, transpose_y = qk_13_transpose_y_0, x = transpose_70, y = transpose_69); | |
| tensor<fp16, [1, 12, 1500, 1500]> var_789_cast = softmax(axis = var_724, x = qk_13_cast); | |
| tensor<bool, []> var_791_transpose_x_0 = const()[name = tensor<string, []>("op_791_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> var_791_transpose_y_0 = const()[name = tensor<string, []>("op_791_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_71 = transpose(perm = var_785, x = var_784_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> var_791_cast = matmul(transpose_x = var_791_transpose_x_0, transpose_y = var_791_transpose_y_0, x = var_789_cast, y = transpose_71); | |
| tensor<int32, [4]> var_792 = const()[name = tensor<string, []>("op_792"), 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, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> transpose_68 = transpose(perm = var_792, x = var_791_cast); | |
| tensor<fp16, [1, 1500, 768]> x_83_cast = reshape(shape = concat_6, x = transpose_68); | |
| tensor<fp16, [768, 768]> var_797_to_fp16 = const()[name = tensor<string, []>("op_797_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94822784)))]; | |
| tensor<fp16, [768]> var_798_to_fp16 = const()[name = tensor<string, []>("op_798_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96002496)))]; | |
| tensor<fp16, [1, 1500, 768]> var_799_cast = linear(bias = var_798_to_fp16, weight = var_797_to_fp16, x = x_83_cast); | |
| tensor<fp16, [1, 1500, 768]> x_85_cast = add(x = x_79_cast, y = var_799_cast); | |
| tensor<int32, [1]> var_805_axes_0 = const()[name = tensor<string, []>("op_805_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_6_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_6_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96004096)))]; | |
| tensor<fp16, [768]> blocks_6_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_6_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96005696)))]; | |
| tensor<fp16, [1, 1500, 768]> var_805_cast = layer_norm(axes = var_805_axes_0, beta = blocks_6_mlp_ln_bias_to_fp16, epsilon = var_730_to_fp16, gamma = blocks_6_mlp_ln_weight_to_fp16, x = x_85_cast); | |
| tensor<fp16, [3072, 768]> var_814_to_fp16 = const()[name = tensor<string, []>("op_814_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96007296)))]; | |
| tensor<fp16, [3072]> var_815_to_fp16 = const()[name = tensor<string, []>("op_815_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100725952)))]; | |
| tensor<fp16, [1, 1500, 3072]> input_57_cast = linear(bias = var_815_to_fp16, weight = var_814_to_fp16, x = var_805_cast); | |
| tensor<string, []> x_89_mode_0 = const()[name = tensor<string, []>("x_89_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_89_cast = gelu(mode = x_89_mode_0, x = input_57_cast); | |
| tensor<fp16, [768, 3072]> var_820_to_fp16 = const()[name = tensor<string, []>("op_820_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100732160)))]; | |
| tensor<fp16, [768]> var_821_to_fp16 = const()[name = tensor<string, []>("op_821_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105450816)))]; | |
| tensor<fp16, [1, 1500, 768]> var_822_cast = linear(bias = var_821_to_fp16, weight = var_820_to_fp16, x = x_89_cast); | |
| tensor<fp16, [1, 1500, 768]> x_91_cast = add(x = x_85_cast, y = var_822_cast); | |
| tensor<int32, []> var_831 = const()[name = tensor<string, []>("op_831"), val = tensor<int32, []>(-1)]; | |
| tensor<int32, [1]> var_848_axes_0 = const()[name = tensor<string, []>("op_848_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_7_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_7_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105452416)))]; | |
| tensor<fp16, [768]> blocks_7_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_7_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105454016)))]; | |
| tensor<fp16, []> var_837_to_fp16 = const()[name = tensor<string, []>("op_837_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_848_cast = layer_norm(axes = var_848_axes_0, beta = blocks_7_attn_ln_bias_to_fp16, epsilon = var_837_to_fp16, gamma = blocks_7_attn_ln_weight_to_fp16, x = x_91_cast); | |
| tensor<fp16, [768, 768]> var_859_to_fp16 = const()[name = tensor<string, []>("op_859_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105455616)))]; | |
| tensor<fp16, [768]> var_860_to_fp16 = const()[name = tensor<string, []>("op_860_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106635328)))]; | |
| tensor<fp16, [1, 1500, 768]> q_29_cast = linear(bias = var_860_to_fp16, weight = var_859_to_fp16, x = var_848_cast); | |
| tensor<fp16, [768, 768]> var_863_to_fp16 = const()[name = tensor<string, []>("op_863_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106636928)))]; | |
| tensor<fp16, [768]> k_29_bias_0_to_fp16 = const()[name = tensor<string, []>("k_29_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107816640)))]; | |
| tensor<fp16, [1, 1500, 768]> k_29_cast = linear(bias = k_29_bias_0_to_fp16, weight = var_863_to_fp16, x = var_848_cast); | |
| tensor<fp16, [768, 768]> var_867_to_fp16 = const()[name = tensor<string, []>("op_867_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107818240)))]; | |
| tensor<fp16, [768]> var_868_to_fp16 = const()[name = tensor<string, []>("op_868_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108997952)))]; | |
| tensor<fp16, [1, 1500, 768]> v_29_cast = linear(bias = var_868_to_fp16, weight = var_867_to_fp16, x = var_848_cast); | |
| tensor<int32, [4]> var_876 = const()[name = tensor<string, []>("op_876"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_877_cast = reshape(shape = var_876, x = q_29_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_98_to_fp16 = const()[name = tensor<string, []>("const_98_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_31_cast = mul(x = var_877_cast, y = const_98_to_fp16); | |
| tensor<int32, [4]> var_883 = const()[name = tensor<string, []>("op_883"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_884_cast = reshape(shape = var_883, x = k_29_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_99_to_fp16 = const()[name = tensor<string, []>("const_99_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_31_cast = mul(x = var_884_cast, y = const_99_to_fp16); | |
| tensor<int32, [4]> var_890 = const()[name = tensor<string, []>("op_890"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_891_cast = reshape(shape = var_890, x = v_29_cast); | |
| tensor<int32, [4]> var_892 = const()[name = tensor<string, []>("op_892"), 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_38_perm_0 = const()[name = tensor<string, []>("transpose_38_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> transpose_39_perm_0 = const()[name = tensor<string, []>("transpose_39_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_65 = transpose(perm = transpose_39_perm_0, x = k_31_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_66 = transpose(perm = transpose_38_perm_0, x = q_31_cast); | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_15_cast = matmul(transpose_x = qk_15_transpose_x_0, transpose_y = qk_15_transpose_y_0, x = transpose_66, y = transpose_65); | |
| tensor<fp16, [1, 12, 1500, 1500]> var_896_cast = softmax(axis = var_831, x = qk_15_cast); | |
| tensor<bool, []> var_898_transpose_x_0 = const()[name = tensor<string, []>("op_898_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> var_898_transpose_y_0 = const()[name = tensor<string, []>("op_898_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_67 = transpose(perm = var_892, x = var_891_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> var_898_cast = matmul(transpose_x = var_898_transpose_x_0, transpose_y = var_898_transpose_y_0, x = var_896_cast, y = transpose_67); | |
| tensor<int32, [4]> var_899 = const()[name = tensor<string, []>("op_899"), 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, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> transpose_64 = transpose(perm = var_899, x = var_898_cast); | |
| tensor<fp16, [1, 1500, 768]> x_95_cast = reshape(shape = concat_7, x = transpose_64); | |
| tensor<fp16, [768, 768]> var_904_to_fp16 = const()[name = tensor<string, []>("op_904_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108999552)))]; | |
| tensor<fp16, [768]> var_905_to_fp16 = const()[name = tensor<string, []>("op_905_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110179264)))]; | |
| tensor<fp16, [1, 1500, 768]> var_906_cast = linear(bias = var_905_to_fp16, weight = var_904_to_fp16, x = x_95_cast); | |
| tensor<fp16, [1, 1500, 768]> x_97_cast = add(x = x_91_cast, y = var_906_cast); | |
| tensor<int32, [1]> var_912_axes_0 = const()[name = tensor<string, []>("op_912_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_7_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_7_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110180864)))]; | |
| tensor<fp16, [768]> blocks_7_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_7_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110182464)))]; | |
| tensor<fp16, [1, 1500, 768]> var_912_cast = layer_norm(axes = var_912_axes_0, beta = blocks_7_mlp_ln_bias_to_fp16, epsilon = var_837_to_fp16, gamma = blocks_7_mlp_ln_weight_to_fp16, x = x_97_cast); | |
| tensor<fp16, [3072, 768]> var_921_to_fp16 = const()[name = tensor<string, []>("op_921_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(110184064)))]; | |
| tensor<fp16, [3072]> var_922_to_fp16 = const()[name = tensor<string, []>("op_922_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114902720)))]; | |
| tensor<fp16, [1, 1500, 3072]> input_65_cast = linear(bias = var_922_to_fp16, weight = var_921_to_fp16, x = var_912_cast); | |
| tensor<string, []> x_101_mode_0 = const()[name = tensor<string, []>("x_101_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_101_cast = gelu(mode = x_101_mode_0, x = input_65_cast); | |
| tensor<fp16, [768, 3072]> var_927_to_fp16 = const()[name = tensor<string, []>("op_927_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(114908928)))]; | |
| tensor<fp16, [768]> var_928_to_fp16 = const()[name = tensor<string, []>("op_928_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119627584)))]; | |
| tensor<fp16, [1, 1500, 768]> var_929_cast = linear(bias = var_928_to_fp16, weight = var_927_to_fp16, x = x_101_cast); | |
| tensor<fp16, [1, 1500, 768]> x_103_cast = add(x = x_97_cast, y = var_929_cast); | |
| tensor<int32, []> var_938 = const()[name = tensor<string, []>("op_938"), val = tensor<int32, []>(-1)]; | |
| tensor<int32, [1]> var_955_axes_0 = const()[name = tensor<string, []>("op_955_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_8_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_8_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119629184)))]; | |
| tensor<fp16, [768]> blocks_8_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_8_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119630784)))]; | |
| tensor<fp16, []> var_944_to_fp16 = const()[name = tensor<string, []>("op_944_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_955_cast = layer_norm(axes = var_955_axes_0, beta = blocks_8_attn_ln_bias_to_fp16, epsilon = var_944_to_fp16, gamma = blocks_8_attn_ln_weight_to_fp16, x = x_103_cast); | |
| tensor<fp16, [768, 768]> var_966_to_fp16 = const()[name = tensor<string, []>("op_966_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(119632384)))]; | |
| tensor<fp16, [768]> var_967_to_fp16 = const()[name = tensor<string, []>("op_967_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120812096)))]; | |
| tensor<fp16, [1, 1500, 768]> q_33_cast = linear(bias = var_967_to_fp16, weight = var_966_to_fp16, x = var_955_cast); | |
| tensor<fp16, [768, 768]> var_970_to_fp16 = const()[name = tensor<string, []>("op_970_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(120813696)))]; | |
| tensor<fp16, [768]> k_33_bias_0_to_fp16 = const()[name = tensor<string, []>("k_33_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121993408)))]; | |
| tensor<fp16, [1, 1500, 768]> k_33_cast = linear(bias = k_33_bias_0_to_fp16, weight = var_970_to_fp16, x = var_955_cast); | |
| tensor<fp16, [768, 768]> var_974_to_fp16 = const()[name = tensor<string, []>("op_974_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(121995008)))]; | |
| tensor<fp16, [768]> var_975_to_fp16 = const()[name = tensor<string, []>("op_975_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123174720)))]; | |
| tensor<fp16, [1, 1500, 768]> v_33_cast = linear(bias = var_975_to_fp16, weight = var_974_to_fp16, x = var_955_cast); | |
| tensor<int32, [4]> var_983 = const()[name = tensor<string, []>("op_983"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_984_cast = reshape(shape = var_983, x = q_33_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_100_to_fp16 = const()[name = tensor<string, []>("const_100_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_35_cast = mul(x = var_984_cast, y = const_100_to_fp16); | |
| tensor<int32, [4]> var_990 = const()[name = tensor<string, []>("op_990"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_991_cast = reshape(shape = var_990, x = k_33_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_101_to_fp16 = const()[name = tensor<string, []>("const_101_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_35_cast = mul(x = var_991_cast, y = const_101_to_fp16); | |
| tensor<int32, [4]> var_997 = const()[name = tensor<string, []>("op_997"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_998_cast = reshape(shape = var_997, x = v_33_cast); | |
| tensor<int32, [4]> var_999 = const()[name = tensor<string, []>("op_999"), 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_40_perm_0 = const()[name = tensor<string, []>("transpose_40_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> transpose_41_perm_0 = const()[name = tensor<string, []>("transpose_41_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_61 = transpose(perm = transpose_41_perm_0, x = k_35_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_62 = transpose(perm = transpose_40_perm_0, x = q_35_cast); | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_17_cast = matmul(transpose_x = qk_17_transpose_x_0, transpose_y = qk_17_transpose_y_0, x = transpose_62, y = transpose_61); | |
| tensor<fp16, [1, 12, 1500, 1500]> var_1003_cast = softmax(axis = var_938, x = qk_17_cast); | |
| tensor<bool, []> var_1005_transpose_x_0 = const()[name = tensor<string, []>("op_1005_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> var_1005_transpose_y_0 = const()[name = tensor<string, []>("op_1005_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_63 = transpose(perm = var_999, x = var_998_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> var_1005_cast = matmul(transpose_x = var_1005_transpose_x_0, transpose_y = var_1005_transpose_y_0, x = var_1003_cast, y = transpose_63); | |
| tensor<int32, [4]> var_1006 = const()[name = tensor<string, []>("op_1006"), 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, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> transpose_60 = transpose(perm = var_1006, x = var_1005_cast); | |
| tensor<fp16, [1, 1500, 768]> x_107_cast = reshape(shape = concat_8, x = transpose_60); | |
| tensor<fp16, [768, 768]> var_1011_to_fp16 = const()[name = tensor<string, []>("op_1011_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(123176320)))]; | |
| tensor<fp16, [768]> var_1012_to_fp16 = const()[name = tensor<string, []>("op_1012_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124356032)))]; | |
| tensor<fp16, [1, 1500, 768]> var_1013_cast = linear(bias = var_1012_to_fp16, weight = var_1011_to_fp16, x = x_107_cast); | |
| tensor<fp16, [1, 1500, 768]> x_109_cast = add(x = x_103_cast, y = var_1013_cast); | |
| tensor<int32, [1]> var_1019_axes_0 = const()[name = tensor<string, []>("op_1019_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_8_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_8_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124357632)))]; | |
| tensor<fp16, [768]> blocks_8_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_8_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124359232)))]; | |
| tensor<fp16, [1, 1500, 768]> var_1019_cast = layer_norm(axes = var_1019_axes_0, beta = blocks_8_mlp_ln_bias_to_fp16, epsilon = var_944_to_fp16, gamma = blocks_8_mlp_ln_weight_to_fp16, x = x_109_cast); | |
| tensor<fp16, [3072, 768]> var_1028_to_fp16 = const()[name = tensor<string, []>("op_1028_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(124360832)))]; | |
| tensor<fp16, [3072]> var_1029_to_fp16 = const()[name = tensor<string, []>("op_1029_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129079488)))]; | |
| tensor<fp16, [1, 1500, 3072]> input_73_cast = linear(bias = var_1029_to_fp16, weight = var_1028_to_fp16, x = var_1019_cast); | |
| tensor<string, []> x_113_mode_0 = const()[name = tensor<string, []>("x_113_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_113_cast = gelu(mode = x_113_mode_0, x = input_73_cast); | |
| tensor<fp16, [768, 3072]> var_1034_to_fp16 = const()[name = tensor<string, []>("op_1034_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(129085696)))]; | |
| tensor<fp16, [768]> var_1035_to_fp16 = const()[name = tensor<string, []>("op_1035_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133804352)))]; | |
| tensor<fp16, [1, 1500, 768]> var_1036_cast = linear(bias = var_1035_to_fp16, weight = var_1034_to_fp16, x = x_113_cast); | |
| tensor<fp16, [1, 1500, 768]> x_115_cast = add(x = x_109_cast, y = var_1036_cast); | |
| tensor<int32, []> var_1045 = const()[name = tensor<string, []>("op_1045"), val = tensor<int32, []>(-1)]; | |
| tensor<int32, [1]> var_1062_axes_0 = const()[name = tensor<string, []>("op_1062_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_9_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_9_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133805952)))]; | |
| tensor<fp16, [768]> blocks_9_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_9_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133807552)))]; | |
| tensor<fp16, []> var_1051_to_fp16 = const()[name = tensor<string, []>("op_1051_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_1062_cast = layer_norm(axes = var_1062_axes_0, beta = blocks_9_attn_ln_bias_to_fp16, epsilon = var_1051_to_fp16, gamma = blocks_9_attn_ln_weight_to_fp16, x = x_115_cast); | |
| tensor<fp16, [768, 768]> var_1073_to_fp16 = const()[name = tensor<string, []>("op_1073_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(133809152)))]; | |
| tensor<fp16, [768]> var_1074_to_fp16 = const()[name = tensor<string, []>("op_1074_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134988864)))]; | |
| tensor<fp16, [1, 1500, 768]> q_37_cast = linear(bias = var_1074_to_fp16, weight = var_1073_to_fp16, x = var_1062_cast); | |
| tensor<fp16, [768, 768]> var_1077_to_fp16 = const()[name = tensor<string, []>("op_1077_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(134990464)))]; | |
| tensor<fp16, [768]> k_37_bias_0_to_fp16 = const()[name = tensor<string, []>("k_37_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136170176)))]; | |
| tensor<fp16, [1, 1500, 768]> k_37_cast = linear(bias = k_37_bias_0_to_fp16, weight = var_1077_to_fp16, x = var_1062_cast); | |
| tensor<fp16, [768, 768]> var_1081_to_fp16 = const()[name = tensor<string, []>("op_1081_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(136171776)))]; | |
| tensor<fp16, [768]> var_1082_to_fp16 = const()[name = tensor<string, []>("op_1082_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137351488)))]; | |
| tensor<fp16, [1, 1500, 768]> v_37_cast = linear(bias = var_1082_to_fp16, weight = var_1081_to_fp16, x = var_1062_cast); | |
| tensor<int32, [4]> var_1090 = const()[name = tensor<string, []>("op_1090"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_1091_cast = reshape(shape = var_1090, x = q_37_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_102_to_fp16 = const()[name = tensor<string, []>("const_102_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_39_cast = mul(x = var_1091_cast, y = const_102_to_fp16); | |
| tensor<int32, [4]> var_1097 = const()[name = tensor<string, []>("op_1097"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_1098_cast = reshape(shape = var_1097, x = k_37_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_103_to_fp16 = const()[name = tensor<string, []>("const_103_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_39_cast = mul(x = var_1098_cast, y = const_103_to_fp16); | |
| tensor<int32, [4]> var_1104 = const()[name = tensor<string, []>("op_1104"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_1105_cast = reshape(shape = var_1104, x = v_37_cast); | |
| tensor<int32, [4]> var_1106 = const()[name = tensor<string, []>("op_1106"), 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_42_perm_0 = const()[name = tensor<string, []>("transpose_42_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> transpose_43_perm_0 = const()[name = tensor<string, []>("transpose_43_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_57 = transpose(perm = transpose_43_perm_0, x = k_39_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_58 = transpose(perm = transpose_42_perm_0, x = q_39_cast); | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_19_cast = matmul(transpose_x = qk_19_transpose_x_0, transpose_y = qk_19_transpose_y_0, x = transpose_58, y = transpose_57); | |
| tensor<fp16, [1, 12, 1500, 1500]> var_1110_cast = softmax(axis = var_1045, x = qk_19_cast); | |
| tensor<bool, []> var_1112_transpose_x_0 = const()[name = tensor<string, []>("op_1112_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> var_1112_transpose_y_0 = const()[name = tensor<string, []>("op_1112_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_59 = transpose(perm = var_1106, x = var_1105_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> var_1112_cast = matmul(transpose_x = var_1112_transpose_x_0, transpose_y = var_1112_transpose_y_0, x = var_1110_cast, y = transpose_59); | |
| tensor<int32, [4]> var_1113 = const()[name = tensor<string, []>("op_1113"), 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, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> transpose_56 = transpose(perm = var_1113, x = var_1112_cast); | |
| tensor<fp16, [1, 1500, 768]> x_119_cast = reshape(shape = concat_9, x = transpose_56); | |
| tensor<fp16, [768, 768]> var_1118_to_fp16 = const()[name = tensor<string, []>("op_1118_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(137353088)))]; | |
| tensor<fp16, [768]> var_1119_to_fp16 = const()[name = tensor<string, []>("op_1119_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138532800)))]; | |
| tensor<fp16, [1, 1500, 768]> var_1120_cast = linear(bias = var_1119_to_fp16, weight = var_1118_to_fp16, x = x_119_cast); | |
| tensor<fp16, [1, 1500, 768]> x_121_cast = add(x = x_115_cast, y = var_1120_cast); | |
| tensor<int32, [1]> var_1126_axes_0 = const()[name = tensor<string, []>("op_1126_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_9_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_9_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138534400)))]; | |
| tensor<fp16, [768]> blocks_9_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_9_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138536000)))]; | |
| tensor<fp16, [1, 1500, 768]> var_1126_cast = layer_norm(axes = var_1126_axes_0, beta = blocks_9_mlp_ln_bias_to_fp16, epsilon = var_1051_to_fp16, gamma = blocks_9_mlp_ln_weight_to_fp16, x = x_121_cast); | |
| tensor<fp16, [3072, 768]> var_1135_to_fp16 = const()[name = tensor<string, []>("op_1135_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(138537600)))]; | |
| tensor<fp16, [3072]> var_1136_to_fp16 = const()[name = tensor<string, []>("op_1136_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(143256256)))]; | |
| tensor<fp16, [1, 1500, 3072]> input_81_cast = linear(bias = var_1136_to_fp16, weight = var_1135_to_fp16, x = var_1126_cast); | |
| tensor<string, []> x_125_mode_0 = const()[name = tensor<string, []>("x_125_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_125_cast = gelu(mode = x_125_mode_0, x = input_81_cast); | |
| tensor<fp16, [768, 3072]> var_1141_to_fp16 = const()[name = tensor<string, []>("op_1141_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(143262464)))]; | |
| tensor<fp16, [768]> var_1142_to_fp16 = const()[name = tensor<string, []>("op_1142_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147981120)))]; | |
| tensor<fp16, [1, 1500, 768]> var_1143_cast = linear(bias = var_1142_to_fp16, weight = var_1141_to_fp16, x = x_125_cast); | |
| tensor<fp16, [1, 1500, 768]> x_127_cast = add(x = x_121_cast, y = var_1143_cast); | |
| tensor<int32, []> var_1152 = const()[name = tensor<string, []>("op_1152"), val = tensor<int32, []>(-1)]; | |
| tensor<int32, [1]> var_1169_axes_0 = const()[name = tensor<string, []>("op_1169_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_10_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_10_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147982720)))]; | |
| tensor<fp16, [768]> blocks_10_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_10_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147984320)))]; | |
| tensor<fp16, []> var_1158_to_fp16 = const()[name = tensor<string, []>("op_1158_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_1169_cast = layer_norm(axes = var_1169_axes_0, beta = blocks_10_attn_ln_bias_to_fp16, epsilon = var_1158_to_fp16, gamma = blocks_10_attn_ln_weight_to_fp16, x = x_127_cast); | |
| tensor<fp16, [768, 768]> var_1180_to_fp16 = const()[name = tensor<string, []>("op_1180_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(147985920)))]; | |
| tensor<fp16, [768]> var_1181_to_fp16 = const()[name = tensor<string, []>("op_1181_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149165632)))]; | |
| tensor<fp16, [1, 1500, 768]> q_41_cast = linear(bias = var_1181_to_fp16, weight = var_1180_to_fp16, x = var_1169_cast); | |
| tensor<fp16, [768, 768]> var_1184_to_fp16 = const()[name = tensor<string, []>("op_1184_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(149167232)))]; | |
| tensor<fp16, [768]> k_41_bias_0_to_fp16 = const()[name = tensor<string, []>("k_41_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150346944)))]; | |
| tensor<fp16, [1, 1500, 768]> k_41_cast = linear(bias = k_41_bias_0_to_fp16, weight = var_1184_to_fp16, x = var_1169_cast); | |
| tensor<fp16, [768, 768]> var_1188_to_fp16 = const()[name = tensor<string, []>("op_1188_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(150348544)))]; | |
| tensor<fp16, [768]> var_1189_to_fp16 = const()[name = tensor<string, []>("op_1189_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151528256)))]; | |
| tensor<fp16, [1, 1500, 768]> v_41_cast = linear(bias = var_1189_to_fp16, weight = var_1188_to_fp16, x = var_1169_cast); | |
| tensor<int32, [4]> var_1197 = const()[name = tensor<string, []>("op_1197"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_1198_cast = reshape(shape = var_1197, x = q_41_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_104_to_fp16 = const()[name = tensor<string, []>("const_104_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_43_cast = mul(x = var_1198_cast, y = const_104_to_fp16); | |
| tensor<int32, [4]> var_1204 = const()[name = tensor<string, []>("op_1204"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_1205_cast = reshape(shape = var_1204, x = k_41_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_105_to_fp16 = const()[name = tensor<string, []>("const_105_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_43_cast = mul(x = var_1205_cast, y = const_105_to_fp16); | |
| tensor<int32, [4]> var_1211 = const()[name = tensor<string, []>("op_1211"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_1212_cast = reshape(shape = var_1211, x = v_41_cast); | |
| tensor<int32, [4]> var_1213 = const()[name = tensor<string, []>("op_1213"), 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_44_perm_0 = const()[name = tensor<string, []>("transpose_44_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> transpose_45_perm_0 = const()[name = tensor<string, []>("transpose_45_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_53 = transpose(perm = transpose_45_perm_0, x = k_43_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_54 = transpose(perm = transpose_44_perm_0, x = q_43_cast); | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_21_cast = matmul(transpose_x = qk_21_transpose_x_0, transpose_y = qk_21_transpose_y_0, x = transpose_54, y = transpose_53); | |
| tensor<fp16, [1, 12, 1500, 1500]> var_1217_cast = softmax(axis = var_1152, x = qk_21_cast); | |
| tensor<bool, []> var_1219_transpose_x_0 = const()[name = tensor<string, []>("op_1219_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> var_1219_transpose_y_0 = const()[name = tensor<string, []>("op_1219_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_55 = transpose(perm = var_1213, x = var_1212_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> var_1219_cast = matmul(transpose_x = var_1219_transpose_x_0, transpose_y = var_1219_transpose_y_0, x = var_1217_cast, y = transpose_55); | |
| tensor<int32, [4]> var_1220 = const()[name = tensor<string, []>("op_1220"), 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, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> transpose_52 = transpose(perm = var_1220, x = var_1219_cast); | |
| tensor<fp16, [1, 1500, 768]> x_131_cast = reshape(shape = concat_10, x = transpose_52); | |
| tensor<fp16, [768, 768]> var_1225_to_fp16 = const()[name = tensor<string, []>("op_1225_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(151529856)))]; | |
| tensor<fp16, [768]> var_1226_to_fp16 = const()[name = tensor<string, []>("op_1226_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152709568)))]; | |
| tensor<fp16, [1, 1500, 768]> var_1227_cast = linear(bias = var_1226_to_fp16, weight = var_1225_to_fp16, x = x_131_cast); | |
| tensor<fp16, [1, 1500, 768]> x_133_cast = add(x = x_127_cast, y = var_1227_cast); | |
| tensor<int32, [1]> var_1233_axes_0 = const()[name = tensor<string, []>("op_1233_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_10_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_10_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152711168)))]; | |
| tensor<fp16, [768]> blocks_10_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_10_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152712768)))]; | |
| tensor<fp16, [1, 1500, 768]> var_1233_cast = layer_norm(axes = var_1233_axes_0, beta = blocks_10_mlp_ln_bias_to_fp16, epsilon = var_1158_to_fp16, gamma = blocks_10_mlp_ln_weight_to_fp16, x = x_133_cast); | |
| tensor<fp16, [3072, 768]> var_1242_to_fp16 = const()[name = tensor<string, []>("op_1242_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(152714368)))]; | |
| tensor<fp16, [3072]> var_1243_to_fp16 = const()[name = tensor<string, []>("op_1243_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(157433024)))]; | |
| tensor<fp16, [1, 1500, 3072]> input_89_cast = linear(bias = var_1243_to_fp16, weight = var_1242_to_fp16, x = var_1233_cast); | |
| tensor<string, []> x_137_mode_0 = const()[name = tensor<string, []>("x_137_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_137_cast = gelu(mode = x_137_mode_0, x = input_89_cast); | |
| tensor<fp16, [768, 3072]> var_1248_to_fp16 = const()[name = tensor<string, []>("op_1248_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(157439232)))]; | |
| tensor<fp16, [768]> var_1249_to_fp16 = const()[name = tensor<string, []>("op_1249_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(162157888)))]; | |
| tensor<fp16, [1, 1500, 768]> var_1250_cast = linear(bias = var_1249_to_fp16, weight = var_1248_to_fp16, x = x_137_cast); | |
| tensor<fp16, [1, 1500, 768]> x_139_cast = add(x = x_133_cast, y = var_1250_cast); | |
| tensor<int32, []> var_1259 = const()[name = tensor<string, []>("op_1259"), val = tensor<int32, []>(-1)]; | |
| tensor<int32, [1]> var_1276_axes_0 = const()[name = tensor<string, []>("op_1276_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_11_attn_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_11_attn_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(162159488)))]; | |
| tensor<fp16, [768]> blocks_11_attn_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_11_attn_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(162161088)))]; | |
| tensor<fp16, []> var_1265_to_fp16 = const()[name = tensor<string, []>("op_1265_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_1276_cast = layer_norm(axes = var_1276_axes_0, beta = blocks_11_attn_ln_bias_to_fp16, epsilon = var_1265_to_fp16, gamma = blocks_11_attn_ln_weight_to_fp16, x = x_139_cast); | |
| tensor<fp16, [768, 768]> var_1287_to_fp16 = const()[name = tensor<string, []>("op_1287_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(162162688)))]; | |
| tensor<fp16, [768]> var_1288_to_fp16 = const()[name = tensor<string, []>("op_1288_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163342400)))]; | |
| tensor<fp16, [1, 1500, 768]> q_45_cast = linear(bias = var_1288_to_fp16, weight = var_1287_to_fp16, x = var_1276_cast); | |
| tensor<fp16, [768, 768]> var_1291_to_fp16 = const()[name = tensor<string, []>("op_1291_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(163344000)))]; | |
| tensor<fp16, [768]> k_45_bias_0_to_fp16 = const()[name = tensor<string, []>("k_45_bias_0_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164523712)))]; | |
| tensor<fp16, [1, 1500, 768]> k_45_cast = linear(bias = k_45_bias_0_to_fp16, weight = var_1291_to_fp16, x = var_1276_cast); | |
| tensor<fp16, [768, 768]> var_1295_to_fp16 = const()[name = tensor<string, []>("op_1295_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(164525312)))]; | |
| tensor<fp16, [768]> var_1296_to_fp16 = const()[name = tensor<string, []>("op_1296_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165705024)))]; | |
| tensor<fp16, [1, 1500, 768]> v_45_cast = linear(bias = var_1296_to_fp16, weight = var_1295_to_fp16, x = var_1276_cast); | |
| tensor<int32, [4]> var_1304 = const()[name = tensor<string, []>("op_1304"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_1305_cast = reshape(shape = var_1304, x = q_45_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_106_to_fp16 = const()[name = tensor<string, []>("const_106_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> q_cast = mul(x = var_1305_cast, y = const_106_to_fp16); | |
| tensor<int32, [4]> var_1311 = const()[name = tensor<string, []>("op_1311"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_1312_cast = reshape(shape = var_1311, x = k_45_cast); | |
| tensor<fp16, [1, 1, 1, 1]> const_107_to_fp16 = const()[name = tensor<string, []>("const_107_to_fp16"), val = tensor<fp16, [1, 1, 1, 1]>([[[[0x1.6ap-2]]]])]; | |
| tensor<fp16, [1, 1500, 12, 64]> k_cast = mul(x = var_1312_cast, y = const_107_to_fp16); | |
| tensor<int32, [4]> var_1318 = const()[name = tensor<string, []>("op_1318"), val = tensor<int32, [4]>([1, 1500, 12, -1])]; | |
| tensor<fp16, [1, 1500, 12, 64]> var_1319_cast = reshape(shape = var_1318, x = v_45_cast); | |
| tensor<int32, [4]> var_1320 = const()[name = tensor<string, []>("op_1320"), 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_46_perm_0 = const()[name = tensor<string, []>("transpose_46_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])]; | |
| tensor<int32, [4]> transpose_47_perm_0 = const()[name = tensor<string, []>("transpose_47_perm_0"), val = tensor<int32, [4]>([0, 2, 3, 1])]; | |
| tensor<fp16, [1, 12, 64, 1500]> transpose_49 = transpose(perm = transpose_47_perm_0, x = k_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_50 = transpose(perm = transpose_46_perm_0, x = q_cast); | |
| tensor<fp16, [1, 12, 1500, 1500]> qk_cast = matmul(transpose_x = qk_transpose_x_0, transpose_y = qk_transpose_y_0, x = transpose_50, y = transpose_49); | |
| tensor<fp16, [1, 12, 1500, 1500]> var_1324_cast = softmax(axis = var_1259, x = qk_cast); | |
| tensor<bool, []> var_1326_transpose_x_0 = const()[name = tensor<string, []>("op_1326_transpose_x_0"), val = tensor<bool, []>(false)]; | |
| tensor<bool, []> var_1326_transpose_y_0 = const()[name = tensor<string, []>("op_1326_transpose_y_0"), val = tensor<bool, []>(false)]; | |
| tensor<fp16, [1, 12, 1500, 64]> transpose_51 = transpose(perm = var_1320, x = var_1319_cast); | |
| tensor<fp16, [1, 12, 1500, 64]> var_1326_cast = matmul(transpose_x = var_1326_transpose_x_0, transpose_y = var_1326_transpose_y_0, x = var_1324_cast, y = transpose_51); | |
| tensor<int32, [4]> var_1327 = const()[name = tensor<string, []>("op_1327"), 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, 768])]; | |
| tensor<fp16, [1, 1500, 12, 64]> transpose_48 = transpose(perm = var_1327, x = var_1326_cast); | |
| tensor<fp16, [1, 1500, 768]> x_143_cast = reshape(shape = concat_11, x = transpose_48); | |
| tensor<fp16, [768, 768]> var_1332_to_fp16 = const()[name = tensor<string, []>("op_1332_to_fp16"), val = tensor<fp16, [768, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(165706624)))]; | |
| tensor<fp16, [768]> var_1333_to_fp16 = const()[name = tensor<string, []>("op_1333_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166886336)))]; | |
| tensor<fp16, [1, 1500, 768]> var_1334_cast = linear(bias = var_1333_to_fp16, weight = var_1332_to_fp16, x = x_143_cast); | |
| tensor<fp16, [1, 1500, 768]> x_145_cast = add(x = x_139_cast, y = var_1334_cast); | |
| tensor<int32, [1]> var_1340_axes_0 = const()[name = tensor<string, []>("op_1340_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> blocks_11_mlp_ln_weight_to_fp16 = const()[name = tensor<string, []>("blocks_11_mlp_ln_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166887936)))]; | |
| tensor<fp16, [768]> blocks_11_mlp_ln_bias_to_fp16 = const()[name = tensor<string, []>("blocks_11_mlp_ln_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166889536)))]; | |
| tensor<fp16, [1, 1500, 768]> var_1340_cast = layer_norm(axes = var_1340_axes_0, beta = blocks_11_mlp_ln_bias_to_fp16, epsilon = var_1265_to_fp16, gamma = blocks_11_mlp_ln_weight_to_fp16, x = x_145_cast); | |
| tensor<fp16, [3072, 768]> var_1349_to_fp16 = const()[name = tensor<string, []>("op_1349_to_fp16"), val = tensor<fp16, [3072, 768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(166891136)))]; | |
| tensor<fp16, [3072]> var_1350_to_fp16 = const()[name = tensor<string, []>("op_1350_to_fp16"), val = tensor<fp16, [3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(171609792)))]; | |
| tensor<fp16, [1, 1500, 3072]> input_97_cast = linear(bias = var_1350_to_fp16, weight = var_1349_to_fp16, x = var_1340_cast); | |
| tensor<string, []> x_149_mode_0 = const()[name = tensor<string, []>("x_149_mode_0"), val = tensor<string, []>("EXACT")]; | |
| tensor<fp16, [1, 1500, 3072]> x_149_cast = gelu(mode = x_149_mode_0, x = input_97_cast); | |
| tensor<fp16, [768, 3072]> var_1355_to_fp16 = const()[name = tensor<string, []>("op_1355_to_fp16"), val = tensor<fp16, [768, 3072]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(171616000)))]; | |
| tensor<fp16, [768]> var_1356_to_fp16 = const()[name = tensor<string, []>("op_1356_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(176334656)))]; | |
| tensor<fp16, [1, 1500, 768]> var_1357_cast = linear(bias = var_1356_to_fp16, weight = var_1355_to_fp16, x = x_149_cast); | |
| tensor<fp16, [1, 1500, 768]> x_cast = add(x = x_145_cast, y = var_1357_cast); | |
| tensor<int32, [1]> var_1370_axes_0 = const()[name = tensor<string, []>("op_1370_axes_0"), val = tensor<int32, [1]>([-1])]; | |
| tensor<fp16, [768]> ln_post_weight_to_fp16 = const()[name = tensor<string, []>("ln_post_weight_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(176336256)))]; | |
| tensor<fp16, [768]> ln_post_bias_to_fp16 = const()[name = tensor<string, []>("ln_post_bias_to_fp16"), val = tensor<fp16, [768]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(176337856)))]; | |
| tensor<fp16, []> var_1361_to_fp16 = const()[name = tensor<string, []>("op_1361_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)]; | |
| tensor<fp16, [1, 1500, 768]> var_1370_cast = layer_norm(axes = var_1370_axes_0, beta = ln_post_bias_to_fp16, epsilon = var_1361_to_fp16, gamma = ln_post_weight_to_fp16, x = x_cast); | |
| tensor<string, []> var_1370_cast_to_fp32_dtype_0 = const()[name = tensor<string, []>("op_1370_cast_to_fp32_dtype_0"), val = tensor<string, []>("fp32")]; | |
| tensor<fp32, [1, 1500, 768]> output = cast(dtype = var_1370_cast_to_fp32_dtype_0, x = var_1370_cast); | |
| } -> (output); | |
| } |