File size: 210,515 Bytes
a2d601b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 | program(1.3)
[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3510.2.1"}, {"coremlc-version", "3500.32.1"}, {"coremltools-component-torch", "2.7.0"}, {"coremltools-source-dialect", "TorchScript"}, {"coremltools-version", "9.0"}})]
{
func main<ios18>(tensor<fp32, [1, 1, 390, 390]> attention_mask, tensor<fp32, [1, 128, 3000]> mel_features) {
tensor<int32, [4]> var_669 = const()[name = string("op_669"), val = tensor<int32, [4]>([1, 128, 30, 100])];
string mel_features_to_fp16_dtype_0 = const()[name = string("mel_features_to_fp16_dtype_0"), val = string("fp16")];
tensor<fp16, [1, 128, 3000]> mel_features_to_fp16 = cast(dtype = mel_features_to_fp16_dtype_0, x = mel_features)[name = string("cast_78")];
tensor<fp16, [1, 128, 30, 100]> var_670_cast_fp16 = reshape(shape = var_669, x = mel_features_to_fp16)[name = string("op_670_cast_fp16")];
tensor<int32, [4]> var_675 = const()[name = string("op_675"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [4]> var_681 = const()[name = string("op_681"), val = tensor<int32, [4]>([30, 1, 128, 100])];
tensor<fp16, [1, 30, 128, 100]> var_676_cast_fp16 = transpose(perm = var_675, x = var_670_cast_fp16)[name = string("transpose_163")];
tensor<fp16, [30, 1, 128, 100]> input_1_cast_fp16 = reshape(shape = var_681, x = var_676_cast_fp16)[name = string("input_1_cast_fp16")];
string var_694_pad_type_0 = const()[name = string("op_694_pad_type_0"), val = string("custom")];
tensor<int32, [4]> var_694_pad_0 = const()[name = string("op_694_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
tensor<int32, [2]> var_694_strides_0 = const()[name = string("op_694_strides_0"), val = tensor<int32, [2]>([2, 2])];
tensor<int32, [2]> var_694_dilations_0 = const()[name = string("op_694_dilations_0"), val = tensor<int32, [2]>([1, 1])];
int32 var_694_groups_0 = const()[name = string("op_694_groups_0"), val = int32(1)];
tensor<fp16, [480, 1, 3, 3]> tower_conv2d1_weight_to_fp16 = const()[name = string("tower_conv2d1_weight_to_fp16"), val = tensor<fp16, [480, 1, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64)))];
tensor<fp16, [480]> tower_conv2d1_bias_to_fp16 = const()[name = string("tower_conv2d1_bias_to_fp16"), val = tensor<fp16, [480]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8768)))];
tensor<fp16, [30, 480, 64, 50]> var_694_cast_fp16 = conv(bias = tower_conv2d1_bias_to_fp16, dilations = var_694_dilations_0, groups = var_694_groups_0, pad = var_694_pad_0, pad_type = var_694_pad_type_0, strides = var_694_strides_0, weight = tower_conv2d1_weight_to_fp16, x = input_1_cast_fp16)[name = string("op_694_cast_fp16")];
string input_3_mode_0 = const()[name = string("input_3_mode_0"), val = string("EXACT")];
tensor<fp16, [30, 480, 64, 50]> input_3_cast_fp16 = gelu(mode = input_3_mode_0, x = var_694_cast_fp16)[name = string("input_3_cast_fp16")];
string var_708_pad_type_0 = const()[name = string("op_708_pad_type_0"), val = string("custom")];
tensor<int32, [4]> var_708_pad_0 = const()[name = string("op_708_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
tensor<int32, [2]> var_708_strides_0 = const()[name = string("op_708_strides_0"), val = tensor<int32, [2]>([2, 2])];
tensor<int32, [2]> var_708_dilations_0 = const()[name = string("op_708_dilations_0"), val = tensor<int32, [2]>([1, 1])];
int32 var_708_groups_0 = const()[name = string("op_708_groups_0"), val = int32(1)];
tensor<fp16, [480, 480, 3, 3]> tower_conv2d2_weight_to_fp16 = const()[name = string("tower_conv2d2_weight_to_fp16"), val = tensor<fp16, [480, 480, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(9792)))];
tensor<fp16, [480]> tower_conv2d2_bias_to_fp16 = const()[name = string("tower_conv2d2_bias_to_fp16"), val = tensor<fp16, [480]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4157056)))];
tensor<fp16, [30, 480, 32, 25]> var_708_cast_fp16 = conv(bias = tower_conv2d2_bias_to_fp16, dilations = var_708_dilations_0, groups = var_708_groups_0, pad = var_708_pad_0, pad_type = var_708_pad_type_0, strides = var_708_strides_0, weight = tower_conv2d2_weight_to_fp16, x = input_3_cast_fp16)[name = string("op_708_cast_fp16")];
string input_5_mode_0 = const()[name = string("input_5_mode_0"), val = string("EXACT")];
tensor<fp16, [30, 480, 32, 25]> input_5_cast_fp16 = gelu(mode = input_5_mode_0, x = var_708_cast_fp16)[name = string("input_5_cast_fp16")];
string var_722_pad_type_0 = const()[name = string("op_722_pad_type_0"), val = string("custom")];
tensor<int32, [4]> var_722_pad_0 = const()[name = string("op_722_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
tensor<int32, [2]> var_722_strides_0 = const()[name = string("op_722_strides_0"), val = tensor<int32, [2]>([2, 2])];
tensor<int32, [2]> var_722_dilations_0 = const()[name = string("op_722_dilations_0"), val = tensor<int32, [2]>([1, 1])];
int32 var_722_groups_0 = const()[name = string("op_722_groups_0"), val = int32(1)];
tensor<fp16, [480, 480, 3, 3]> tower_conv2d3_weight_to_fp16 = const()[name = string("tower_conv2d3_weight_to_fp16"), val = tensor<fp16, [480, 480, 3, 3]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4158080)))];
tensor<fp16, [480]> tower_conv2d3_bias_to_fp16 = const()[name = string("tower_conv2d3_bias_to_fp16"), val = tensor<fp16, [480]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8305344)))];
tensor<fp16, [30, 480, 16, 13]> var_722_cast_fp16 = conv(bias = tower_conv2d3_bias_to_fp16, dilations = var_722_dilations_0, groups = var_722_groups_0, pad = var_722_pad_0, pad_type = var_722_pad_type_0, strides = var_722_strides_0, weight = tower_conv2d3_weight_to_fp16, x = input_5_cast_fp16)[name = string("op_722_cast_fp16")];
string hidden_1_mode_0 = const()[name = string("hidden_1_mode_0"), val = string("EXACT")];
tensor<fp16, [30, 480, 16, 13]> hidden_1_cast_fp16 = gelu(mode = hidden_1_mode_0, x = var_722_cast_fp16)[name = string("hidden_1_cast_fp16")];
tensor<int32, [4]> var_729 = const()[name = string("op_729"), val = tensor<int32, [4]>([0, 3, 1, 2])];
tensor<int32, [3]> var_734 = const()[name = string("op_734"), val = tensor<int32, [3]>([30, 13, 7680])];
tensor<fp16, [30, 13, 480, 16]> var_730_cast_fp16 = transpose(perm = var_729, x = hidden_1_cast_fp16)[name = string("transpose_162")];
tensor<fp16, [30, 13, 7680]> input_7_cast_fp16 = reshape(shape = var_734, x = var_730_cast_fp16)[name = string("input_7_cast_fp16")];
tensor<fp16, [896, 7680]> tower_conv_out_weight_to_fp16 = const()[name = string("tower_conv_out_weight_to_fp16"), val = tensor<fp16, [896, 7680]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(8306368)))];
tensor<fp16, [896]> linear_0_bias_0_to_fp16 = const()[name = string("linear_0_bias_0_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22068992)))];
tensor<fp16, [30, 13, 896]> linear_0_cast_fp16 = linear(bias = linear_0_bias_0_to_fp16, weight = tower_conv_out_weight_to_fp16, x = input_7_cast_fp16)[name = string("linear_0_cast_fp16")];
tensor<fp16, [13, 896]> const_0_to_fp16 = const()[name = string("const_0_to_fp16"), val = tensor<fp16, [13, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22070848)))];
tensor<fp16, [30, 13, 896]> hidden_5_cast_fp16 = add(x = linear_0_cast_fp16, y = const_0_to_fp16)[name = string("hidden_5_cast_fp16")];
tensor<int32, [3]> var_754 = const()[name = string("op_754"), val = tensor<int32, [3]>([1, 390, 896])];
tensor<fp16, [1, 390, 896]> input_9_cast_fp16 = reshape(shape = var_754, x = hidden_5_cast_fp16)[name = string("input_9_cast_fp16")];
tensor<int32, [1]> input_11_axes_0 = const()[name = string("input_11_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_0_self_attn_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_0_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22094208)))];
tensor<fp16, [896]> tower_layers_0_self_attn_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_0_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22096064)))];
fp16 var_757_to_fp16 = const()[name = string("op_757_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_11_cast_fp16 = layer_norm(axes = input_11_axes_0, beta = tower_layers_0_self_attn_layer_norm_bias_to_fp16, epsilon = var_757_to_fp16, gamma = tower_layers_0_self_attn_layer_norm_weight_to_fp16, x = input_9_cast_fp16)[name = string("input_11_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_0_self_attn_q_proj_weight_to_fp16 = const()[name = string("tower_layers_0_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(22097920)))];
tensor<fp16, [896]> tower_layers_0_self_attn_q_proj_bias_to_fp16 = const()[name = string("tower_layers_0_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23703616)))];
tensor<fp16, [1, 390, 896]> linear_1_cast_fp16 = linear(bias = tower_layers_0_self_attn_q_proj_bias_to_fp16, weight = tower_layers_0_self_attn_q_proj_weight_to_fp16, x = input_11_cast_fp16)[name = string("linear_1_cast_fp16")];
tensor<int32, [4]> var_770 = const()[name = string("op_770"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> query_1_cast_fp16 = reshape(shape = var_770, x = linear_1_cast_fp16)[name = string("query_1_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_0_self_attn_k_proj_weight_to_fp16 = const()[name = string("tower_layers_0_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(23705472)))];
tensor<fp16, [896]> tower_layers_0_self_attn_k_proj_bias_to_fp16 = const()[name = string("tower_layers_0_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25311168)))];
tensor<fp16, [1, 390, 896]> linear_2_cast_fp16 = linear(bias = tower_layers_0_self_attn_k_proj_bias_to_fp16, weight = tower_layers_0_self_attn_k_proj_weight_to_fp16, x = input_11_cast_fp16)[name = string("linear_2_cast_fp16")];
tensor<int32, [4]> var_779 = const()[name = string("op_779"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> key_1_cast_fp16 = reshape(shape = var_779, x = linear_2_cast_fp16)[name = string("key_1_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_0_self_attn_v_proj_weight_to_fp16 = const()[name = string("tower_layers_0_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(25313024)))];
tensor<fp16, [896]> tower_layers_0_self_attn_v_proj_bias_to_fp16 = const()[name = string("tower_layers_0_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26918720)))];
tensor<fp16, [1, 390, 896]> linear_3_cast_fp16 = linear(bias = tower_layers_0_self_attn_v_proj_bias_to_fp16, weight = tower_layers_0_self_attn_v_proj_weight_to_fp16, x = input_11_cast_fp16)[name = string("linear_3_cast_fp16")];
tensor<int32, [4]> var_788 = const()[name = string("op_788"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> value_1_cast_fp16 = reshape(shape = var_788, x = linear_3_cast_fp16)[name = string("value_1_cast_fp16")];
tensor<int32, [4]> var_806 = const()[name = string("op_806"), val = tensor<int32, [4]>([0, 2, 1, 3])];
bool var_811_transpose_x_0 = const()[name = string("op_811_transpose_x_0"), val = bool(false)];
bool var_811_transpose_y_0 = const()[name = string("op_811_transpose_y_0"), val = bool(false)];
tensor<int32, [4]> transpose_54_perm_0 = const()[name = string("transpose_54_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
tensor<int32, [4]> transpose_55_perm_0 = const()[name = string("transpose_55_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
tensor<fp16, [1, 14, 64, 390]> transpose_55 = transpose(perm = transpose_55_perm_0, x = key_1_cast_fp16)[name = string("transpose_159")];
tensor<fp16, [1, 14, 390, 64]> transpose_54 = transpose(perm = transpose_54_perm_0, x = query_1_cast_fp16)[name = string("transpose_160")];
tensor<fp16, [1, 14, 390, 390]> var_811_cast_fp16 = matmul(transpose_x = var_811_transpose_x_0, transpose_y = var_811_transpose_y_0, x = transpose_54, y = transpose_55)[name = string("op_811_cast_fp16")];
fp16 var_812_to_fp16 = const()[name = string("op_812_to_fp16"), val = fp16(0x1p-3)];
tensor<fp16, [1, 14, 390, 390]> scores_1_cast_fp16 = mul(x = var_811_cast_fp16, y = var_812_to_fp16)[name = string("scores_1_cast_fp16")];
string attention_mask_to_fp16_dtype_0 = const()[name = string("attention_mask_to_fp16_dtype_0"), val = string("fp16")];
tensor<fp16, [1, 1, 390, 390]> attention_mask_to_fp16 = cast(dtype = attention_mask_to_fp16_dtype_0, x = attention_mask)[name = string("cast_77")];
tensor<fp16, [1, 14, 390, 390]> scores_3_cast_fp16 = add(x = scores_1_cast_fp16, y = attention_mask_to_fp16)[name = string("scores_3_cast_fp16")];
int32 var_821 = const()[name = string("op_821"), val = int32(-1)];
tensor<fp16, [1, 14, 390, 390]> var_823_cast_fp16 = softmax(axis = var_821, x = scores_3_cast_fp16)[name = string("op_823_cast_fp16")];
bool output_1_transpose_x_0 = const()[name = string("output_1_transpose_x_0"), val = bool(false)];
bool output_1_transpose_y_0 = const()[name = string("output_1_transpose_y_0"), val = bool(false)];
tensor<fp16, [1, 14, 390, 64]> value_3_cast_fp16 = transpose(perm = var_806, x = value_1_cast_fp16)[name = string("transpose_161")];
tensor<fp16, [1, 14, 390, 64]> output_1_cast_fp16 = matmul(transpose_x = output_1_transpose_x_0, transpose_y = output_1_transpose_y_0, x = var_823_cast_fp16, y = value_3_cast_fp16)[name = string("output_1_cast_fp16")];
tensor<int32, [4]> var_834 = const()[name = string("op_834"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_839 = const()[name = string("op_839"), val = tensor<int32, [3]>([1, 390, 896])];
tensor<fp16, [1, 390, 14, 64]> var_835_cast_fp16 = transpose(perm = var_834, x = output_1_cast_fp16)[name = string("transpose_158")];
tensor<fp16, [1, 390, 896]> input_13_cast_fp16 = reshape(shape = var_839, x = var_835_cast_fp16)[name = string("input_13_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_0_self_attn_out_proj_weight_to_fp16 = const()[name = string("tower_layers_0_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(26920576)))];
tensor<fp16, [896]> tower_layers_0_self_attn_out_proj_bias_to_fp16 = const()[name = string("tower_layers_0_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28526272)))];
tensor<fp16, [1, 390, 896]> linear_4_cast_fp16 = linear(bias = tower_layers_0_self_attn_out_proj_bias_to_fp16, weight = tower_layers_0_self_attn_out_proj_weight_to_fp16, x = input_13_cast_fp16)[name = string("linear_4_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_15_cast_fp16 = add(x = input_9_cast_fp16, y = linear_4_cast_fp16)[name = string("input_15_cast_fp16")];
tensor<int32, [1]> input_17_axes_0 = const()[name = string("input_17_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_0_final_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_0_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28528128)))];
tensor<fp16, [896]> tower_layers_0_final_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_0_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28529984)))];
fp16 var_847_to_fp16 = const()[name = string("op_847_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_17_cast_fp16 = layer_norm(axes = input_17_axes_0, beta = tower_layers_0_final_layer_norm_bias_to_fp16, epsilon = var_847_to_fp16, gamma = tower_layers_0_final_layer_norm_weight_to_fp16, x = input_15_cast_fp16)[name = string("input_17_cast_fp16")];
tensor<fp16, [3584, 896]> tower_layers_0_fc1_weight_to_fp16 = const()[name = string("tower_layers_0_fc1_weight_to_fp16"), val = tensor<fp16, [3584, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(28531840)))];
tensor<fp16, [3584]> tower_layers_0_fc1_bias_to_fp16 = const()[name = string("tower_layers_0_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34954432)))];
tensor<fp16, [1, 390, 3584]> linear_5_cast_fp16 = linear(bias = tower_layers_0_fc1_bias_to_fp16, weight = tower_layers_0_fc1_weight_to_fp16, x = input_17_cast_fp16)[name = string("linear_5_cast_fp16")];
string input_21_mode_0 = const()[name = string("input_21_mode_0"), val = string("EXACT")];
tensor<fp16, [1, 390, 3584]> input_21_cast_fp16 = gelu(mode = input_21_mode_0, x = linear_5_cast_fp16)[name = string("input_21_cast_fp16")];
tensor<fp16, [896, 3584]> tower_layers_0_fc2_weight_to_fp16 = const()[name = string("tower_layers_0_fc2_weight_to_fp16"), val = tensor<fp16, [896, 3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(34961664)))];
tensor<fp16, [896]> tower_layers_0_fc2_bias_to_fp16 = const()[name = string("tower_layers_0_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41384256)))];
tensor<fp16, [1, 390, 896]> linear_6_cast_fp16 = linear(bias = tower_layers_0_fc2_bias_to_fp16, weight = tower_layers_0_fc2_weight_to_fp16, x = input_21_cast_fp16)[name = string("linear_6_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_23_cast_fp16 = add(x = input_15_cast_fp16, y = linear_6_cast_fp16)[name = string("input_23_cast_fp16")];
tensor<int32, [1]> input_25_axes_0 = const()[name = string("input_25_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_1_self_attn_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_1_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41386112)))];
tensor<fp16, [896]> tower_layers_1_self_attn_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_1_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41387968)))];
fp16 var_864_to_fp16 = const()[name = string("op_864_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_25_cast_fp16 = layer_norm(axes = input_25_axes_0, beta = tower_layers_1_self_attn_layer_norm_bias_to_fp16, epsilon = var_864_to_fp16, gamma = tower_layers_1_self_attn_layer_norm_weight_to_fp16, x = input_23_cast_fp16)[name = string("input_25_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_1_self_attn_q_proj_weight_to_fp16 = const()[name = string("tower_layers_1_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(41389824)))];
tensor<fp16, [896]> tower_layers_1_self_attn_q_proj_bias_to_fp16 = const()[name = string("tower_layers_1_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42995520)))];
tensor<fp16, [1, 390, 896]> linear_7_cast_fp16 = linear(bias = tower_layers_1_self_attn_q_proj_bias_to_fp16, weight = tower_layers_1_self_attn_q_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("linear_7_cast_fp16")];
tensor<int32, [4]> var_877 = const()[name = string("op_877"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> query_5_cast_fp16 = reshape(shape = var_877, x = linear_7_cast_fp16)[name = string("query_5_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_1_self_attn_k_proj_weight_to_fp16 = const()[name = string("tower_layers_1_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(42997376)))];
tensor<fp16, [896]> tower_layers_1_self_attn_k_proj_bias_to_fp16 = const()[name = string("tower_layers_1_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44603072)))];
tensor<fp16, [1, 390, 896]> linear_8_cast_fp16 = linear(bias = tower_layers_1_self_attn_k_proj_bias_to_fp16, weight = tower_layers_1_self_attn_k_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("linear_8_cast_fp16")];
tensor<int32, [4]> var_886 = const()[name = string("op_886"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> key_5_cast_fp16 = reshape(shape = var_886, x = linear_8_cast_fp16)[name = string("key_5_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_1_self_attn_v_proj_weight_to_fp16 = const()[name = string("tower_layers_1_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(44604928)))];
tensor<fp16, [896]> tower_layers_1_self_attn_v_proj_bias_to_fp16 = const()[name = string("tower_layers_1_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46210624)))];
tensor<fp16, [1, 390, 896]> linear_9_cast_fp16 = linear(bias = tower_layers_1_self_attn_v_proj_bias_to_fp16, weight = tower_layers_1_self_attn_v_proj_weight_to_fp16, x = input_25_cast_fp16)[name = string("linear_9_cast_fp16")];
tensor<int32, [4]> var_895 = const()[name = string("op_895"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> value_5_cast_fp16 = reshape(shape = var_895, x = linear_9_cast_fp16)[name = string("value_5_cast_fp16")];
tensor<int32, [4]> var_913 = const()[name = string("op_913"), val = tensor<int32, [4]>([0, 2, 1, 3])];
bool var_918_transpose_x_0 = const()[name = string("op_918_transpose_x_0"), val = bool(false)];
bool var_918_transpose_y_0 = const()[name = string("op_918_transpose_y_0"), val = bool(false)];
tensor<int32, [4]> transpose_56_perm_0 = const()[name = string("transpose_56_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
tensor<int32, [4]> transpose_57_perm_0 = const()[name = string("transpose_57_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
tensor<fp16, [1, 14, 64, 390]> transpose_57 = transpose(perm = transpose_57_perm_0, x = key_5_cast_fp16)[name = string("transpose_155")];
tensor<fp16, [1, 14, 390, 64]> transpose_56 = transpose(perm = transpose_56_perm_0, x = query_5_cast_fp16)[name = string("transpose_156")];
tensor<fp16, [1, 14, 390, 390]> var_918_cast_fp16 = matmul(transpose_x = var_918_transpose_x_0, transpose_y = var_918_transpose_y_0, x = transpose_56, y = transpose_57)[name = string("op_918_cast_fp16")];
fp16 var_919_to_fp16 = const()[name = string("op_919_to_fp16"), val = fp16(0x1p-3)];
tensor<fp16, [1, 14, 390, 390]> scores_7_cast_fp16 = mul(x = var_918_cast_fp16, y = var_919_to_fp16)[name = string("scores_7_cast_fp16")];
tensor<fp16, [1, 14, 390, 390]> scores_9_cast_fp16 = add(x = scores_7_cast_fp16, y = attention_mask_to_fp16)[name = string("scores_9_cast_fp16")];
int32 var_928 = const()[name = string("op_928"), val = int32(-1)];
tensor<fp16, [1, 14, 390, 390]> var_930_cast_fp16 = softmax(axis = var_928, x = scores_9_cast_fp16)[name = string("op_930_cast_fp16")];
bool output_3_transpose_x_0 = const()[name = string("output_3_transpose_x_0"), val = bool(false)];
bool output_3_transpose_y_0 = const()[name = string("output_3_transpose_y_0"), val = bool(false)];
tensor<fp16, [1, 14, 390, 64]> value_7_cast_fp16 = transpose(perm = var_913, x = value_5_cast_fp16)[name = string("transpose_157")];
tensor<fp16, [1, 14, 390, 64]> output_3_cast_fp16 = matmul(transpose_x = output_3_transpose_x_0, transpose_y = output_3_transpose_y_0, x = var_930_cast_fp16, y = value_7_cast_fp16)[name = string("output_3_cast_fp16")];
tensor<int32, [4]> var_941 = const()[name = string("op_941"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_946 = const()[name = string("op_946"), val = tensor<int32, [3]>([1, 390, 896])];
tensor<fp16, [1, 390, 14, 64]> var_942_cast_fp16 = transpose(perm = var_941, x = output_3_cast_fp16)[name = string("transpose_154")];
tensor<fp16, [1, 390, 896]> input_27_cast_fp16 = reshape(shape = var_946, x = var_942_cast_fp16)[name = string("input_27_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_1_self_attn_out_proj_weight_to_fp16 = const()[name = string("tower_layers_1_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(46212480)))];
tensor<fp16, [896]> tower_layers_1_self_attn_out_proj_bias_to_fp16 = const()[name = string("tower_layers_1_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47818176)))];
tensor<fp16, [1, 390, 896]> linear_10_cast_fp16 = linear(bias = tower_layers_1_self_attn_out_proj_bias_to_fp16, weight = tower_layers_1_self_attn_out_proj_weight_to_fp16, x = input_27_cast_fp16)[name = string("linear_10_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_29_cast_fp16 = add(x = input_23_cast_fp16, y = linear_10_cast_fp16)[name = string("input_29_cast_fp16")];
tensor<int32, [1]> input_31_axes_0 = const()[name = string("input_31_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_1_final_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_1_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47820032)))];
tensor<fp16, [896]> tower_layers_1_final_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_1_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47821888)))];
fp16 var_954_to_fp16 = const()[name = string("op_954_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_31_cast_fp16 = layer_norm(axes = input_31_axes_0, beta = tower_layers_1_final_layer_norm_bias_to_fp16, epsilon = var_954_to_fp16, gamma = tower_layers_1_final_layer_norm_weight_to_fp16, x = input_29_cast_fp16)[name = string("input_31_cast_fp16")];
tensor<fp16, [3584, 896]> tower_layers_1_fc1_weight_to_fp16 = const()[name = string("tower_layers_1_fc1_weight_to_fp16"), val = tensor<fp16, [3584, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(47823744)))];
tensor<fp16, [3584]> tower_layers_1_fc1_bias_to_fp16 = const()[name = string("tower_layers_1_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54246336)))];
tensor<fp16, [1, 390, 3584]> linear_11_cast_fp16 = linear(bias = tower_layers_1_fc1_bias_to_fp16, weight = tower_layers_1_fc1_weight_to_fp16, x = input_31_cast_fp16)[name = string("linear_11_cast_fp16")];
string input_35_mode_0 = const()[name = string("input_35_mode_0"), val = string("EXACT")];
tensor<fp16, [1, 390, 3584]> input_35_cast_fp16 = gelu(mode = input_35_mode_0, x = linear_11_cast_fp16)[name = string("input_35_cast_fp16")];
tensor<fp16, [896, 3584]> tower_layers_1_fc2_weight_to_fp16 = const()[name = string("tower_layers_1_fc2_weight_to_fp16"), val = tensor<fp16, [896, 3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(54253568)))];
tensor<fp16, [896]> tower_layers_1_fc2_bias_to_fp16 = const()[name = string("tower_layers_1_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60676160)))];
tensor<fp16, [1, 390, 896]> linear_12_cast_fp16 = linear(bias = tower_layers_1_fc2_bias_to_fp16, weight = tower_layers_1_fc2_weight_to_fp16, x = input_35_cast_fp16)[name = string("linear_12_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_37_cast_fp16 = add(x = input_29_cast_fp16, y = linear_12_cast_fp16)[name = string("input_37_cast_fp16")];
tensor<int32, [1]> input_39_axes_0 = const()[name = string("input_39_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_2_self_attn_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_2_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60678016)))];
tensor<fp16, [896]> tower_layers_2_self_attn_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_2_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60679872)))];
fp16 var_971_to_fp16 = const()[name = string("op_971_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_39_cast_fp16 = layer_norm(axes = input_39_axes_0, beta = tower_layers_2_self_attn_layer_norm_bias_to_fp16, epsilon = var_971_to_fp16, gamma = tower_layers_2_self_attn_layer_norm_weight_to_fp16, x = input_37_cast_fp16)[name = string("input_39_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_2_self_attn_q_proj_weight_to_fp16 = const()[name = string("tower_layers_2_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(60681728)))];
tensor<fp16, [896]> tower_layers_2_self_attn_q_proj_bias_to_fp16 = const()[name = string("tower_layers_2_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62287424)))];
tensor<fp16, [1, 390, 896]> linear_13_cast_fp16 = linear(bias = tower_layers_2_self_attn_q_proj_bias_to_fp16, weight = tower_layers_2_self_attn_q_proj_weight_to_fp16, x = input_39_cast_fp16)[name = string("linear_13_cast_fp16")];
tensor<int32, [4]> var_984 = const()[name = string("op_984"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> query_9_cast_fp16 = reshape(shape = var_984, x = linear_13_cast_fp16)[name = string("query_9_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_2_self_attn_k_proj_weight_to_fp16 = const()[name = string("tower_layers_2_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(62289280)))];
tensor<fp16, [896]> tower_layers_2_self_attn_k_proj_bias_to_fp16 = const()[name = string("tower_layers_2_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63894976)))];
tensor<fp16, [1, 390, 896]> linear_14_cast_fp16 = linear(bias = tower_layers_2_self_attn_k_proj_bias_to_fp16, weight = tower_layers_2_self_attn_k_proj_weight_to_fp16, x = input_39_cast_fp16)[name = string("linear_14_cast_fp16")];
tensor<int32, [4]> var_993 = const()[name = string("op_993"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> key_9_cast_fp16 = reshape(shape = var_993, x = linear_14_cast_fp16)[name = string("key_9_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_2_self_attn_v_proj_weight_to_fp16 = const()[name = string("tower_layers_2_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(63896832)))];
tensor<fp16, [896]> tower_layers_2_self_attn_v_proj_bias_to_fp16 = const()[name = string("tower_layers_2_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65502528)))];
tensor<fp16, [1, 390, 896]> linear_15_cast_fp16 = linear(bias = tower_layers_2_self_attn_v_proj_bias_to_fp16, weight = tower_layers_2_self_attn_v_proj_weight_to_fp16, x = input_39_cast_fp16)[name = string("linear_15_cast_fp16")];
tensor<int32, [4]> var_1002 = const()[name = string("op_1002"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> value_9_cast_fp16 = reshape(shape = var_1002, x = linear_15_cast_fp16)[name = string("value_9_cast_fp16")];
tensor<int32, [4]> var_1020 = const()[name = string("op_1020"), val = tensor<int32, [4]>([0, 2, 1, 3])];
bool var_1025_transpose_x_0 = const()[name = string("op_1025_transpose_x_0"), val = bool(false)];
bool var_1025_transpose_y_0 = const()[name = string("op_1025_transpose_y_0"), val = bool(false)];
tensor<int32, [4]> transpose_58_perm_0 = const()[name = string("transpose_58_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
tensor<int32, [4]> transpose_59_perm_0 = const()[name = string("transpose_59_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
tensor<fp16, [1, 14, 64, 390]> transpose_59 = transpose(perm = transpose_59_perm_0, x = key_9_cast_fp16)[name = string("transpose_151")];
tensor<fp16, [1, 14, 390, 64]> transpose_58 = transpose(perm = transpose_58_perm_0, x = query_9_cast_fp16)[name = string("transpose_152")];
tensor<fp16, [1, 14, 390, 390]> var_1025_cast_fp16 = matmul(transpose_x = var_1025_transpose_x_0, transpose_y = var_1025_transpose_y_0, x = transpose_58, y = transpose_59)[name = string("op_1025_cast_fp16")];
fp16 var_1026_to_fp16 = const()[name = string("op_1026_to_fp16"), val = fp16(0x1p-3)];
tensor<fp16, [1, 14, 390, 390]> scores_13_cast_fp16 = mul(x = var_1025_cast_fp16, y = var_1026_to_fp16)[name = string("scores_13_cast_fp16")];
tensor<fp16, [1, 14, 390, 390]> scores_15_cast_fp16 = add(x = scores_13_cast_fp16, y = attention_mask_to_fp16)[name = string("scores_15_cast_fp16")];
int32 var_1035 = const()[name = string("op_1035"), val = int32(-1)];
tensor<fp16, [1, 14, 390, 390]> var_1037_cast_fp16 = softmax(axis = var_1035, x = scores_15_cast_fp16)[name = string("op_1037_cast_fp16")];
bool output_5_transpose_x_0 = const()[name = string("output_5_transpose_x_0"), val = bool(false)];
bool output_5_transpose_y_0 = const()[name = string("output_5_transpose_y_0"), val = bool(false)];
tensor<fp16, [1, 14, 390, 64]> value_11_cast_fp16 = transpose(perm = var_1020, x = value_9_cast_fp16)[name = string("transpose_153")];
tensor<fp16, [1, 14, 390, 64]> output_5_cast_fp16 = matmul(transpose_x = output_5_transpose_x_0, transpose_y = output_5_transpose_y_0, x = var_1037_cast_fp16, y = value_11_cast_fp16)[name = string("output_5_cast_fp16")];
tensor<int32, [4]> var_1048 = const()[name = string("op_1048"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_1053 = const()[name = string("op_1053"), val = tensor<int32, [3]>([1, 390, 896])];
tensor<fp16, [1, 390, 14, 64]> var_1049_cast_fp16 = transpose(perm = var_1048, x = output_5_cast_fp16)[name = string("transpose_150")];
tensor<fp16, [1, 390, 896]> input_41_cast_fp16 = reshape(shape = var_1053, x = var_1049_cast_fp16)[name = string("input_41_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_2_self_attn_out_proj_weight_to_fp16 = const()[name = string("tower_layers_2_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(65504384)))];
tensor<fp16, [896]> tower_layers_2_self_attn_out_proj_bias_to_fp16 = const()[name = string("tower_layers_2_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67110080)))];
tensor<fp16, [1, 390, 896]> linear_16_cast_fp16 = linear(bias = tower_layers_2_self_attn_out_proj_bias_to_fp16, weight = tower_layers_2_self_attn_out_proj_weight_to_fp16, x = input_41_cast_fp16)[name = string("linear_16_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_43_cast_fp16 = add(x = input_37_cast_fp16, y = linear_16_cast_fp16)[name = string("input_43_cast_fp16")];
tensor<int32, [1]> input_45_axes_0 = const()[name = string("input_45_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_2_final_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_2_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67111936)))];
tensor<fp16, [896]> tower_layers_2_final_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_2_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67113792)))];
fp16 var_1061_to_fp16 = const()[name = string("op_1061_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_45_cast_fp16 = layer_norm(axes = input_45_axes_0, beta = tower_layers_2_final_layer_norm_bias_to_fp16, epsilon = var_1061_to_fp16, gamma = tower_layers_2_final_layer_norm_weight_to_fp16, x = input_43_cast_fp16)[name = string("input_45_cast_fp16")];
tensor<fp16, [3584, 896]> tower_layers_2_fc1_weight_to_fp16 = const()[name = string("tower_layers_2_fc1_weight_to_fp16"), val = tensor<fp16, [3584, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(67115648)))];
tensor<fp16, [3584]> tower_layers_2_fc1_bias_to_fp16 = const()[name = string("tower_layers_2_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73538240)))];
tensor<fp16, [1, 390, 3584]> linear_17_cast_fp16 = linear(bias = tower_layers_2_fc1_bias_to_fp16, weight = tower_layers_2_fc1_weight_to_fp16, x = input_45_cast_fp16)[name = string("linear_17_cast_fp16")];
string input_49_mode_0 = const()[name = string("input_49_mode_0"), val = string("EXACT")];
tensor<fp16, [1, 390, 3584]> input_49_cast_fp16 = gelu(mode = input_49_mode_0, x = linear_17_cast_fp16)[name = string("input_49_cast_fp16")];
tensor<fp16, [896, 3584]> tower_layers_2_fc2_weight_to_fp16 = const()[name = string("tower_layers_2_fc2_weight_to_fp16"), val = tensor<fp16, [896, 3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(73545472)))];
tensor<fp16, [896]> tower_layers_2_fc2_bias_to_fp16 = const()[name = string("tower_layers_2_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79968064)))];
tensor<fp16, [1, 390, 896]> linear_18_cast_fp16 = linear(bias = tower_layers_2_fc2_bias_to_fp16, weight = tower_layers_2_fc2_weight_to_fp16, x = input_49_cast_fp16)[name = string("linear_18_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_51_cast_fp16 = add(x = input_43_cast_fp16, y = linear_18_cast_fp16)[name = string("input_51_cast_fp16")];
tensor<int32, [1]> input_53_axes_0 = const()[name = string("input_53_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_3_self_attn_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_3_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79969920)))];
tensor<fp16, [896]> tower_layers_3_self_attn_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_3_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79971776)))];
fp16 var_1078_to_fp16 = const()[name = string("op_1078_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_53_cast_fp16 = layer_norm(axes = input_53_axes_0, beta = tower_layers_3_self_attn_layer_norm_bias_to_fp16, epsilon = var_1078_to_fp16, gamma = tower_layers_3_self_attn_layer_norm_weight_to_fp16, x = input_51_cast_fp16)[name = string("input_53_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_3_self_attn_q_proj_weight_to_fp16 = const()[name = string("tower_layers_3_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(79973632)))];
tensor<fp16, [896]> tower_layers_3_self_attn_q_proj_bias_to_fp16 = const()[name = string("tower_layers_3_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81579328)))];
tensor<fp16, [1, 390, 896]> linear_19_cast_fp16 = linear(bias = tower_layers_3_self_attn_q_proj_bias_to_fp16, weight = tower_layers_3_self_attn_q_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("linear_19_cast_fp16")];
tensor<int32, [4]> var_1091 = const()[name = string("op_1091"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> query_13_cast_fp16 = reshape(shape = var_1091, x = linear_19_cast_fp16)[name = string("query_13_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_3_self_attn_k_proj_weight_to_fp16 = const()[name = string("tower_layers_3_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(81581184)))];
tensor<fp16, [896]> tower_layers_3_self_attn_k_proj_bias_to_fp16 = const()[name = string("tower_layers_3_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83186880)))];
tensor<fp16, [1, 390, 896]> linear_20_cast_fp16 = linear(bias = tower_layers_3_self_attn_k_proj_bias_to_fp16, weight = tower_layers_3_self_attn_k_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("linear_20_cast_fp16")];
tensor<int32, [4]> var_1100 = const()[name = string("op_1100"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> key_13_cast_fp16 = reshape(shape = var_1100, x = linear_20_cast_fp16)[name = string("key_13_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_3_self_attn_v_proj_weight_to_fp16 = const()[name = string("tower_layers_3_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(83188736)))];
tensor<fp16, [896]> tower_layers_3_self_attn_v_proj_bias_to_fp16 = const()[name = string("tower_layers_3_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(84794432)))];
tensor<fp16, [1, 390, 896]> linear_21_cast_fp16 = linear(bias = tower_layers_3_self_attn_v_proj_bias_to_fp16, weight = tower_layers_3_self_attn_v_proj_weight_to_fp16, x = input_53_cast_fp16)[name = string("linear_21_cast_fp16")];
tensor<int32, [4]> var_1109 = const()[name = string("op_1109"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> value_13_cast_fp16 = reshape(shape = var_1109, x = linear_21_cast_fp16)[name = string("value_13_cast_fp16")];
tensor<int32, [4]> var_1127 = const()[name = string("op_1127"), val = tensor<int32, [4]>([0, 2, 1, 3])];
bool var_1132_transpose_x_0 = const()[name = string("op_1132_transpose_x_0"), val = bool(false)];
bool var_1132_transpose_y_0 = const()[name = string("op_1132_transpose_y_0"), val = bool(false)];
tensor<int32, [4]> transpose_60_perm_0 = const()[name = string("transpose_60_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
tensor<int32, [4]> transpose_61_perm_0 = const()[name = string("transpose_61_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
tensor<fp16, [1, 14, 64, 390]> transpose_61 = transpose(perm = transpose_61_perm_0, x = key_13_cast_fp16)[name = string("transpose_147")];
tensor<fp16, [1, 14, 390, 64]> transpose_60 = transpose(perm = transpose_60_perm_0, x = query_13_cast_fp16)[name = string("transpose_148")];
tensor<fp16, [1, 14, 390, 390]> var_1132_cast_fp16 = matmul(transpose_x = var_1132_transpose_x_0, transpose_y = var_1132_transpose_y_0, x = transpose_60, y = transpose_61)[name = string("op_1132_cast_fp16")];
fp16 var_1133_to_fp16 = const()[name = string("op_1133_to_fp16"), val = fp16(0x1p-3)];
tensor<fp16, [1, 14, 390, 390]> scores_19_cast_fp16 = mul(x = var_1132_cast_fp16, y = var_1133_to_fp16)[name = string("scores_19_cast_fp16")];
tensor<fp16, [1, 14, 390, 390]> scores_21_cast_fp16 = add(x = scores_19_cast_fp16, y = attention_mask_to_fp16)[name = string("scores_21_cast_fp16")];
int32 var_1142 = const()[name = string("op_1142"), val = int32(-1)];
tensor<fp16, [1, 14, 390, 390]> var_1144_cast_fp16 = softmax(axis = var_1142, x = scores_21_cast_fp16)[name = string("op_1144_cast_fp16")];
bool output_7_transpose_x_0 = const()[name = string("output_7_transpose_x_0"), val = bool(false)];
bool output_7_transpose_y_0 = const()[name = string("output_7_transpose_y_0"), val = bool(false)];
tensor<fp16, [1, 14, 390, 64]> value_15_cast_fp16 = transpose(perm = var_1127, x = value_13_cast_fp16)[name = string("transpose_149")];
tensor<fp16, [1, 14, 390, 64]> output_7_cast_fp16 = matmul(transpose_x = output_7_transpose_x_0, transpose_y = output_7_transpose_y_0, x = var_1144_cast_fp16, y = value_15_cast_fp16)[name = string("output_7_cast_fp16")];
tensor<int32, [4]> var_1155 = const()[name = string("op_1155"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_1160 = const()[name = string("op_1160"), val = tensor<int32, [3]>([1, 390, 896])];
tensor<fp16, [1, 390, 14, 64]> var_1156_cast_fp16 = transpose(perm = var_1155, x = output_7_cast_fp16)[name = string("transpose_146")];
tensor<fp16, [1, 390, 896]> input_55_cast_fp16 = reshape(shape = var_1160, x = var_1156_cast_fp16)[name = string("input_55_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_3_self_attn_out_proj_weight_to_fp16 = const()[name = string("tower_layers_3_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(84796288)))];
tensor<fp16, [896]> tower_layers_3_self_attn_out_proj_bias_to_fp16 = const()[name = string("tower_layers_3_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86401984)))];
tensor<fp16, [1, 390, 896]> linear_22_cast_fp16 = linear(bias = tower_layers_3_self_attn_out_proj_bias_to_fp16, weight = tower_layers_3_self_attn_out_proj_weight_to_fp16, x = input_55_cast_fp16)[name = string("linear_22_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_57_cast_fp16 = add(x = input_51_cast_fp16, y = linear_22_cast_fp16)[name = string("input_57_cast_fp16")];
tensor<int32, [1]> input_59_axes_0 = const()[name = string("input_59_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_3_final_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_3_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86403840)))];
tensor<fp16, [896]> tower_layers_3_final_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_3_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86405696)))];
fp16 var_1168_to_fp16 = const()[name = string("op_1168_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_59_cast_fp16 = layer_norm(axes = input_59_axes_0, beta = tower_layers_3_final_layer_norm_bias_to_fp16, epsilon = var_1168_to_fp16, gamma = tower_layers_3_final_layer_norm_weight_to_fp16, x = input_57_cast_fp16)[name = string("input_59_cast_fp16")];
tensor<fp16, [3584, 896]> tower_layers_3_fc1_weight_to_fp16 = const()[name = string("tower_layers_3_fc1_weight_to_fp16"), val = tensor<fp16, [3584, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(86407552)))];
tensor<fp16, [3584]> tower_layers_3_fc1_bias_to_fp16 = const()[name = string("tower_layers_3_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92830144)))];
tensor<fp16, [1, 390, 3584]> linear_23_cast_fp16 = linear(bias = tower_layers_3_fc1_bias_to_fp16, weight = tower_layers_3_fc1_weight_to_fp16, x = input_59_cast_fp16)[name = string("linear_23_cast_fp16")];
string input_63_mode_0 = const()[name = string("input_63_mode_0"), val = string("EXACT")];
tensor<fp16, [1, 390, 3584]> input_63_cast_fp16 = gelu(mode = input_63_mode_0, x = linear_23_cast_fp16)[name = string("input_63_cast_fp16")];
tensor<fp16, [896, 3584]> tower_layers_3_fc2_weight_to_fp16 = const()[name = string("tower_layers_3_fc2_weight_to_fp16"), val = tensor<fp16, [896, 3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(92837376)))];
tensor<fp16, [896]> tower_layers_3_fc2_bias_to_fp16 = const()[name = string("tower_layers_3_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99259968)))];
tensor<fp16, [1, 390, 896]> linear_24_cast_fp16 = linear(bias = tower_layers_3_fc2_bias_to_fp16, weight = tower_layers_3_fc2_weight_to_fp16, x = input_63_cast_fp16)[name = string("linear_24_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_65_cast_fp16 = add(x = input_57_cast_fp16, y = linear_24_cast_fp16)[name = string("input_65_cast_fp16")];
tensor<int32, [1]> input_67_axes_0 = const()[name = string("input_67_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_4_self_attn_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_4_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99261824)))];
tensor<fp16, [896]> tower_layers_4_self_attn_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_4_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99263680)))];
fp16 var_1185_to_fp16 = const()[name = string("op_1185_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_67_cast_fp16 = layer_norm(axes = input_67_axes_0, beta = tower_layers_4_self_attn_layer_norm_bias_to_fp16, epsilon = var_1185_to_fp16, gamma = tower_layers_4_self_attn_layer_norm_weight_to_fp16, x = input_65_cast_fp16)[name = string("input_67_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_4_self_attn_q_proj_weight_to_fp16 = const()[name = string("tower_layers_4_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(99265536)))];
tensor<fp16, [896]> tower_layers_4_self_attn_q_proj_bias_to_fp16 = const()[name = string("tower_layers_4_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100871232)))];
tensor<fp16, [1, 390, 896]> linear_25_cast_fp16 = linear(bias = tower_layers_4_self_attn_q_proj_bias_to_fp16, weight = tower_layers_4_self_attn_q_proj_weight_to_fp16, x = input_67_cast_fp16)[name = string("linear_25_cast_fp16")];
tensor<int32, [4]> var_1198 = const()[name = string("op_1198"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> query_17_cast_fp16 = reshape(shape = var_1198, x = linear_25_cast_fp16)[name = string("query_17_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_4_self_attn_k_proj_weight_to_fp16 = const()[name = string("tower_layers_4_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(100873088)))];
tensor<fp16, [896]> tower_layers_4_self_attn_k_proj_bias_to_fp16 = const()[name = string("tower_layers_4_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102478784)))];
tensor<fp16, [1, 390, 896]> linear_26_cast_fp16 = linear(bias = tower_layers_4_self_attn_k_proj_bias_to_fp16, weight = tower_layers_4_self_attn_k_proj_weight_to_fp16, x = input_67_cast_fp16)[name = string("linear_26_cast_fp16")];
tensor<int32, [4]> var_1207 = const()[name = string("op_1207"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> key_17_cast_fp16 = reshape(shape = var_1207, x = linear_26_cast_fp16)[name = string("key_17_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_4_self_attn_v_proj_weight_to_fp16 = const()[name = string("tower_layers_4_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(102480640)))];
tensor<fp16, [896]> tower_layers_4_self_attn_v_proj_bias_to_fp16 = const()[name = string("tower_layers_4_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104086336)))];
tensor<fp16, [1, 390, 896]> linear_27_cast_fp16 = linear(bias = tower_layers_4_self_attn_v_proj_bias_to_fp16, weight = tower_layers_4_self_attn_v_proj_weight_to_fp16, x = input_67_cast_fp16)[name = string("linear_27_cast_fp16")];
tensor<int32, [4]> var_1216 = const()[name = string("op_1216"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> value_17_cast_fp16 = reshape(shape = var_1216, x = linear_27_cast_fp16)[name = string("value_17_cast_fp16")];
tensor<int32, [4]> var_1234 = const()[name = string("op_1234"), val = tensor<int32, [4]>([0, 2, 1, 3])];
bool var_1239_transpose_x_0 = const()[name = string("op_1239_transpose_x_0"), val = bool(false)];
bool var_1239_transpose_y_0 = const()[name = string("op_1239_transpose_y_0"), val = bool(false)];
tensor<int32, [4]> transpose_62_perm_0 = const()[name = string("transpose_62_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
tensor<int32, [4]> transpose_63_perm_0 = const()[name = string("transpose_63_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
tensor<fp16, [1, 14, 64, 390]> transpose_63 = transpose(perm = transpose_63_perm_0, x = key_17_cast_fp16)[name = string("transpose_143")];
tensor<fp16, [1, 14, 390, 64]> transpose_62 = transpose(perm = transpose_62_perm_0, x = query_17_cast_fp16)[name = string("transpose_144")];
tensor<fp16, [1, 14, 390, 390]> var_1239_cast_fp16 = matmul(transpose_x = var_1239_transpose_x_0, transpose_y = var_1239_transpose_y_0, x = transpose_62, y = transpose_63)[name = string("op_1239_cast_fp16")];
fp16 var_1240_to_fp16 = const()[name = string("op_1240_to_fp16"), val = fp16(0x1p-3)];
tensor<fp16, [1, 14, 390, 390]> scores_25_cast_fp16 = mul(x = var_1239_cast_fp16, y = var_1240_to_fp16)[name = string("scores_25_cast_fp16")];
tensor<fp16, [1, 14, 390, 390]> scores_27_cast_fp16 = add(x = scores_25_cast_fp16, y = attention_mask_to_fp16)[name = string("scores_27_cast_fp16")];
int32 var_1249 = const()[name = string("op_1249"), val = int32(-1)];
tensor<fp16, [1, 14, 390, 390]> var_1251_cast_fp16 = softmax(axis = var_1249, x = scores_27_cast_fp16)[name = string("op_1251_cast_fp16")];
bool output_9_transpose_x_0 = const()[name = string("output_9_transpose_x_0"), val = bool(false)];
bool output_9_transpose_y_0 = const()[name = string("output_9_transpose_y_0"), val = bool(false)];
tensor<fp16, [1, 14, 390, 64]> value_19_cast_fp16 = transpose(perm = var_1234, x = value_17_cast_fp16)[name = string("transpose_145")];
tensor<fp16, [1, 14, 390, 64]> output_9_cast_fp16 = matmul(transpose_x = output_9_transpose_x_0, transpose_y = output_9_transpose_y_0, x = var_1251_cast_fp16, y = value_19_cast_fp16)[name = string("output_9_cast_fp16")];
tensor<int32, [4]> var_1262 = const()[name = string("op_1262"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_1267 = const()[name = string("op_1267"), val = tensor<int32, [3]>([1, 390, 896])];
tensor<fp16, [1, 390, 14, 64]> var_1263_cast_fp16 = transpose(perm = var_1262, x = output_9_cast_fp16)[name = string("transpose_142")];
tensor<fp16, [1, 390, 896]> input_69_cast_fp16 = reshape(shape = var_1267, x = var_1263_cast_fp16)[name = string("input_69_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_4_self_attn_out_proj_weight_to_fp16 = const()[name = string("tower_layers_4_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(104088192)))];
tensor<fp16, [896]> tower_layers_4_self_attn_out_proj_bias_to_fp16 = const()[name = string("tower_layers_4_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105693888)))];
tensor<fp16, [1, 390, 896]> linear_28_cast_fp16 = linear(bias = tower_layers_4_self_attn_out_proj_bias_to_fp16, weight = tower_layers_4_self_attn_out_proj_weight_to_fp16, x = input_69_cast_fp16)[name = string("linear_28_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_71_cast_fp16 = add(x = input_65_cast_fp16, y = linear_28_cast_fp16)[name = string("input_71_cast_fp16")];
tensor<int32, [1]> input_73_axes_0 = const()[name = string("input_73_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_4_final_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_4_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105695744)))];
tensor<fp16, [896]> tower_layers_4_final_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_4_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105697600)))];
fp16 var_1275_to_fp16 = const()[name = string("op_1275_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_73_cast_fp16 = layer_norm(axes = input_73_axes_0, beta = tower_layers_4_final_layer_norm_bias_to_fp16, epsilon = var_1275_to_fp16, gamma = tower_layers_4_final_layer_norm_weight_to_fp16, x = input_71_cast_fp16)[name = string("input_73_cast_fp16")];
tensor<fp16, [3584, 896]> tower_layers_4_fc1_weight_to_fp16 = const()[name = string("tower_layers_4_fc1_weight_to_fp16"), val = tensor<fp16, [3584, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(105699456)))];
tensor<fp16, [3584]> tower_layers_4_fc1_bias_to_fp16 = const()[name = string("tower_layers_4_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112122048)))];
tensor<fp16, [1, 390, 3584]> linear_29_cast_fp16 = linear(bias = tower_layers_4_fc1_bias_to_fp16, weight = tower_layers_4_fc1_weight_to_fp16, x = input_73_cast_fp16)[name = string("linear_29_cast_fp16")];
string input_77_mode_0 = const()[name = string("input_77_mode_0"), val = string("EXACT")];
tensor<fp16, [1, 390, 3584]> input_77_cast_fp16 = gelu(mode = input_77_mode_0, x = linear_29_cast_fp16)[name = string("input_77_cast_fp16")];
tensor<fp16, [896, 3584]> tower_layers_4_fc2_weight_to_fp16 = const()[name = string("tower_layers_4_fc2_weight_to_fp16"), val = tensor<fp16, [896, 3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(112129280)))];
tensor<fp16, [896]> tower_layers_4_fc2_bias_to_fp16 = const()[name = string("tower_layers_4_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118551872)))];
tensor<fp16, [1, 390, 896]> linear_30_cast_fp16 = linear(bias = tower_layers_4_fc2_bias_to_fp16, weight = tower_layers_4_fc2_weight_to_fp16, x = input_77_cast_fp16)[name = string("linear_30_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_79_cast_fp16 = add(x = input_71_cast_fp16, y = linear_30_cast_fp16)[name = string("input_79_cast_fp16")];
tensor<int32, [1]> input_81_axes_0 = const()[name = string("input_81_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_5_self_attn_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_5_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118553728)))];
tensor<fp16, [896]> tower_layers_5_self_attn_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_5_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118555584)))];
fp16 var_1292_to_fp16 = const()[name = string("op_1292_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_81_cast_fp16 = layer_norm(axes = input_81_axes_0, beta = tower_layers_5_self_attn_layer_norm_bias_to_fp16, epsilon = var_1292_to_fp16, gamma = tower_layers_5_self_attn_layer_norm_weight_to_fp16, x = input_79_cast_fp16)[name = string("input_81_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_5_self_attn_q_proj_weight_to_fp16 = const()[name = string("tower_layers_5_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(118557440)))];
tensor<fp16, [896]> tower_layers_5_self_attn_q_proj_bias_to_fp16 = const()[name = string("tower_layers_5_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120163136)))];
tensor<fp16, [1, 390, 896]> linear_31_cast_fp16 = linear(bias = tower_layers_5_self_attn_q_proj_bias_to_fp16, weight = tower_layers_5_self_attn_q_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("linear_31_cast_fp16")];
tensor<int32, [4]> var_1305 = const()[name = string("op_1305"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> query_21_cast_fp16 = reshape(shape = var_1305, x = linear_31_cast_fp16)[name = string("query_21_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_5_self_attn_k_proj_weight_to_fp16 = const()[name = string("tower_layers_5_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(120164992)))];
tensor<fp16, [896]> tower_layers_5_self_attn_k_proj_bias_to_fp16 = const()[name = string("tower_layers_5_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121770688)))];
tensor<fp16, [1, 390, 896]> linear_32_cast_fp16 = linear(bias = tower_layers_5_self_attn_k_proj_bias_to_fp16, weight = tower_layers_5_self_attn_k_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("linear_32_cast_fp16")];
tensor<int32, [4]> var_1314 = const()[name = string("op_1314"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> key_21_cast_fp16 = reshape(shape = var_1314, x = linear_32_cast_fp16)[name = string("key_21_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_5_self_attn_v_proj_weight_to_fp16 = const()[name = string("tower_layers_5_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(121772544)))];
tensor<fp16, [896]> tower_layers_5_self_attn_v_proj_bias_to_fp16 = const()[name = string("tower_layers_5_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123378240)))];
tensor<fp16, [1, 390, 896]> linear_33_cast_fp16 = linear(bias = tower_layers_5_self_attn_v_proj_bias_to_fp16, weight = tower_layers_5_self_attn_v_proj_weight_to_fp16, x = input_81_cast_fp16)[name = string("linear_33_cast_fp16")];
tensor<int32, [4]> var_1323 = const()[name = string("op_1323"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> value_21_cast_fp16 = reshape(shape = var_1323, x = linear_33_cast_fp16)[name = string("value_21_cast_fp16")];
tensor<int32, [4]> var_1341 = const()[name = string("op_1341"), val = tensor<int32, [4]>([0, 2, 1, 3])];
bool var_1346_transpose_x_0 = const()[name = string("op_1346_transpose_x_0"), val = bool(false)];
bool var_1346_transpose_y_0 = const()[name = string("op_1346_transpose_y_0"), val = bool(false)];
tensor<int32, [4]> transpose_64_perm_0 = const()[name = string("transpose_64_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
tensor<int32, [4]> transpose_65_perm_0 = const()[name = string("transpose_65_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
tensor<fp16, [1, 14, 64, 390]> transpose_65 = transpose(perm = transpose_65_perm_0, x = key_21_cast_fp16)[name = string("transpose_139")];
tensor<fp16, [1, 14, 390, 64]> transpose_64 = transpose(perm = transpose_64_perm_0, x = query_21_cast_fp16)[name = string("transpose_140")];
tensor<fp16, [1, 14, 390, 390]> var_1346_cast_fp16 = matmul(transpose_x = var_1346_transpose_x_0, transpose_y = var_1346_transpose_y_0, x = transpose_64, y = transpose_65)[name = string("op_1346_cast_fp16")];
fp16 var_1347_to_fp16 = const()[name = string("op_1347_to_fp16"), val = fp16(0x1p-3)];
tensor<fp16, [1, 14, 390, 390]> scores_31_cast_fp16 = mul(x = var_1346_cast_fp16, y = var_1347_to_fp16)[name = string("scores_31_cast_fp16")];
tensor<fp16, [1, 14, 390, 390]> scores_33_cast_fp16 = add(x = scores_31_cast_fp16, y = attention_mask_to_fp16)[name = string("scores_33_cast_fp16")];
int32 var_1356 = const()[name = string("op_1356"), val = int32(-1)];
tensor<fp16, [1, 14, 390, 390]> var_1358_cast_fp16 = softmax(axis = var_1356, x = scores_33_cast_fp16)[name = string("op_1358_cast_fp16")];
bool output_11_transpose_x_0 = const()[name = string("output_11_transpose_x_0"), val = bool(false)];
bool output_11_transpose_y_0 = const()[name = string("output_11_transpose_y_0"), val = bool(false)];
tensor<fp16, [1, 14, 390, 64]> value_23_cast_fp16 = transpose(perm = var_1341, x = value_21_cast_fp16)[name = string("transpose_141")];
tensor<fp16, [1, 14, 390, 64]> output_11_cast_fp16 = matmul(transpose_x = output_11_transpose_x_0, transpose_y = output_11_transpose_y_0, x = var_1358_cast_fp16, y = value_23_cast_fp16)[name = string("output_11_cast_fp16")];
tensor<int32, [4]> var_1369 = const()[name = string("op_1369"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_1374 = const()[name = string("op_1374"), val = tensor<int32, [3]>([1, 390, 896])];
tensor<fp16, [1, 390, 14, 64]> var_1370_cast_fp16 = transpose(perm = var_1369, x = output_11_cast_fp16)[name = string("transpose_138")];
tensor<fp16, [1, 390, 896]> input_83_cast_fp16 = reshape(shape = var_1374, x = var_1370_cast_fp16)[name = string("input_83_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_5_self_attn_out_proj_weight_to_fp16 = const()[name = string("tower_layers_5_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(123380096)))];
tensor<fp16, [896]> tower_layers_5_self_attn_out_proj_bias_to_fp16 = const()[name = string("tower_layers_5_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124985792)))];
tensor<fp16, [1, 390, 896]> linear_34_cast_fp16 = linear(bias = tower_layers_5_self_attn_out_proj_bias_to_fp16, weight = tower_layers_5_self_attn_out_proj_weight_to_fp16, x = input_83_cast_fp16)[name = string("linear_34_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_85_cast_fp16 = add(x = input_79_cast_fp16, y = linear_34_cast_fp16)[name = string("input_85_cast_fp16")];
tensor<int32, [1]> input_87_axes_0 = const()[name = string("input_87_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_5_final_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_5_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124987648)))];
tensor<fp16, [896]> tower_layers_5_final_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_5_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124989504)))];
fp16 var_1382_to_fp16 = const()[name = string("op_1382_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_87_cast_fp16 = layer_norm(axes = input_87_axes_0, beta = tower_layers_5_final_layer_norm_bias_to_fp16, epsilon = var_1382_to_fp16, gamma = tower_layers_5_final_layer_norm_weight_to_fp16, x = input_85_cast_fp16)[name = string("input_87_cast_fp16")];
tensor<fp16, [3584, 896]> tower_layers_5_fc1_weight_to_fp16 = const()[name = string("tower_layers_5_fc1_weight_to_fp16"), val = tensor<fp16, [3584, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(124991360)))];
tensor<fp16, [3584]> tower_layers_5_fc1_bias_to_fp16 = const()[name = string("tower_layers_5_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131413952)))];
tensor<fp16, [1, 390, 3584]> linear_35_cast_fp16 = linear(bias = tower_layers_5_fc1_bias_to_fp16, weight = tower_layers_5_fc1_weight_to_fp16, x = input_87_cast_fp16)[name = string("linear_35_cast_fp16")];
string input_91_mode_0 = const()[name = string("input_91_mode_0"), val = string("EXACT")];
tensor<fp16, [1, 390, 3584]> input_91_cast_fp16 = gelu(mode = input_91_mode_0, x = linear_35_cast_fp16)[name = string("input_91_cast_fp16")];
tensor<fp16, [896, 3584]> tower_layers_5_fc2_weight_to_fp16 = const()[name = string("tower_layers_5_fc2_weight_to_fp16"), val = tensor<fp16, [896, 3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(131421184)))];
tensor<fp16, [896]> tower_layers_5_fc2_bias_to_fp16 = const()[name = string("tower_layers_5_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137843776)))];
tensor<fp16, [1, 390, 896]> linear_36_cast_fp16 = linear(bias = tower_layers_5_fc2_bias_to_fp16, weight = tower_layers_5_fc2_weight_to_fp16, x = input_91_cast_fp16)[name = string("linear_36_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_93_cast_fp16 = add(x = input_85_cast_fp16, y = linear_36_cast_fp16)[name = string("input_93_cast_fp16")];
tensor<int32, [1]> input_95_axes_0 = const()[name = string("input_95_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_6_self_attn_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_6_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137845632)))];
tensor<fp16, [896]> tower_layers_6_self_attn_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_6_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137847488)))];
fp16 var_1399_to_fp16 = const()[name = string("op_1399_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_95_cast_fp16 = layer_norm(axes = input_95_axes_0, beta = tower_layers_6_self_attn_layer_norm_bias_to_fp16, epsilon = var_1399_to_fp16, gamma = tower_layers_6_self_attn_layer_norm_weight_to_fp16, x = input_93_cast_fp16)[name = string("input_95_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_6_self_attn_q_proj_weight_to_fp16 = const()[name = string("tower_layers_6_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(137849344)))];
tensor<fp16, [896]> tower_layers_6_self_attn_q_proj_bias_to_fp16 = const()[name = string("tower_layers_6_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139455040)))];
tensor<fp16, [1, 390, 896]> linear_37_cast_fp16 = linear(bias = tower_layers_6_self_attn_q_proj_bias_to_fp16, weight = tower_layers_6_self_attn_q_proj_weight_to_fp16, x = input_95_cast_fp16)[name = string("linear_37_cast_fp16")];
tensor<int32, [4]> var_1412 = const()[name = string("op_1412"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> query_25_cast_fp16 = reshape(shape = var_1412, x = linear_37_cast_fp16)[name = string("query_25_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_6_self_attn_k_proj_weight_to_fp16 = const()[name = string("tower_layers_6_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(139456896)))];
tensor<fp16, [896]> tower_layers_6_self_attn_k_proj_bias_to_fp16 = const()[name = string("tower_layers_6_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141062592)))];
tensor<fp16, [1, 390, 896]> linear_38_cast_fp16 = linear(bias = tower_layers_6_self_attn_k_proj_bias_to_fp16, weight = tower_layers_6_self_attn_k_proj_weight_to_fp16, x = input_95_cast_fp16)[name = string("linear_38_cast_fp16")];
tensor<int32, [4]> var_1421 = const()[name = string("op_1421"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> key_25_cast_fp16 = reshape(shape = var_1421, x = linear_38_cast_fp16)[name = string("key_25_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_6_self_attn_v_proj_weight_to_fp16 = const()[name = string("tower_layers_6_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(141064448)))];
tensor<fp16, [896]> tower_layers_6_self_attn_v_proj_bias_to_fp16 = const()[name = string("tower_layers_6_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142670144)))];
tensor<fp16, [1, 390, 896]> linear_39_cast_fp16 = linear(bias = tower_layers_6_self_attn_v_proj_bias_to_fp16, weight = tower_layers_6_self_attn_v_proj_weight_to_fp16, x = input_95_cast_fp16)[name = string("linear_39_cast_fp16")];
tensor<int32, [4]> var_1430 = const()[name = string("op_1430"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> value_25_cast_fp16 = reshape(shape = var_1430, x = linear_39_cast_fp16)[name = string("value_25_cast_fp16")];
tensor<int32, [4]> var_1448 = const()[name = string("op_1448"), val = tensor<int32, [4]>([0, 2, 1, 3])];
bool var_1453_transpose_x_0 = const()[name = string("op_1453_transpose_x_0"), val = bool(false)];
bool var_1453_transpose_y_0 = const()[name = string("op_1453_transpose_y_0"), val = bool(false)];
tensor<int32, [4]> transpose_66_perm_0 = const()[name = string("transpose_66_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
tensor<int32, [4]> transpose_67_perm_0 = const()[name = string("transpose_67_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
tensor<fp16, [1, 14, 64, 390]> transpose_67 = transpose(perm = transpose_67_perm_0, x = key_25_cast_fp16)[name = string("transpose_135")];
tensor<fp16, [1, 14, 390, 64]> transpose_66 = transpose(perm = transpose_66_perm_0, x = query_25_cast_fp16)[name = string("transpose_136")];
tensor<fp16, [1, 14, 390, 390]> var_1453_cast_fp16 = matmul(transpose_x = var_1453_transpose_x_0, transpose_y = var_1453_transpose_y_0, x = transpose_66, y = transpose_67)[name = string("op_1453_cast_fp16")];
fp16 var_1454_to_fp16 = const()[name = string("op_1454_to_fp16"), val = fp16(0x1p-3)];
tensor<fp16, [1, 14, 390, 390]> scores_37_cast_fp16 = mul(x = var_1453_cast_fp16, y = var_1454_to_fp16)[name = string("scores_37_cast_fp16")];
tensor<fp16, [1, 14, 390, 390]> scores_39_cast_fp16 = add(x = scores_37_cast_fp16, y = attention_mask_to_fp16)[name = string("scores_39_cast_fp16")];
int32 var_1463 = const()[name = string("op_1463"), val = int32(-1)];
tensor<fp16, [1, 14, 390, 390]> var_1465_cast_fp16 = softmax(axis = var_1463, x = scores_39_cast_fp16)[name = string("op_1465_cast_fp16")];
bool output_13_transpose_x_0 = const()[name = string("output_13_transpose_x_0"), val = bool(false)];
bool output_13_transpose_y_0 = const()[name = string("output_13_transpose_y_0"), val = bool(false)];
tensor<fp16, [1, 14, 390, 64]> value_27_cast_fp16 = transpose(perm = var_1448, x = value_25_cast_fp16)[name = string("transpose_137")];
tensor<fp16, [1, 14, 390, 64]> output_13_cast_fp16 = matmul(transpose_x = output_13_transpose_x_0, transpose_y = output_13_transpose_y_0, x = var_1465_cast_fp16, y = value_27_cast_fp16)[name = string("output_13_cast_fp16")];
tensor<int32, [4]> var_1476 = const()[name = string("op_1476"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_1481 = const()[name = string("op_1481"), val = tensor<int32, [3]>([1, 390, 896])];
tensor<fp16, [1, 390, 14, 64]> var_1477_cast_fp16 = transpose(perm = var_1476, x = output_13_cast_fp16)[name = string("transpose_134")];
tensor<fp16, [1, 390, 896]> input_97_cast_fp16 = reshape(shape = var_1481, x = var_1477_cast_fp16)[name = string("input_97_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_6_self_attn_out_proj_weight_to_fp16 = const()[name = string("tower_layers_6_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(142672000)))];
tensor<fp16, [896]> tower_layers_6_self_attn_out_proj_bias_to_fp16 = const()[name = string("tower_layers_6_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144277696)))];
tensor<fp16, [1, 390, 896]> linear_40_cast_fp16 = linear(bias = tower_layers_6_self_attn_out_proj_bias_to_fp16, weight = tower_layers_6_self_attn_out_proj_weight_to_fp16, x = input_97_cast_fp16)[name = string("linear_40_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_99_cast_fp16 = add(x = input_93_cast_fp16, y = linear_40_cast_fp16)[name = string("input_99_cast_fp16")];
tensor<int32, [1]> input_101_axes_0 = const()[name = string("input_101_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_6_final_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_6_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144279552)))];
tensor<fp16, [896]> tower_layers_6_final_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_6_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144281408)))];
fp16 var_1489_to_fp16 = const()[name = string("op_1489_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_101_cast_fp16 = layer_norm(axes = input_101_axes_0, beta = tower_layers_6_final_layer_norm_bias_to_fp16, epsilon = var_1489_to_fp16, gamma = tower_layers_6_final_layer_norm_weight_to_fp16, x = input_99_cast_fp16)[name = string("input_101_cast_fp16")];
tensor<fp16, [3584, 896]> tower_layers_6_fc1_weight_to_fp16 = const()[name = string("tower_layers_6_fc1_weight_to_fp16"), val = tensor<fp16, [3584, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(144283264)))];
tensor<fp16, [3584]> tower_layers_6_fc1_bias_to_fp16 = const()[name = string("tower_layers_6_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150705856)))];
tensor<fp16, [1, 390, 3584]> linear_41_cast_fp16 = linear(bias = tower_layers_6_fc1_bias_to_fp16, weight = tower_layers_6_fc1_weight_to_fp16, x = input_101_cast_fp16)[name = string("linear_41_cast_fp16")];
string input_105_mode_0 = const()[name = string("input_105_mode_0"), val = string("EXACT")];
tensor<fp16, [1, 390, 3584]> input_105_cast_fp16 = gelu(mode = input_105_mode_0, x = linear_41_cast_fp16)[name = string("input_105_cast_fp16")];
tensor<fp16, [896, 3584]> tower_layers_6_fc2_weight_to_fp16 = const()[name = string("tower_layers_6_fc2_weight_to_fp16"), val = tensor<fp16, [896, 3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(150713088)))];
tensor<fp16, [896]> tower_layers_6_fc2_bias_to_fp16 = const()[name = string("tower_layers_6_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157135680)))];
tensor<fp16, [1, 390, 896]> linear_42_cast_fp16 = linear(bias = tower_layers_6_fc2_bias_to_fp16, weight = tower_layers_6_fc2_weight_to_fp16, x = input_105_cast_fp16)[name = string("linear_42_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_107_cast_fp16 = add(x = input_99_cast_fp16, y = linear_42_cast_fp16)[name = string("input_107_cast_fp16")];
tensor<int32, [1]> input_109_axes_0 = const()[name = string("input_109_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_7_self_attn_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_7_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157137536)))];
tensor<fp16, [896]> tower_layers_7_self_attn_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_7_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157139392)))];
fp16 var_1506_to_fp16 = const()[name = string("op_1506_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_109_cast_fp16 = layer_norm(axes = input_109_axes_0, beta = tower_layers_7_self_attn_layer_norm_bias_to_fp16, epsilon = var_1506_to_fp16, gamma = tower_layers_7_self_attn_layer_norm_weight_to_fp16, x = input_107_cast_fp16)[name = string("input_109_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_7_self_attn_q_proj_weight_to_fp16 = const()[name = string("tower_layers_7_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(157141248)))];
tensor<fp16, [896]> tower_layers_7_self_attn_q_proj_bias_to_fp16 = const()[name = string("tower_layers_7_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158746944)))];
tensor<fp16, [1, 390, 896]> linear_43_cast_fp16 = linear(bias = tower_layers_7_self_attn_q_proj_bias_to_fp16, weight = tower_layers_7_self_attn_q_proj_weight_to_fp16, x = input_109_cast_fp16)[name = string("linear_43_cast_fp16")];
tensor<int32, [4]> var_1519 = const()[name = string("op_1519"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> query_29_cast_fp16 = reshape(shape = var_1519, x = linear_43_cast_fp16)[name = string("query_29_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_7_self_attn_k_proj_weight_to_fp16 = const()[name = string("tower_layers_7_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(158748800)))];
tensor<fp16, [896]> tower_layers_7_self_attn_k_proj_bias_to_fp16 = const()[name = string("tower_layers_7_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160354496)))];
tensor<fp16, [1, 390, 896]> linear_44_cast_fp16 = linear(bias = tower_layers_7_self_attn_k_proj_bias_to_fp16, weight = tower_layers_7_self_attn_k_proj_weight_to_fp16, x = input_109_cast_fp16)[name = string("linear_44_cast_fp16")];
tensor<int32, [4]> var_1528 = const()[name = string("op_1528"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> key_29_cast_fp16 = reshape(shape = var_1528, x = linear_44_cast_fp16)[name = string("key_29_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_7_self_attn_v_proj_weight_to_fp16 = const()[name = string("tower_layers_7_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(160356352)))];
tensor<fp16, [896]> tower_layers_7_self_attn_v_proj_bias_to_fp16 = const()[name = string("tower_layers_7_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161962048)))];
tensor<fp16, [1, 390, 896]> linear_45_cast_fp16 = linear(bias = tower_layers_7_self_attn_v_proj_bias_to_fp16, weight = tower_layers_7_self_attn_v_proj_weight_to_fp16, x = input_109_cast_fp16)[name = string("linear_45_cast_fp16")];
tensor<int32, [4]> var_1537 = const()[name = string("op_1537"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> value_29_cast_fp16 = reshape(shape = var_1537, x = linear_45_cast_fp16)[name = string("value_29_cast_fp16")];
tensor<int32, [4]> var_1555 = const()[name = string("op_1555"), val = tensor<int32, [4]>([0, 2, 1, 3])];
bool var_1560_transpose_x_0 = const()[name = string("op_1560_transpose_x_0"), val = bool(false)];
bool var_1560_transpose_y_0 = const()[name = string("op_1560_transpose_y_0"), val = bool(false)];
tensor<int32, [4]> transpose_68_perm_0 = const()[name = string("transpose_68_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
tensor<int32, [4]> transpose_69_perm_0 = const()[name = string("transpose_69_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
tensor<fp16, [1, 14, 64, 390]> transpose_69 = transpose(perm = transpose_69_perm_0, x = key_29_cast_fp16)[name = string("transpose_131")];
tensor<fp16, [1, 14, 390, 64]> transpose_68 = transpose(perm = transpose_68_perm_0, x = query_29_cast_fp16)[name = string("transpose_132")];
tensor<fp16, [1, 14, 390, 390]> var_1560_cast_fp16 = matmul(transpose_x = var_1560_transpose_x_0, transpose_y = var_1560_transpose_y_0, x = transpose_68, y = transpose_69)[name = string("op_1560_cast_fp16")];
fp16 var_1561_to_fp16 = const()[name = string("op_1561_to_fp16"), val = fp16(0x1p-3)];
tensor<fp16, [1, 14, 390, 390]> scores_43_cast_fp16 = mul(x = var_1560_cast_fp16, y = var_1561_to_fp16)[name = string("scores_43_cast_fp16")];
tensor<fp16, [1, 14, 390, 390]> scores_45_cast_fp16 = add(x = scores_43_cast_fp16, y = attention_mask_to_fp16)[name = string("scores_45_cast_fp16")];
int32 var_1570 = const()[name = string("op_1570"), val = int32(-1)];
tensor<fp16, [1, 14, 390, 390]> var_1572_cast_fp16 = softmax(axis = var_1570, x = scores_45_cast_fp16)[name = string("op_1572_cast_fp16")];
bool output_15_transpose_x_0 = const()[name = string("output_15_transpose_x_0"), val = bool(false)];
bool output_15_transpose_y_0 = const()[name = string("output_15_transpose_y_0"), val = bool(false)];
tensor<fp16, [1, 14, 390, 64]> value_31_cast_fp16 = transpose(perm = var_1555, x = value_29_cast_fp16)[name = string("transpose_133")];
tensor<fp16, [1, 14, 390, 64]> output_15_cast_fp16 = matmul(transpose_x = output_15_transpose_x_0, transpose_y = output_15_transpose_y_0, x = var_1572_cast_fp16, y = value_31_cast_fp16)[name = string("output_15_cast_fp16")];
tensor<int32, [4]> var_1583 = const()[name = string("op_1583"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_1588 = const()[name = string("op_1588"), val = tensor<int32, [3]>([1, 390, 896])];
tensor<fp16, [1, 390, 14, 64]> var_1584_cast_fp16 = transpose(perm = var_1583, x = output_15_cast_fp16)[name = string("transpose_130")];
tensor<fp16, [1, 390, 896]> input_111_cast_fp16 = reshape(shape = var_1588, x = var_1584_cast_fp16)[name = string("input_111_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_7_self_attn_out_proj_weight_to_fp16 = const()[name = string("tower_layers_7_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(161963904)))];
tensor<fp16, [896]> tower_layers_7_self_attn_out_proj_bias_to_fp16 = const()[name = string("tower_layers_7_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163569600)))];
tensor<fp16, [1, 390, 896]> linear_46_cast_fp16 = linear(bias = tower_layers_7_self_attn_out_proj_bias_to_fp16, weight = tower_layers_7_self_attn_out_proj_weight_to_fp16, x = input_111_cast_fp16)[name = string("linear_46_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_113_cast_fp16 = add(x = input_107_cast_fp16, y = linear_46_cast_fp16)[name = string("input_113_cast_fp16")];
tensor<int32, [1]> input_115_axes_0 = const()[name = string("input_115_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_7_final_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_7_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163571456)))];
tensor<fp16, [896]> tower_layers_7_final_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_7_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163573312)))];
fp16 var_1596_to_fp16 = const()[name = string("op_1596_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_115_cast_fp16 = layer_norm(axes = input_115_axes_0, beta = tower_layers_7_final_layer_norm_bias_to_fp16, epsilon = var_1596_to_fp16, gamma = tower_layers_7_final_layer_norm_weight_to_fp16, x = input_113_cast_fp16)[name = string("input_115_cast_fp16")];
tensor<fp16, [3584, 896]> tower_layers_7_fc1_weight_to_fp16 = const()[name = string("tower_layers_7_fc1_weight_to_fp16"), val = tensor<fp16, [3584, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(163575168)))];
tensor<fp16, [3584]> tower_layers_7_fc1_bias_to_fp16 = const()[name = string("tower_layers_7_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(169997760)))];
tensor<fp16, [1, 390, 3584]> linear_47_cast_fp16 = linear(bias = tower_layers_7_fc1_bias_to_fp16, weight = tower_layers_7_fc1_weight_to_fp16, x = input_115_cast_fp16)[name = string("linear_47_cast_fp16")];
string input_119_mode_0 = const()[name = string("input_119_mode_0"), val = string("EXACT")];
tensor<fp16, [1, 390, 3584]> input_119_cast_fp16 = gelu(mode = input_119_mode_0, x = linear_47_cast_fp16)[name = string("input_119_cast_fp16")];
tensor<fp16, [896, 3584]> tower_layers_7_fc2_weight_to_fp16 = const()[name = string("tower_layers_7_fc2_weight_to_fp16"), val = tensor<fp16, [896, 3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(170004992)))];
tensor<fp16, [896]> tower_layers_7_fc2_bias_to_fp16 = const()[name = string("tower_layers_7_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176427584)))];
tensor<fp16, [1, 390, 896]> linear_48_cast_fp16 = linear(bias = tower_layers_7_fc2_bias_to_fp16, weight = tower_layers_7_fc2_weight_to_fp16, x = input_119_cast_fp16)[name = string("linear_48_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_121_cast_fp16 = add(x = input_113_cast_fp16, y = linear_48_cast_fp16)[name = string("input_121_cast_fp16")];
tensor<int32, [1]> input_123_axes_0 = const()[name = string("input_123_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_8_self_attn_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_8_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176429440)))];
tensor<fp16, [896]> tower_layers_8_self_attn_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_8_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176431296)))];
fp16 var_1613_to_fp16 = const()[name = string("op_1613_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_123_cast_fp16 = layer_norm(axes = input_123_axes_0, beta = tower_layers_8_self_attn_layer_norm_bias_to_fp16, epsilon = var_1613_to_fp16, gamma = tower_layers_8_self_attn_layer_norm_weight_to_fp16, x = input_121_cast_fp16)[name = string("input_123_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_8_self_attn_q_proj_weight_to_fp16 = const()[name = string("tower_layers_8_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(176433152)))];
tensor<fp16, [896]> tower_layers_8_self_attn_q_proj_bias_to_fp16 = const()[name = string("tower_layers_8_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178038848)))];
tensor<fp16, [1, 390, 896]> linear_49_cast_fp16 = linear(bias = tower_layers_8_self_attn_q_proj_bias_to_fp16, weight = tower_layers_8_self_attn_q_proj_weight_to_fp16, x = input_123_cast_fp16)[name = string("linear_49_cast_fp16")];
tensor<int32, [4]> var_1626 = const()[name = string("op_1626"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> query_33_cast_fp16 = reshape(shape = var_1626, x = linear_49_cast_fp16)[name = string("query_33_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_8_self_attn_k_proj_weight_to_fp16 = const()[name = string("tower_layers_8_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(178040704)))];
tensor<fp16, [896]> tower_layers_8_self_attn_k_proj_bias_to_fp16 = const()[name = string("tower_layers_8_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179646400)))];
tensor<fp16, [1, 390, 896]> linear_50_cast_fp16 = linear(bias = tower_layers_8_self_attn_k_proj_bias_to_fp16, weight = tower_layers_8_self_attn_k_proj_weight_to_fp16, x = input_123_cast_fp16)[name = string("linear_50_cast_fp16")];
tensor<int32, [4]> var_1635 = const()[name = string("op_1635"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> key_33_cast_fp16 = reshape(shape = var_1635, x = linear_50_cast_fp16)[name = string("key_33_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_8_self_attn_v_proj_weight_to_fp16 = const()[name = string("tower_layers_8_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(179648256)))];
tensor<fp16, [896]> tower_layers_8_self_attn_v_proj_bias_to_fp16 = const()[name = string("tower_layers_8_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181253952)))];
tensor<fp16, [1, 390, 896]> linear_51_cast_fp16 = linear(bias = tower_layers_8_self_attn_v_proj_bias_to_fp16, weight = tower_layers_8_self_attn_v_proj_weight_to_fp16, x = input_123_cast_fp16)[name = string("linear_51_cast_fp16")];
tensor<int32, [4]> var_1644 = const()[name = string("op_1644"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> value_33_cast_fp16 = reshape(shape = var_1644, x = linear_51_cast_fp16)[name = string("value_33_cast_fp16")];
tensor<int32, [4]> var_1662 = const()[name = string("op_1662"), val = tensor<int32, [4]>([0, 2, 1, 3])];
bool var_1667_transpose_x_0 = const()[name = string("op_1667_transpose_x_0"), val = bool(false)];
bool var_1667_transpose_y_0 = const()[name = string("op_1667_transpose_y_0"), val = bool(false)];
tensor<int32, [4]> transpose_70_perm_0 = const()[name = string("transpose_70_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
tensor<int32, [4]> transpose_71_perm_0 = const()[name = string("transpose_71_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
tensor<fp16, [1, 14, 64, 390]> transpose_71 = transpose(perm = transpose_71_perm_0, x = key_33_cast_fp16)[name = string("transpose_127")];
tensor<fp16, [1, 14, 390, 64]> transpose_70 = transpose(perm = transpose_70_perm_0, x = query_33_cast_fp16)[name = string("transpose_128")];
tensor<fp16, [1, 14, 390, 390]> var_1667_cast_fp16 = matmul(transpose_x = var_1667_transpose_x_0, transpose_y = var_1667_transpose_y_0, x = transpose_70, y = transpose_71)[name = string("op_1667_cast_fp16")];
fp16 var_1668_to_fp16 = const()[name = string("op_1668_to_fp16"), val = fp16(0x1p-3)];
tensor<fp16, [1, 14, 390, 390]> scores_49_cast_fp16 = mul(x = var_1667_cast_fp16, y = var_1668_to_fp16)[name = string("scores_49_cast_fp16")];
tensor<fp16, [1, 14, 390, 390]> scores_51_cast_fp16 = add(x = scores_49_cast_fp16, y = attention_mask_to_fp16)[name = string("scores_51_cast_fp16")];
int32 var_1677 = const()[name = string("op_1677"), val = int32(-1)];
tensor<fp16, [1, 14, 390, 390]> var_1679_cast_fp16 = softmax(axis = var_1677, x = scores_51_cast_fp16)[name = string("op_1679_cast_fp16")];
bool output_17_transpose_x_0 = const()[name = string("output_17_transpose_x_0"), val = bool(false)];
bool output_17_transpose_y_0 = const()[name = string("output_17_transpose_y_0"), val = bool(false)];
tensor<fp16, [1, 14, 390, 64]> value_35_cast_fp16 = transpose(perm = var_1662, x = value_33_cast_fp16)[name = string("transpose_129")];
tensor<fp16, [1, 14, 390, 64]> output_17_cast_fp16 = matmul(transpose_x = output_17_transpose_x_0, transpose_y = output_17_transpose_y_0, x = var_1679_cast_fp16, y = value_35_cast_fp16)[name = string("output_17_cast_fp16")];
tensor<int32, [4]> var_1690 = const()[name = string("op_1690"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_1695 = const()[name = string("op_1695"), val = tensor<int32, [3]>([1, 390, 896])];
tensor<fp16, [1, 390, 14, 64]> var_1691_cast_fp16 = transpose(perm = var_1690, x = output_17_cast_fp16)[name = string("transpose_126")];
tensor<fp16, [1, 390, 896]> input_125_cast_fp16 = reshape(shape = var_1695, x = var_1691_cast_fp16)[name = string("input_125_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_8_self_attn_out_proj_weight_to_fp16 = const()[name = string("tower_layers_8_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(181255808)))];
tensor<fp16, [896]> tower_layers_8_self_attn_out_proj_bias_to_fp16 = const()[name = string("tower_layers_8_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182861504)))];
tensor<fp16, [1, 390, 896]> linear_52_cast_fp16 = linear(bias = tower_layers_8_self_attn_out_proj_bias_to_fp16, weight = tower_layers_8_self_attn_out_proj_weight_to_fp16, x = input_125_cast_fp16)[name = string("linear_52_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_127_cast_fp16 = add(x = input_121_cast_fp16, y = linear_52_cast_fp16)[name = string("input_127_cast_fp16")];
tensor<int32, [1]> input_129_axes_0 = const()[name = string("input_129_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_8_final_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_8_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182863360)))];
tensor<fp16, [896]> tower_layers_8_final_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_8_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182865216)))];
fp16 var_1703_to_fp16 = const()[name = string("op_1703_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_129_cast_fp16 = layer_norm(axes = input_129_axes_0, beta = tower_layers_8_final_layer_norm_bias_to_fp16, epsilon = var_1703_to_fp16, gamma = tower_layers_8_final_layer_norm_weight_to_fp16, x = input_127_cast_fp16)[name = string("input_129_cast_fp16")];
tensor<fp16, [3584, 896]> tower_layers_8_fc1_weight_to_fp16 = const()[name = string("tower_layers_8_fc1_weight_to_fp16"), val = tensor<fp16, [3584, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(182867072)))];
tensor<fp16, [3584]> tower_layers_8_fc1_bias_to_fp16 = const()[name = string("tower_layers_8_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189289664)))];
tensor<fp16, [1, 390, 3584]> linear_53_cast_fp16 = linear(bias = tower_layers_8_fc1_bias_to_fp16, weight = tower_layers_8_fc1_weight_to_fp16, x = input_129_cast_fp16)[name = string("linear_53_cast_fp16")];
string input_133_mode_0 = const()[name = string("input_133_mode_0"), val = string("EXACT")];
tensor<fp16, [1, 390, 3584]> input_133_cast_fp16 = gelu(mode = input_133_mode_0, x = linear_53_cast_fp16)[name = string("input_133_cast_fp16")];
tensor<fp16, [896, 3584]> tower_layers_8_fc2_weight_to_fp16 = const()[name = string("tower_layers_8_fc2_weight_to_fp16"), val = tensor<fp16, [896, 3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(189296896)))];
tensor<fp16, [896]> tower_layers_8_fc2_bias_to_fp16 = const()[name = string("tower_layers_8_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195719488)))];
tensor<fp16, [1, 390, 896]> linear_54_cast_fp16 = linear(bias = tower_layers_8_fc2_bias_to_fp16, weight = tower_layers_8_fc2_weight_to_fp16, x = input_133_cast_fp16)[name = string("linear_54_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_135_cast_fp16 = add(x = input_127_cast_fp16, y = linear_54_cast_fp16)[name = string("input_135_cast_fp16")];
tensor<int32, [1]> input_137_axes_0 = const()[name = string("input_137_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_9_self_attn_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_9_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195721344)))];
tensor<fp16, [896]> tower_layers_9_self_attn_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_9_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195723200)))];
fp16 var_1720_to_fp16 = const()[name = string("op_1720_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_137_cast_fp16 = layer_norm(axes = input_137_axes_0, beta = tower_layers_9_self_attn_layer_norm_bias_to_fp16, epsilon = var_1720_to_fp16, gamma = tower_layers_9_self_attn_layer_norm_weight_to_fp16, x = input_135_cast_fp16)[name = string("input_137_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_9_self_attn_q_proj_weight_to_fp16 = const()[name = string("tower_layers_9_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(195725056)))];
tensor<fp16, [896]> tower_layers_9_self_attn_q_proj_bias_to_fp16 = const()[name = string("tower_layers_9_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197330752)))];
tensor<fp16, [1, 390, 896]> linear_55_cast_fp16 = linear(bias = tower_layers_9_self_attn_q_proj_bias_to_fp16, weight = tower_layers_9_self_attn_q_proj_weight_to_fp16, x = input_137_cast_fp16)[name = string("linear_55_cast_fp16")];
tensor<int32, [4]> var_1733 = const()[name = string("op_1733"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> query_37_cast_fp16 = reshape(shape = var_1733, x = linear_55_cast_fp16)[name = string("query_37_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_9_self_attn_k_proj_weight_to_fp16 = const()[name = string("tower_layers_9_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(197332608)))];
tensor<fp16, [896]> tower_layers_9_self_attn_k_proj_bias_to_fp16 = const()[name = string("tower_layers_9_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198938304)))];
tensor<fp16, [1, 390, 896]> linear_56_cast_fp16 = linear(bias = tower_layers_9_self_attn_k_proj_bias_to_fp16, weight = tower_layers_9_self_attn_k_proj_weight_to_fp16, x = input_137_cast_fp16)[name = string("linear_56_cast_fp16")];
tensor<int32, [4]> var_1742 = const()[name = string("op_1742"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> key_37_cast_fp16 = reshape(shape = var_1742, x = linear_56_cast_fp16)[name = string("key_37_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_9_self_attn_v_proj_weight_to_fp16 = const()[name = string("tower_layers_9_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(198940160)))];
tensor<fp16, [896]> tower_layers_9_self_attn_v_proj_bias_to_fp16 = const()[name = string("tower_layers_9_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200545856)))];
tensor<fp16, [1, 390, 896]> linear_57_cast_fp16 = linear(bias = tower_layers_9_self_attn_v_proj_bias_to_fp16, weight = tower_layers_9_self_attn_v_proj_weight_to_fp16, x = input_137_cast_fp16)[name = string("linear_57_cast_fp16")];
tensor<int32, [4]> var_1751 = const()[name = string("op_1751"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> value_37_cast_fp16 = reshape(shape = var_1751, x = linear_57_cast_fp16)[name = string("value_37_cast_fp16")];
tensor<int32, [4]> var_1769 = const()[name = string("op_1769"), val = tensor<int32, [4]>([0, 2, 1, 3])];
bool var_1774_transpose_x_0 = const()[name = string("op_1774_transpose_x_0"), val = bool(false)];
bool var_1774_transpose_y_0 = const()[name = string("op_1774_transpose_y_0"), val = bool(false)];
tensor<int32, [4]> transpose_72_perm_0 = const()[name = string("transpose_72_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
tensor<int32, [4]> transpose_73_perm_0 = const()[name = string("transpose_73_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
tensor<fp16, [1, 14, 64, 390]> transpose_73 = transpose(perm = transpose_73_perm_0, x = key_37_cast_fp16)[name = string("transpose_123")];
tensor<fp16, [1, 14, 390, 64]> transpose_72 = transpose(perm = transpose_72_perm_0, x = query_37_cast_fp16)[name = string("transpose_124")];
tensor<fp16, [1, 14, 390, 390]> var_1774_cast_fp16 = matmul(transpose_x = var_1774_transpose_x_0, transpose_y = var_1774_transpose_y_0, x = transpose_72, y = transpose_73)[name = string("op_1774_cast_fp16")];
fp16 var_1775_to_fp16 = const()[name = string("op_1775_to_fp16"), val = fp16(0x1p-3)];
tensor<fp16, [1, 14, 390, 390]> scores_55_cast_fp16 = mul(x = var_1774_cast_fp16, y = var_1775_to_fp16)[name = string("scores_55_cast_fp16")];
tensor<fp16, [1, 14, 390, 390]> scores_57_cast_fp16 = add(x = scores_55_cast_fp16, y = attention_mask_to_fp16)[name = string("scores_57_cast_fp16")];
int32 var_1784 = const()[name = string("op_1784"), val = int32(-1)];
tensor<fp16, [1, 14, 390, 390]> var_1786_cast_fp16 = softmax(axis = var_1784, x = scores_57_cast_fp16)[name = string("op_1786_cast_fp16")];
bool output_19_transpose_x_0 = const()[name = string("output_19_transpose_x_0"), val = bool(false)];
bool output_19_transpose_y_0 = const()[name = string("output_19_transpose_y_0"), val = bool(false)];
tensor<fp16, [1, 14, 390, 64]> value_39_cast_fp16 = transpose(perm = var_1769, x = value_37_cast_fp16)[name = string("transpose_125")];
tensor<fp16, [1, 14, 390, 64]> output_19_cast_fp16 = matmul(transpose_x = output_19_transpose_x_0, transpose_y = output_19_transpose_y_0, x = var_1786_cast_fp16, y = value_39_cast_fp16)[name = string("output_19_cast_fp16")];
tensor<int32, [4]> var_1797 = const()[name = string("op_1797"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_1802 = const()[name = string("op_1802"), val = tensor<int32, [3]>([1, 390, 896])];
tensor<fp16, [1, 390, 14, 64]> var_1798_cast_fp16 = transpose(perm = var_1797, x = output_19_cast_fp16)[name = string("transpose_122")];
tensor<fp16, [1, 390, 896]> input_139_cast_fp16 = reshape(shape = var_1802, x = var_1798_cast_fp16)[name = string("input_139_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_9_self_attn_out_proj_weight_to_fp16 = const()[name = string("tower_layers_9_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(200547712)))];
tensor<fp16, [896]> tower_layers_9_self_attn_out_proj_bias_to_fp16 = const()[name = string("tower_layers_9_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202153408)))];
tensor<fp16, [1, 390, 896]> linear_58_cast_fp16 = linear(bias = tower_layers_9_self_attn_out_proj_bias_to_fp16, weight = tower_layers_9_self_attn_out_proj_weight_to_fp16, x = input_139_cast_fp16)[name = string("linear_58_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_141_cast_fp16 = add(x = input_135_cast_fp16, y = linear_58_cast_fp16)[name = string("input_141_cast_fp16")];
tensor<int32, [1]> input_143_axes_0 = const()[name = string("input_143_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_9_final_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_9_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202155264)))];
tensor<fp16, [896]> tower_layers_9_final_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_9_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202157120)))];
fp16 var_1810_to_fp16 = const()[name = string("op_1810_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_143_cast_fp16 = layer_norm(axes = input_143_axes_0, beta = tower_layers_9_final_layer_norm_bias_to_fp16, epsilon = var_1810_to_fp16, gamma = tower_layers_9_final_layer_norm_weight_to_fp16, x = input_141_cast_fp16)[name = string("input_143_cast_fp16")];
tensor<fp16, [3584, 896]> tower_layers_9_fc1_weight_to_fp16 = const()[name = string("tower_layers_9_fc1_weight_to_fp16"), val = tensor<fp16, [3584, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(202158976)))];
tensor<fp16, [3584]> tower_layers_9_fc1_bias_to_fp16 = const()[name = string("tower_layers_9_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208581568)))];
tensor<fp16, [1, 390, 3584]> linear_59_cast_fp16 = linear(bias = tower_layers_9_fc1_bias_to_fp16, weight = tower_layers_9_fc1_weight_to_fp16, x = input_143_cast_fp16)[name = string("linear_59_cast_fp16")];
string input_147_mode_0 = const()[name = string("input_147_mode_0"), val = string("EXACT")];
tensor<fp16, [1, 390, 3584]> input_147_cast_fp16 = gelu(mode = input_147_mode_0, x = linear_59_cast_fp16)[name = string("input_147_cast_fp16")];
tensor<fp16, [896, 3584]> tower_layers_9_fc2_weight_to_fp16 = const()[name = string("tower_layers_9_fc2_weight_to_fp16"), val = tensor<fp16, [896, 3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(208588800)))];
tensor<fp16, [896]> tower_layers_9_fc2_bias_to_fp16 = const()[name = string("tower_layers_9_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215011392)))];
tensor<fp16, [1, 390, 896]> linear_60_cast_fp16 = linear(bias = tower_layers_9_fc2_bias_to_fp16, weight = tower_layers_9_fc2_weight_to_fp16, x = input_147_cast_fp16)[name = string("linear_60_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_149_cast_fp16 = add(x = input_141_cast_fp16, y = linear_60_cast_fp16)[name = string("input_149_cast_fp16")];
tensor<int32, [1]> input_151_axes_0 = const()[name = string("input_151_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_10_self_attn_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_10_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215013248)))];
tensor<fp16, [896]> tower_layers_10_self_attn_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_10_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215015104)))];
fp16 var_1827_to_fp16 = const()[name = string("op_1827_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_151_cast_fp16 = layer_norm(axes = input_151_axes_0, beta = tower_layers_10_self_attn_layer_norm_bias_to_fp16, epsilon = var_1827_to_fp16, gamma = tower_layers_10_self_attn_layer_norm_weight_to_fp16, x = input_149_cast_fp16)[name = string("input_151_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_10_self_attn_q_proj_weight_to_fp16 = const()[name = string("tower_layers_10_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(215016960)))];
tensor<fp16, [896]> tower_layers_10_self_attn_q_proj_bias_to_fp16 = const()[name = string("tower_layers_10_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216622656)))];
tensor<fp16, [1, 390, 896]> linear_61_cast_fp16 = linear(bias = tower_layers_10_self_attn_q_proj_bias_to_fp16, weight = tower_layers_10_self_attn_q_proj_weight_to_fp16, x = input_151_cast_fp16)[name = string("linear_61_cast_fp16")];
tensor<int32, [4]> var_1840 = const()[name = string("op_1840"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> query_41_cast_fp16 = reshape(shape = var_1840, x = linear_61_cast_fp16)[name = string("query_41_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_10_self_attn_k_proj_weight_to_fp16 = const()[name = string("tower_layers_10_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(216624512)))];
tensor<fp16, [896]> tower_layers_10_self_attn_k_proj_bias_to_fp16 = const()[name = string("tower_layers_10_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218230208)))];
tensor<fp16, [1, 390, 896]> linear_62_cast_fp16 = linear(bias = tower_layers_10_self_attn_k_proj_bias_to_fp16, weight = tower_layers_10_self_attn_k_proj_weight_to_fp16, x = input_151_cast_fp16)[name = string("linear_62_cast_fp16")];
tensor<int32, [4]> var_1849 = const()[name = string("op_1849"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> key_41_cast_fp16 = reshape(shape = var_1849, x = linear_62_cast_fp16)[name = string("key_41_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_10_self_attn_v_proj_weight_to_fp16 = const()[name = string("tower_layers_10_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(218232064)))];
tensor<fp16, [896]> tower_layers_10_self_attn_v_proj_bias_to_fp16 = const()[name = string("tower_layers_10_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219837760)))];
tensor<fp16, [1, 390, 896]> linear_63_cast_fp16 = linear(bias = tower_layers_10_self_attn_v_proj_bias_to_fp16, weight = tower_layers_10_self_attn_v_proj_weight_to_fp16, x = input_151_cast_fp16)[name = string("linear_63_cast_fp16")];
tensor<int32, [4]> var_1858 = const()[name = string("op_1858"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> value_41_cast_fp16 = reshape(shape = var_1858, x = linear_63_cast_fp16)[name = string("value_41_cast_fp16")];
tensor<int32, [4]> var_1876 = const()[name = string("op_1876"), val = tensor<int32, [4]>([0, 2, 1, 3])];
bool var_1881_transpose_x_0 = const()[name = string("op_1881_transpose_x_0"), val = bool(false)];
bool var_1881_transpose_y_0 = const()[name = string("op_1881_transpose_y_0"), val = bool(false)];
tensor<int32, [4]> transpose_74_perm_0 = const()[name = string("transpose_74_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
tensor<int32, [4]> transpose_75_perm_0 = const()[name = string("transpose_75_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
tensor<fp16, [1, 14, 64, 390]> transpose_75 = transpose(perm = transpose_75_perm_0, x = key_41_cast_fp16)[name = string("transpose_119")];
tensor<fp16, [1, 14, 390, 64]> transpose_74 = transpose(perm = transpose_74_perm_0, x = query_41_cast_fp16)[name = string("transpose_120")];
tensor<fp16, [1, 14, 390, 390]> var_1881_cast_fp16 = matmul(transpose_x = var_1881_transpose_x_0, transpose_y = var_1881_transpose_y_0, x = transpose_74, y = transpose_75)[name = string("op_1881_cast_fp16")];
fp16 var_1882_to_fp16 = const()[name = string("op_1882_to_fp16"), val = fp16(0x1p-3)];
tensor<fp16, [1, 14, 390, 390]> scores_61_cast_fp16 = mul(x = var_1881_cast_fp16, y = var_1882_to_fp16)[name = string("scores_61_cast_fp16")];
tensor<fp16, [1, 14, 390, 390]> scores_63_cast_fp16 = add(x = scores_61_cast_fp16, y = attention_mask_to_fp16)[name = string("scores_63_cast_fp16")];
int32 var_1891 = const()[name = string("op_1891"), val = int32(-1)];
tensor<fp16, [1, 14, 390, 390]> var_1893_cast_fp16 = softmax(axis = var_1891, x = scores_63_cast_fp16)[name = string("op_1893_cast_fp16")];
bool output_21_transpose_x_0 = const()[name = string("output_21_transpose_x_0"), val = bool(false)];
bool output_21_transpose_y_0 = const()[name = string("output_21_transpose_y_0"), val = bool(false)];
tensor<fp16, [1, 14, 390, 64]> value_43_cast_fp16 = transpose(perm = var_1876, x = value_41_cast_fp16)[name = string("transpose_121")];
tensor<fp16, [1, 14, 390, 64]> output_21_cast_fp16 = matmul(transpose_x = output_21_transpose_x_0, transpose_y = output_21_transpose_y_0, x = var_1893_cast_fp16, y = value_43_cast_fp16)[name = string("output_21_cast_fp16")];
tensor<int32, [4]> var_1904 = const()[name = string("op_1904"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_1909 = const()[name = string("op_1909"), val = tensor<int32, [3]>([1, 390, 896])];
tensor<fp16, [1, 390, 14, 64]> var_1905_cast_fp16 = transpose(perm = var_1904, x = output_21_cast_fp16)[name = string("transpose_118")];
tensor<fp16, [1, 390, 896]> input_153_cast_fp16 = reshape(shape = var_1909, x = var_1905_cast_fp16)[name = string("input_153_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_10_self_attn_out_proj_weight_to_fp16 = const()[name = string("tower_layers_10_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(219839616)))];
tensor<fp16, [896]> tower_layers_10_self_attn_out_proj_bias_to_fp16 = const()[name = string("tower_layers_10_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221445312)))];
tensor<fp16, [1, 390, 896]> linear_64_cast_fp16 = linear(bias = tower_layers_10_self_attn_out_proj_bias_to_fp16, weight = tower_layers_10_self_attn_out_proj_weight_to_fp16, x = input_153_cast_fp16)[name = string("linear_64_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_155_cast_fp16 = add(x = input_149_cast_fp16, y = linear_64_cast_fp16)[name = string("input_155_cast_fp16")];
tensor<int32, [1]> input_157_axes_0 = const()[name = string("input_157_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_10_final_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_10_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221447168)))];
tensor<fp16, [896]> tower_layers_10_final_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_10_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221449024)))];
fp16 var_1917_to_fp16 = const()[name = string("op_1917_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_157_cast_fp16 = layer_norm(axes = input_157_axes_0, beta = tower_layers_10_final_layer_norm_bias_to_fp16, epsilon = var_1917_to_fp16, gamma = tower_layers_10_final_layer_norm_weight_to_fp16, x = input_155_cast_fp16)[name = string("input_157_cast_fp16")];
tensor<fp16, [3584, 896]> tower_layers_10_fc1_weight_to_fp16 = const()[name = string("tower_layers_10_fc1_weight_to_fp16"), val = tensor<fp16, [3584, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(221450880)))];
tensor<fp16, [3584]> tower_layers_10_fc1_bias_to_fp16 = const()[name = string("tower_layers_10_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227873472)))];
tensor<fp16, [1, 390, 3584]> linear_65_cast_fp16 = linear(bias = tower_layers_10_fc1_bias_to_fp16, weight = tower_layers_10_fc1_weight_to_fp16, x = input_157_cast_fp16)[name = string("linear_65_cast_fp16")];
string input_161_mode_0 = const()[name = string("input_161_mode_0"), val = string("EXACT")];
tensor<fp16, [1, 390, 3584]> input_161_cast_fp16 = gelu(mode = input_161_mode_0, x = linear_65_cast_fp16)[name = string("input_161_cast_fp16")];
tensor<fp16, [896, 3584]> tower_layers_10_fc2_weight_to_fp16 = const()[name = string("tower_layers_10_fc2_weight_to_fp16"), val = tensor<fp16, [896, 3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(227880704)))];
tensor<fp16, [896]> tower_layers_10_fc2_bias_to_fp16 = const()[name = string("tower_layers_10_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234303296)))];
tensor<fp16, [1, 390, 896]> linear_66_cast_fp16 = linear(bias = tower_layers_10_fc2_bias_to_fp16, weight = tower_layers_10_fc2_weight_to_fp16, x = input_161_cast_fp16)[name = string("linear_66_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_163_cast_fp16 = add(x = input_155_cast_fp16, y = linear_66_cast_fp16)[name = string("input_163_cast_fp16")];
tensor<int32, [1]> input_165_axes_0 = const()[name = string("input_165_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_11_self_attn_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_11_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234305152)))];
tensor<fp16, [896]> tower_layers_11_self_attn_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_11_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234307008)))];
fp16 var_1934_to_fp16 = const()[name = string("op_1934_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_165_cast_fp16 = layer_norm(axes = input_165_axes_0, beta = tower_layers_11_self_attn_layer_norm_bias_to_fp16, epsilon = var_1934_to_fp16, gamma = tower_layers_11_self_attn_layer_norm_weight_to_fp16, x = input_163_cast_fp16)[name = string("input_165_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_11_self_attn_q_proj_weight_to_fp16 = const()[name = string("tower_layers_11_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(234308864)))];
tensor<fp16, [896]> tower_layers_11_self_attn_q_proj_bias_to_fp16 = const()[name = string("tower_layers_11_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235914560)))];
tensor<fp16, [1, 390, 896]> linear_67_cast_fp16 = linear(bias = tower_layers_11_self_attn_q_proj_bias_to_fp16, weight = tower_layers_11_self_attn_q_proj_weight_to_fp16, x = input_165_cast_fp16)[name = string("linear_67_cast_fp16")];
tensor<int32, [4]> var_1947 = const()[name = string("op_1947"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> query_45_cast_fp16 = reshape(shape = var_1947, x = linear_67_cast_fp16)[name = string("query_45_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_11_self_attn_k_proj_weight_to_fp16 = const()[name = string("tower_layers_11_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(235916416)))];
tensor<fp16, [896]> tower_layers_11_self_attn_k_proj_bias_to_fp16 = const()[name = string("tower_layers_11_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237522112)))];
tensor<fp16, [1, 390, 896]> linear_68_cast_fp16 = linear(bias = tower_layers_11_self_attn_k_proj_bias_to_fp16, weight = tower_layers_11_self_attn_k_proj_weight_to_fp16, x = input_165_cast_fp16)[name = string("linear_68_cast_fp16")];
tensor<int32, [4]> var_1956 = const()[name = string("op_1956"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> key_45_cast_fp16 = reshape(shape = var_1956, x = linear_68_cast_fp16)[name = string("key_45_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_11_self_attn_v_proj_weight_to_fp16 = const()[name = string("tower_layers_11_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(237523968)))];
tensor<fp16, [896]> tower_layers_11_self_attn_v_proj_bias_to_fp16 = const()[name = string("tower_layers_11_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239129664)))];
tensor<fp16, [1, 390, 896]> linear_69_cast_fp16 = linear(bias = tower_layers_11_self_attn_v_proj_bias_to_fp16, weight = tower_layers_11_self_attn_v_proj_weight_to_fp16, x = input_165_cast_fp16)[name = string("linear_69_cast_fp16")];
tensor<int32, [4]> var_1965 = const()[name = string("op_1965"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> value_45_cast_fp16 = reshape(shape = var_1965, x = linear_69_cast_fp16)[name = string("value_45_cast_fp16")];
tensor<int32, [4]> var_1983 = const()[name = string("op_1983"), val = tensor<int32, [4]>([0, 2, 1, 3])];
bool var_1988_transpose_x_0 = const()[name = string("op_1988_transpose_x_0"), val = bool(false)];
bool var_1988_transpose_y_0 = const()[name = string("op_1988_transpose_y_0"), val = bool(false)];
tensor<int32, [4]> transpose_76_perm_0 = const()[name = string("transpose_76_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
tensor<int32, [4]> transpose_77_perm_0 = const()[name = string("transpose_77_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
tensor<fp16, [1, 14, 64, 390]> transpose_77 = transpose(perm = transpose_77_perm_0, x = key_45_cast_fp16)[name = string("transpose_115")];
tensor<fp16, [1, 14, 390, 64]> transpose_76 = transpose(perm = transpose_76_perm_0, x = query_45_cast_fp16)[name = string("transpose_116")];
tensor<fp16, [1, 14, 390, 390]> var_1988_cast_fp16 = matmul(transpose_x = var_1988_transpose_x_0, transpose_y = var_1988_transpose_y_0, x = transpose_76, y = transpose_77)[name = string("op_1988_cast_fp16")];
fp16 var_1989_to_fp16 = const()[name = string("op_1989_to_fp16"), val = fp16(0x1p-3)];
tensor<fp16, [1, 14, 390, 390]> scores_67_cast_fp16 = mul(x = var_1988_cast_fp16, y = var_1989_to_fp16)[name = string("scores_67_cast_fp16")];
tensor<fp16, [1, 14, 390, 390]> scores_69_cast_fp16 = add(x = scores_67_cast_fp16, y = attention_mask_to_fp16)[name = string("scores_69_cast_fp16")];
int32 var_1998 = const()[name = string("op_1998"), val = int32(-1)];
tensor<fp16, [1, 14, 390, 390]> var_2000_cast_fp16 = softmax(axis = var_1998, x = scores_69_cast_fp16)[name = string("op_2000_cast_fp16")];
bool output_23_transpose_x_0 = const()[name = string("output_23_transpose_x_0"), val = bool(false)];
bool output_23_transpose_y_0 = const()[name = string("output_23_transpose_y_0"), val = bool(false)];
tensor<fp16, [1, 14, 390, 64]> value_47_cast_fp16 = transpose(perm = var_1983, x = value_45_cast_fp16)[name = string("transpose_117")];
tensor<fp16, [1, 14, 390, 64]> output_23_cast_fp16 = matmul(transpose_x = output_23_transpose_x_0, transpose_y = output_23_transpose_y_0, x = var_2000_cast_fp16, y = value_47_cast_fp16)[name = string("output_23_cast_fp16")];
tensor<int32, [4]> var_2011 = const()[name = string("op_2011"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_2016 = const()[name = string("op_2016"), val = tensor<int32, [3]>([1, 390, 896])];
tensor<fp16, [1, 390, 14, 64]> var_2012_cast_fp16 = transpose(perm = var_2011, x = output_23_cast_fp16)[name = string("transpose_114")];
tensor<fp16, [1, 390, 896]> input_167_cast_fp16 = reshape(shape = var_2016, x = var_2012_cast_fp16)[name = string("input_167_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_11_self_attn_out_proj_weight_to_fp16 = const()[name = string("tower_layers_11_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(239131520)))];
tensor<fp16, [896]> tower_layers_11_self_attn_out_proj_bias_to_fp16 = const()[name = string("tower_layers_11_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240737216)))];
tensor<fp16, [1, 390, 896]> linear_70_cast_fp16 = linear(bias = tower_layers_11_self_attn_out_proj_bias_to_fp16, weight = tower_layers_11_self_attn_out_proj_weight_to_fp16, x = input_167_cast_fp16)[name = string("linear_70_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_169_cast_fp16 = add(x = input_163_cast_fp16, y = linear_70_cast_fp16)[name = string("input_169_cast_fp16")];
tensor<int32, [1]> input_171_axes_0 = const()[name = string("input_171_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_11_final_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_11_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240739072)))];
tensor<fp16, [896]> tower_layers_11_final_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_11_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240740928)))];
fp16 var_2024_to_fp16 = const()[name = string("op_2024_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_171_cast_fp16 = layer_norm(axes = input_171_axes_0, beta = tower_layers_11_final_layer_norm_bias_to_fp16, epsilon = var_2024_to_fp16, gamma = tower_layers_11_final_layer_norm_weight_to_fp16, x = input_169_cast_fp16)[name = string("input_171_cast_fp16")];
tensor<fp16, [3584, 896]> tower_layers_11_fc1_weight_to_fp16 = const()[name = string("tower_layers_11_fc1_weight_to_fp16"), val = tensor<fp16, [3584, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(240742784)))];
tensor<fp16, [3584]> tower_layers_11_fc1_bias_to_fp16 = const()[name = string("tower_layers_11_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247165376)))];
tensor<fp16, [1, 390, 3584]> linear_71_cast_fp16 = linear(bias = tower_layers_11_fc1_bias_to_fp16, weight = tower_layers_11_fc1_weight_to_fp16, x = input_171_cast_fp16)[name = string("linear_71_cast_fp16")];
string input_175_mode_0 = const()[name = string("input_175_mode_0"), val = string("EXACT")];
tensor<fp16, [1, 390, 3584]> input_175_cast_fp16 = gelu(mode = input_175_mode_0, x = linear_71_cast_fp16)[name = string("input_175_cast_fp16")];
tensor<fp16, [896, 3584]> tower_layers_11_fc2_weight_to_fp16 = const()[name = string("tower_layers_11_fc2_weight_to_fp16"), val = tensor<fp16, [896, 3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(247172608)))];
tensor<fp16, [896]> tower_layers_11_fc2_bias_to_fp16 = const()[name = string("tower_layers_11_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253595200)))];
tensor<fp16, [1, 390, 896]> linear_72_cast_fp16 = linear(bias = tower_layers_11_fc2_bias_to_fp16, weight = tower_layers_11_fc2_weight_to_fp16, x = input_175_cast_fp16)[name = string("linear_72_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_177_cast_fp16 = add(x = input_169_cast_fp16, y = linear_72_cast_fp16)[name = string("input_177_cast_fp16")];
tensor<int32, [1]> input_179_axes_0 = const()[name = string("input_179_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_12_self_attn_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_12_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253597056)))];
tensor<fp16, [896]> tower_layers_12_self_attn_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_12_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253598912)))];
fp16 var_2041_to_fp16 = const()[name = string("op_2041_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_179_cast_fp16 = layer_norm(axes = input_179_axes_0, beta = tower_layers_12_self_attn_layer_norm_bias_to_fp16, epsilon = var_2041_to_fp16, gamma = tower_layers_12_self_attn_layer_norm_weight_to_fp16, x = input_177_cast_fp16)[name = string("input_179_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_12_self_attn_q_proj_weight_to_fp16 = const()[name = string("tower_layers_12_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(253600768)))];
tensor<fp16, [896]> tower_layers_12_self_attn_q_proj_bias_to_fp16 = const()[name = string("tower_layers_12_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(255206464)))];
tensor<fp16, [1, 390, 896]> linear_73_cast_fp16 = linear(bias = tower_layers_12_self_attn_q_proj_bias_to_fp16, weight = tower_layers_12_self_attn_q_proj_weight_to_fp16, x = input_179_cast_fp16)[name = string("linear_73_cast_fp16")];
tensor<int32, [4]> var_2054 = const()[name = string("op_2054"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> query_49_cast_fp16 = reshape(shape = var_2054, x = linear_73_cast_fp16)[name = string("query_49_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_12_self_attn_k_proj_weight_to_fp16 = const()[name = string("tower_layers_12_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(255208320)))];
tensor<fp16, [896]> tower_layers_12_self_attn_k_proj_bias_to_fp16 = const()[name = string("tower_layers_12_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256814016)))];
tensor<fp16, [1, 390, 896]> linear_74_cast_fp16 = linear(bias = tower_layers_12_self_attn_k_proj_bias_to_fp16, weight = tower_layers_12_self_attn_k_proj_weight_to_fp16, x = input_179_cast_fp16)[name = string("linear_74_cast_fp16")];
tensor<int32, [4]> var_2063 = const()[name = string("op_2063"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> key_49_cast_fp16 = reshape(shape = var_2063, x = linear_74_cast_fp16)[name = string("key_49_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_12_self_attn_v_proj_weight_to_fp16 = const()[name = string("tower_layers_12_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(256815872)))];
tensor<fp16, [896]> tower_layers_12_self_attn_v_proj_bias_to_fp16 = const()[name = string("tower_layers_12_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258421568)))];
tensor<fp16, [1, 390, 896]> linear_75_cast_fp16 = linear(bias = tower_layers_12_self_attn_v_proj_bias_to_fp16, weight = tower_layers_12_self_attn_v_proj_weight_to_fp16, x = input_179_cast_fp16)[name = string("linear_75_cast_fp16")];
tensor<int32, [4]> var_2072 = const()[name = string("op_2072"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> value_49_cast_fp16 = reshape(shape = var_2072, x = linear_75_cast_fp16)[name = string("value_49_cast_fp16")];
tensor<int32, [4]> var_2090 = const()[name = string("op_2090"), val = tensor<int32, [4]>([0, 2, 1, 3])];
bool var_2095_transpose_x_0 = const()[name = string("op_2095_transpose_x_0"), val = bool(false)];
bool var_2095_transpose_y_0 = const()[name = string("op_2095_transpose_y_0"), val = bool(false)];
tensor<int32, [4]> transpose_78_perm_0 = const()[name = string("transpose_78_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
tensor<int32, [4]> transpose_79_perm_0 = const()[name = string("transpose_79_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
tensor<fp16, [1, 14, 64, 390]> transpose_79 = transpose(perm = transpose_79_perm_0, x = key_49_cast_fp16)[name = string("transpose_111")];
tensor<fp16, [1, 14, 390, 64]> transpose_78 = transpose(perm = transpose_78_perm_0, x = query_49_cast_fp16)[name = string("transpose_112")];
tensor<fp16, [1, 14, 390, 390]> var_2095_cast_fp16 = matmul(transpose_x = var_2095_transpose_x_0, transpose_y = var_2095_transpose_y_0, x = transpose_78, y = transpose_79)[name = string("op_2095_cast_fp16")];
fp16 var_2096_to_fp16 = const()[name = string("op_2096_to_fp16"), val = fp16(0x1p-3)];
tensor<fp16, [1, 14, 390, 390]> scores_73_cast_fp16 = mul(x = var_2095_cast_fp16, y = var_2096_to_fp16)[name = string("scores_73_cast_fp16")];
tensor<fp16, [1, 14, 390, 390]> scores_75_cast_fp16 = add(x = scores_73_cast_fp16, y = attention_mask_to_fp16)[name = string("scores_75_cast_fp16")];
int32 var_2105 = const()[name = string("op_2105"), val = int32(-1)];
tensor<fp16, [1, 14, 390, 390]> var_2107_cast_fp16 = softmax(axis = var_2105, x = scores_75_cast_fp16)[name = string("op_2107_cast_fp16")];
bool output_25_transpose_x_0 = const()[name = string("output_25_transpose_x_0"), val = bool(false)];
bool output_25_transpose_y_0 = const()[name = string("output_25_transpose_y_0"), val = bool(false)];
tensor<fp16, [1, 14, 390, 64]> value_51_cast_fp16 = transpose(perm = var_2090, x = value_49_cast_fp16)[name = string("transpose_113")];
tensor<fp16, [1, 14, 390, 64]> output_25_cast_fp16 = matmul(transpose_x = output_25_transpose_x_0, transpose_y = output_25_transpose_y_0, x = var_2107_cast_fp16, y = value_51_cast_fp16)[name = string("output_25_cast_fp16")];
tensor<int32, [4]> var_2118 = const()[name = string("op_2118"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_2123 = const()[name = string("op_2123"), val = tensor<int32, [3]>([1, 390, 896])];
tensor<fp16, [1, 390, 14, 64]> var_2119_cast_fp16 = transpose(perm = var_2118, x = output_25_cast_fp16)[name = string("transpose_110")];
tensor<fp16, [1, 390, 896]> input_181_cast_fp16 = reshape(shape = var_2123, x = var_2119_cast_fp16)[name = string("input_181_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_12_self_attn_out_proj_weight_to_fp16 = const()[name = string("tower_layers_12_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(258423424)))];
tensor<fp16, [896]> tower_layers_12_self_attn_out_proj_bias_to_fp16 = const()[name = string("tower_layers_12_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260029120)))];
tensor<fp16, [1, 390, 896]> linear_76_cast_fp16 = linear(bias = tower_layers_12_self_attn_out_proj_bias_to_fp16, weight = tower_layers_12_self_attn_out_proj_weight_to_fp16, x = input_181_cast_fp16)[name = string("linear_76_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_183_cast_fp16 = add(x = input_177_cast_fp16, y = linear_76_cast_fp16)[name = string("input_183_cast_fp16")];
tensor<int32, [1]> input_185_axes_0 = const()[name = string("input_185_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_12_final_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_12_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260030976)))];
tensor<fp16, [896]> tower_layers_12_final_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_12_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260032832)))];
fp16 var_2131_to_fp16 = const()[name = string("op_2131_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_185_cast_fp16 = layer_norm(axes = input_185_axes_0, beta = tower_layers_12_final_layer_norm_bias_to_fp16, epsilon = var_2131_to_fp16, gamma = tower_layers_12_final_layer_norm_weight_to_fp16, x = input_183_cast_fp16)[name = string("input_185_cast_fp16")];
tensor<fp16, [3584, 896]> tower_layers_12_fc1_weight_to_fp16 = const()[name = string("tower_layers_12_fc1_weight_to_fp16"), val = tensor<fp16, [3584, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(260034688)))];
tensor<fp16, [3584]> tower_layers_12_fc1_bias_to_fp16 = const()[name = string("tower_layers_12_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(266457280)))];
tensor<fp16, [1, 390, 3584]> linear_77_cast_fp16 = linear(bias = tower_layers_12_fc1_bias_to_fp16, weight = tower_layers_12_fc1_weight_to_fp16, x = input_185_cast_fp16)[name = string("linear_77_cast_fp16")];
string input_189_mode_0 = const()[name = string("input_189_mode_0"), val = string("EXACT")];
tensor<fp16, [1, 390, 3584]> input_189_cast_fp16 = gelu(mode = input_189_mode_0, x = linear_77_cast_fp16)[name = string("input_189_cast_fp16")];
tensor<fp16, [896, 3584]> tower_layers_12_fc2_weight_to_fp16 = const()[name = string("tower_layers_12_fc2_weight_to_fp16"), val = tensor<fp16, [896, 3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(266464512)))];
tensor<fp16, [896]> tower_layers_12_fc2_bias_to_fp16 = const()[name = string("tower_layers_12_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272887104)))];
tensor<fp16, [1, 390, 896]> linear_78_cast_fp16 = linear(bias = tower_layers_12_fc2_bias_to_fp16, weight = tower_layers_12_fc2_weight_to_fp16, x = input_189_cast_fp16)[name = string("linear_78_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_191_cast_fp16 = add(x = input_183_cast_fp16, y = linear_78_cast_fp16)[name = string("input_191_cast_fp16")];
tensor<int32, [1]> input_193_axes_0 = const()[name = string("input_193_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_13_self_attn_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_13_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272888960)))];
tensor<fp16, [896]> tower_layers_13_self_attn_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_13_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272890816)))];
fp16 var_2148_to_fp16 = const()[name = string("op_2148_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_193_cast_fp16 = layer_norm(axes = input_193_axes_0, beta = tower_layers_13_self_attn_layer_norm_bias_to_fp16, epsilon = var_2148_to_fp16, gamma = tower_layers_13_self_attn_layer_norm_weight_to_fp16, x = input_191_cast_fp16)[name = string("input_193_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_13_self_attn_q_proj_weight_to_fp16 = const()[name = string("tower_layers_13_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(272892672)))];
tensor<fp16, [896]> tower_layers_13_self_attn_q_proj_bias_to_fp16 = const()[name = string("tower_layers_13_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274498368)))];
tensor<fp16, [1, 390, 896]> linear_79_cast_fp16 = linear(bias = tower_layers_13_self_attn_q_proj_bias_to_fp16, weight = tower_layers_13_self_attn_q_proj_weight_to_fp16, x = input_193_cast_fp16)[name = string("linear_79_cast_fp16")];
tensor<int32, [4]> var_2161 = const()[name = string("op_2161"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> query_53_cast_fp16 = reshape(shape = var_2161, x = linear_79_cast_fp16)[name = string("query_53_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_13_self_attn_k_proj_weight_to_fp16 = const()[name = string("tower_layers_13_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(274500224)))];
tensor<fp16, [896]> tower_layers_13_self_attn_k_proj_bias_to_fp16 = const()[name = string("tower_layers_13_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276105920)))];
tensor<fp16, [1, 390, 896]> linear_80_cast_fp16 = linear(bias = tower_layers_13_self_attn_k_proj_bias_to_fp16, weight = tower_layers_13_self_attn_k_proj_weight_to_fp16, x = input_193_cast_fp16)[name = string("linear_80_cast_fp16")];
tensor<int32, [4]> var_2170 = const()[name = string("op_2170"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> key_53_cast_fp16 = reshape(shape = var_2170, x = linear_80_cast_fp16)[name = string("key_53_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_13_self_attn_v_proj_weight_to_fp16 = const()[name = string("tower_layers_13_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(276107776)))];
tensor<fp16, [896]> tower_layers_13_self_attn_v_proj_bias_to_fp16 = const()[name = string("tower_layers_13_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277713472)))];
tensor<fp16, [1, 390, 896]> linear_81_cast_fp16 = linear(bias = tower_layers_13_self_attn_v_proj_bias_to_fp16, weight = tower_layers_13_self_attn_v_proj_weight_to_fp16, x = input_193_cast_fp16)[name = string("linear_81_cast_fp16")];
tensor<int32, [4]> var_2179 = const()[name = string("op_2179"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> value_53_cast_fp16 = reshape(shape = var_2179, x = linear_81_cast_fp16)[name = string("value_53_cast_fp16")];
tensor<int32, [4]> var_2197 = const()[name = string("op_2197"), val = tensor<int32, [4]>([0, 2, 1, 3])];
bool var_2202_transpose_x_0 = const()[name = string("op_2202_transpose_x_0"), val = bool(false)];
bool var_2202_transpose_y_0 = const()[name = string("op_2202_transpose_y_0"), val = bool(false)];
tensor<int32, [4]> transpose_80_perm_0 = const()[name = string("transpose_80_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
tensor<int32, [4]> transpose_81_perm_0 = const()[name = string("transpose_81_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
tensor<fp16, [1, 14, 64, 390]> transpose_81 = transpose(perm = transpose_81_perm_0, x = key_53_cast_fp16)[name = string("transpose_107")];
tensor<fp16, [1, 14, 390, 64]> transpose_80 = transpose(perm = transpose_80_perm_0, x = query_53_cast_fp16)[name = string("transpose_108")];
tensor<fp16, [1, 14, 390, 390]> var_2202_cast_fp16 = matmul(transpose_x = var_2202_transpose_x_0, transpose_y = var_2202_transpose_y_0, x = transpose_80, y = transpose_81)[name = string("op_2202_cast_fp16")];
fp16 var_2203_to_fp16 = const()[name = string("op_2203_to_fp16"), val = fp16(0x1p-3)];
tensor<fp16, [1, 14, 390, 390]> scores_79_cast_fp16 = mul(x = var_2202_cast_fp16, y = var_2203_to_fp16)[name = string("scores_79_cast_fp16")];
tensor<fp16, [1, 14, 390, 390]> scores_81_cast_fp16 = add(x = scores_79_cast_fp16, y = attention_mask_to_fp16)[name = string("scores_81_cast_fp16")];
int32 var_2212 = const()[name = string("op_2212"), val = int32(-1)];
tensor<fp16, [1, 14, 390, 390]> var_2214_cast_fp16 = softmax(axis = var_2212, x = scores_81_cast_fp16)[name = string("op_2214_cast_fp16")];
bool output_27_transpose_x_0 = const()[name = string("output_27_transpose_x_0"), val = bool(false)];
bool output_27_transpose_y_0 = const()[name = string("output_27_transpose_y_0"), val = bool(false)];
tensor<fp16, [1, 14, 390, 64]> value_55_cast_fp16 = transpose(perm = var_2197, x = value_53_cast_fp16)[name = string("transpose_109")];
tensor<fp16, [1, 14, 390, 64]> output_27_cast_fp16 = matmul(transpose_x = output_27_transpose_x_0, transpose_y = output_27_transpose_y_0, x = var_2214_cast_fp16, y = value_55_cast_fp16)[name = string("output_27_cast_fp16")];
tensor<int32, [4]> var_2225 = const()[name = string("op_2225"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_2230 = const()[name = string("op_2230"), val = tensor<int32, [3]>([1, 390, 896])];
tensor<fp16, [1, 390, 14, 64]> var_2226_cast_fp16 = transpose(perm = var_2225, x = output_27_cast_fp16)[name = string("transpose_106")];
tensor<fp16, [1, 390, 896]> input_195_cast_fp16 = reshape(shape = var_2230, x = var_2226_cast_fp16)[name = string("input_195_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_13_self_attn_out_proj_weight_to_fp16 = const()[name = string("tower_layers_13_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(277715328)))];
tensor<fp16, [896]> tower_layers_13_self_attn_out_proj_bias_to_fp16 = const()[name = string("tower_layers_13_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279321024)))];
tensor<fp16, [1, 390, 896]> linear_82_cast_fp16 = linear(bias = tower_layers_13_self_attn_out_proj_bias_to_fp16, weight = tower_layers_13_self_attn_out_proj_weight_to_fp16, x = input_195_cast_fp16)[name = string("linear_82_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_197_cast_fp16 = add(x = input_191_cast_fp16, y = linear_82_cast_fp16)[name = string("input_197_cast_fp16")];
tensor<int32, [1]> input_199_axes_0 = const()[name = string("input_199_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_13_final_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_13_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279322880)))];
tensor<fp16, [896]> tower_layers_13_final_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_13_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279324736)))];
fp16 var_2238_to_fp16 = const()[name = string("op_2238_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_199_cast_fp16 = layer_norm(axes = input_199_axes_0, beta = tower_layers_13_final_layer_norm_bias_to_fp16, epsilon = var_2238_to_fp16, gamma = tower_layers_13_final_layer_norm_weight_to_fp16, x = input_197_cast_fp16)[name = string("input_199_cast_fp16")];
tensor<fp16, [3584, 896]> tower_layers_13_fc1_weight_to_fp16 = const()[name = string("tower_layers_13_fc1_weight_to_fp16"), val = tensor<fp16, [3584, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(279326592)))];
tensor<fp16, [3584]> tower_layers_13_fc1_bias_to_fp16 = const()[name = string("tower_layers_13_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285749184)))];
tensor<fp16, [1, 390, 3584]> linear_83_cast_fp16 = linear(bias = tower_layers_13_fc1_bias_to_fp16, weight = tower_layers_13_fc1_weight_to_fp16, x = input_199_cast_fp16)[name = string("linear_83_cast_fp16")];
string input_203_mode_0 = const()[name = string("input_203_mode_0"), val = string("EXACT")];
tensor<fp16, [1, 390, 3584]> input_203_cast_fp16 = gelu(mode = input_203_mode_0, x = linear_83_cast_fp16)[name = string("input_203_cast_fp16")];
tensor<fp16, [896, 3584]> tower_layers_13_fc2_weight_to_fp16 = const()[name = string("tower_layers_13_fc2_weight_to_fp16"), val = tensor<fp16, [896, 3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(285756416)))];
tensor<fp16, [896]> tower_layers_13_fc2_bias_to_fp16 = const()[name = string("tower_layers_13_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292179008)))];
tensor<fp16, [1, 390, 896]> linear_84_cast_fp16 = linear(bias = tower_layers_13_fc2_bias_to_fp16, weight = tower_layers_13_fc2_weight_to_fp16, x = input_203_cast_fp16)[name = string("linear_84_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_205_cast_fp16 = add(x = input_197_cast_fp16, y = linear_84_cast_fp16)[name = string("input_205_cast_fp16")];
tensor<int32, [1]> input_207_axes_0 = const()[name = string("input_207_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_14_self_attn_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_14_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292180864)))];
tensor<fp16, [896]> tower_layers_14_self_attn_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_14_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292182720)))];
fp16 var_2255_to_fp16 = const()[name = string("op_2255_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_207_cast_fp16 = layer_norm(axes = input_207_axes_0, beta = tower_layers_14_self_attn_layer_norm_bias_to_fp16, epsilon = var_2255_to_fp16, gamma = tower_layers_14_self_attn_layer_norm_weight_to_fp16, x = input_205_cast_fp16)[name = string("input_207_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_14_self_attn_q_proj_weight_to_fp16 = const()[name = string("tower_layers_14_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(292184576)))];
tensor<fp16, [896]> tower_layers_14_self_attn_q_proj_bias_to_fp16 = const()[name = string("tower_layers_14_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293790272)))];
tensor<fp16, [1, 390, 896]> linear_85_cast_fp16 = linear(bias = tower_layers_14_self_attn_q_proj_bias_to_fp16, weight = tower_layers_14_self_attn_q_proj_weight_to_fp16, x = input_207_cast_fp16)[name = string("linear_85_cast_fp16")];
tensor<int32, [4]> var_2268 = const()[name = string("op_2268"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> query_57_cast_fp16 = reshape(shape = var_2268, x = linear_85_cast_fp16)[name = string("query_57_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_14_self_attn_k_proj_weight_to_fp16 = const()[name = string("tower_layers_14_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(293792128)))];
tensor<fp16, [896]> tower_layers_14_self_attn_k_proj_bias_to_fp16 = const()[name = string("tower_layers_14_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295397824)))];
tensor<fp16, [1, 390, 896]> linear_86_cast_fp16 = linear(bias = tower_layers_14_self_attn_k_proj_bias_to_fp16, weight = tower_layers_14_self_attn_k_proj_weight_to_fp16, x = input_207_cast_fp16)[name = string("linear_86_cast_fp16")];
tensor<int32, [4]> var_2277 = const()[name = string("op_2277"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> key_57_cast_fp16 = reshape(shape = var_2277, x = linear_86_cast_fp16)[name = string("key_57_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_14_self_attn_v_proj_weight_to_fp16 = const()[name = string("tower_layers_14_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(295399680)))];
tensor<fp16, [896]> tower_layers_14_self_attn_v_proj_bias_to_fp16 = const()[name = string("tower_layers_14_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297005376)))];
tensor<fp16, [1, 390, 896]> linear_87_cast_fp16 = linear(bias = tower_layers_14_self_attn_v_proj_bias_to_fp16, weight = tower_layers_14_self_attn_v_proj_weight_to_fp16, x = input_207_cast_fp16)[name = string("linear_87_cast_fp16")];
tensor<int32, [4]> var_2286 = const()[name = string("op_2286"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> value_57_cast_fp16 = reshape(shape = var_2286, x = linear_87_cast_fp16)[name = string("value_57_cast_fp16")];
tensor<int32, [4]> var_2304 = const()[name = string("op_2304"), val = tensor<int32, [4]>([0, 2, 1, 3])];
bool var_2309_transpose_x_0 = const()[name = string("op_2309_transpose_x_0"), val = bool(false)];
bool var_2309_transpose_y_0 = const()[name = string("op_2309_transpose_y_0"), val = bool(false)];
tensor<int32, [4]> transpose_82_perm_0 = const()[name = string("transpose_82_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
tensor<int32, [4]> transpose_83_perm_0 = const()[name = string("transpose_83_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
tensor<fp16, [1, 14, 64, 390]> transpose_83 = transpose(perm = transpose_83_perm_0, x = key_57_cast_fp16)[name = string("transpose_103")];
tensor<fp16, [1, 14, 390, 64]> transpose_82 = transpose(perm = transpose_82_perm_0, x = query_57_cast_fp16)[name = string("transpose_104")];
tensor<fp16, [1, 14, 390, 390]> var_2309_cast_fp16 = matmul(transpose_x = var_2309_transpose_x_0, transpose_y = var_2309_transpose_y_0, x = transpose_82, y = transpose_83)[name = string("op_2309_cast_fp16")];
fp16 var_2310_to_fp16 = const()[name = string("op_2310_to_fp16"), val = fp16(0x1p-3)];
tensor<fp16, [1, 14, 390, 390]> scores_85_cast_fp16 = mul(x = var_2309_cast_fp16, y = var_2310_to_fp16)[name = string("scores_85_cast_fp16")];
tensor<fp16, [1, 14, 390, 390]> scores_87_cast_fp16 = add(x = scores_85_cast_fp16, y = attention_mask_to_fp16)[name = string("scores_87_cast_fp16")];
int32 var_2319 = const()[name = string("op_2319"), val = int32(-1)];
tensor<fp16, [1, 14, 390, 390]> var_2321_cast_fp16 = softmax(axis = var_2319, x = scores_87_cast_fp16)[name = string("op_2321_cast_fp16")];
bool output_29_transpose_x_0 = const()[name = string("output_29_transpose_x_0"), val = bool(false)];
bool output_29_transpose_y_0 = const()[name = string("output_29_transpose_y_0"), val = bool(false)];
tensor<fp16, [1, 14, 390, 64]> value_59_cast_fp16 = transpose(perm = var_2304, x = value_57_cast_fp16)[name = string("transpose_105")];
tensor<fp16, [1, 14, 390, 64]> output_29_cast_fp16 = matmul(transpose_x = output_29_transpose_x_0, transpose_y = output_29_transpose_y_0, x = var_2321_cast_fp16, y = value_59_cast_fp16)[name = string("output_29_cast_fp16")];
tensor<int32, [4]> var_2332 = const()[name = string("op_2332"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_2337 = const()[name = string("op_2337"), val = tensor<int32, [3]>([1, 390, 896])];
tensor<fp16, [1, 390, 14, 64]> var_2333_cast_fp16 = transpose(perm = var_2332, x = output_29_cast_fp16)[name = string("transpose_102")];
tensor<fp16, [1, 390, 896]> input_209_cast_fp16 = reshape(shape = var_2337, x = var_2333_cast_fp16)[name = string("input_209_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_14_self_attn_out_proj_weight_to_fp16 = const()[name = string("tower_layers_14_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(297007232)))];
tensor<fp16, [896]> tower_layers_14_self_attn_out_proj_bias_to_fp16 = const()[name = string("tower_layers_14_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298612928)))];
tensor<fp16, [1, 390, 896]> linear_88_cast_fp16 = linear(bias = tower_layers_14_self_attn_out_proj_bias_to_fp16, weight = tower_layers_14_self_attn_out_proj_weight_to_fp16, x = input_209_cast_fp16)[name = string("linear_88_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_211_cast_fp16 = add(x = input_205_cast_fp16, y = linear_88_cast_fp16)[name = string("input_211_cast_fp16")];
tensor<int32, [1]> input_213_axes_0 = const()[name = string("input_213_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_14_final_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_14_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298614784)))];
tensor<fp16, [896]> tower_layers_14_final_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_14_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298616640)))];
fp16 var_2345_to_fp16 = const()[name = string("op_2345_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_213_cast_fp16 = layer_norm(axes = input_213_axes_0, beta = tower_layers_14_final_layer_norm_bias_to_fp16, epsilon = var_2345_to_fp16, gamma = tower_layers_14_final_layer_norm_weight_to_fp16, x = input_211_cast_fp16)[name = string("input_213_cast_fp16")];
tensor<fp16, [3584, 896]> tower_layers_14_fc1_weight_to_fp16 = const()[name = string("tower_layers_14_fc1_weight_to_fp16"), val = tensor<fp16, [3584, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(298618496)))];
tensor<fp16, [3584]> tower_layers_14_fc1_bias_to_fp16 = const()[name = string("tower_layers_14_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305041088)))];
tensor<fp16, [1, 390, 3584]> linear_89_cast_fp16 = linear(bias = tower_layers_14_fc1_bias_to_fp16, weight = tower_layers_14_fc1_weight_to_fp16, x = input_213_cast_fp16)[name = string("linear_89_cast_fp16")];
string input_217_mode_0 = const()[name = string("input_217_mode_0"), val = string("EXACT")];
tensor<fp16, [1, 390, 3584]> input_217_cast_fp16 = gelu(mode = input_217_mode_0, x = linear_89_cast_fp16)[name = string("input_217_cast_fp16")];
tensor<fp16, [896, 3584]> tower_layers_14_fc2_weight_to_fp16 = const()[name = string("tower_layers_14_fc2_weight_to_fp16"), val = tensor<fp16, [896, 3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(305048320)))];
tensor<fp16, [896]> tower_layers_14_fc2_bias_to_fp16 = const()[name = string("tower_layers_14_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311470912)))];
tensor<fp16, [1, 390, 896]> linear_90_cast_fp16 = linear(bias = tower_layers_14_fc2_bias_to_fp16, weight = tower_layers_14_fc2_weight_to_fp16, x = input_217_cast_fp16)[name = string("linear_90_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_219_cast_fp16 = add(x = input_211_cast_fp16, y = linear_90_cast_fp16)[name = string("input_219_cast_fp16")];
tensor<int32, [1]> input_221_axes_0 = const()[name = string("input_221_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_15_self_attn_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_15_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311472768)))];
tensor<fp16, [896]> tower_layers_15_self_attn_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_15_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311474624)))];
fp16 var_2362_to_fp16 = const()[name = string("op_2362_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_221_cast_fp16 = layer_norm(axes = input_221_axes_0, beta = tower_layers_15_self_attn_layer_norm_bias_to_fp16, epsilon = var_2362_to_fp16, gamma = tower_layers_15_self_attn_layer_norm_weight_to_fp16, x = input_219_cast_fp16)[name = string("input_221_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_15_self_attn_q_proj_weight_to_fp16 = const()[name = string("tower_layers_15_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(311476480)))];
tensor<fp16, [896]> tower_layers_15_self_attn_q_proj_bias_to_fp16 = const()[name = string("tower_layers_15_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313082176)))];
tensor<fp16, [1, 390, 896]> linear_91_cast_fp16 = linear(bias = tower_layers_15_self_attn_q_proj_bias_to_fp16, weight = tower_layers_15_self_attn_q_proj_weight_to_fp16, x = input_221_cast_fp16)[name = string("linear_91_cast_fp16")];
tensor<int32, [4]> var_2375 = const()[name = string("op_2375"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> query_61_cast_fp16 = reshape(shape = var_2375, x = linear_91_cast_fp16)[name = string("query_61_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_15_self_attn_k_proj_weight_to_fp16 = const()[name = string("tower_layers_15_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(313084032)))];
tensor<fp16, [896]> tower_layers_15_self_attn_k_proj_bias_to_fp16 = const()[name = string("tower_layers_15_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314689728)))];
tensor<fp16, [1, 390, 896]> linear_92_cast_fp16 = linear(bias = tower_layers_15_self_attn_k_proj_bias_to_fp16, weight = tower_layers_15_self_attn_k_proj_weight_to_fp16, x = input_221_cast_fp16)[name = string("linear_92_cast_fp16")];
tensor<int32, [4]> var_2384 = const()[name = string("op_2384"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> key_61_cast_fp16 = reshape(shape = var_2384, x = linear_92_cast_fp16)[name = string("key_61_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_15_self_attn_v_proj_weight_to_fp16 = const()[name = string("tower_layers_15_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(314691584)))];
tensor<fp16, [896]> tower_layers_15_self_attn_v_proj_bias_to_fp16 = const()[name = string("tower_layers_15_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316297280)))];
tensor<fp16, [1, 390, 896]> linear_93_cast_fp16 = linear(bias = tower_layers_15_self_attn_v_proj_bias_to_fp16, weight = tower_layers_15_self_attn_v_proj_weight_to_fp16, x = input_221_cast_fp16)[name = string("linear_93_cast_fp16")];
tensor<int32, [4]> var_2393 = const()[name = string("op_2393"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> value_61_cast_fp16 = reshape(shape = var_2393, x = linear_93_cast_fp16)[name = string("value_61_cast_fp16")];
tensor<int32, [4]> var_2411 = const()[name = string("op_2411"), val = tensor<int32, [4]>([0, 2, 1, 3])];
bool var_2416_transpose_x_0 = const()[name = string("op_2416_transpose_x_0"), val = bool(false)];
bool var_2416_transpose_y_0 = const()[name = string("op_2416_transpose_y_0"), val = bool(false)];
tensor<int32, [4]> transpose_84_perm_0 = const()[name = string("transpose_84_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
tensor<int32, [4]> transpose_85_perm_0 = const()[name = string("transpose_85_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
tensor<fp16, [1, 14, 64, 390]> transpose_85 = transpose(perm = transpose_85_perm_0, x = key_61_cast_fp16)[name = string("transpose_99")];
tensor<fp16, [1, 14, 390, 64]> transpose_84 = transpose(perm = transpose_84_perm_0, x = query_61_cast_fp16)[name = string("transpose_100")];
tensor<fp16, [1, 14, 390, 390]> var_2416_cast_fp16 = matmul(transpose_x = var_2416_transpose_x_0, transpose_y = var_2416_transpose_y_0, x = transpose_84, y = transpose_85)[name = string("op_2416_cast_fp16")];
fp16 var_2417_to_fp16 = const()[name = string("op_2417_to_fp16"), val = fp16(0x1p-3)];
tensor<fp16, [1, 14, 390, 390]> scores_91_cast_fp16 = mul(x = var_2416_cast_fp16, y = var_2417_to_fp16)[name = string("scores_91_cast_fp16")];
tensor<fp16, [1, 14, 390, 390]> scores_93_cast_fp16 = add(x = scores_91_cast_fp16, y = attention_mask_to_fp16)[name = string("scores_93_cast_fp16")];
int32 var_2426 = const()[name = string("op_2426"), val = int32(-1)];
tensor<fp16, [1, 14, 390, 390]> var_2428_cast_fp16 = softmax(axis = var_2426, x = scores_93_cast_fp16)[name = string("op_2428_cast_fp16")];
bool output_31_transpose_x_0 = const()[name = string("output_31_transpose_x_0"), val = bool(false)];
bool output_31_transpose_y_0 = const()[name = string("output_31_transpose_y_0"), val = bool(false)];
tensor<fp16, [1, 14, 390, 64]> value_63_cast_fp16 = transpose(perm = var_2411, x = value_61_cast_fp16)[name = string("transpose_101")];
tensor<fp16, [1, 14, 390, 64]> output_31_cast_fp16 = matmul(transpose_x = output_31_transpose_x_0, transpose_y = output_31_transpose_y_0, x = var_2428_cast_fp16, y = value_63_cast_fp16)[name = string("output_31_cast_fp16")];
tensor<int32, [4]> var_2439 = const()[name = string("op_2439"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_2444 = const()[name = string("op_2444"), val = tensor<int32, [3]>([1, 390, 896])];
tensor<fp16, [1, 390, 14, 64]> var_2440_cast_fp16 = transpose(perm = var_2439, x = output_31_cast_fp16)[name = string("transpose_98")];
tensor<fp16, [1, 390, 896]> input_223_cast_fp16 = reshape(shape = var_2444, x = var_2440_cast_fp16)[name = string("input_223_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_15_self_attn_out_proj_weight_to_fp16 = const()[name = string("tower_layers_15_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(316299136)))];
tensor<fp16, [896]> tower_layers_15_self_attn_out_proj_bias_to_fp16 = const()[name = string("tower_layers_15_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(317904832)))];
tensor<fp16, [1, 390, 896]> linear_94_cast_fp16 = linear(bias = tower_layers_15_self_attn_out_proj_bias_to_fp16, weight = tower_layers_15_self_attn_out_proj_weight_to_fp16, x = input_223_cast_fp16)[name = string("linear_94_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_225_cast_fp16 = add(x = input_219_cast_fp16, y = linear_94_cast_fp16)[name = string("input_225_cast_fp16")];
tensor<int32, [1]> input_227_axes_0 = const()[name = string("input_227_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_15_final_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_15_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(317906688)))];
tensor<fp16, [896]> tower_layers_15_final_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_15_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(317908544)))];
fp16 var_2452_to_fp16 = const()[name = string("op_2452_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_227_cast_fp16 = layer_norm(axes = input_227_axes_0, beta = tower_layers_15_final_layer_norm_bias_to_fp16, epsilon = var_2452_to_fp16, gamma = tower_layers_15_final_layer_norm_weight_to_fp16, x = input_225_cast_fp16)[name = string("input_227_cast_fp16")];
tensor<fp16, [3584, 896]> tower_layers_15_fc1_weight_to_fp16 = const()[name = string("tower_layers_15_fc1_weight_to_fp16"), val = tensor<fp16, [3584, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(317910400)))];
tensor<fp16, [3584]> tower_layers_15_fc1_bias_to_fp16 = const()[name = string("tower_layers_15_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(324332992)))];
tensor<fp16, [1, 390, 3584]> linear_95_cast_fp16 = linear(bias = tower_layers_15_fc1_bias_to_fp16, weight = tower_layers_15_fc1_weight_to_fp16, x = input_227_cast_fp16)[name = string("linear_95_cast_fp16")];
string input_231_mode_0 = const()[name = string("input_231_mode_0"), val = string("EXACT")];
tensor<fp16, [1, 390, 3584]> input_231_cast_fp16 = gelu(mode = input_231_mode_0, x = linear_95_cast_fp16)[name = string("input_231_cast_fp16")];
tensor<fp16, [896, 3584]> tower_layers_15_fc2_weight_to_fp16 = const()[name = string("tower_layers_15_fc2_weight_to_fp16"), val = tensor<fp16, [896, 3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(324340224)))];
tensor<fp16, [896]> tower_layers_15_fc2_bias_to_fp16 = const()[name = string("tower_layers_15_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(330762816)))];
tensor<fp16, [1, 390, 896]> linear_96_cast_fp16 = linear(bias = tower_layers_15_fc2_bias_to_fp16, weight = tower_layers_15_fc2_weight_to_fp16, x = input_231_cast_fp16)[name = string("linear_96_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_233_cast_fp16 = add(x = input_225_cast_fp16, y = linear_96_cast_fp16)[name = string("input_233_cast_fp16")];
tensor<int32, [1]> input_235_axes_0 = const()[name = string("input_235_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_16_self_attn_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_16_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(330764672)))];
tensor<fp16, [896]> tower_layers_16_self_attn_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_16_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(330766528)))];
fp16 var_2469_to_fp16 = const()[name = string("op_2469_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_235_cast_fp16 = layer_norm(axes = input_235_axes_0, beta = tower_layers_16_self_attn_layer_norm_bias_to_fp16, epsilon = var_2469_to_fp16, gamma = tower_layers_16_self_attn_layer_norm_weight_to_fp16, x = input_233_cast_fp16)[name = string("input_235_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_16_self_attn_q_proj_weight_to_fp16 = const()[name = string("tower_layers_16_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(330768384)))];
tensor<fp16, [896]> tower_layers_16_self_attn_q_proj_bias_to_fp16 = const()[name = string("tower_layers_16_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(332374080)))];
tensor<fp16, [1, 390, 896]> linear_97_cast_fp16 = linear(bias = tower_layers_16_self_attn_q_proj_bias_to_fp16, weight = tower_layers_16_self_attn_q_proj_weight_to_fp16, x = input_235_cast_fp16)[name = string("linear_97_cast_fp16")];
tensor<int32, [4]> var_2482 = const()[name = string("op_2482"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> query_65_cast_fp16 = reshape(shape = var_2482, x = linear_97_cast_fp16)[name = string("query_65_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_16_self_attn_k_proj_weight_to_fp16 = const()[name = string("tower_layers_16_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(332375936)))];
tensor<fp16, [896]> tower_layers_16_self_attn_k_proj_bias_to_fp16 = const()[name = string("tower_layers_16_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(333981632)))];
tensor<fp16, [1, 390, 896]> linear_98_cast_fp16 = linear(bias = tower_layers_16_self_attn_k_proj_bias_to_fp16, weight = tower_layers_16_self_attn_k_proj_weight_to_fp16, x = input_235_cast_fp16)[name = string("linear_98_cast_fp16")];
tensor<int32, [4]> var_2491 = const()[name = string("op_2491"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> key_65_cast_fp16 = reshape(shape = var_2491, x = linear_98_cast_fp16)[name = string("key_65_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_16_self_attn_v_proj_weight_to_fp16 = const()[name = string("tower_layers_16_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(333983488)))];
tensor<fp16, [896]> tower_layers_16_self_attn_v_proj_bias_to_fp16 = const()[name = string("tower_layers_16_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(335589184)))];
tensor<fp16, [1, 390, 896]> linear_99_cast_fp16 = linear(bias = tower_layers_16_self_attn_v_proj_bias_to_fp16, weight = tower_layers_16_self_attn_v_proj_weight_to_fp16, x = input_235_cast_fp16)[name = string("linear_99_cast_fp16")];
tensor<int32, [4]> var_2500 = const()[name = string("op_2500"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> value_65_cast_fp16 = reshape(shape = var_2500, x = linear_99_cast_fp16)[name = string("value_65_cast_fp16")];
tensor<int32, [4]> var_2518 = const()[name = string("op_2518"), val = tensor<int32, [4]>([0, 2, 1, 3])];
bool var_2523_transpose_x_0 = const()[name = string("op_2523_transpose_x_0"), val = bool(false)];
bool var_2523_transpose_y_0 = const()[name = string("op_2523_transpose_y_0"), val = bool(false)];
tensor<int32, [4]> transpose_86_perm_0 = const()[name = string("transpose_86_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
tensor<int32, [4]> transpose_87_perm_0 = const()[name = string("transpose_87_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
tensor<fp16, [1, 14, 64, 390]> transpose_87 = transpose(perm = transpose_87_perm_0, x = key_65_cast_fp16)[name = string("transpose_95")];
tensor<fp16, [1, 14, 390, 64]> transpose_86 = transpose(perm = transpose_86_perm_0, x = query_65_cast_fp16)[name = string("transpose_96")];
tensor<fp16, [1, 14, 390, 390]> var_2523_cast_fp16 = matmul(transpose_x = var_2523_transpose_x_0, transpose_y = var_2523_transpose_y_0, x = transpose_86, y = transpose_87)[name = string("op_2523_cast_fp16")];
fp16 var_2524_to_fp16 = const()[name = string("op_2524_to_fp16"), val = fp16(0x1p-3)];
tensor<fp16, [1, 14, 390, 390]> scores_97_cast_fp16 = mul(x = var_2523_cast_fp16, y = var_2524_to_fp16)[name = string("scores_97_cast_fp16")];
tensor<fp16, [1, 14, 390, 390]> scores_99_cast_fp16 = add(x = scores_97_cast_fp16, y = attention_mask_to_fp16)[name = string("scores_99_cast_fp16")];
int32 var_2533 = const()[name = string("op_2533"), val = int32(-1)];
tensor<fp16, [1, 14, 390, 390]> var_2535_cast_fp16 = softmax(axis = var_2533, x = scores_99_cast_fp16)[name = string("op_2535_cast_fp16")];
bool output_33_transpose_x_0 = const()[name = string("output_33_transpose_x_0"), val = bool(false)];
bool output_33_transpose_y_0 = const()[name = string("output_33_transpose_y_0"), val = bool(false)];
tensor<fp16, [1, 14, 390, 64]> value_67_cast_fp16 = transpose(perm = var_2518, x = value_65_cast_fp16)[name = string("transpose_97")];
tensor<fp16, [1, 14, 390, 64]> output_33_cast_fp16 = matmul(transpose_x = output_33_transpose_x_0, transpose_y = output_33_transpose_y_0, x = var_2535_cast_fp16, y = value_67_cast_fp16)[name = string("output_33_cast_fp16")];
tensor<int32, [4]> var_2546 = const()[name = string("op_2546"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_2551 = const()[name = string("op_2551"), val = tensor<int32, [3]>([1, 390, 896])];
tensor<fp16, [1, 390, 14, 64]> var_2547_cast_fp16 = transpose(perm = var_2546, x = output_33_cast_fp16)[name = string("transpose_94")];
tensor<fp16, [1, 390, 896]> input_237_cast_fp16 = reshape(shape = var_2551, x = var_2547_cast_fp16)[name = string("input_237_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_16_self_attn_out_proj_weight_to_fp16 = const()[name = string("tower_layers_16_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(335591040)))];
tensor<fp16, [896]> tower_layers_16_self_attn_out_proj_bias_to_fp16 = const()[name = string("tower_layers_16_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337196736)))];
tensor<fp16, [1, 390, 896]> linear_100_cast_fp16 = linear(bias = tower_layers_16_self_attn_out_proj_bias_to_fp16, weight = tower_layers_16_self_attn_out_proj_weight_to_fp16, x = input_237_cast_fp16)[name = string("linear_100_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_239_cast_fp16 = add(x = input_233_cast_fp16, y = linear_100_cast_fp16)[name = string("input_239_cast_fp16")];
tensor<int32, [1]> input_241_axes_0 = const()[name = string("input_241_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_16_final_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_16_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337198592)))];
tensor<fp16, [896]> tower_layers_16_final_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_16_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337200448)))];
fp16 var_2559_to_fp16 = const()[name = string("op_2559_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_241_cast_fp16 = layer_norm(axes = input_241_axes_0, beta = tower_layers_16_final_layer_norm_bias_to_fp16, epsilon = var_2559_to_fp16, gamma = tower_layers_16_final_layer_norm_weight_to_fp16, x = input_239_cast_fp16)[name = string("input_241_cast_fp16")];
tensor<fp16, [3584, 896]> tower_layers_16_fc1_weight_to_fp16 = const()[name = string("tower_layers_16_fc1_weight_to_fp16"), val = tensor<fp16, [3584, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(337202304)))];
tensor<fp16, [3584]> tower_layers_16_fc1_bias_to_fp16 = const()[name = string("tower_layers_16_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(343624896)))];
tensor<fp16, [1, 390, 3584]> linear_101_cast_fp16 = linear(bias = tower_layers_16_fc1_bias_to_fp16, weight = tower_layers_16_fc1_weight_to_fp16, x = input_241_cast_fp16)[name = string("linear_101_cast_fp16")];
string input_245_mode_0 = const()[name = string("input_245_mode_0"), val = string("EXACT")];
tensor<fp16, [1, 390, 3584]> input_245_cast_fp16 = gelu(mode = input_245_mode_0, x = linear_101_cast_fp16)[name = string("input_245_cast_fp16")];
tensor<fp16, [896, 3584]> tower_layers_16_fc2_weight_to_fp16 = const()[name = string("tower_layers_16_fc2_weight_to_fp16"), val = tensor<fp16, [896, 3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(343632128)))];
tensor<fp16, [896]> tower_layers_16_fc2_bias_to_fp16 = const()[name = string("tower_layers_16_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(350054720)))];
tensor<fp16, [1, 390, 896]> linear_102_cast_fp16 = linear(bias = tower_layers_16_fc2_bias_to_fp16, weight = tower_layers_16_fc2_weight_to_fp16, x = input_245_cast_fp16)[name = string("linear_102_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_247_cast_fp16 = add(x = input_239_cast_fp16, y = linear_102_cast_fp16)[name = string("input_247_cast_fp16")];
tensor<int32, [1]> input_249_axes_0 = const()[name = string("input_249_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_17_self_attn_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_17_self_attn_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(350056576)))];
tensor<fp16, [896]> tower_layers_17_self_attn_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_17_self_attn_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(350058432)))];
fp16 var_2576_to_fp16 = const()[name = string("op_2576_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_249_cast_fp16 = layer_norm(axes = input_249_axes_0, beta = tower_layers_17_self_attn_layer_norm_bias_to_fp16, epsilon = var_2576_to_fp16, gamma = tower_layers_17_self_attn_layer_norm_weight_to_fp16, x = input_247_cast_fp16)[name = string("input_249_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_17_self_attn_q_proj_weight_to_fp16 = const()[name = string("tower_layers_17_self_attn_q_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(350060288)))];
tensor<fp16, [896]> tower_layers_17_self_attn_q_proj_bias_to_fp16 = const()[name = string("tower_layers_17_self_attn_q_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(351665984)))];
tensor<fp16, [1, 390, 896]> linear_103_cast_fp16 = linear(bias = tower_layers_17_self_attn_q_proj_bias_to_fp16, weight = tower_layers_17_self_attn_q_proj_weight_to_fp16, x = input_249_cast_fp16)[name = string("linear_103_cast_fp16")];
tensor<int32, [4]> var_2589 = const()[name = string("op_2589"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> query_69_cast_fp16 = reshape(shape = var_2589, x = linear_103_cast_fp16)[name = string("query_69_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_17_self_attn_k_proj_weight_to_fp16 = const()[name = string("tower_layers_17_self_attn_k_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(351667840)))];
tensor<fp16, [896]> tower_layers_17_self_attn_k_proj_bias_to_fp16 = const()[name = string("tower_layers_17_self_attn_k_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(353273536)))];
tensor<fp16, [1, 390, 896]> linear_104_cast_fp16 = linear(bias = tower_layers_17_self_attn_k_proj_bias_to_fp16, weight = tower_layers_17_self_attn_k_proj_weight_to_fp16, x = input_249_cast_fp16)[name = string("linear_104_cast_fp16")];
tensor<int32, [4]> var_2598 = const()[name = string("op_2598"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> key_69_cast_fp16 = reshape(shape = var_2598, x = linear_104_cast_fp16)[name = string("key_69_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_17_self_attn_v_proj_weight_to_fp16 = const()[name = string("tower_layers_17_self_attn_v_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(353275392)))];
tensor<fp16, [896]> tower_layers_17_self_attn_v_proj_bias_to_fp16 = const()[name = string("tower_layers_17_self_attn_v_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354881088)))];
tensor<fp16, [1, 390, 896]> linear_105_cast_fp16 = linear(bias = tower_layers_17_self_attn_v_proj_bias_to_fp16, weight = tower_layers_17_self_attn_v_proj_weight_to_fp16, x = input_249_cast_fp16)[name = string("linear_105_cast_fp16")];
tensor<int32, [4]> var_2607 = const()[name = string("op_2607"), val = tensor<int32, [4]>([1, 390, 14, 64])];
tensor<fp16, [1, 390, 14, 64]> value_69_cast_fp16 = reshape(shape = var_2607, x = linear_105_cast_fp16)[name = string("value_69_cast_fp16")];
tensor<int32, [4]> var_2625 = const()[name = string("op_2625"), val = tensor<int32, [4]>([0, 2, 1, 3])];
bool var_2630_transpose_x_0 = const()[name = string("op_2630_transpose_x_0"), val = bool(false)];
bool var_2630_transpose_y_0 = const()[name = string("op_2630_transpose_y_0"), val = bool(false)];
tensor<int32, [4]> transpose_88_perm_0 = const()[name = string("transpose_88_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
tensor<int32, [4]> transpose_89_perm_0 = const()[name = string("transpose_89_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
tensor<fp16, [1, 14, 64, 390]> transpose_89 = transpose(perm = transpose_89_perm_0, x = key_69_cast_fp16)[name = string("transpose_91")];
tensor<fp16, [1, 14, 390, 64]> transpose_88 = transpose(perm = transpose_88_perm_0, x = query_69_cast_fp16)[name = string("transpose_92")];
tensor<fp16, [1, 14, 390, 390]> var_2630_cast_fp16 = matmul(transpose_x = var_2630_transpose_x_0, transpose_y = var_2630_transpose_y_0, x = transpose_88, y = transpose_89)[name = string("op_2630_cast_fp16")];
fp16 var_2631_to_fp16 = const()[name = string("op_2631_to_fp16"), val = fp16(0x1p-3)];
tensor<fp16, [1, 14, 390, 390]> scores_103_cast_fp16 = mul(x = var_2630_cast_fp16, y = var_2631_to_fp16)[name = string("scores_103_cast_fp16")];
tensor<fp16, [1, 14, 390, 390]> scores_105_cast_fp16 = add(x = scores_103_cast_fp16, y = attention_mask_to_fp16)[name = string("scores_105_cast_fp16")];
int32 var_2640 = const()[name = string("op_2640"), val = int32(-1)];
tensor<fp16, [1, 14, 390, 390]> var_2642_cast_fp16 = softmax(axis = var_2640, x = scores_105_cast_fp16)[name = string("op_2642_cast_fp16")];
bool output_transpose_x_0 = const()[name = string("output_transpose_x_0"), val = bool(false)];
bool output_transpose_y_0 = const()[name = string("output_transpose_y_0"), val = bool(false)];
tensor<fp16, [1, 14, 390, 64]> value_cast_fp16 = transpose(perm = var_2625, x = value_69_cast_fp16)[name = string("transpose_93")];
tensor<fp16, [1, 14, 390, 64]> output_cast_fp16 = matmul(transpose_x = output_transpose_x_0, transpose_y = output_transpose_y_0, x = var_2642_cast_fp16, y = value_cast_fp16)[name = string("output_cast_fp16")];
tensor<int32, [4]> var_2653 = const()[name = string("op_2653"), val = tensor<int32, [4]>([0, 2, 1, 3])];
tensor<int32, [3]> var_2658 = const()[name = string("op_2658"), val = tensor<int32, [3]>([1, 390, 896])];
tensor<fp16, [1, 390, 14, 64]> var_2654_cast_fp16 = transpose(perm = var_2653, x = output_cast_fp16)[name = string("transpose_90")];
tensor<fp16, [1, 390, 896]> input_251_cast_fp16 = reshape(shape = var_2658, x = var_2654_cast_fp16)[name = string("input_251_cast_fp16")];
tensor<fp16, [896, 896]> tower_layers_17_self_attn_out_proj_weight_to_fp16 = const()[name = string("tower_layers_17_self_attn_out_proj_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(354882944)))];
tensor<fp16, [896]> tower_layers_17_self_attn_out_proj_bias_to_fp16 = const()[name = string("tower_layers_17_self_attn_out_proj_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(356488640)))];
tensor<fp16, [1, 390, 896]> linear_106_cast_fp16 = linear(bias = tower_layers_17_self_attn_out_proj_bias_to_fp16, weight = tower_layers_17_self_attn_out_proj_weight_to_fp16, x = input_251_cast_fp16)[name = string("linear_106_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_253_cast_fp16 = add(x = input_247_cast_fp16, y = linear_106_cast_fp16)[name = string("input_253_cast_fp16")];
tensor<int32, [1]> input_255_axes_0 = const()[name = string("input_255_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_layers_17_final_layer_norm_weight_to_fp16 = const()[name = string("tower_layers_17_final_layer_norm_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(356490496)))];
tensor<fp16, [896]> tower_layers_17_final_layer_norm_bias_to_fp16 = const()[name = string("tower_layers_17_final_layer_norm_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(356492352)))];
fp16 var_2666_to_fp16 = const()[name = string("op_2666_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_255_cast_fp16 = layer_norm(axes = input_255_axes_0, beta = tower_layers_17_final_layer_norm_bias_to_fp16, epsilon = var_2666_to_fp16, gamma = tower_layers_17_final_layer_norm_weight_to_fp16, x = input_253_cast_fp16)[name = string("input_255_cast_fp16")];
tensor<fp16, [3584, 896]> tower_layers_17_fc1_weight_to_fp16 = const()[name = string("tower_layers_17_fc1_weight_to_fp16"), val = tensor<fp16, [3584, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(356494208)))];
tensor<fp16, [3584]> tower_layers_17_fc1_bias_to_fp16 = const()[name = string("tower_layers_17_fc1_bias_to_fp16"), val = tensor<fp16, [3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(362916800)))];
tensor<fp16, [1, 390, 3584]> linear_107_cast_fp16 = linear(bias = tower_layers_17_fc1_bias_to_fp16, weight = tower_layers_17_fc1_weight_to_fp16, x = input_255_cast_fp16)[name = string("linear_107_cast_fp16")];
string input_259_mode_0 = const()[name = string("input_259_mode_0"), val = string("EXACT")];
tensor<fp16, [1, 390, 3584]> input_259_cast_fp16 = gelu(mode = input_259_mode_0, x = linear_107_cast_fp16)[name = string("input_259_cast_fp16")];
tensor<fp16, [896, 3584]> tower_layers_17_fc2_weight_to_fp16 = const()[name = string("tower_layers_17_fc2_weight_to_fp16"), val = tensor<fp16, [896, 3584]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(362924032)))];
tensor<fp16, [896]> tower_layers_17_fc2_bias_to_fp16 = const()[name = string("tower_layers_17_fc2_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(369346624)))];
tensor<fp16, [1, 390, 896]> linear_108_cast_fp16 = linear(bias = tower_layers_17_fc2_bias_to_fp16, weight = tower_layers_17_fc2_weight_to_fp16, x = input_259_cast_fp16)[name = string("linear_108_cast_fp16")];
tensor<fp16, [1, 390, 896]> input_261_cast_fp16 = add(x = input_253_cast_fp16, y = linear_108_cast_fp16)[name = string("input_261_cast_fp16")];
tensor<int32, [1]> input_263_axes_0 = const()[name = string("input_263_axes_0"), val = tensor<int32, [1]>([-1])];
tensor<fp16, [896]> tower_ln_post_weight_to_fp16 = const()[name = string("tower_ln_post_weight_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(369348480)))];
tensor<fp16, [896]> tower_ln_post_bias_to_fp16 = const()[name = string("tower_ln_post_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(369350336)))];
fp16 var_2683_to_fp16 = const()[name = string("op_2683_to_fp16"), val = fp16(0x1.5p-17)];
tensor<fp16, [1, 390, 896]> input_263_cast_fp16 = layer_norm(axes = input_263_axes_0, beta = tower_ln_post_bias_to_fp16, epsilon = var_2683_to_fp16, gamma = tower_ln_post_weight_to_fp16, x = input_261_cast_fp16)[name = string("input_263_cast_fp16")];
tensor<fp16, [896, 896]> projector_linear_1_weight_to_fp16 = const()[name = string("projector_linear_1_weight_to_fp16"), val = tensor<fp16, [896, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(369352192)))];
tensor<fp16, [896]> projector_linear_1_bias_to_fp16 = const()[name = string("projector_linear_1_bias_to_fp16"), val = tensor<fp16, [896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370957888)))];
tensor<fp16, [1, 390, 896]> linear_109_cast_fp16 = linear(bias = projector_linear_1_bias_to_fp16, weight = projector_linear_1_weight_to_fp16, x = input_263_cast_fp16)[name = string("linear_109_cast_fp16")];
string input_mode_0 = const()[name = string("input_mode_0"), val = string("EXACT")];
tensor<fp16, [1, 390, 896]> input_cast_fp16 = gelu(mode = input_mode_0, x = linear_109_cast_fp16)[name = string("input_cast_fp16")];
tensor<fp16, [1024, 896]> projector_linear_2_weight_to_fp16 = const()[name = string("projector_linear_2_weight_to_fp16"), val = tensor<fp16, [1024, 896]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(370959744)))];
tensor<fp16, [1024]> projector_linear_2_bias_to_fp16 = const()[name = string("projector_linear_2_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(372794816)))];
tensor<fp16, [1, 390, 1024]> linear_110_cast_fp16 = linear(bias = projector_linear_2_bias_to_fp16, weight = projector_linear_2_weight_to_fp16, x = input_cast_fp16)[name = string("linear_110_cast_fp16")];
string linear_110_cast_fp16_to_fp32_dtype_0 = const()[name = string("linear_110_cast_fp16_to_fp32_dtype_0"), val = string("fp32")];
tensor<fp32, [1, 390, 1024]> audio_embeddings = cast(dtype = linear_110_cast_fp16_to_fp32_dtype_0, x = linear_110_cast_fp16)[name = string("cast_76")];
} -> (audio_embeddings);
} |