program(1.0)
[buildInfo = dict<tensor<string, []>, tensor<string, []>>({{"coremlc-component-MIL", "3520.4.1"}, {"coremlc-version", "3520.5.1"}})]
{
    func main<ios17>(tensor<fp32, [1, 128, 1000]> audio_signal, tensor<int32, [1]> length) {
            tensor<int32, []> var_19 = const()[name = tensor<string, []>("op_19"), val = tensor<int32, []>(-1)];
            tensor<int32, [3]> x_1_perm_0 = const()[name = tensor<string, []>("x_1_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> audio_signal_to_fp16_dtype_0 = const()[name = tensor<string, []>("audio_signal_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
            tensor<int32, [1]> tensor_1_axes_0 = const()[name = tensor<string, []>("tensor_1_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [1, 128, 1000]> audio_signal_to_fp16 = cast(dtype = audio_signal_to_fp16_dtype_0, x = audio_signal)[name = tensor<string, []>("cast_9")];
            tensor<fp16, [1, 1000, 128]> x_1_cast_fp16 = transpose(perm = x_1_perm_0, x = audio_signal_to_fp16)[name = tensor<string, []>("transpose_223")];
            tensor<fp16, [1, 1, 1000, 128]> tensor_1_cast_fp16 = expand_dims(axes = tensor_1_axes_0, x = x_1_cast_fp16)[name = tensor<string, []>("tensor_1_cast_fp16")];
            tensor<int32, [1, 1000]> expand_dims_0 = const()[name = tensor<string, []>("expand_dims_0"), val = tensor<int32, [1, 1000]>([[0, 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]])];
            tensor<int32, [1]> var_83_axes_0 = const()[name = tensor<string, []>("op_83_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1, 1]> var_83 = expand_dims(axes = var_83_axes_0, x = length)[name = tensor<string, []>("op_83")];
            tensor<bool, [1, 1000]> time_mask_1 = less(x = expand_dims_0, y = var_83)[name = tensor<string, []>("time_mask_1")];
            tensor<int32, [1]> var_85_axes_0 = const()[name = tensor<string, []>("op_85_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, [1, 1000, 1]> var_85 = expand_dims(axes = var_85_axes_0, x = time_mask_1)[name = tensor<string, []>("op_85")];
            tensor<int32, [3]> var_87_reps_0 = const()[name = tensor<string, []>("op_87_reps_0"), val = tensor<int32, [3]>([1, 1, 128])];
            tensor<bool, [1, 1000, 128]> var_87 = tile(reps = var_87_reps_0, x = var_85)[name = tensor<string, []>("op_87")];
            tensor<int32, [1]> var_93_axes_0 = const()[name = tensor<string, []>("op_93_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<string, []> mask_1_to_fp16_dtype_0 = const()[name = tensor<string, []>("mask_1_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
            tensor<fp16, [1, 1000, 128]> var_87_to_fp16 = cast(dtype = mask_1_to_fp16_dtype_0, x = var_87)[name = tensor<string, []>("cast_8")];
            tensor<fp16, [1, 1, 1000, 128]> var_93_cast_fp16 = expand_dims(axes = var_93_axes_0, x = var_87_to_fp16)[name = tensor<string, []>("op_93_cast_fp16")];
            tensor<fp16, [1, 1, 1000, 128]> input_1_cast_fp16 = mul(x = tensor_1_cast_fp16, y = var_93_cast_fp16)[name = tensor<string, []>("input_1_cast_fp16")];
            tensor<string, []> tensor_3_pad_type_0 = const()[name = tensor<string, []>("tensor_3_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> tensor_3_pad_0 = const()[name = tensor<string, []>("tensor_3_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> tensor_3_strides_0 = const()[name = tensor<string, []>("tensor_3_strides_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, [2]> tensor_3_dilations_0 = const()[name = tensor<string, []>("tensor_3_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> tensor_3_groups_0 = const()[name = tensor<string, []>("tensor_3_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [256, 1, 3, 3]> encoder_pre_encode_conv_0_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_pre_encode_conv_0_weight_to_fp16_quantized"), quantized_data = tensor<int8, [256, 1, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64))), scale = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2752))), zero_point = tensor<int8, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2432)))];
            tensor<fp16, [256]> encoder_pre_encode_conv_0_bias_to_fp16 = const()[name = tensor<string, []>("encoder_pre_encode_conv_0_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3328)))];
            tensor<fp16, [1, 256, 500, 64]> tensor_3_cast_fp16 = conv(bias = encoder_pre_encode_conv_0_bias_to_fp16, dilations = tensor_3_dilations_0, groups = tensor_3_groups_0, pad = tensor_3_pad_0, pad_type = tensor_3_pad_type_0, strides = tensor_3_strides_0, weight = encoder_pre_encode_conv_0_weight_to_fp16_quantized, x = input_1_cast_fp16)[name = tensor<string, []>("tensor_3_cast_fp16")];
            tensor<string, []> current_lengths_1_to_fp16_dtype_0 = const()[name = tensor<string, []>("current_lengths_1_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
            tensor<fp16, []> var_104_promoted_to_fp16 = const()[name = tensor<string, []>("op_104_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
            tensor<fp16, [1]> length_to_fp16 = cast(dtype = current_lengths_1_to_fp16_dtype_0, x = length)[name = tensor<string, []>("cast_7")];
            tensor<fp16, [1]> var_105_cast_fp16 = add(x = length_to_fp16, y = var_104_promoted_to_fp16)[name = tensor<string, []>("op_105_cast_fp16")];
            tensor<fp16, []> var_106_promoted_to_fp16 = const()[name = tensor<string, []>("op_106_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
            tensor<fp16, [1]> var_107_cast_fp16 = add(x = var_105_cast_fp16, y = var_106_promoted_to_fp16)[name = tensor<string, []>("op_107_cast_fp16")];
            tensor<fp16, []> var_108_promoted_to_fp16 = const()[name = tensor<string, []>("op_108_promoted_to_fp16"), val = tensor<fp16, []>(0x1.8p+1)];
            tensor<fp16, [1]> var_109_cast_fp16 = sub(x = var_107_cast_fp16, y = var_108_promoted_to_fp16)[name = tensor<string, []>("op_109_cast_fp16")];
            tensor<fp16, []> var_17_promoted_to_fp16 = const()[name = tensor<string, []>("op_17_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1]> floor_div_0_cast_fp16 = floor_div(x = var_109_cast_fp16, y = var_17_promoted_to_fp16)[name = tensor<string, []>("floor_div_0_cast_fp16")];
            tensor<fp16, []> var_111_promoted_to_fp16 = const()[name = tensor<string, []>("op_111_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
            tensor<fp16, [1]> current_lengths_3_cast_fp16 = add(x = floor_div_0_cast_fp16, y = var_111_promoted_to_fp16)[name = tensor<string, []>("current_lengths_3_cast_fp16")];
            tensor<string, []> lengths_19_dtype_0 = const()[name = tensor<string, []>("lengths_19_dtype_0"), val = tensor<string, []>("int32")];
            tensor<int32, [1, 500]> expand_dims_1 = const()[name = tensor<string, []>("expand_dims_1"), val = tensor<int32, [1, 500]>([[0, 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]])];
            tensor<int32, [1]> var_120_axes_0 = const()[name = tensor<string, []>("op_120_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> current_lengths_3_cast_fp16_to_int32 = cast(dtype = lengths_19_dtype_0, x = current_lengths_3_cast_fp16)[name = tensor<string, []>("cast_6")];
            tensor<int32, [1, 1]> var_120 = expand_dims(axes = var_120_axes_0, x = current_lengths_3_cast_fp16_to_int32)[name = tensor<string, []>("op_120")];
            tensor<bool, [1, 500]> time_mask_3 = less(x = expand_dims_1, y = var_120)[name = tensor<string, []>("time_mask_3")];
            tensor<int32, [1]> var_122_axes_0 = const()[name = tensor<string, []>("op_122_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, [1, 500, 1]> var_122 = expand_dims(axes = var_122_axes_0, x = time_mask_3)[name = tensor<string, []>("op_122")];
            tensor<int32, [3]> var_124_reps_0 = const()[name = tensor<string, []>("op_124_reps_0"), val = tensor<int32, [3]>([1, 1, 64])];
            tensor<bool, [1, 500, 64]> var_124 = tile(reps = var_124_reps_0, x = var_122)[name = tensor<string, []>("op_124")];
            tensor<int32, [1]> var_130_axes_0 = const()[name = tensor<string, []>("op_130_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<string, []> mask_3_to_fp16_dtype_0 = const()[name = tensor<string, []>("mask_3_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
            tensor<fp16, [1, 500, 64]> var_124_to_fp16 = cast(dtype = mask_3_to_fp16_dtype_0, x = var_124)[name = tensor<string, []>("cast_5")];
            tensor<fp16, [1, 1, 500, 64]> var_130_cast_fp16 = expand_dims(axes = var_130_axes_0, x = var_124_to_fp16)[name = tensor<string, []>("op_130_cast_fp16")];
            tensor<int32, [4]> expanded_mask_3_reps_0 = const()[name = tensor<string, []>("expanded_mask_3_reps_0"), val = tensor<int32, [4]>([1, 256, 1, 1])];
            tensor<fp16, [1, 256, 500, 64]> expanded_mask_3_cast_fp16 = tile(reps = expanded_mask_3_reps_0, x = var_130_cast_fp16)[name = tensor<string, []>("expanded_mask_3_cast_fp16")];
            tensor<fp16, [1, 256, 500, 64]> input_3_cast_fp16 = mul(x = tensor_3_cast_fp16, y = expanded_mask_3_cast_fp16)[name = tensor<string, []>("input_3_cast_fp16")];
            tensor<fp16, [1, 256, 500, 64]> tensor_5_cast_fp16 = relu(x = input_3_cast_fp16)[name = tensor<string, []>("tensor_5_cast_fp16")];
            tensor<fp16, [1, 256, 500, 64]> input_5_cast_fp16 = mul(x = tensor_5_cast_fp16, y = expanded_mask_3_cast_fp16)[name = tensor<string, []>("input_5_cast_fp16")];
            tensor<string, []> tensor_7_pad_type_0 = const()[name = tensor<string, []>("tensor_7_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> tensor_7_pad_0 = const()[name = tensor<string, []>("tensor_7_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> tensor_7_strides_0 = const()[name = tensor<string, []>("tensor_7_strides_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, []> tensor_7_groups_0 = const()[name = tensor<string, []>("tensor_7_groups_0"), val = tensor<int32, []>(256)];
            tensor<int32, [2]> tensor_7_dilations_0 = const()[name = tensor<string, []>("tensor_7_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [256, 1, 3, 3]> encoder_pre_encode_conv_2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_pre_encode_conv_2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [256, 1, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3904))), scale = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6272))), zero_point = tensor<int8, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2432)))];
            tensor<fp16, [256]> encoder_pre_encode_conv_2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_pre_encode_conv_2_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6848)))];
            tensor<fp16, [1, 256, 250, 32]> tensor_7_cast_fp16 = conv(bias = encoder_pre_encode_conv_2_bias_to_fp16, dilations = tensor_7_dilations_0, groups = tensor_7_groups_0, pad = tensor_7_pad_0, pad_type = tensor_7_pad_type_0, strides = tensor_7_strides_0, weight = encoder_pre_encode_conv_2_weight_to_fp16_quantized, x = input_5_cast_fp16)[name = tensor<string, []>("tensor_7_cast_fp16")];
            tensor<fp16, []> var_150_promoted_to_fp16 = const()[name = tensor<string, []>("op_150_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
            tensor<fp16, [1]> var_151_cast_fp16 = add(x = current_lengths_3_cast_fp16, y = var_150_promoted_to_fp16)[name = tensor<string, []>("op_151_cast_fp16")];
            tensor<fp16, []> var_152_promoted_to_fp16 = const()[name = tensor<string, []>("op_152_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
            tensor<fp16, [1]> var_153_cast_fp16 = add(x = var_151_cast_fp16, y = var_152_promoted_to_fp16)[name = tensor<string, []>("op_153_cast_fp16")];
            tensor<fp16, []> var_154_promoted_to_fp16 = const()[name = tensor<string, []>("op_154_promoted_to_fp16"), val = tensor<fp16, []>(0x1.8p+1)];
            tensor<fp16, [1]> var_155_cast_fp16 = sub(x = var_153_cast_fp16, y = var_154_promoted_to_fp16)[name = tensor<string, []>("op_155_cast_fp16")];
            tensor<fp16, []> var_17_promoted_1_to_fp16 = const()[name = tensor<string, []>("op_17_promoted_1_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1]> floor_div_1_cast_fp16 = floor_div(x = var_155_cast_fp16, y = var_17_promoted_1_to_fp16)[name = tensor<string, []>("floor_div_1_cast_fp16")];
            tensor<fp16, []> var_157_promoted_to_fp16 = const()[name = tensor<string, []>("op_157_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
            tensor<fp16, [1]> current_lengths_5_cast_fp16 = add(x = floor_div_1_cast_fp16, y = var_157_promoted_to_fp16)[name = tensor<string, []>("current_lengths_5_cast_fp16")];
            tensor<string, []> lengths_21_dtype_0 = const()[name = tensor<string, []>("lengths_21_dtype_0"), val = tensor<string, []>("int32")];
            tensor<int32, [1, 250]> expand_dims_2 = const()[name = tensor<string, []>("expand_dims_2"), val = tensor<int32, [1, 250]>([[0, 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]])];
            tensor<int32, [1]> var_166_axes_0 = const()[name = tensor<string, []>("op_166_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> current_lengths_5_cast_fp16_to_int32 = cast(dtype = lengths_21_dtype_0, x = current_lengths_5_cast_fp16)[name = tensor<string, []>("cast_4")];
            tensor<int32, [1, 1]> var_166 = expand_dims(axes = var_166_axes_0, x = current_lengths_5_cast_fp16_to_int32)[name = tensor<string, []>("op_166")];
            tensor<bool, [1, 250]> time_mask_5 = less(x = expand_dims_2, y = var_166)[name = tensor<string, []>("time_mask_5")];
            tensor<int32, [1]> var_168_axes_0 = const()[name = tensor<string, []>("op_168_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, [1, 250, 1]> var_168 = expand_dims(axes = var_168_axes_0, x = time_mask_5)[name = tensor<string, []>("op_168")];
            tensor<int32, [3]> var_170_reps_0 = const()[name = tensor<string, []>("op_170_reps_0"), val = tensor<int32, [3]>([1, 1, 32])];
            tensor<bool, [1, 250, 32]> var_170 = tile(reps = var_170_reps_0, x = var_168)[name = tensor<string, []>("op_170")];
            tensor<int32, [1]> var_176_axes_0 = const()[name = tensor<string, []>("op_176_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<string, []> mask_5_to_fp16_dtype_0 = const()[name = tensor<string, []>("mask_5_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
            tensor<fp16, [1, 250, 32]> var_170_to_fp16 = cast(dtype = mask_5_to_fp16_dtype_0, x = var_170)[name = tensor<string, []>("cast_3")];
            tensor<fp16, [1, 1, 250, 32]> var_176_cast_fp16 = expand_dims(axes = var_176_axes_0, x = var_170_to_fp16)[name = tensor<string, []>("op_176_cast_fp16")];
            tensor<int32, [4]> expanded_mask_7_reps_0 = const()[name = tensor<string, []>("expanded_mask_7_reps_0"), val = tensor<int32, [4]>([1, 256, 1, 1])];
            tensor<fp16, [1, 256, 250, 32]> expanded_mask_7_cast_fp16 = tile(reps = expanded_mask_7_reps_0, x = var_176_cast_fp16)[name = tensor<string, []>("expanded_mask_7_cast_fp16")];
            tensor<fp16, [1, 256, 250, 32]> input_7_cast_fp16 = mul(x = tensor_7_cast_fp16, y = expanded_mask_7_cast_fp16)[name = tensor<string, []>("input_7_cast_fp16")];
            tensor<string, []> tensor_9_pad_type_0 = const()[name = tensor<string, []>("tensor_9_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> tensor_9_strides_0 = const()[name = tensor<string, []>("tensor_9_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> tensor_9_pad_0 = const()[name = tensor<string, []>("tensor_9_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> tensor_9_dilations_0 = const()[name = tensor<string, []>("tensor_9_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> tensor_9_groups_0 = const()[name = tensor<string, []>("tensor_9_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [256, 256, 1, 1]> encoder_pre_encode_conv_3_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_pre_encode_conv_3_weight_to_fp16_quantized"), quantized_data = tensor<int8, [256, 256, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7424))), scale = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73024))), zero_point = tensor<int8, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2432)))];
            tensor<fp16, [256]> encoder_pre_encode_conv_3_bias_to_fp16 = const()[name = tensor<string, []>("encoder_pre_encode_conv_3_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73600)))];
            tensor<fp16, [1, 256, 250, 32]> tensor_9_cast_fp16 = conv(bias = encoder_pre_encode_conv_3_bias_to_fp16, dilations = tensor_9_dilations_0, groups = tensor_9_groups_0, pad = tensor_9_pad_0, pad_type = tensor_9_pad_type_0, strides = tensor_9_strides_0, weight = encoder_pre_encode_conv_3_weight_to_fp16_quantized, x = input_7_cast_fp16)[name = tensor<string, []>("tensor_9_cast_fp16")];
            tensor<fp16, [1, 256, 250, 32]> input_9_cast_fp16 = mul(x = tensor_9_cast_fp16, y = expanded_mask_7_cast_fp16)[name = tensor<string, []>("input_9_cast_fp16")];
            tensor<fp16, [1, 256, 250, 32]> tensor_11_cast_fp16 = relu(x = input_9_cast_fp16)[name = tensor<string, []>("tensor_11_cast_fp16")];
            tensor<fp16, [1, 256, 250, 32]> input_11_cast_fp16 = mul(x = tensor_11_cast_fp16, y = expanded_mask_7_cast_fp16)[name = tensor<string, []>("input_11_cast_fp16")];
            tensor<string, []> tensor_13_pad_type_0 = const()[name = tensor<string, []>("tensor_13_pad_type_0"), val = tensor<string, []>("custom")];
            tensor<int32, [4]> tensor_13_pad_0 = const()[name = tensor<string, []>("tensor_13_pad_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<int32, [2]> tensor_13_strides_0 = const()[name = tensor<string, []>("tensor_13_strides_0"), val = tensor<int32, [2]>([2, 2])];
            tensor<int32, []> tensor_13_groups_0 = const()[name = tensor<string, []>("tensor_13_groups_0"), val = tensor<int32, []>(256)];
            tensor<int32, [2]> tensor_13_dilations_0 = const()[name = tensor<string, []>("tensor_13_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<fp16, [256, 1, 3, 3]> encoder_pre_encode_conv_5_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_pre_encode_conv_5_weight_to_fp16_quantized"), quantized_data = tensor<int8, [256, 1, 3, 3]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74176))), scale = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76544))), zero_point = tensor<int8, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2432)))];
            tensor<fp16, [256]> encoder_pre_encode_conv_5_bias_to_fp16 = const()[name = tensor<string, []>("encoder_pre_encode_conv_5_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77120)))];
            tensor<fp16, [1, 256, 125, 16]> tensor_13_cast_fp16 = conv(bias = encoder_pre_encode_conv_5_bias_to_fp16, dilations = tensor_13_dilations_0, groups = tensor_13_groups_0, pad = tensor_13_pad_0, pad_type = tensor_13_pad_type_0, strides = tensor_13_strides_0, weight = encoder_pre_encode_conv_5_weight_to_fp16_quantized, x = input_11_cast_fp16)[name = tensor<string, []>("tensor_13_cast_fp16")];
            tensor<fp16, []> var_211_promoted_to_fp16 = const()[name = tensor<string, []>("op_211_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
            tensor<fp16, [1]> var_212_cast_fp16 = add(x = current_lengths_5_cast_fp16, y = var_211_promoted_to_fp16)[name = tensor<string, []>("op_212_cast_fp16")];
            tensor<fp16, []> var_213_promoted_to_fp16 = const()[name = tensor<string, []>("op_213_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
            tensor<fp16, [1]> var_214_cast_fp16 = add(x = var_212_cast_fp16, y = var_213_promoted_to_fp16)[name = tensor<string, []>("op_214_cast_fp16")];
            tensor<fp16, []> var_215_promoted_to_fp16 = const()[name = tensor<string, []>("op_215_promoted_to_fp16"), val = tensor<fp16, []>(0x1.8p+1)];
            tensor<fp16, [1]> var_216_cast_fp16 = sub(x = var_214_cast_fp16, y = var_215_promoted_to_fp16)[name = tensor<string, []>("op_216_cast_fp16")];
            tensor<fp16, []> var_17_promoted_2_to_fp16 = const()[name = tensor<string, []>("op_17_promoted_2_to_fp16"), val = tensor<fp16, []>(0x1p+1)];
            tensor<fp16, [1]> floor_div_2_cast_fp16 = floor_div(x = var_216_cast_fp16, y = var_17_promoted_2_to_fp16)[name = tensor<string, []>("floor_div_2_cast_fp16")];
            tensor<fp16, []> var_218_promoted_to_fp16 = const()[name = tensor<string, []>("op_218_promoted_to_fp16"), val = tensor<fp16, []>(0x1p+0)];
            tensor<fp16, [1]> current_lengths_cast_fp16 = add(x = floor_div_2_cast_fp16, y = var_218_promoted_to_fp16)[name = tensor<string, []>("current_lengths_cast_fp16")];
            tensor<string, []> lengths_dtype_0 = const()[name = tensor<string, []>("lengths_dtype_0"), val = tensor<string, []>("int32")];
            tensor<int32, [1, 125]> expand_dims_3 = const()[name = tensor<string, []>("expand_dims_3"), val = tensor<int32, [1, 125]>([[0, 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]])];
            tensor<int32, [1]> var_227_axes_0 = const()[name = tensor<string, []>("op_227_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [1]> current_lengths_cast_fp16_to_int32 = cast(dtype = lengths_dtype_0, x = current_lengths_cast_fp16)[name = tensor<string, []>("cast_2")];
            tensor<int32, [1, 1]> var_227 = expand_dims(axes = var_227_axes_0, x = current_lengths_cast_fp16_to_int32)[name = tensor<string, []>("op_227")];
            tensor<bool, [1, 125]> time_mask = less(x = expand_dims_3, y = var_227)[name = tensor<string, []>("time_mask")];
            tensor<int32, [1]> var_229_axes_0 = const()[name = tensor<string, []>("op_229_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<bool, [1, 125, 1]> var_229 = expand_dims(axes = var_229_axes_0, x = time_mask)[name = tensor<string, []>("op_229")];
            tensor<int32, [3]> var_231_reps_0 = const()[name = tensor<string, []>("op_231_reps_0"), val = tensor<int32, [3]>([1, 1, 16])];
            tensor<bool, [1, 125, 16]> var_231 = tile(reps = var_231_reps_0, x = var_229)[name = tensor<string, []>("op_231")];
            tensor<int32, [1]> var_237_axes_0 = const()[name = tensor<string, []>("op_237_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<string, []> mask_7_to_fp16_dtype_0 = const()[name = tensor<string, []>("mask_7_to_fp16_dtype_0"), val = tensor<string, []>("fp16")];
            tensor<fp16, [1, 125, 16]> var_231_to_fp16 = cast(dtype = mask_7_to_fp16_dtype_0, x = var_231)[name = tensor<string, []>("cast_1")];
            tensor<fp16, [1, 1, 125, 16]> var_237_cast_fp16 = expand_dims(axes = var_237_axes_0, x = var_231_to_fp16)[name = tensor<string, []>("op_237_cast_fp16")];
            tensor<int32, [4]> expanded_mask_13_reps_0 = const()[name = tensor<string, []>("expanded_mask_13_reps_0"), val = tensor<int32, [4]>([1, 256, 1, 1])];
            tensor<fp16, [1, 256, 125, 16]> expanded_mask_13_cast_fp16 = tile(reps = expanded_mask_13_reps_0, x = var_237_cast_fp16)[name = tensor<string, []>("expanded_mask_13_cast_fp16")];
            tensor<fp16, [1, 256, 125, 16]> input_13_cast_fp16 = mul(x = tensor_13_cast_fp16, y = expanded_mask_13_cast_fp16)[name = tensor<string, []>("input_13_cast_fp16")];
            tensor<string, []> tensor_15_pad_type_0 = const()[name = tensor<string, []>("tensor_15_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [2]> tensor_15_strides_0 = const()[name = tensor<string, []>("tensor_15_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> tensor_15_pad_0 = const()[name = tensor<string, []>("tensor_15_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> tensor_15_dilations_0 = const()[name = tensor<string, []>("tensor_15_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, []> tensor_15_groups_0 = const()[name = tensor<string, []>("tensor_15_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [256, 256, 1, 1]> encoder_pre_encode_conv_6_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_pre_encode_conv_6_weight_to_fp16_quantized"), quantized_data = tensor<int8, [256, 256, 1, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77696))), scale = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(143296))), zero_point = tensor<int8, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2432)))];
            tensor<fp16, [256]> encoder_pre_encode_conv_6_bias_to_fp16 = const()[name = tensor<string, []>("encoder_pre_encode_conv_6_bias_to_fp16"), val = tensor<fp16, [256]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(143872)))];
            tensor<fp16, [1, 256, 125, 16]> tensor_15_cast_fp16 = conv(bias = encoder_pre_encode_conv_6_bias_to_fp16, dilations = tensor_15_dilations_0, groups = tensor_15_groups_0, pad = tensor_15_pad_0, pad_type = tensor_15_pad_type_0, strides = tensor_15_strides_0, weight = encoder_pre_encode_conv_6_weight_to_fp16_quantized, x = input_13_cast_fp16)[name = tensor<string, []>("tensor_15_cast_fp16")];
            tensor<fp16, [1, 256, 125, 16]> input_15_cast_fp16 = mul(x = tensor_15_cast_fp16, y = expanded_mask_13_cast_fp16)[name = tensor<string, []>("input_15_cast_fp16")];
            tensor<fp16, [1, 256, 125, 16]> tensor_cast_fp16 = relu(x = input_15_cast_fp16)[name = tensor<string, []>("tensor_cast_fp16")];
            tensor<fp16, [1, 256, 125, 16]> x_3_cast_fp16 = mul(x = tensor_cast_fp16, y = expanded_mask_13_cast_fp16)[name = tensor<string, []>("x_3_cast_fp16")];
            tensor<int32, [4]> var_271_perm_0 = const()[name = tensor<string, []>("op_271_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_272 = const()[name = tensor<string, []>("op_272"), val = tensor<int32, [3]>([1, 125, -1])];
            tensor<fp16, [1, 125, 256, 16]> var_271_cast_fp16 = transpose(perm = var_271_perm_0, x = x_3_cast_fp16)[name = tensor<string, []>("transpose_222")];
            tensor<fp16, [1, 125, 4096]> input_17_cast_fp16 = reshape(shape = var_272, x = var_271_cast_fp16)[name = tensor<string, []>("input_17_cast_fp16")];
            tensor<fp16, [512, 4096]> encoder_pre_encode_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_pre_encode_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 4096]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(144448))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2242240))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_pre_encode_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_pre_encode_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2243328)))];
            tensor<fp16, [1, 125, 512]> linear_0_cast_fp16 = linear(bias = encoder_pre_encode_out_bias_to_fp16, weight = encoder_pre_encode_out_weight_to_fp16_quantized, x = input_17_cast_fp16)[name = tensor<string, []>("linear_0_cast_fp16")];
            tensor<int32, [1]> var_310_axes_0 = const()[name = tensor<string, []>("op_310_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<int32, [1, 1]> var_310 = expand_dims(axes = var_310_axes_0, x = current_lengths_cast_fp16_to_int32)[name = tensor<string, []>("op_310")];
            tensor<bool, [1, 125]> pad_mask_1 = less(x = expand_dims_3, y = var_310)[name = tensor<string, []>("pad_mask_1")];
            tensor<int32, [1]> var_312_axes_0 = const()[name = tensor<string, []>("op_312_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<bool, [1, 1, 125]> var_312 = expand_dims(axes = var_312_axes_0, x = pad_mask_1)[name = tensor<string, []>("op_312")];
            tensor<int32, [3]> var_313 = const()[name = tensor<string, []>("op_313"), val = tensor<int32, [3]>([1, 125, 1])];
            tensor<bool, [1, 125, 125]> pad_mask_for_att_mask_1 = tile(reps = var_313, x = var_312)[name = tensor<string, []>("pad_mask_for_att_mask_1")];
            tensor<int32, [3]> var_315_perm_0 = const()[name = tensor<string, []>("op_315_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<bool, [1, 125, 125]> var_315 = transpose(perm = var_315_perm_0, x = pad_mask_for_att_mask_1)[name = tensor<string, []>("transpose_221")];
            tensor<bool, [1, 125, 125]> pad_mask_for_att_mask = logical_and(x = pad_mask_for_att_mask_1, y = var_315)[name = tensor<string, []>("pad_mask_for_att_mask")];
            tensor<bool, [1, 125, 125]> const_63 = const()[name = tensor<string, []>("const_63"), val = tensor<bool, [1, 125, 125]>([[[true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true], [true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true, true]]])];
            tensor<bool, [1, 125, 125]> att_mask = logical_and(x = pad_mask_for_att_mask, y = const_63)[name = tensor<string, []>("att_mask")];
            tensor<bool, [1, 125, 125]> mask_9 = logical_not(x = att_mask)[name = tensor<string, []>("mask_9")];
            tensor<bool, [1, 125]> pad_mask = logical_not(x = pad_mask_1)[name = tensor<string, []>("pad_mask")];
            tensor<int32, [1]> input_21_axes_0 = const()[name = tensor<string, []>("input_21_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_0_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2244416)))];
            tensor<fp16, [512]> encoder_layers_0_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2245504)))];
            tensor<fp16, []> var_5_to_fp16 = const()[name = tensor<string, []>("op_5_to_fp16"), val = tensor<fp16, []>(0x1.5p-17)];
            tensor<fp16, [1, 125, 512]> input_21_cast_fp16 = layer_norm(axes = input_21_axes_0, beta = encoder_layers_0_norm_feed_forward1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_0_norm_feed_forward1_weight_to_fp16, x = linear_0_cast_fp16)[name = tensor<string, []>("input_21_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_0_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_0_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2246592))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3297344))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_0_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3301504)))];
            tensor<fp16, [1, 125, 2048]> linear_1_cast_fp16 = linear(bias = encoder_layers_0_feed_forward1_linear1_bias_to_fp16, weight = encoder_layers_0_feed_forward1_linear1_weight_to_fp16_quantized, x = input_21_cast_fp16)[name = tensor<string, []>("linear_1_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_25_cast_fp16 = silu(x = linear_1_cast_fp16)[name = tensor<string, []>("input_25_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_0_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_0_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3305664))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4354304))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_0_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4355392)))];
            tensor<fp16, [1, 125, 512]> linear_2_cast_fp16 = linear(bias = encoder_layers_0_feed_forward1_linear2_bias_to_fp16, weight = encoder_layers_0_feed_forward1_linear2_weight_to_fp16_quantized, x = input_25_cast_fp16)[name = tensor<string, []>("linear_2_cast_fp16")];
            tensor<fp16, []> var_348_to_fp16 = const()[name = tensor<string, []>("op_348_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_349_cast_fp16 = mul(x = linear_2_cast_fp16, y = var_348_to_fp16)[name = tensor<string, []>("op_349_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_31_cast_fp16 = add(x = linear_0_cast_fp16, y = var_349_cast_fp16)[name = tensor<string, []>("input_31_cast_fp16")];
            tensor<int32, [1]> query_1_axes_0 = const()[name = tensor<string, []>("query_1_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_0_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_norm_self_att_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4356480)))];
            tensor<fp16, [512]> encoder_layers_0_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_norm_self_att_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4357568)))];
            tensor<fp16, [1, 125, 512]> query_1_cast_fp16 = layer_norm(axes = query_1_axes_0, beta = encoder_layers_0_norm_self_att_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_0_norm_self_att_weight_to_fp16, x = input_31_cast_fp16)[name = tensor<string, []>("query_1_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_0_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_0_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4358656))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4620864))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_0_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4621952)))];
            tensor<fp16, [1, 125, 512]> linear_3_cast_fp16 = linear(bias = encoder_layers_0_self_attn_linear_q_bias_to_fp16, weight = encoder_layers_0_self_attn_linear_q_weight_to_fp16_quantized, x = query_1_cast_fp16)[name = tensor<string, []>("linear_3_cast_fp16")];
            tensor<int32, [4]> var_366 = const()[name = tensor<string, []>("op_366"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> q_1_cast_fp16 = reshape(shape = var_366, x = linear_3_cast_fp16)[name = tensor<string, []>("q_1_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_0_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_0_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4623040))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4885248))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_0_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4886336)))];
            tensor<fp16, [1, 125, 512]> linear_4_cast_fp16 = linear(bias = encoder_layers_0_self_attn_linear_k_bias_to_fp16, weight = encoder_layers_0_self_attn_linear_k_weight_to_fp16_quantized, x = query_1_cast_fp16)[name = tensor<string, []>("linear_4_cast_fp16")];
            tensor<int32, [4]> var_371 = const()[name = tensor<string, []>("op_371"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> k_1_cast_fp16 = reshape(shape = var_371, x = linear_4_cast_fp16)[name = tensor<string, []>("k_1_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_0_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_0_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(4887424))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5149632))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_0_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5150720)))];
            tensor<fp16, [1, 125, 512]> linear_5_cast_fp16 = linear(bias = encoder_layers_0_self_attn_linear_v_bias_to_fp16, weight = encoder_layers_0_self_attn_linear_v_weight_to_fp16_quantized, x = query_1_cast_fp16)[name = tensor<string, []>("linear_5_cast_fp16")];
            tensor<int32, [4]> var_376 = const()[name = tensor<string, []>("op_376"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> v_1_cast_fp16 = reshape(shape = var_376, x = linear_5_cast_fp16)[name = tensor<string, []>("v_1_cast_fp16")];
            tensor<int32, [4]> value_1_perm_0 = const()[name = tensor<string, []>("value_1_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 64]> encoder_layers_0_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5151808)))];
            tensor<fp16, [1, 125, 8, 64]> var_388_cast_fp16 = add(x = q_1_cast_fp16, y = encoder_layers_0_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_388_cast_fp16")];
            tensor<fp16, [8, 64]> encoder_layers_0_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5152896)))];
            tensor<fp16, [1, 125, 8, 64]> var_390_cast_fp16 = add(x = q_1_cast_fp16, y = encoder_layers_0_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_390_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_1_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_1_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<bool, []> x_7_transpose_x_0 = const()[name = tensor<string, []>("x_7_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_7_transpose_y_0 = const()[name = tensor<string, []>("x_7_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 64, 249]> op_392_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(3), name = tensor<string, []>("op_392_to_fp16_quantized"), quantized_data = tensor<int8, [1, 8, 64, 249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5153984))), scale = tensor<fp16, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5281856))), zero_point = tensor<int8, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5281536)))];
            tensor<fp16, [1, 8, 125, 64]> q_with_bias_v_1_cast_fp16 = transpose(perm = q_with_bias_v_1_perm_0, x = var_390_cast_fp16)[name = tensor<string, []>("transpose_220")];
            tensor<fp16, [1, 8, 125, 249]> x_7_cast_fp16 = matmul(transpose_x = x_7_transpose_x_0, transpose_y = x_7_transpose_y_0, x = q_with_bias_v_1_cast_fp16, y = op_392_to_fp16_quantized)[name = tensor<string, []>("x_7_cast_fp16")];
            tensor<int32, [8]> x_9_pad_0 = const()[name = tensor<string, []>("x_9_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            tensor<string, []> x_9_mode_0 = const()[name = tensor<string, []>("x_9_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_70_to_fp16 = const()[name = tensor<string, []>("const_70_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 8, 125, 250]> x_9_cast_fp16 = pad(constant_val = const_70_to_fp16, mode = x_9_mode_0, pad = x_9_pad_0, x = x_7_cast_fp16)[name = tensor<string, []>("x_9_cast_fp16")];
            tensor<int32, [4]> var_400 = const()[name = tensor<string, []>("op_400"), val = tensor<int32, [4]>([1, 8, -1, 125])];
            tensor<fp16, [1, 8, 250, 125]> x_11_cast_fp16 = reshape(shape = var_400, x = x_9_cast_fp16)[name = tensor<string, []>("x_11_cast_fp16")];
            tensor<int32, [4]> var_404_begin_0 = const()[name = tensor<string, []>("op_404_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_404_end_0 = const()[name = tensor<string, []>("op_404_end_0"), val = tensor<int32, [4]>([1, 8, 250, 125])];
            tensor<bool, [4]> var_404_end_mask_0 = const()[name = tensor<string, []>("op_404_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 249, 125]> var_404_cast_fp16 = slice_by_index(begin = var_404_begin_0, end = var_404_end_0, end_mask = var_404_end_mask_0, x = x_11_cast_fp16)[name = tensor<string, []>("op_404_cast_fp16")];
            tensor<int32, [4]> var_405 = const()[name = tensor<string, []>("op_405"), val = tensor<int32, [4]>([1, 8, 125, 249])];
            tensor<fp16, [1, 8, 125, 249]> matrix_bd_1_cast_fp16 = reshape(shape = var_405, x = var_404_cast_fp16)[name = tensor<string, []>("matrix_bd_1_cast_fp16")];
            tensor<bool, []> matrix_ac_1_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_1_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matrix_ac_1_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_1_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_68_perm_0 = const()[name = tensor<string, []>("transpose_68_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_69_perm_0 = const()[name = tensor<string, []>("transpose_69_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 64, 125]> transpose_69 = transpose(perm = transpose_69_perm_0, x = k_1_cast_fp16)[name = tensor<string, []>("transpose_218")];
            tensor<fp16, [1, 8, 125, 64]> transpose_68 = transpose(perm = transpose_68_perm_0, x = var_388_cast_fp16)[name = tensor<string, []>("transpose_219")];
            tensor<fp16, [1, 8, 125, 125]> matrix_ac_1_cast_fp16 = matmul(transpose_x = matrix_ac_1_transpose_x_0, transpose_y = matrix_ac_1_transpose_y_0, x = transpose_68, y = transpose_69)[name = tensor<string, []>("matrix_ac_1_cast_fp16")];
            tensor<int32, [4]> matrix_bd_3_begin_0 = const()[name = tensor<string, []>("matrix_bd_3_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_3_end_0 = const()[name = tensor<string, []>("matrix_bd_3_end_0"), val = tensor<int32, [4]>([1, 8, 125, 125])];
            tensor<bool, [4]> matrix_bd_3_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_3_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 125, 125]> matrix_bd_3_cast_fp16 = slice_by_index(begin = matrix_bd_3_begin_0, end = matrix_bd_3_end_0, end_mask = matrix_bd_3_end_mask_0, x = matrix_bd_1_cast_fp16)[name = tensor<string, []>("matrix_bd_3_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_414_cast_fp16 = add(x = matrix_ac_1_cast_fp16, y = matrix_bd_3_cast_fp16)[name = tensor<string, []>("op_414_cast_fp16")];
            tensor<fp16, []> _inversed_scores_1_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_1_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 8, 125, 125]> _inversed_scores_1_cast_fp16 = mul(x = var_414_cast_fp16, y = _inversed_scores_1_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_1_cast_fp16")];
            tensor<int32, [1]> mask_11_axes_0 = const()[name = tensor<string, []>("mask_11_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<bool, [1, 1, 125, 125]> mask_11 = expand_dims(axes = mask_11_axes_0, x = mask_9)[name = tensor<string, []>("mask_11")];
            tensor<fp16, []> var_8_to_fp16 = const()[name = tensor<string, []>("op_8_to_fp16"), val = tensor<fp16, []>(-0x1.388p+13)];
            tensor<fp16, [1, 8, 125, 125]> scores_3_cast_fp16 = select(a = var_8_to_fp16, b = _inversed_scores_1_cast_fp16, cond = mask_11)[name = tensor<string, []>("scores_3_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_420_cast_fp16 = softmax(axis = var_19, x = scores_3_cast_fp16)[name = tensor<string, []>("op_420_cast_fp16")];
            tensor<fp16, []> var_7_to_fp16 = const()[name = tensor<string, []>("op_7_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 8, 125, 125]> input_33_cast_fp16 = select(a = var_7_to_fp16, b = var_420_cast_fp16, cond = mask_11)[name = tensor<string, []>("input_33_cast_fp16")];
            tensor<bool, []> x_13_transpose_x_0 = const()[name = tensor<string, []>("x_13_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_13_transpose_y_0 = const()[name = tensor<string, []>("x_13_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 125, 64]> value_1_cast_fp16 = transpose(perm = value_1_perm_0, x = v_1_cast_fp16)[name = tensor<string, []>("transpose_217")];
            tensor<fp16, [1, 8, 125, 64]> x_13_cast_fp16 = matmul(transpose_x = x_13_transpose_x_0, transpose_y = x_13_transpose_y_0, x = input_33_cast_fp16, y = value_1_cast_fp16)[name = tensor<string, []>("x_13_cast_fp16")];
            tensor<int32, [4]> var_424_perm_0 = const()[name = tensor<string, []>("op_424_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_425 = const()[name = tensor<string, []>("op_425"), val = tensor<int32, [3]>([1, -1, 512])];
            tensor<fp16, [1, 125, 8, 64]> var_424_cast_fp16 = transpose(perm = var_424_perm_0, x = x_13_cast_fp16)[name = tensor<string, []>("transpose_216")];
            tensor<fp16, [1, 125, 512]> input_35_cast_fp16 = reshape(shape = var_425, x = var_424_cast_fp16)[name = tensor<string, []>("input_35_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_0_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_0_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5282432))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5544640))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_0_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5545728)))];
            tensor<fp16, [1, 125, 512]> linear_7_cast_fp16 = linear(bias = encoder_layers_0_self_attn_linear_out_bias_to_fp16, weight = encoder_layers_0_self_attn_linear_out_weight_to_fp16_quantized, x = input_35_cast_fp16)[name = tensor<string, []>("linear_7_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_39_cast_fp16 = add(x = input_31_cast_fp16, y = linear_7_cast_fp16)[name = tensor<string, []>("input_39_cast_fp16")];
            tensor<int32, [1]> x_17_axes_0 = const()[name = tensor<string, []>("x_17_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_0_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_norm_conv_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5546816)))];
            tensor<fp16, [512]> encoder_layers_0_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_norm_conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5547904)))];
            tensor<fp16, [1, 125, 512]> x_17_cast_fp16 = layer_norm(axes = x_17_axes_0, beta = encoder_layers_0_norm_conv_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_0_norm_conv_weight_to_fp16, x = input_39_cast_fp16)[name = tensor<string, []>("x_17_cast_fp16")];
            tensor<int32, [3]> input_41_perm_0 = const()[name = tensor<string, []>("input_41_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> input_43_pad_type_0 = const()[name = tensor<string, []>("input_43_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> input_43_strides_0 = const()[name = tensor<string, []>("input_43_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_43_pad_0 = const()[name = tensor<string, []>("input_43_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_43_dilations_0 = const()[name = tensor<string, []>("input_43_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> input_43_groups_0 = const()[name = tensor<string, []>("input_43_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 512, 1]> encoder_layers_0_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_0_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1024, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5548992))), scale = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6074432))), zero_point = tensor<int8, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6073344)))];
            tensor<fp16, [1024]> encoder_layers_0_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6076544)))];
            tensor<fp16, [1, 512, 125]> input_41_cast_fp16 = transpose(perm = input_41_perm_0, x = x_17_cast_fp16)[name = tensor<string, []>("transpose_215")];
            tensor<fp16, [1, 1024, 125]> input_43_cast_fp16 = conv(bias = encoder_layers_0_conv_pointwise_conv1_bias_to_fp16, dilations = input_43_dilations_0, groups = input_43_groups_0, pad = input_43_pad_0, pad_type = input_43_pad_type_0, strides = input_43_strides_0, weight = encoder_layers_0_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_41_cast_fp16)[name = tensor<string, []>("input_43_cast_fp16")];
            tensor<int32, []> x_19_split_num_splits_0 = const()[name = tensor<string, []>("x_19_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> x_19_split_axis_0 = const()[name = tensor<string, []>("x_19_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 512, 125]> x_19_split_cast_fp16_0, tensor<fp16, [1, 512, 125]> x_19_split_cast_fp16_1 = split(axis = x_19_split_axis_0, num_splits = x_19_split_num_splits_0, x = input_43_cast_fp16)[name = tensor<string, []>("x_19_split_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_19_split_1_sigmoid_cast_fp16 = sigmoid(x = x_19_split_cast_fp16_1)[name = tensor<string, []>("x_19_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_19_cast_fp16 = mul(x = x_19_split_cast_fp16_0, y = x_19_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_19_cast_fp16")];
            tensor<int32, [1]> var_449_axes_0 = const()[name = tensor<string, []>("op_449_axes_0"), val = tensor<int32, [1]>([1])];
            tensor<bool, [1, 1, 125]> var_449 = expand_dims(axes = var_449_axes_0, x = pad_mask)[name = tensor<string, []>("op_449")];
            tensor<fp16, [1, 512, 125]> input_45_cast_fp16 = select(a = var_7_to_fp16, b = x_19_cast_fp16, cond = var_449)[name = tensor<string, []>("input_45_cast_fp16")];
            tensor<int32, [6]> input_47_pad_0 = const()[name = tensor<string, []>("input_47_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            tensor<string, []> input_47_mode_0 = const()[name = tensor<string, []>("input_47_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_73_to_fp16 = const()[name = tensor<string, []>("const_73_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 512, 133]> input_47_cast_fp16 = pad(constant_val = const_73_to_fp16, mode = input_47_mode_0, pad = input_47_pad_0, x = input_45_cast_fp16)[name = tensor<string, []>("input_47_cast_fp16")];
            tensor<string, []> input_49_pad_type_0 = const()[name = tensor<string, []>("input_49_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, []> input_49_groups_0 = const()[name = tensor<string, []>("input_49_groups_0"), val = tensor<int32, []>(512)];
            tensor<int32, [1]> input_49_strides_0 = const()[name = tensor<string, []>("input_49_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_49_pad_0 = const()[name = tensor<string, []>("input_49_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_49_dilations_0 = const()[name = tensor<string, []>("input_49_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 9]> const_237_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("const_237_to_fp16_quantized"), quantized_data = tensor<int8, [512, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6078656))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6083328))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> const_238_to_fp16 = const()[name = tensor<string, []>("const_238_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6084416)))];
            tensor<fp16, [1, 512, 125]> input_51_cast_fp16 = conv(bias = const_238_to_fp16, dilations = input_49_dilations_0, groups = input_49_groups_0, pad = input_49_pad_0, pad_type = input_49_pad_type_0, strides = input_49_strides_0, weight = const_237_to_fp16_quantized, x = input_47_cast_fp16)[name = tensor<string, []>("input_51_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_53_cast_fp16 = silu(x = input_51_cast_fp16)[name = tensor<string, []>("input_53_cast_fp16")];
            tensor<string, []> x_21_pad_type_0 = const()[name = tensor<string, []>("x_21_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> x_21_strides_0 = const()[name = tensor<string, []>("x_21_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_21_pad_0 = const()[name = tensor<string, []>("x_21_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_21_dilations_0 = const()[name = tensor<string, []>("x_21_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> x_21_groups_0 = const()[name = tensor<string, []>("x_21_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [512, 512, 1]> encoder_layers_0_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_0_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6085504))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6347712))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_0_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6348800)))];
            tensor<fp16, [1, 512, 125]> x_21_cast_fp16 = conv(bias = encoder_layers_0_conv_pointwise_conv2_bias_to_fp16, dilations = x_21_dilations_0, groups = x_21_groups_0, pad = x_21_pad_0, pad_type = x_21_pad_type_0, strides = x_21_strides_0, weight = encoder_layers_0_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_53_cast_fp16)[name = tensor<string, []>("x_21_cast_fp16")];
            tensor<int32, [3]> input_55_perm_0 = const()[name = tensor<string, []>("input_55_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 125, 512]> input_55_cast_fp16 = transpose(perm = input_55_perm_0, x = x_21_cast_fp16)[name = tensor<string, []>("transpose_214")];
            tensor<fp16, [1, 125, 512]> input_57_cast_fp16 = add(x = input_39_cast_fp16, y = input_55_cast_fp16)[name = tensor<string, []>("input_57_cast_fp16")];
            tensor<int32, [1]> input_59_axes_0 = const()[name = tensor<string, []>("input_59_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_0_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6349888)))];
            tensor<fp16, [512]> encoder_layers_0_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6350976)))];
            tensor<fp16, [1, 125, 512]> input_59_cast_fp16 = layer_norm(axes = input_59_axes_0, beta = encoder_layers_0_norm_feed_forward2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_0_norm_feed_forward2_weight_to_fp16, x = input_57_cast_fp16)[name = tensor<string, []>("input_59_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_0_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_0_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6352064))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7400704))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_0_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7404864)))];
            tensor<fp16, [1, 125, 2048]> linear_8_cast_fp16 = linear(bias = encoder_layers_0_feed_forward2_linear1_bias_to_fp16, weight = encoder_layers_0_feed_forward2_linear1_weight_to_fp16_quantized, x = input_59_cast_fp16)[name = tensor<string, []>("linear_8_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_63_cast_fp16 = silu(x = linear_8_cast_fp16)[name = tensor<string, []>("input_63_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_0_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_0_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(7409024))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8457664))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_0_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8458752)))];
            tensor<fp16, [1, 125, 512]> linear_9_cast_fp16 = linear(bias = encoder_layers_0_feed_forward2_linear2_bias_to_fp16, weight = encoder_layers_0_feed_forward2_linear2_weight_to_fp16_quantized, x = input_63_cast_fp16)[name = tensor<string, []>("linear_9_cast_fp16")];
            tensor<fp16, []> var_491_to_fp16 = const()[name = tensor<string, []>("op_491_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_492_cast_fp16 = mul(x = linear_9_cast_fp16, y = var_491_to_fp16)[name = tensor<string, []>("op_492_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_69_cast_fp16 = add(x = input_57_cast_fp16, y = var_492_cast_fp16)[name = tensor<string, []>("input_69_cast_fp16")];
            tensor<int32, [1]> input_71_axes_0 = const()[name = tensor<string, []>("input_71_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_0_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_norm_out_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8459840)))];
            tensor<fp16, [512]> encoder_layers_0_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_0_norm_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8460928)))];
            tensor<fp16, [1, 125, 512]> input_71_cast_fp16 = layer_norm(axes = input_71_axes_0, beta = encoder_layers_0_norm_out_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_0_norm_out_weight_to_fp16, x = input_69_cast_fp16)[name = tensor<string, []>("input_71_cast_fp16")];
            tensor<int32, [1]> input_73_axes_0 = const()[name = tensor<string, []>("input_73_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_1_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8462016)))];
            tensor<fp16, [512]> encoder_layers_1_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8463104)))];
            tensor<fp16, [1, 125, 512]> input_73_cast_fp16 = layer_norm(axes = input_73_axes_0, beta = encoder_layers_1_norm_feed_forward1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_1_norm_feed_forward1_weight_to_fp16, x = input_71_cast_fp16)[name = tensor<string, []>("input_73_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_1_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_1_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(8464192))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9512832))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_1_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9516992)))];
            tensor<fp16, [1, 125, 2048]> linear_10_cast_fp16 = linear(bias = encoder_layers_1_feed_forward1_linear1_bias_to_fp16, weight = encoder_layers_1_feed_forward1_linear1_weight_to_fp16_quantized, x = input_73_cast_fp16)[name = tensor<string, []>("linear_10_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_77_cast_fp16 = silu(x = linear_10_cast_fp16)[name = tensor<string, []>("input_77_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_1_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_1_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(9521152))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10569792))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_1_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10570880)))];
            tensor<fp16, [1, 125, 512]> linear_11_cast_fp16 = linear(bias = encoder_layers_1_feed_forward1_linear2_bias_to_fp16, weight = encoder_layers_1_feed_forward1_linear2_weight_to_fp16_quantized, x = input_77_cast_fp16)[name = tensor<string, []>("linear_11_cast_fp16")];
            tensor<fp16, []> var_522_to_fp16 = const()[name = tensor<string, []>("op_522_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_523_cast_fp16 = mul(x = linear_11_cast_fp16, y = var_522_to_fp16)[name = tensor<string, []>("op_523_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_83_cast_fp16 = add(x = input_71_cast_fp16, y = var_523_cast_fp16)[name = tensor<string, []>("input_83_cast_fp16")];
            tensor<int32, [1]> query_3_axes_0 = const()[name = tensor<string, []>("query_3_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_1_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_norm_self_att_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10571968)))];
            tensor<fp16, [512]> encoder_layers_1_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_norm_self_att_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10573056)))];
            tensor<fp16, [1, 125, 512]> query_3_cast_fp16 = layer_norm(axes = query_3_axes_0, beta = encoder_layers_1_norm_self_att_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_1_norm_self_att_weight_to_fp16, x = input_83_cast_fp16)[name = tensor<string, []>("query_3_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_1_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_1_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10574144))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10836352))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_1_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10837440)))];
            tensor<fp16, [1, 125, 512]> linear_12_cast_fp16 = linear(bias = encoder_layers_1_self_attn_linear_q_bias_to_fp16, weight = encoder_layers_1_self_attn_linear_q_weight_to_fp16_quantized, x = query_3_cast_fp16)[name = tensor<string, []>("linear_12_cast_fp16")];
            tensor<int32, [4]> var_540 = const()[name = tensor<string, []>("op_540"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> q_7_cast_fp16 = reshape(shape = var_540, x = linear_12_cast_fp16)[name = tensor<string, []>("q_7_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_1_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_1_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(10838528))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11100736))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_1_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11101824)))];
            tensor<fp16, [1, 125, 512]> linear_13_cast_fp16 = linear(bias = encoder_layers_1_self_attn_linear_k_bias_to_fp16, weight = encoder_layers_1_self_attn_linear_k_weight_to_fp16_quantized, x = query_3_cast_fp16)[name = tensor<string, []>("linear_13_cast_fp16")];
            tensor<int32, [4]> var_545 = const()[name = tensor<string, []>("op_545"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> k_5_cast_fp16 = reshape(shape = var_545, x = linear_13_cast_fp16)[name = tensor<string, []>("k_5_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_1_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_1_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11102912))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11365120))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_1_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11366208)))];
            tensor<fp16, [1, 125, 512]> linear_14_cast_fp16 = linear(bias = encoder_layers_1_self_attn_linear_v_bias_to_fp16, weight = encoder_layers_1_self_attn_linear_v_weight_to_fp16_quantized, x = query_3_cast_fp16)[name = tensor<string, []>("linear_14_cast_fp16")];
            tensor<int32, [4]> var_550 = const()[name = tensor<string, []>("op_550"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> v_3_cast_fp16 = reshape(shape = var_550, x = linear_14_cast_fp16)[name = tensor<string, []>("v_3_cast_fp16")];
            tensor<int32, [4]> value_3_perm_0 = const()[name = tensor<string, []>("value_3_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 64]> encoder_layers_1_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11367296)))];
            tensor<fp16, [1, 125, 8, 64]> var_562_cast_fp16 = add(x = q_7_cast_fp16, y = encoder_layers_1_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_562_cast_fp16")];
            tensor<fp16, [8, 64]> encoder_layers_1_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11368384)))];
            tensor<fp16, [1, 125, 8, 64]> var_564_cast_fp16 = add(x = q_7_cast_fp16, y = encoder_layers_1_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_564_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_3_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_3_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<bool, []> x_29_transpose_x_0 = const()[name = tensor<string, []>("x_29_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_29_transpose_y_0 = const()[name = tensor<string, []>("x_29_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 64, 249]> op_566_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(3), name = tensor<string, []>("op_566_to_fp16_quantized"), quantized_data = tensor<int8, [1, 8, 64, 249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11369472))), scale = tensor<fp16, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11497024))), zero_point = tensor<int8, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5281536)))];
            tensor<fp16, [1, 8, 125, 64]> q_with_bias_v_3_cast_fp16 = transpose(perm = q_with_bias_v_3_perm_0, x = var_564_cast_fp16)[name = tensor<string, []>("transpose_213")];
            tensor<fp16, [1, 8, 125, 249]> x_29_cast_fp16 = matmul(transpose_x = x_29_transpose_x_0, transpose_y = x_29_transpose_y_0, x = q_with_bias_v_3_cast_fp16, y = op_566_to_fp16_quantized)[name = tensor<string, []>("x_29_cast_fp16")];
            tensor<int32, [8]> x_31_pad_0 = const()[name = tensor<string, []>("x_31_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            tensor<string, []> x_31_mode_0 = const()[name = tensor<string, []>("x_31_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_80_to_fp16 = const()[name = tensor<string, []>("const_80_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 8, 125, 250]> x_31_cast_fp16 = pad(constant_val = const_80_to_fp16, mode = x_31_mode_0, pad = x_31_pad_0, x = x_29_cast_fp16)[name = tensor<string, []>("x_31_cast_fp16")];
            tensor<int32, [4]> var_574 = const()[name = tensor<string, []>("op_574"), val = tensor<int32, [4]>([1, 8, -1, 125])];
            tensor<fp16, [1, 8, 250, 125]> x_33_cast_fp16 = reshape(shape = var_574, x = x_31_cast_fp16)[name = tensor<string, []>("x_33_cast_fp16")];
            tensor<int32, [4]> var_578_begin_0 = const()[name = tensor<string, []>("op_578_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_578_end_0 = const()[name = tensor<string, []>("op_578_end_0"), val = tensor<int32, [4]>([1, 8, 250, 125])];
            tensor<bool, [4]> var_578_end_mask_0 = const()[name = tensor<string, []>("op_578_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 249, 125]> var_578_cast_fp16 = slice_by_index(begin = var_578_begin_0, end = var_578_end_0, end_mask = var_578_end_mask_0, x = x_33_cast_fp16)[name = tensor<string, []>("op_578_cast_fp16")];
            tensor<int32, [4]> var_579 = const()[name = tensor<string, []>("op_579"), val = tensor<int32, [4]>([1, 8, 125, 249])];
            tensor<fp16, [1, 8, 125, 249]> matrix_bd_5_cast_fp16 = reshape(shape = var_579, x = var_578_cast_fp16)[name = tensor<string, []>("matrix_bd_5_cast_fp16")];
            tensor<bool, []> matrix_ac_3_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_3_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matrix_ac_3_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_3_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_70_perm_0 = const()[name = tensor<string, []>("transpose_70_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_71_perm_0 = const()[name = tensor<string, []>("transpose_71_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 64, 125]> transpose_71 = transpose(perm = transpose_71_perm_0, x = k_5_cast_fp16)[name = tensor<string, []>("transpose_211")];
            tensor<fp16, [1, 8, 125, 64]> transpose_70 = transpose(perm = transpose_70_perm_0, x = var_562_cast_fp16)[name = tensor<string, []>("transpose_212")];
            tensor<fp16, [1, 8, 125, 125]> matrix_ac_3_cast_fp16 = matmul(transpose_x = matrix_ac_3_transpose_x_0, transpose_y = matrix_ac_3_transpose_y_0, x = transpose_70, y = transpose_71)[name = tensor<string, []>("matrix_ac_3_cast_fp16")];
            tensor<int32, [4]> matrix_bd_7_begin_0 = const()[name = tensor<string, []>("matrix_bd_7_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_7_end_0 = const()[name = tensor<string, []>("matrix_bd_7_end_0"), val = tensor<int32, [4]>([1, 8, 125, 125])];
            tensor<bool, [4]> matrix_bd_7_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_7_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 125, 125]> matrix_bd_7_cast_fp16 = slice_by_index(begin = matrix_bd_7_begin_0, end = matrix_bd_7_end_0, end_mask = matrix_bd_7_end_mask_0, x = matrix_bd_5_cast_fp16)[name = tensor<string, []>("matrix_bd_7_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_588_cast_fp16 = add(x = matrix_ac_3_cast_fp16, y = matrix_bd_7_cast_fp16)[name = tensor<string, []>("op_588_cast_fp16")];
            tensor<fp16, []> _inversed_scores_5_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_5_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 8, 125, 125]> _inversed_scores_5_cast_fp16 = mul(x = var_588_cast_fp16, y = _inversed_scores_5_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_5_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> scores_7_cast_fp16 = select(a = var_8_to_fp16, b = _inversed_scores_5_cast_fp16, cond = mask_11)[name = tensor<string, []>("scores_7_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_594_cast_fp16 = softmax(axis = var_19, x = scores_7_cast_fp16)[name = tensor<string, []>("op_594_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> input_85_cast_fp16 = select(a = var_7_to_fp16, b = var_594_cast_fp16, cond = mask_11)[name = tensor<string, []>("input_85_cast_fp16")];
            tensor<bool, []> x_35_transpose_x_0 = const()[name = tensor<string, []>("x_35_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_35_transpose_y_0 = const()[name = tensor<string, []>("x_35_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 125, 64]> value_3_cast_fp16 = transpose(perm = value_3_perm_0, x = v_3_cast_fp16)[name = tensor<string, []>("transpose_210")];
            tensor<fp16, [1, 8, 125, 64]> x_35_cast_fp16 = matmul(transpose_x = x_35_transpose_x_0, transpose_y = x_35_transpose_y_0, x = input_85_cast_fp16, y = value_3_cast_fp16)[name = tensor<string, []>("x_35_cast_fp16")];
            tensor<int32, [4]> var_598_perm_0 = const()[name = tensor<string, []>("op_598_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_599 = const()[name = tensor<string, []>("op_599"), val = tensor<int32, [3]>([1, -1, 512])];
            tensor<fp16, [1, 125, 8, 64]> var_598_cast_fp16 = transpose(perm = var_598_perm_0, x = x_35_cast_fp16)[name = tensor<string, []>("transpose_209")];
            tensor<fp16, [1, 125, 512]> input_87_cast_fp16 = reshape(shape = var_599, x = var_598_cast_fp16)[name = tensor<string, []>("input_87_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_1_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_1_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11497600))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11759808))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_1_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11760896)))];
            tensor<fp16, [1, 125, 512]> linear_16_cast_fp16 = linear(bias = encoder_layers_1_self_attn_linear_out_bias_to_fp16, weight = encoder_layers_1_self_attn_linear_out_weight_to_fp16_quantized, x = input_87_cast_fp16)[name = tensor<string, []>("linear_16_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_91_cast_fp16 = add(x = input_83_cast_fp16, y = linear_16_cast_fp16)[name = tensor<string, []>("input_91_cast_fp16")];
            tensor<int32, [1]> x_39_axes_0 = const()[name = tensor<string, []>("x_39_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_1_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_norm_conv_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11761984)))];
            tensor<fp16, [512]> encoder_layers_1_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_norm_conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11763072)))];
            tensor<fp16, [1, 125, 512]> x_39_cast_fp16 = layer_norm(axes = x_39_axes_0, beta = encoder_layers_1_norm_conv_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_1_norm_conv_weight_to_fp16, x = input_91_cast_fp16)[name = tensor<string, []>("x_39_cast_fp16")];
            tensor<int32, [3]> input_93_perm_0 = const()[name = tensor<string, []>("input_93_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> input_95_pad_type_0 = const()[name = tensor<string, []>("input_95_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> input_95_strides_0 = const()[name = tensor<string, []>("input_95_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_95_pad_0 = const()[name = tensor<string, []>("input_95_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_95_dilations_0 = const()[name = tensor<string, []>("input_95_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> input_95_groups_0 = const()[name = tensor<string, []>("input_95_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 512, 1]> encoder_layers_1_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_1_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1024, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(11764160))), scale = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12288512))), zero_point = tensor<int8, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6073344)))];
            tensor<fp16, [1024]> encoder_layers_1_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12290624)))];
            tensor<fp16, [1, 512, 125]> input_93_cast_fp16 = transpose(perm = input_93_perm_0, x = x_39_cast_fp16)[name = tensor<string, []>("transpose_208")];
            tensor<fp16, [1, 1024, 125]> input_95_cast_fp16 = conv(bias = encoder_layers_1_conv_pointwise_conv1_bias_to_fp16, dilations = input_95_dilations_0, groups = input_95_groups_0, pad = input_95_pad_0, pad_type = input_95_pad_type_0, strides = input_95_strides_0, weight = encoder_layers_1_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_93_cast_fp16)[name = tensor<string, []>("input_95_cast_fp16")];
            tensor<int32, []> x_41_split_num_splits_0 = const()[name = tensor<string, []>("x_41_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> x_41_split_axis_0 = const()[name = tensor<string, []>("x_41_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 512, 125]> x_41_split_cast_fp16_0, tensor<fp16, [1, 512, 125]> x_41_split_cast_fp16_1 = split(axis = x_41_split_axis_0, num_splits = x_41_split_num_splits_0, x = input_95_cast_fp16)[name = tensor<string, []>("x_41_split_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_41_split_1_sigmoid_cast_fp16 = sigmoid(x = x_41_split_cast_fp16_1)[name = tensor<string, []>("x_41_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_41_cast_fp16 = mul(x = x_41_split_cast_fp16_0, y = x_41_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_41_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_97_cast_fp16 = select(a = var_7_to_fp16, b = x_41_cast_fp16, cond = var_449)[name = tensor<string, []>("input_97_cast_fp16")];
            tensor<int32, [6]> input_99_pad_0 = const()[name = tensor<string, []>("input_99_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            tensor<string, []> input_99_mode_0 = const()[name = tensor<string, []>("input_99_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_83_to_fp16 = const()[name = tensor<string, []>("const_83_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 512, 133]> input_99_cast_fp16 = pad(constant_val = const_83_to_fp16, mode = input_99_mode_0, pad = input_99_pad_0, x = input_97_cast_fp16)[name = tensor<string, []>("input_99_cast_fp16")];
            tensor<string, []> input_101_pad_type_0 = const()[name = tensor<string, []>("input_101_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, []> input_101_groups_0 = const()[name = tensor<string, []>("input_101_groups_0"), val = tensor<int32, []>(512)];
            tensor<int32, [1]> input_101_strides_0 = const()[name = tensor<string, []>("input_101_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_101_pad_0 = const()[name = tensor<string, []>("input_101_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_101_dilations_0 = const()[name = tensor<string, []>("input_101_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 9]> const_239_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("const_239_to_fp16_quantized"), quantized_data = tensor<int8, [512, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12292736))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12297408))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> const_240_to_fp16 = const()[name = tensor<string, []>("const_240_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12298496)))];
            tensor<fp16, [1, 512, 125]> input_103_cast_fp16 = conv(bias = const_240_to_fp16, dilations = input_101_dilations_0, groups = input_101_groups_0, pad = input_101_pad_0, pad_type = input_101_pad_type_0, strides = input_101_strides_0, weight = const_239_to_fp16_quantized, x = input_99_cast_fp16)[name = tensor<string, []>("input_103_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_105_cast_fp16 = silu(x = input_103_cast_fp16)[name = tensor<string, []>("input_105_cast_fp16")];
            tensor<string, []> x_43_pad_type_0 = const()[name = tensor<string, []>("x_43_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> x_43_strides_0 = const()[name = tensor<string, []>("x_43_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_43_pad_0 = const()[name = tensor<string, []>("x_43_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_43_dilations_0 = const()[name = tensor<string, []>("x_43_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> x_43_groups_0 = const()[name = tensor<string, []>("x_43_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [512, 512, 1]> encoder_layers_1_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_1_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12299584))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12561792))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_1_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12562880)))];
            tensor<fp16, [1, 512, 125]> x_43_cast_fp16 = conv(bias = encoder_layers_1_conv_pointwise_conv2_bias_to_fp16, dilations = x_43_dilations_0, groups = x_43_groups_0, pad = x_43_pad_0, pad_type = x_43_pad_type_0, strides = x_43_strides_0, weight = encoder_layers_1_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_105_cast_fp16)[name = tensor<string, []>("x_43_cast_fp16")];
            tensor<int32, [3]> input_107_perm_0 = const()[name = tensor<string, []>("input_107_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 125, 512]> input_107_cast_fp16 = transpose(perm = input_107_perm_0, x = x_43_cast_fp16)[name = tensor<string, []>("transpose_207")];
            tensor<fp16, [1, 125, 512]> input_109_cast_fp16 = add(x = input_91_cast_fp16, y = input_107_cast_fp16)[name = tensor<string, []>("input_109_cast_fp16")];
            tensor<int32, [1]> input_111_axes_0 = const()[name = tensor<string, []>("input_111_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_1_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12563968)))];
            tensor<fp16, [512]> encoder_layers_1_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12565056)))];
            tensor<fp16, [1, 125, 512]> input_111_cast_fp16 = layer_norm(axes = input_111_axes_0, beta = encoder_layers_1_norm_feed_forward2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_1_norm_feed_forward2_weight_to_fp16, x = input_109_cast_fp16)[name = tensor<string, []>("input_111_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_1_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_1_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(12566144))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13614784))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_1_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13618944)))];
            tensor<fp16, [1, 125, 2048]> linear_17_cast_fp16 = linear(bias = encoder_layers_1_feed_forward2_linear1_bias_to_fp16, weight = encoder_layers_1_feed_forward2_linear1_weight_to_fp16_quantized, x = input_111_cast_fp16)[name = tensor<string, []>("linear_17_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_115_cast_fp16 = silu(x = linear_17_cast_fp16)[name = tensor<string, []>("input_115_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_1_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_1_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(13623104))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14671744))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_1_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14672832)))];
            tensor<fp16, [1, 125, 512]> linear_18_cast_fp16 = linear(bias = encoder_layers_1_feed_forward2_linear2_bias_to_fp16, weight = encoder_layers_1_feed_forward2_linear2_weight_to_fp16_quantized, x = input_115_cast_fp16)[name = tensor<string, []>("linear_18_cast_fp16")];
            tensor<fp16, []> var_665_to_fp16 = const()[name = tensor<string, []>("op_665_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_666_cast_fp16 = mul(x = linear_18_cast_fp16, y = var_665_to_fp16)[name = tensor<string, []>("op_666_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_121_cast_fp16 = add(x = input_109_cast_fp16, y = var_666_cast_fp16)[name = tensor<string, []>("input_121_cast_fp16")];
            tensor<int32, [1]> input_123_axes_0 = const()[name = tensor<string, []>("input_123_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_1_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_norm_out_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14673920)))];
            tensor<fp16, [512]> encoder_layers_1_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_1_norm_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14675008)))];
            tensor<fp16, [1, 125, 512]> input_123_cast_fp16 = layer_norm(axes = input_123_axes_0, beta = encoder_layers_1_norm_out_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_1_norm_out_weight_to_fp16, x = input_121_cast_fp16)[name = tensor<string, []>("input_123_cast_fp16")];
            tensor<int32, [1]> input_125_axes_0 = const()[name = tensor<string, []>("input_125_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_2_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14676096)))];
            tensor<fp16, [512]> encoder_layers_2_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14677184)))];
            tensor<fp16, [1, 125, 512]> input_125_cast_fp16 = layer_norm(axes = input_125_axes_0, beta = encoder_layers_2_norm_feed_forward1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_2_norm_feed_forward1_weight_to_fp16, x = input_123_cast_fp16)[name = tensor<string, []>("input_125_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_2_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_2_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(14678272))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15726912))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_2_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15731072)))];
            tensor<fp16, [1, 125, 2048]> linear_19_cast_fp16 = linear(bias = encoder_layers_2_feed_forward1_linear1_bias_to_fp16, weight = encoder_layers_2_feed_forward1_linear1_weight_to_fp16_quantized, x = input_125_cast_fp16)[name = tensor<string, []>("linear_19_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_129_cast_fp16 = silu(x = linear_19_cast_fp16)[name = tensor<string, []>("input_129_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_2_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_2_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(15735232))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16783872))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_2_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16784960)))];
            tensor<fp16, [1, 125, 512]> linear_20_cast_fp16 = linear(bias = encoder_layers_2_feed_forward1_linear2_bias_to_fp16, weight = encoder_layers_2_feed_forward1_linear2_weight_to_fp16_quantized, x = input_129_cast_fp16)[name = tensor<string, []>("linear_20_cast_fp16")];
            tensor<fp16, []> var_696_to_fp16 = const()[name = tensor<string, []>("op_696_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_697_cast_fp16 = mul(x = linear_20_cast_fp16, y = var_696_to_fp16)[name = tensor<string, []>("op_697_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_135_cast_fp16 = add(x = input_123_cast_fp16, y = var_697_cast_fp16)[name = tensor<string, []>("input_135_cast_fp16")];
            tensor<int32, [1]> query_5_axes_0 = const()[name = tensor<string, []>("query_5_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_2_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_norm_self_att_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16786048)))];
            tensor<fp16, [512]> encoder_layers_2_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_norm_self_att_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16787136)))];
            tensor<fp16, [1, 125, 512]> query_5_cast_fp16 = layer_norm(axes = query_5_axes_0, beta = encoder_layers_2_norm_self_att_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_2_norm_self_att_weight_to_fp16, x = input_135_cast_fp16)[name = tensor<string, []>("query_5_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_2_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_2_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(16788224))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17050432))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_2_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17051520)))];
            tensor<fp16, [1, 125, 512]> linear_21_cast_fp16 = linear(bias = encoder_layers_2_self_attn_linear_q_bias_to_fp16, weight = encoder_layers_2_self_attn_linear_q_weight_to_fp16_quantized, x = query_5_cast_fp16)[name = tensor<string, []>("linear_21_cast_fp16")];
            tensor<int32, [4]> var_714 = const()[name = tensor<string, []>("op_714"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> q_13_cast_fp16 = reshape(shape = var_714, x = linear_21_cast_fp16)[name = tensor<string, []>("q_13_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_2_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_2_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17052608))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17314816))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_2_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17315904)))];
            tensor<fp16, [1, 125, 512]> linear_22_cast_fp16 = linear(bias = encoder_layers_2_self_attn_linear_k_bias_to_fp16, weight = encoder_layers_2_self_attn_linear_k_weight_to_fp16_quantized, x = query_5_cast_fp16)[name = tensor<string, []>("linear_22_cast_fp16")];
            tensor<int32, [4]> var_719 = const()[name = tensor<string, []>("op_719"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> k_9_cast_fp16 = reshape(shape = var_719, x = linear_22_cast_fp16)[name = tensor<string, []>("k_9_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_2_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_2_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17316992))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17579200))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_2_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17580288)))];
            tensor<fp16, [1, 125, 512]> linear_23_cast_fp16 = linear(bias = encoder_layers_2_self_attn_linear_v_bias_to_fp16, weight = encoder_layers_2_self_attn_linear_v_weight_to_fp16_quantized, x = query_5_cast_fp16)[name = tensor<string, []>("linear_23_cast_fp16")];
            tensor<int32, [4]> var_724 = const()[name = tensor<string, []>("op_724"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> v_5_cast_fp16 = reshape(shape = var_724, x = linear_23_cast_fp16)[name = tensor<string, []>("v_5_cast_fp16")];
            tensor<int32, [4]> value_5_perm_0 = const()[name = tensor<string, []>("value_5_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 64]> encoder_layers_2_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17581376)))];
            tensor<fp16, [1, 125, 8, 64]> var_736_cast_fp16 = add(x = q_13_cast_fp16, y = encoder_layers_2_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_736_cast_fp16")];
            tensor<fp16, [8, 64]> encoder_layers_2_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17582464)))];
            tensor<fp16, [1, 125, 8, 64]> var_738_cast_fp16 = add(x = q_13_cast_fp16, y = encoder_layers_2_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_738_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_5_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_5_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<bool, []> x_51_transpose_x_0 = const()[name = tensor<string, []>("x_51_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_51_transpose_y_0 = const()[name = tensor<string, []>("x_51_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 64, 249]> op_740_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(3), name = tensor<string, []>("op_740_to_fp16_quantized"), quantized_data = tensor<int8, [1, 8, 64, 249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17583552))), scale = tensor<fp16, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17711104))), zero_point = tensor<int8, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5281536)))];
            tensor<fp16, [1, 8, 125, 64]> q_with_bias_v_5_cast_fp16 = transpose(perm = q_with_bias_v_5_perm_0, x = var_738_cast_fp16)[name = tensor<string, []>("transpose_206")];
            tensor<fp16, [1, 8, 125, 249]> x_51_cast_fp16 = matmul(transpose_x = x_51_transpose_x_0, transpose_y = x_51_transpose_y_0, x = q_with_bias_v_5_cast_fp16, y = op_740_to_fp16_quantized)[name = tensor<string, []>("x_51_cast_fp16")];
            tensor<int32, [8]> x_53_pad_0 = const()[name = tensor<string, []>("x_53_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            tensor<string, []> x_53_mode_0 = const()[name = tensor<string, []>("x_53_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_90_to_fp16 = const()[name = tensor<string, []>("const_90_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 8, 125, 250]> x_53_cast_fp16 = pad(constant_val = const_90_to_fp16, mode = x_53_mode_0, pad = x_53_pad_0, x = x_51_cast_fp16)[name = tensor<string, []>("x_53_cast_fp16")];
            tensor<int32, [4]> var_748 = const()[name = tensor<string, []>("op_748"), val = tensor<int32, [4]>([1, 8, -1, 125])];
            tensor<fp16, [1, 8, 250, 125]> x_55_cast_fp16 = reshape(shape = var_748, x = x_53_cast_fp16)[name = tensor<string, []>("x_55_cast_fp16")];
            tensor<int32, [4]> var_752_begin_0 = const()[name = tensor<string, []>("op_752_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_752_end_0 = const()[name = tensor<string, []>("op_752_end_0"), val = tensor<int32, [4]>([1, 8, 250, 125])];
            tensor<bool, [4]> var_752_end_mask_0 = const()[name = tensor<string, []>("op_752_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 249, 125]> var_752_cast_fp16 = slice_by_index(begin = var_752_begin_0, end = var_752_end_0, end_mask = var_752_end_mask_0, x = x_55_cast_fp16)[name = tensor<string, []>("op_752_cast_fp16")];
            tensor<int32, [4]> var_753 = const()[name = tensor<string, []>("op_753"), val = tensor<int32, [4]>([1, 8, 125, 249])];
            tensor<fp16, [1, 8, 125, 249]> matrix_bd_9_cast_fp16 = reshape(shape = var_753, x = var_752_cast_fp16)[name = tensor<string, []>("matrix_bd_9_cast_fp16")];
            tensor<bool, []> matrix_ac_5_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_5_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matrix_ac_5_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_5_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_72_perm_0 = const()[name = tensor<string, []>("transpose_72_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_73_perm_0 = const()[name = tensor<string, []>("transpose_73_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 64, 125]> transpose_73 = transpose(perm = transpose_73_perm_0, x = k_9_cast_fp16)[name = tensor<string, []>("transpose_204")];
            tensor<fp16, [1, 8, 125, 64]> transpose_72 = transpose(perm = transpose_72_perm_0, x = var_736_cast_fp16)[name = tensor<string, []>("transpose_205")];
            tensor<fp16, [1, 8, 125, 125]> matrix_ac_5_cast_fp16 = matmul(transpose_x = matrix_ac_5_transpose_x_0, transpose_y = matrix_ac_5_transpose_y_0, x = transpose_72, y = transpose_73)[name = tensor<string, []>("matrix_ac_5_cast_fp16")];
            tensor<int32, [4]> matrix_bd_11_begin_0 = const()[name = tensor<string, []>("matrix_bd_11_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_11_end_0 = const()[name = tensor<string, []>("matrix_bd_11_end_0"), val = tensor<int32, [4]>([1, 8, 125, 125])];
            tensor<bool, [4]> matrix_bd_11_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_11_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 125, 125]> matrix_bd_11_cast_fp16 = slice_by_index(begin = matrix_bd_11_begin_0, end = matrix_bd_11_end_0, end_mask = matrix_bd_11_end_mask_0, x = matrix_bd_9_cast_fp16)[name = tensor<string, []>("matrix_bd_11_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_762_cast_fp16 = add(x = matrix_ac_5_cast_fp16, y = matrix_bd_11_cast_fp16)[name = tensor<string, []>("op_762_cast_fp16")];
            tensor<fp16, []> _inversed_scores_9_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_9_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 8, 125, 125]> _inversed_scores_9_cast_fp16 = mul(x = var_762_cast_fp16, y = _inversed_scores_9_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_9_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> scores_11_cast_fp16 = select(a = var_8_to_fp16, b = _inversed_scores_9_cast_fp16, cond = mask_11)[name = tensor<string, []>("scores_11_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_768_cast_fp16 = softmax(axis = var_19, x = scores_11_cast_fp16)[name = tensor<string, []>("op_768_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> input_137_cast_fp16 = select(a = var_7_to_fp16, b = var_768_cast_fp16, cond = mask_11)[name = tensor<string, []>("input_137_cast_fp16")];
            tensor<bool, []> x_57_transpose_x_0 = const()[name = tensor<string, []>("x_57_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_57_transpose_y_0 = const()[name = tensor<string, []>("x_57_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 125, 64]> value_5_cast_fp16 = transpose(perm = value_5_perm_0, x = v_5_cast_fp16)[name = tensor<string, []>("transpose_203")];
            tensor<fp16, [1, 8, 125, 64]> x_57_cast_fp16 = matmul(transpose_x = x_57_transpose_x_0, transpose_y = x_57_transpose_y_0, x = input_137_cast_fp16, y = value_5_cast_fp16)[name = tensor<string, []>("x_57_cast_fp16")];
            tensor<int32, [4]> var_772_perm_0 = const()[name = tensor<string, []>("op_772_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_773 = const()[name = tensor<string, []>("op_773"), val = tensor<int32, [3]>([1, -1, 512])];
            tensor<fp16, [1, 125, 8, 64]> var_772_cast_fp16 = transpose(perm = var_772_perm_0, x = x_57_cast_fp16)[name = tensor<string, []>("transpose_202")];
            tensor<fp16, [1, 125, 512]> input_139_cast_fp16 = reshape(shape = var_773, x = var_772_cast_fp16)[name = tensor<string, []>("input_139_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_2_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_2_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17711680))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17973888))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_2_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17974976)))];
            tensor<fp16, [1, 125, 512]> linear_25_cast_fp16 = linear(bias = encoder_layers_2_self_attn_linear_out_bias_to_fp16, weight = encoder_layers_2_self_attn_linear_out_weight_to_fp16_quantized, x = input_139_cast_fp16)[name = tensor<string, []>("linear_25_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_143_cast_fp16 = add(x = input_135_cast_fp16, y = linear_25_cast_fp16)[name = tensor<string, []>("input_143_cast_fp16")];
            tensor<int32, [1]> x_61_axes_0 = const()[name = tensor<string, []>("x_61_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_2_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_norm_conv_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17976064)))];
            tensor<fp16, [512]> encoder_layers_2_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_norm_conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17977152)))];
            tensor<fp16, [1, 125, 512]> x_61_cast_fp16 = layer_norm(axes = x_61_axes_0, beta = encoder_layers_2_norm_conv_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_2_norm_conv_weight_to_fp16, x = input_143_cast_fp16)[name = tensor<string, []>("x_61_cast_fp16")];
            tensor<int32, [3]> input_145_perm_0 = const()[name = tensor<string, []>("input_145_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> input_147_pad_type_0 = const()[name = tensor<string, []>("input_147_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> input_147_strides_0 = const()[name = tensor<string, []>("input_147_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_147_pad_0 = const()[name = tensor<string, []>("input_147_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_147_dilations_0 = const()[name = tensor<string, []>("input_147_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> input_147_groups_0 = const()[name = tensor<string, []>("input_147_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 512, 1]> encoder_layers_2_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_2_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1024, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(17978240))), scale = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18502592))), zero_point = tensor<int8, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6073344)))];
            tensor<fp16, [1024]> encoder_layers_2_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18504704)))];
            tensor<fp16, [1, 512, 125]> input_145_cast_fp16 = transpose(perm = input_145_perm_0, x = x_61_cast_fp16)[name = tensor<string, []>("transpose_201")];
            tensor<fp16, [1, 1024, 125]> input_147_cast_fp16 = conv(bias = encoder_layers_2_conv_pointwise_conv1_bias_to_fp16, dilations = input_147_dilations_0, groups = input_147_groups_0, pad = input_147_pad_0, pad_type = input_147_pad_type_0, strides = input_147_strides_0, weight = encoder_layers_2_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_145_cast_fp16)[name = tensor<string, []>("input_147_cast_fp16")];
            tensor<int32, []> x_63_split_num_splits_0 = const()[name = tensor<string, []>("x_63_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> x_63_split_axis_0 = const()[name = tensor<string, []>("x_63_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 512, 125]> x_63_split_cast_fp16_0, tensor<fp16, [1, 512, 125]> x_63_split_cast_fp16_1 = split(axis = x_63_split_axis_0, num_splits = x_63_split_num_splits_0, x = input_147_cast_fp16)[name = tensor<string, []>("x_63_split_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_63_split_1_sigmoid_cast_fp16 = sigmoid(x = x_63_split_cast_fp16_1)[name = tensor<string, []>("x_63_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_63_cast_fp16 = mul(x = x_63_split_cast_fp16_0, y = x_63_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_63_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_149_cast_fp16 = select(a = var_7_to_fp16, b = x_63_cast_fp16, cond = var_449)[name = tensor<string, []>("input_149_cast_fp16")];
            tensor<int32, [6]> input_151_pad_0 = const()[name = tensor<string, []>("input_151_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            tensor<string, []> input_151_mode_0 = const()[name = tensor<string, []>("input_151_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_93_to_fp16 = const()[name = tensor<string, []>("const_93_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 512, 133]> input_151_cast_fp16 = pad(constant_val = const_93_to_fp16, mode = input_151_mode_0, pad = input_151_pad_0, x = input_149_cast_fp16)[name = tensor<string, []>("input_151_cast_fp16")];
            tensor<string, []> input_153_pad_type_0 = const()[name = tensor<string, []>("input_153_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, []> input_153_groups_0 = const()[name = tensor<string, []>("input_153_groups_0"), val = tensor<int32, []>(512)];
            tensor<int32, [1]> input_153_strides_0 = const()[name = tensor<string, []>("input_153_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_153_pad_0 = const()[name = tensor<string, []>("input_153_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_153_dilations_0 = const()[name = tensor<string, []>("input_153_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 9]> const_241_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("const_241_to_fp16_quantized"), quantized_data = tensor<int8, [512, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18506816))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18511488))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> const_242_to_fp16 = const()[name = tensor<string, []>("const_242_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18512576)))];
            tensor<fp16, [1, 512, 125]> input_155_cast_fp16 = conv(bias = const_242_to_fp16, dilations = input_153_dilations_0, groups = input_153_groups_0, pad = input_153_pad_0, pad_type = input_153_pad_type_0, strides = input_153_strides_0, weight = const_241_to_fp16_quantized, x = input_151_cast_fp16)[name = tensor<string, []>("input_155_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_157_cast_fp16 = silu(x = input_155_cast_fp16)[name = tensor<string, []>("input_157_cast_fp16")];
            tensor<string, []> x_65_pad_type_0 = const()[name = tensor<string, []>("x_65_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> x_65_strides_0 = const()[name = tensor<string, []>("x_65_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_65_pad_0 = const()[name = tensor<string, []>("x_65_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_65_dilations_0 = const()[name = tensor<string, []>("x_65_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> x_65_groups_0 = const()[name = tensor<string, []>("x_65_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [512, 512, 1]> encoder_layers_2_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_2_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18513664))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18775872))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_2_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18776960)))];
            tensor<fp16, [1, 512, 125]> x_65_cast_fp16 = conv(bias = encoder_layers_2_conv_pointwise_conv2_bias_to_fp16, dilations = x_65_dilations_0, groups = x_65_groups_0, pad = x_65_pad_0, pad_type = x_65_pad_type_0, strides = x_65_strides_0, weight = encoder_layers_2_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_157_cast_fp16)[name = tensor<string, []>("x_65_cast_fp16")];
            tensor<int32, [3]> input_159_perm_0 = const()[name = tensor<string, []>("input_159_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 125, 512]> input_159_cast_fp16 = transpose(perm = input_159_perm_0, x = x_65_cast_fp16)[name = tensor<string, []>("transpose_200")];
            tensor<fp16, [1, 125, 512]> input_161_cast_fp16 = add(x = input_143_cast_fp16, y = input_159_cast_fp16)[name = tensor<string, []>("input_161_cast_fp16")];
            tensor<int32, [1]> input_163_axes_0 = const()[name = tensor<string, []>("input_163_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_2_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18778048)))];
            tensor<fp16, [512]> encoder_layers_2_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18779136)))];
            tensor<fp16, [1, 125, 512]> input_163_cast_fp16 = layer_norm(axes = input_163_axes_0, beta = encoder_layers_2_norm_feed_forward2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_2_norm_feed_forward2_weight_to_fp16, x = input_161_cast_fp16)[name = tensor<string, []>("input_163_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_2_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_2_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(18780224))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19828864))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_2_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19833024)))];
            tensor<fp16, [1, 125, 2048]> linear_26_cast_fp16 = linear(bias = encoder_layers_2_feed_forward2_linear1_bias_to_fp16, weight = encoder_layers_2_feed_forward2_linear1_weight_to_fp16_quantized, x = input_163_cast_fp16)[name = tensor<string, []>("linear_26_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_167_cast_fp16 = silu(x = linear_26_cast_fp16)[name = tensor<string, []>("input_167_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_2_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_2_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(19837184))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20885824))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_2_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20886912)))];
            tensor<fp16, [1, 125, 512]> linear_27_cast_fp16 = linear(bias = encoder_layers_2_feed_forward2_linear2_bias_to_fp16, weight = encoder_layers_2_feed_forward2_linear2_weight_to_fp16_quantized, x = input_167_cast_fp16)[name = tensor<string, []>("linear_27_cast_fp16")];
            tensor<fp16, []> var_839_to_fp16 = const()[name = tensor<string, []>("op_839_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_840_cast_fp16 = mul(x = linear_27_cast_fp16, y = var_839_to_fp16)[name = tensor<string, []>("op_840_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_173_cast_fp16 = add(x = input_161_cast_fp16, y = var_840_cast_fp16)[name = tensor<string, []>("input_173_cast_fp16")];
            tensor<int32, [1]> input_175_axes_0 = const()[name = tensor<string, []>("input_175_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_2_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_norm_out_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20888000)))];
            tensor<fp16, [512]> encoder_layers_2_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_2_norm_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20889088)))];
            tensor<fp16, [1, 125, 512]> input_175_cast_fp16 = layer_norm(axes = input_175_axes_0, beta = encoder_layers_2_norm_out_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_2_norm_out_weight_to_fp16, x = input_173_cast_fp16)[name = tensor<string, []>("input_175_cast_fp16")];
            tensor<int32, [1]> input_177_axes_0 = const()[name = tensor<string, []>("input_177_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_3_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20890176)))];
            tensor<fp16, [512]> encoder_layers_3_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20891264)))];
            tensor<fp16, [1, 125, 512]> input_177_cast_fp16 = layer_norm(axes = input_177_axes_0, beta = encoder_layers_3_norm_feed_forward1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_3_norm_feed_forward1_weight_to_fp16, x = input_175_cast_fp16)[name = tensor<string, []>("input_177_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_3_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_3_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(20892352))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21940992))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_3_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21945152)))];
            tensor<fp16, [1, 125, 2048]> linear_28_cast_fp16 = linear(bias = encoder_layers_3_feed_forward1_linear1_bias_to_fp16, weight = encoder_layers_3_feed_forward1_linear1_weight_to_fp16_quantized, x = input_177_cast_fp16)[name = tensor<string, []>("linear_28_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_181_cast_fp16 = silu(x = linear_28_cast_fp16)[name = tensor<string, []>("input_181_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_3_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_3_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(21949312))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22997952))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_3_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(22999040)))];
            tensor<fp16, [1, 125, 512]> linear_29_cast_fp16 = linear(bias = encoder_layers_3_feed_forward1_linear2_bias_to_fp16, weight = encoder_layers_3_feed_forward1_linear2_weight_to_fp16_quantized, x = input_181_cast_fp16)[name = tensor<string, []>("linear_29_cast_fp16")];
            tensor<fp16, []> var_870_to_fp16 = const()[name = tensor<string, []>("op_870_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_871_cast_fp16 = mul(x = linear_29_cast_fp16, y = var_870_to_fp16)[name = tensor<string, []>("op_871_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_187_cast_fp16 = add(x = input_175_cast_fp16, y = var_871_cast_fp16)[name = tensor<string, []>("input_187_cast_fp16")];
            tensor<int32, [1]> query_7_axes_0 = const()[name = tensor<string, []>("query_7_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_3_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_norm_self_att_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23000128)))];
            tensor<fp16, [512]> encoder_layers_3_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_norm_self_att_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23001216)))];
            tensor<fp16, [1, 125, 512]> query_7_cast_fp16 = layer_norm(axes = query_7_axes_0, beta = encoder_layers_3_norm_self_att_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_3_norm_self_att_weight_to_fp16, x = input_187_cast_fp16)[name = tensor<string, []>("query_7_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_3_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_3_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23002304))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23264512))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_3_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23265600)))];
            tensor<fp16, [1, 125, 512]> linear_30_cast_fp16 = linear(bias = encoder_layers_3_self_attn_linear_q_bias_to_fp16, weight = encoder_layers_3_self_attn_linear_q_weight_to_fp16_quantized, x = query_7_cast_fp16)[name = tensor<string, []>("linear_30_cast_fp16")];
            tensor<int32, [4]> var_888 = const()[name = tensor<string, []>("op_888"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> q_19_cast_fp16 = reshape(shape = var_888, x = linear_30_cast_fp16)[name = tensor<string, []>("q_19_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_3_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_3_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23266688))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23528896))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_3_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23529984)))];
            tensor<fp16, [1, 125, 512]> linear_31_cast_fp16 = linear(bias = encoder_layers_3_self_attn_linear_k_bias_to_fp16, weight = encoder_layers_3_self_attn_linear_k_weight_to_fp16_quantized, x = query_7_cast_fp16)[name = tensor<string, []>("linear_31_cast_fp16")];
            tensor<int32, [4]> var_893 = const()[name = tensor<string, []>("op_893"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> k_13_cast_fp16 = reshape(shape = var_893, x = linear_31_cast_fp16)[name = tensor<string, []>("k_13_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_3_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_3_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23531072))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23793280))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_3_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23794368)))];
            tensor<fp16, [1, 125, 512]> linear_32_cast_fp16 = linear(bias = encoder_layers_3_self_attn_linear_v_bias_to_fp16, weight = encoder_layers_3_self_attn_linear_v_weight_to_fp16_quantized, x = query_7_cast_fp16)[name = tensor<string, []>("linear_32_cast_fp16")];
            tensor<int32, [4]> var_898 = const()[name = tensor<string, []>("op_898"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> v_7_cast_fp16 = reshape(shape = var_898, x = linear_32_cast_fp16)[name = tensor<string, []>("v_7_cast_fp16")];
            tensor<int32, [4]> value_7_perm_0 = const()[name = tensor<string, []>("value_7_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 64]> encoder_layers_3_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23795456)))];
            tensor<fp16, [1, 125, 8, 64]> var_910_cast_fp16 = add(x = q_19_cast_fp16, y = encoder_layers_3_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_910_cast_fp16")];
            tensor<fp16, [8, 64]> encoder_layers_3_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23796544)))];
            tensor<fp16, [1, 125, 8, 64]> var_912_cast_fp16 = add(x = q_19_cast_fp16, y = encoder_layers_3_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_912_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_7_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_7_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<bool, []> x_73_transpose_x_0 = const()[name = tensor<string, []>("x_73_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_73_transpose_y_0 = const()[name = tensor<string, []>("x_73_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 64, 249]> op_914_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(3), name = tensor<string, []>("op_914_to_fp16_quantized"), quantized_data = tensor<int8, [1, 8, 64, 249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23797632))), scale = tensor<fp16, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23925184))), zero_point = tensor<int8, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5281536)))];
            tensor<fp16, [1, 8, 125, 64]> q_with_bias_v_7_cast_fp16 = transpose(perm = q_with_bias_v_7_perm_0, x = var_912_cast_fp16)[name = tensor<string, []>("transpose_199")];
            tensor<fp16, [1, 8, 125, 249]> x_73_cast_fp16 = matmul(transpose_x = x_73_transpose_x_0, transpose_y = x_73_transpose_y_0, x = q_with_bias_v_7_cast_fp16, y = op_914_to_fp16_quantized)[name = tensor<string, []>("x_73_cast_fp16")];
            tensor<int32, [8]> x_75_pad_0 = const()[name = tensor<string, []>("x_75_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            tensor<string, []> x_75_mode_0 = const()[name = tensor<string, []>("x_75_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_100_to_fp16 = const()[name = tensor<string, []>("const_100_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 8, 125, 250]> x_75_cast_fp16 = pad(constant_val = const_100_to_fp16, mode = x_75_mode_0, pad = x_75_pad_0, x = x_73_cast_fp16)[name = tensor<string, []>("x_75_cast_fp16")];
            tensor<int32, [4]> var_922 = const()[name = tensor<string, []>("op_922"), val = tensor<int32, [4]>([1, 8, -1, 125])];
            tensor<fp16, [1, 8, 250, 125]> x_77_cast_fp16 = reshape(shape = var_922, x = x_75_cast_fp16)[name = tensor<string, []>("x_77_cast_fp16")];
            tensor<int32, [4]> var_926_begin_0 = const()[name = tensor<string, []>("op_926_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_926_end_0 = const()[name = tensor<string, []>("op_926_end_0"), val = tensor<int32, [4]>([1, 8, 250, 125])];
            tensor<bool, [4]> var_926_end_mask_0 = const()[name = tensor<string, []>("op_926_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 249, 125]> var_926_cast_fp16 = slice_by_index(begin = var_926_begin_0, end = var_926_end_0, end_mask = var_926_end_mask_0, x = x_77_cast_fp16)[name = tensor<string, []>("op_926_cast_fp16")];
            tensor<int32, [4]> var_927 = const()[name = tensor<string, []>("op_927"), val = tensor<int32, [4]>([1, 8, 125, 249])];
            tensor<fp16, [1, 8, 125, 249]> matrix_bd_13_cast_fp16 = reshape(shape = var_927, x = var_926_cast_fp16)[name = tensor<string, []>("matrix_bd_13_cast_fp16")];
            tensor<bool, []> matrix_ac_7_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_7_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matrix_ac_7_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_7_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_74_perm_0 = const()[name = tensor<string, []>("transpose_74_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_75_perm_0 = const()[name = tensor<string, []>("transpose_75_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 64, 125]> transpose_75 = transpose(perm = transpose_75_perm_0, x = k_13_cast_fp16)[name = tensor<string, []>("transpose_197")];
            tensor<fp16, [1, 8, 125, 64]> transpose_74 = transpose(perm = transpose_74_perm_0, x = var_910_cast_fp16)[name = tensor<string, []>("transpose_198")];
            tensor<fp16, [1, 8, 125, 125]> matrix_ac_7_cast_fp16 = matmul(transpose_x = matrix_ac_7_transpose_x_0, transpose_y = matrix_ac_7_transpose_y_0, x = transpose_74, y = transpose_75)[name = tensor<string, []>("matrix_ac_7_cast_fp16")];
            tensor<int32, [4]> matrix_bd_15_begin_0 = const()[name = tensor<string, []>("matrix_bd_15_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_15_end_0 = const()[name = tensor<string, []>("matrix_bd_15_end_0"), val = tensor<int32, [4]>([1, 8, 125, 125])];
            tensor<bool, [4]> matrix_bd_15_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_15_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 125, 125]> matrix_bd_15_cast_fp16 = slice_by_index(begin = matrix_bd_15_begin_0, end = matrix_bd_15_end_0, end_mask = matrix_bd_15_end_mask_0, x = matrix_bd_13_cast_fp16)[name = tensor<string, []>("matrix_bd_15_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_936_cast_fp16 = add(x = matrix_ac_7_cast_fp16, y = matrix_bd_15_cast_fp16)[name = tensor<string, []>("op_936_cast_fp16")];
            tensor<fp16, []> _inversed_scores_13_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_13_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 8, 125, 125]> _inversed_scores_13_cast_fp16 = mul(x = var_936_cast_fp16, y = _inversed_scores_13_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_13_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> scores_15_cast_fp16 = select(a = var_8_to_fp16, b = _inversed_scores_13_cast_fp16, cond = mask_11)[name = tensor<string, []>("scores_15_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_942_cast_fp16 = softmax(axis = var_19, x = scores_15_cast_fp16)[name = tensor<string, []>("op_942_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> input_189_cast_fp16 = select(a = var_7_to_fp16, b = var_942_cast_fp16, cond = mask_11)[name = tensor<string, []>("input_189_cast_fp16")];
            tensor<bool, []> x_79_transpose_x_0 = const()[name = tensor<string, []>("x_79_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_79_transpose_y_0 = const()[name = tensor<string, []>("x_79_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 125, 64]> value_7_cast_fp16 = transpose(perm = value_7_perm_0, x = v_7_cast_fp16)[name = tensor<string, []>("transpose_196")];
            tensor<fp16, [1, 8, 125, 64]> x_79_cast_fp16 = matmul(transpose_x = x_79_transpose_x_0, transpose_y = x_79_transpose_y_0, x = input_189_cast_fp16, y = value_7_cast_fp16)[name = tensor<string, []>("x_79_cast_fp16")];
            tensor<int32, [4]> var_946_perm_0 = const()[name = tensor<string, []>("op_946_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_947 = const()[name = tensor<string, []>("op_947"), val = tensor<int32, [3]>([1, -1, 512])];
            tensor<fp16, [1, 125, 8, 64]> var_946_cast_fp16 = transpose(perm = var_946_perm_0, x = x_79_cast_fp16)[name = tensor<string, []>("transpose_195")];
            tensor<fp16, [1, 125, 512]> input_191_cast_fp16 = reshape(shape = var_947, x = var_946_cast_fp16)[name = tensor<string, []>("input_191_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_3_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_3_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(23925760))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24187968))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_3_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24189056)))];
            tensor<fp16, [1, 125, 512]> linear_34_cast_fp16 = linear(bias = encoder_layers_3_self_attn_linear_out_bias_to_fp16, weight = encoder_layers_3_self_attn_linear_out_weight_to_fp16_quantized, x = input_191_cast_fp16)[name = tensor<string, []>("linear_34_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_195_cast_fp16 = add(x = input_187_cast_fp16, y = linear_34_cast_fp16)[name = tensor<string, []>("input_195_cast_fp16")];
            tensor<int32, [1]> x_83_axes_0 = const()[name = tensor<string, []>("x_83_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_3_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_norm_conv_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24190144)))];
            tensor<fp16, [512]> encoder_layers_3_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_norm_conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24191232)))];
            tensor<fp16, [1, 125, 512]> x_83_cast_fp16 = layer_norm(axes = x_83_axes_0, beta = encoder_layers_3_norm_conv_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_3_norm_conv_weight_to_fp16, x = input_195_cast_fp16)[name = tensor<string, []>("x_83_cast_fp16")];
            tensor<int32, [3]> input_197_perm_0 = const()[name = tensor<string, []>("input_197_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> input_199_pad_type_0 = const()[name = tensor<string, []>("input_199_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> input_199_strides_0 = const()[name = tensor<string, []>("input_199_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_199_pad_0 = const()[name = tensor<string, []>("input_199_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_199_dilations_0 = const()[name = tensor<string, []>("input_199_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> input_199_groups_0 = const()[name = tensor<string, []>("input_199_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 512, 1]> encoder_layers_3_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_3_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1024, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24192320))), scale = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24716672))), zero_point = tensor<int8, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6073344)))];
            tensor<fp16, [1024]> encoder_layers_3_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24718784)))];
            tensor<fp16, [1, 512, 125]> input_197_cast_fp16 = transpose(perm = input_197_perm_0, x = x_83_cast_fp16)[name = tensor<string, []>("transpose_194")];
            tensor<fp16, [1, 1024, 125]> input_199_cast_fp16 = conv(bias = encoder_layers_3_conv_pointwise_conv1_bias_to_fp16, dilations = input_199_dilations_0, groups = input_199_groups_0, pad = input_199_pad_0, pad_type = input_199_pad_type_0, strides = input_199_strides_0, weight = encoder_layers_3_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_197_cast_fp16)[name = tensor<string, []>("input_199_cast_fp16")];
            tensor<int32, []> x_85_split_num_splits_0 = const()[name = tensor<string, []>("x_85_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> x_85_split_axis_0 = const()[name = tensor<string, []>("x_85_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 512, 125]> x_85_split_cast_fp16_0, tensor<fp16, [1, 512, 125]> x_85_split_cast_fp16_1 = split(axis = x_85_split_axis_0, num_splits = x_85_split_num_splits_0, x = input_199_cast_fp16)[name = tensor<string, []>("x_85_split_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_85_split_1_sigmoid_cast_fp16 = sigmoid(x = x_85_split_cast_fp16_1)[name = tensor<string, []>("x_85_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_85_cast_fp16 = mul(x = x_85_split_cast_fp16_0, y = x_85_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_85_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_201_cast_fp16 = select(a = var_7_to_fp16, b = x_85_cast_fp16, cond = var_449)[name = tensor<string, []>("input_201_cast_fp16")];
            tensor<int32, [6]> input_203_pad_0 = const()[name = tensor<string, []>("input_203_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            tensor<string, []> input_203_mode_0 = const()[name = tensor<string, []>("input_203_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_103_to_fp16 = const()[name = tensor<string, []>("const_103_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 512, 133]> input_203_cast_fp16 = pad(constant_val = const_103_to_fp16, mode = input_203_mode_0, pad = input_203_pad_0, x = input_201_cast_fp16)[name = tensor<string, []>("input_203_cast_fp16")];
            tensor<string, []> input_205_pad_type_0 = const()[name = tensor<string, []>("input_205_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, []> input_205_groups_0 = const()[name = tensor<string, []>("input_205_groups_0"), val = tensor<int32, []>(512)];
            tensor<int32, [1]> input_205_strides_0 = const()[name = tensor<string, []>("input_205_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_205_pad_0 = const()[name = tensor<string, []>("input_205_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_205_dilations_0 = const()[name = tensor<string, []>("input_205_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 9]> const_243_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("const_243_to_fp16_quantized"), quantized_data = tensor<int8, [512, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24720896))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24725568))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> const_244_to_fp16 = const()[name = tensor<string, []>("const_244_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24726656)))];
            tensor<fp16, [1, 512, 125]> input_207_cast_fp16 = conv(bias = const_244_to_fp16, dilations = input_205_dilations_0, groups = input_205_groups_0, pad = input_205_pad_0, pad_type = input_205_pad_type_0, strides = input_205_strides_0, weight = const_243_to_fp16_quantized, x = input_203_cast_fp16)[name = tensor<string, []>("input_207_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_209_cast_fp16 = silu(x = input_207_cast_fp16)[name = tensor<string, []>("input_209_cast_fp16")];
            tensor<string, []> x_87_pad_type_0 = const()[name = tensor<string, []>("x_87_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> x_87_strides_0 = const()[name = tensor<string, []>("x_87_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_87_pad_0 = const()[name = tensor<string, []>("x_87_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_87_dilations_0 = const()[name = tensor<string, []>("x_87_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> x_87_groups_0 = const()[name = tensor<string, []>("x_87_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [512, 512, 1]> encoder_layers_3_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_3_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24727744))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24989952))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_3_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24991040)))];
            tensor<fp16, [1, 512, 125]> x_87_cast_fp16 = conv(bias = encoder_layers_3_conv_pointwise_conv2_bias_to_fp16, dilations = x_87_dilations_0, groups = x_87_groups_0, pad = x_87_pad_0, pad_type = x_87_pad_type_0, strides = x_87_strides_0, weight = encoder_layers_3_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_209_cast_fp16)[name = tensor<string, []>("x_87_cast_fp16")];
            tensor<int32, [3]> input_211_perm_0 = const()[name = tensor<string, []>("input_211_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 125, 512]> input_211_cast_fp16 = transpose(perm = input_211_perm_0, x = x_87_cast_fp16)[name = tensor<string, []>("transpose_193")];
            tensor<fp16, [1, 125, 512]> input_213_cast_fp16 = add(x = input_195_cast_fp16, y = input_211_cast_fp16)[name = tensor<string, []>("input_213_cast_fp16")];
            tensor<int32, [1]> input_215_axes_0 = const()[name = tensor<string, []>("input_215_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_3_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24992128)))];
            tensor<fp16, [512]> encoder_layers_3_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24993216)))];
            tensor<fp16, [1, 125, 512]> input_215_cast_fp16 = layer_norm(axes = input_215_axes_0, beta = encoder_layers_3_norm_feed_forward2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_3_norm_feed_forward2_weight_to_fp16, x = input_213_cast_fp16)[name = tensor<string, []>("input_215_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_3_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_3_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(24994304))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26042944))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_3_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26047104)))];
            tensor<fp16, [1, 125, 2048]> linear_35_cast_fp16 = linear(bias = encoder_layers_3_feed_forward2_linear1_bias_to_fp16, weight = encoder_layers_3_feed_forward2_linear1_weight_to_fp16_quantized, x = input_215_cast_fp16)[name = tensor<string, []>("linear_35_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_219_cast_fp16 = silu(x = linear_35_cast_fp16)[name = tensor<string, []>("input_219_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_3_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_3_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(26051264))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27099904))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_3_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27100992)))];
            tensor<fp16, [1, 125, 512]> linear_36_cast_fp16 = linear(bias = encoder_layers_3_feed_forward2_linear2_bias_to_fp16, weight = encoder_layers_3_feed_forward2_linear2_weight_to_fp16_quantized, x = input_219_cast_fp16)[name = tensor<string, []>("linear_36_cast_fp16")];
            tensor<fp16, []> var_1013_to_fp16 = const()[name = tensor<string, []>("op_1013_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_1014_cast_fp16 = mul(x = linear_36_cast_fp16, y = var_1013_to_fp16)[name = tensor<string, []>("op_1014_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_225_cast_fp16 = add(x = input_213_cast_fp16, y = var_1014_cast_fp16)[name = tensor<string, []>("input_225_cast_fp16")];
            tensor<int32, [1]> input_227_axes_0 = const()[name = tensor<string, []>("input_227_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_3_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_norm_out_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27102080)))];
            tensor<fp16, [512]> encoder_layers_3_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_3_norm_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27103168)))];
            tensor<fp16, [1, 125, 512]> input_227_cast_fp16 = layer_norm(axes = input_227_axes_0, beta = encoder_layers_3_norm_out_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_3_norm_out_weight_to_fp16, x = input_225_cast_fp16)[name = tensor<string, []>("input_227_cast_fp16")];
            tensor<int32, [1]> input_229_axes_0 = const()[name = tensor<string, []>("input_229_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_4_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27104256)))];
            tensor<fp16, [512]> encoder_layers_4_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27105344)))];
            tensor<fp16, [1, 125, 512]> input_229_cast_fp16 = layer_norm(axes = input_229_axes_0, beta = encoder_layers_4_norm_feed_forward1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_4_norm_feed_forward1_weight_to_fp16, x = input_227_cast_fp16)[name = tensor<string, []>("input_229_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_4_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_4_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(27106432))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28155072))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_4_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28159232)))];
            tensor<fp16, [1, 125, 2048]> linear_37_cast_fp16 = linear(bias = encoder_layers_4_feed_forward1_linear1_bias_to_fp16, weight = encoder_layers_4_feed_forward1_linear1_weight_to_fp16_quantized, x = input_229_cast_fp16)[name = tensor<string, []>("linear_37_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_233_cast_fp16 = silu(x = linear_37_cast_fp16)[name = tensor<string, []>("input_233_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_4_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_4_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(28163392))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29212032))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_4_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29213120)))];
            tensor<fp16, [1, 125, 512]> linear_38_cast_fp16 = linear(bias = encoder_layers_4_feed_forward1_linear2_bias_to_fp16, weight = encoder_layers_4_feed_forward1_linear2_weight_to_fp16_quantized, x = input_233_cast_fp16)[name = tensor<string, []>("linear_38_cast_fp16")];
            tensor<fp16, []> var_1044_to_fp16 = const()[name = tensor<string, []>("op_1044_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_1045_cast_fp16 = mul(x = linear_38_cast_fp16, y = var_1044_to_fp16)[name = tensor<string, []>("op_1045_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_239_cast_fp16 = add(x = input_227_cast_fp16, y = var_1045_cast_fp16)[name = tensor<string, []>("input_239_cast_fp16")];
            tensor<int32, [1]> query_9_axes_0 = const()[name = tensor<string, []>("query_9_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_4_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_norm_self_att_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29214208)))];
            tensor<fp16, [512]> encoder_layers_4_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_norm_self_att_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29215296)))];
            tensor<fp16, [1, 125, 512]> query_9_cast_fp16 = layer_norm(axes = query_9_axes_0, beta = encoder_layers_4_norm_self_att_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_4_norm_self_att_weight_to_fp16, x = input_239_cast_fp16)[name = tensor<string, []>("query_9_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_4_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_4_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29216384))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29478592))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_4_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29479680)))];
            tensor<fp16, [1, 125, 512]> linear_39_cast_fp16 = linear(bias = encoder_layers_4_self_attn_linear_q_bias_to_fp16, weight = encoder_layers_4_self_attn_linear_q_weight_to_fp16_quantized, x = query_9_cast_fp16)[name = tensor<string, []>("linear_39_cast_fp16")];
            tensor<int32, [4]> var_1062 = const()[name = tensor<string, []>("op_1062"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> q_25_cast_fp16 = reshape(shape = var_1062, x = linear_39_cast_fp16)[name = tensor<string, []>("q_25_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_4_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_4_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29480768))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29742976))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_4_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29744064)))];
            tensor<fp16, [1, 125, 512]> linear_40_cast_fp16 = linear(bias = encoder_layers_4_self_attn_linear_k_bias_to_fp16, weight = encoder_layers_4_self_attn_linear_k_weight_to_fp16_quantized, x = query_9_cast_fp16)[name = tensor<string, []>("linear_40_cast_fp16")];
            tensor<int32, [4]> var_1067 = const()[name = tensor<string, []>("op_1067"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> k_17_cast_fp16 = reshape(shape = var_1067, x = linear_40_cast_fp16)[name = tensor<string, []>("k_17_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_4_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_4_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(29745152))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30007360))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_4_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30008448)))];
            tensor<fp16, [1, 125, 512]> linear_41_cast_fp16 = linear(bias = encoder_layers_4_self_attn_linear_v_bias_to_fp16, weight = encoder_layers_4_self_attn_linear_v_weight_to_fp16_quantized, x = query_9_cast_fp16)[name = tensor<string, []>("linear_41_cast_fp16")];
            tensor<int32, [4]> var_1072 = const()[name = tensor<string, []>("op_1072"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> v_9_cast_fp16 = reshape(shape = var_1072, x = linear_41_cast_fp16)[name = tensor<string, []>("v_9_cast_fp16")];
            tensor<int32, [4]> value_9_perm_0 = const()[name = tensor<string, []>("value_9_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 64]> encoder_layers_4_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30009536)))];
            tensor<fp16, [1, 125, 8, 64]> var_1084_cast_fp16 = add(x = q_25_cast_fp16, y = encoder_layers_4_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_1084_cast_fp16")];
            tensor<fp16, [8, 64]> encoder_layers_4_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30010624)))];
            tensor<fp16, [1, 125, 8, 64]> var_1086_cast_fp16 = add(x = q_25_cast_fp16, y = encoder_layers_4_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_1086_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_9_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_9_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<bool, []> x_95_transpose_x_0 = const()[name = tensor<string, []>("x_95_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_95_transpose_y_0 = const()[name = tensor<string, []>("x_95_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 64, 249]> op_1088_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(3), name = tensor<string, []>("op_1088_to_fp16_quantized"), quantized_data = tensor<int8, [1, 8, 64, 249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30011712))), scale = tensor<fp16, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30139264))), zero_point = tensor<int8, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5281536)))];
            tensor<fp16, [1, 8, 125, 64]> q_with_bias_v_9_cast_fp16 = transpose(perm = q_with_bias_v_9_perm_0, x = var_1086_cast_fp16)[name = tensor<string, []>("transpose_192")];
            tensor<fp16, [1, 8, 125, 249]> x_95_cast_fp16 = matmul(transpose_x = x_95_transpose_x_0, transpose_y = x_95_transpose_y_0, x = q_with_bias_v_9_cast_fp16, y = op_1088_to_fp16_quantized)[name = tensor<string, []>("x_95_cast_fp16")];
            tensor<int32, [8]> x_97_pad_0 = const()[name = tensor<string, []>("x_97_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            tensor<string, []> x_97_mode_0 = const()[name = tensor<string, []>("x_97_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_110_to_fp16 = const()[name = tensor<string, []>("const_110_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 8, 125, 250]> x_97_cast_fp16 = pad(constant_val = const_110_to_fp16, mode = x_97_mode_0, pad = x_97_pad_0, x = x_95_cast_fp16)[name = tensor<string, []>("x_97_cast_fp16")];
            tensor<int32, [4]> var_1096 = const()[name = tensor<string, []>("op_1096"), val = tensor<int32, [4]>([1, 8, -1, 125])];
            tensor<fp16, [1, 8, 250, 125]> x_99_cast_fp16 = reshape(shape = var_1096, x = x_97_cast_fp16)[name = tensor<string, []>("x_99_cast_fp16")];
            tensor<int32, [4]> var_1100_begin_0 = const()[name = tensor<string, []>("op_1100_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_1100_end_0 = const()[name = tensor<string, []>("op_1100_end_0"), val = tensor<int32, [4]>([1, 8, 250, 125])];
            tensor<bool, [4]> var_1100_end_mask_0 = const()[name = tensor<string, []>("op_1100_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 249, 125]> var_1100_cast_fp16 = slice_by_index(begin = var_1100_begin_0, end = var_1100_end_0, end_mask = var_1100_end_mask_0, x = x_99_cast_fp16)[name = tensor<string, []>("op_1100_cast_fp16")];
            tensor<int32, [4]> var_1101 = const()[name = tensor<string, []>("op_1101"), val = tensor<int32, [4]>([1, 8, 125, 249])];
            tensor<fp16, [1, 8, 125, 249]> matrix_bd_17_cast_fp16 = reshape(shape = var_1101, x = var_1100_cast_fp16)[name = tensor<string, []>("matrix_bd_17_cast_fp16")];
            tensor<bool, []> matrix_ac_9_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_9_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matrix_ac_9_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_9_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_76_perm_0 = const()[name = tensor<string, []>("transpose_76_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_77_perm_0 = const()[name = tensor<string, []>("transpose_77_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 64, 125]> transpose_77 = transpose(perm = transpose_77_perm_0, x = k_17_cast_fp16)[name = tensor<string, []>("transpose_190")];
            tensor<fp16, [1, 8, 125, 64]> transpose_76 = transpose(perm = transpose_76_perm_0, x = var_1084_cast_fp16)[name = tensor<string, []>("transpose_191")];
            tensor<fp16, [1, 8, 125, 125]> matrix_ac_9_cast_fp16 = matmul(transpose_x = matrix_ac_9_transpose_x_0, transpose_y = matrix_ac_9_transpose_y_0, x = transpose_76, y = transpose_77)[name = tensor<string, []>("matrix_ac_9_cast_fp16")];
            tensor<int32, [4]> matrix_bd_19_begin_0 = const()[name = tensor<string, []>("matrix_bd_19_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_19_end_0 = const()[name = tensor<string, []>("matrix_bd_19_end_0"), val = tensor<int32, [4]>([1, 8, 125, 125])];
            tensor<bool, [4]> matrix_bd_19_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_19_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 125, 125]> matrix_bd_19_cast_fp16 = slice_by_index(begin = matrix_bd_19_begin_0, end = matrix_bd_19_end_0, end_mask = matrix_bd_19_end_mask_0, x = matrix_bd_17_cast_fp16)[name = tensor<string, []>("matrix_bd_19_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_1110_cast_fp16 = add(x = matrix_ac_9_cast_fp16, y = matrix_bd_19_cast_fp16)[name = tensor<string, []>("op_1110_cast_fp16")];
            tensor<fp16, []> _inversed_scores_17_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_17_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 8, 125, 125]> _inversed_scores_17_cast_fp16 = mul(x = var_1110_cast_fp16, y = _inversed_scores_17_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_17_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> scores_19_cast_fp16 = select(a = var_8_to_fp16, b = _inversed_scores_17_cast_fp16, cond = mask_11)[name = tensor<string, []>("scores_19_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_1116_cast_fp16 = softmax(axis = var_19, x = scores_19_cast_fp16)[name = tensor<string, []>("op_1116_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> input_241_cast_fp16 = select(a = var_7_to_fp16, b = var_1116_cast_fp16, cond = mask_11)[name = tensor<string, []>("input_241_cast_fp16")];
            tensor<bool, []> x_101_transpose_x_0 = const()[name = tensor<string, []>("x_101_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_101_transpose_y_0 = const()[name = tensor<string, []>("x_101_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 125, 64]> value_9_cast_fp16 = transpose(perm = value_9_perm_0, x = v_9_cast_fp16)[name = tensor<string, []>("transpose_189")];
            tensor<fp16, [1, 8, 125, 64]> x_101_cast_fp16 = matmul(transpose_x = x_101_transpose_x_0, transpose_y = x_101_transpose_y_0, x = input_241_cast_fp16, y = value_9_cast_fp16)[name = tensor<string, []>("x_101_cast_fp16")];
            tensor<int32, [4]> var_1120_perm_0 = const()[name = tensor<string, []>("op_1120_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1121 = const()[name = tensor<string, []>("op_1121"), val = tensor<int32, [3]>([1, -1, 512])];
            tensor<fp16, [1, 125, 8, 64]> var_1120_cast_fp16 = transpose(perm = var_1120_perm_0, x = x_101_cast_fp16)[name = tensor<string, []>("transpose_188")];
            tensor<fp16, [1, 125, 512]> input_243_cast_fp16 = reshape(shape = var_1121, x = var_1120_cast_fp16)[name = tensor<string, []>("input_243_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_4_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_4_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30139840))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30402048))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_4_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30403136)))];
            tensor<fp16, [1, 125, 512]> linear_43_cast_fp16 = linear(bias = encoder_layers_4_self_attn_linear_out_bias_to_fp16, weight = encoder_layers_4_self_attn_linear_out_weight_to_fp16_quantized, x = input_243_cast_fp16)[name = tensor<string, []>("linear_43_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_247_cast_fp16 = add(x = input_239_cast_fp16, y = linear_43_cast_fp16)[name = tensor<string, []>("input_247_cast_fp16")];
            tensor<int32, [1]> x_105_axes_0 = const()[name = tensor<string, []>("x_105_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_4_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_norm_conv_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30404224)))];
            tensor<fp16, [512]> encoder_layers_4_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_norm_conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30405312)))];
            tensor<fp16, [1, 125, 512]> x_105_cast_fp16 = layer_norm(axes = x_105_axes_0, beta = encoder_layers_4_norm_conv_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_4_norm_conv_weight_to_fp16, x = input_247_cast_fp16)[name = tensor<string, []>("x_105_cast_fp16")];
            tensor<int32, [3]> input_249_perm_0 = const()[name = tensor<string, []>("input_249_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> input_251_pad_type_0 = const()[name = tensor<string, []>("input_251_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> input_251_strides_0 = const()[name = tensor<string, []>("input_251_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_251_pad_0 = const()[name = tensor<string, []>("input_251_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_251_dilations_0 = const()[name = tensor<string, []>("input_251_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> input_251_groups_0 = const()[name = tensor<string, []>("input_251_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 512, 1]> encoder_layers_4_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_4_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1024, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30406400))), scale = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30930752))), zero_point = tensor<int8, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6073344)))];
            tensor<fp16, [1024]> encoder_layers_4_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30932864)))];
            tensor<fp16, [1, 512, 125]> input_249_cast_fp16 = transpose(perm = input_249_perm_0, x = x_105_cast_fp16)[name = tensor<string, []>("transpose_187")];
            tensor<fp16, [1, 1024, 125]> input_251_cast_fp16 = conv(bias = encoder_layers_4_conv_pointwise_conv1_bias_to_fp16, dilations = input_251_dilations_0, groups = input_251_groups_0, pad = input_251_pad_0, pad_type = input_251_pad_type_0, strides = input_251_strides_0, weight = encoder_layers_4_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_249_cast_fp16)[name = tensor<string, []>("input_251_cast_fp16")];
            tensor<int32, []> x_107_split_num_splits_0 = const()[name = tensor<string, []>("x_107_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> x_107_split_axis_0 = const()[name = tensor<string, []>("x_107_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 512, 125]> x_107_split_cast_fp16_0, tensor<fp16, [1, 512, 125]> x_107_split_cast_fp16_1 = split(axis = x_107_split_axis_0, num_splits = x_107_split_num_splits_0, x = input_251_cast_fp16)[name = tensor<string, []>("x_107_split_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_107_split_1_sigmoid_cast_fp16 = sigmoid(x = x_107_split_cast_fp16_1)[name = tensor<string, []>("x_107_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_107_cast_fp16 = mul(x = x_107_split_cast_fp16_0, y = x_107_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_107_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_253_cast_fp16 = select(a = var_7_to_fp16, b = x_107_cast_fp16, cond = var_449)[name = tensor<string, []>("input_253_cast_fp16")];
            tensor<int32, [6]> input_255_pad_0 = const()[name = tensor<string, []>("input_255_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            tensor<string, []> input_255_mode_0 = const()[name = tensor<string, []>("input_255_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_113_to_fp16 = const()[name = tensor<string, []>("const_113_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 512, 133]> input_255_cast_fp16 = pad(constant_val = const_113_to_fp16, mode = input_255_mode_0, pad = input_255_pad_0, x = input_253_cast_fp16)[name = tensor<string, []>("input_255_cast_fp16")];
            tensor<string, []> input_257_pad_type_0 = const()[name = tensor<string, []>("input_257_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, []> input_257_groups_0 = const()[name = tensor<string, []>("input_257_groups_0"), val = tensor<int32, []>(512)];
            tensor<int32, [1]> input_257_strides_0 = const()[name = tensor<string, []>("input_257_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_257_pad_0 = const()[name = tensor<string, []>("input_257_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_257_dilations_0 = const()[name = tensor<string, []>("input_257_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 9]> const_245_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("const_245_to_fp16_quantized"), quantized_data = tensor<int8, [512, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30934976))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30939648))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> const_246_to_fp16 = const()[name = tensor<string, []>("const_246_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30940736)))];
            tensor<fp16, [1, 512, 125]> input_259_cast_fp16 = conv(bias = const_246_to_fp16, dilations = input_257_dilations_0, groups = input_257_groups_0, pad = input_257_pad_0, pad_type = input_257_pad_type_0, strides = input_257_strides_0, weight = const_245_to_fp16_quantized, x = input_255_cast_fp16)[name = tensor<string, []>("input_259_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_261_cast_fp16 = silu(x = input_259_cast_fp16)[name = tensor<string, []>("input_261_cast_fp16")];
            tensor<string, []> x_109_pad_type_0 = const()[name = tensor<string, []>("x_109_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> x_109_strides_0 = const()[name = tensor<string, []>("x_109_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_109_pad_0 = const()[name = tensor<string, []>("x_109_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_109_dilations_0 = const()[name = tensor<string, []>("x_109_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> x_109_groups_0 = const()[name = tensor<string, []>("x_109_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [512, 512, 1]> encoder_layers_4_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_4_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(30941824))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31204032))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_4_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31205120)))];
            tensor<fp16, [1, 512, 125]> x_109_cast_fp16 = conv(bias = encoder_layers_4_conv_pointwise_conv2_bias_to_fp16, dilations = x_109_dilations_0, groups = x_109_groups_0, pad = x_109_pad_0, pad_type = x_109_pad_type_0, strides = x_109_strides_0, weight = encoder_layers_4_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_261_cast_fp16)[name = tensor<string, []>("x_109_cast_fp16")];
            tensor<int32, [3]> input_263_perm_0 = const()[name = tensor<string, []>("input_263_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 125, 512]> input_263_cast_fp16 = transpose(perm = input_263_perm_0, x = x_109_cast_fp16)[name = tensor<string, []>("transpose_186")];
            tensor<fp16, [1, 125, 512]> input_265_cast_fp16 = add(x = input_247_cast_fp16, y = input_263_cast_fp16)[name = tensor<string, []>("input_265_cast_fp16")];
            tensor<int32, [1]> input_267_axes_0 = const()[name = tensor<string, []>("input_267_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_4_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31206208)))];
            tensor<fp16, [512]> encoder_layers_4_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31207296)))];
            tensor<fp16, [1, 125, 512]> input_267_cast_fp16 = layer_norm(axes = input_267_axes_0, beta = encoder_layers_4_norm_feed_forward2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_4_norm_feed_forward2_weight_to_fp16, x = input_265_cast_fp16)[name = tensor<string, []>("input_267_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_4_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_4_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(31208384))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32257024))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_4_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32261184)))];
            tensor<fp16, [1, 125, 2048]> linear_44_cast_fp16 = linear(bias = encoder_layers_4_feed_forward2_linear1_bias_to_fp16, weight = encoder_layers_4_feed_forward2_linear1_weight_to_fp16_quantized, x = input_267_cast_fp16)[name = tensor<string, []>("linear_44_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_271_cast_fp16 = silu(x = linear_44_cast_fp16)[name = tensor<string, []>("input_271_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_4_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_4_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(32265344))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33313984))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_4_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33315072)))];
            tensor<fp16, [1, 125, 512]> linear_45_cast_fp16 = linear(bias = encoder_layers_4_feed_forward2_linear2_bias_to_fp16, weight = encoder_layers_4_feed_forward2_linear2_weight_to_fp16_quantized, x = input_271_cast_fp16)[name = tensor<string, []>("linear_45_cast_fp16")];
            tensor<fp16, []> var_1187_to_fp16 = const()[name = tensor<string, []>("op_1187_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_1188_cast_fp16 = mul(x = linear_45_cast_fp16, y = var_1187_to_fp16)[name = tensor<string, []>("op_1188_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_277_cast_fp16 = add(x = input_265_cast_fp16, y = var_1188_cast_fp16)[name = tensor<string, []>("input_277_cast_fp16")];
            tensor<int32, [1]> input_279_axes_0 = const()[name = tensor<string, []>("input_279_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_4_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_norm_out_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33316160)))];
            tensor<fp16, [512]> encoder_layers_4_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_4_norm_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33317248)))];
            tensor<fp16, [1, 125, 512]> input_279_cast_fp16 = layer_norm(axes = input_279_axes_0, beta = encoder_layers_4_norm_out_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_4_norm_out_weight_to_fp16, x = input_277_cast_fp16)[name = tensor<string, []>("input_279_cast_fp16")];
            tensor<int32, [1]> input_281_axes_0 = const()[name = tensor<string, []>("input_281_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_5_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33318336)))];
            tensor<fp16, [512]> encoder_layers_5_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33319424)))];
            tensor<fp16, [1, 125, 512]> input_281_cast_fp16 = layer_norm(axes = input_281_axes_0, beta = encoder_layers_5_norm_feed_forward1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_5_norm_feed_forward1_weight_to_fp16, x = input_279_cast_fp16)[name = tensor<string, []>("input_281_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_5_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_5_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(33320512))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34369152))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_5_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34373312)))];
            tensor<fp16, [1, 125, 2048]> linear_46_cast_fp16 = linear(bias = encoder_layers_5_feed_forward1_linear1_bias_to_fp16, weight = encoder_layers_5_feed_forward1_linear1_weight_to_fp16_quantized, x = input_281_cast_fp16)[name = tensor<string, []>("linear_46_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_285_cast_fp16 = silu(x = linear_46_cast_fp16)[name = tensor<string, []>("input_285_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_5_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_5_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(34377472))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35426112))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_5_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35427200)))];
            tensor<fp16, [1, 125, 512]> linear_47_cast_fp16 = linear(bias = encoder_layers_5_feed_forward1_linear2_bias_to_fp16, weight = encoder_layers_5_feed_forward1_linear2_weight_to_fp16_quantized, x = input_285_cast_fp16)[name = tensor<string, []>("linear_47_cast_fp16")];
            tensor<fp16, []> var_1218_to_fp16 = const()[name = tensor<string, []>("op_1218_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_1219_cast_fp16 = mul(x = linear_47_cast_fp16, y = var_1218_to_fp16)[name = tensor<string, []>("op_1219_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_291_cast_fp16 = add(x = input_279_cast_fp16, y = var_1219_cast_fp16)[name = tensor<string, []>("input_291_cast_fp16")];
            tensor<int32, [1]> query_11_axes_0 = const()[name = tensor<string, []>("query_11_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_5_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_norm_self_att_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35428288)))];
            tensor<fp16, [512]> encoder_layers_5_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_norm_self_att_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35429376)))];
            tensor<fp16, [1, 125, 512]> query_11_cast_fp16 = layer_norm(axes = query_11_axes_0, beta = encoder_layers_5_norm_self_att_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_5_norm_self_att_weight_to_fp16, x = input_291_cast_fp16)[name = tensor<string, []>("query_11_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_5_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_5_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35430464))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35692672))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_5_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35693760)))];
            tensor<fp16, [1, 125, 512]> linear_48_cast_fp16 = linear(bias = encoder_layers_5_self_attn_linear_q_bias_to_fp16, weight = encoder_layers_5_self_attn_linear_q_weight_to_fp16_quantized, x = query_11_cast_fp16)[name = tensor<string, []>("linear_48_cast_fp16")];
            tensor<int32, [4]> var_1236 = const()[name = tensor<string, []>("op_1236"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> q_31_cast_fp16 = reshape(shape = var_1236, x = linear_48_cast_fp16)[name = tensor<string, []>("q_31_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_5_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_5_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35694848))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35957056))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_5_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35958144)))];
            tensor<fp16, [1, 125, 512]> linear_49_cast_fp16 = linear(bias = encoder_layers_5_self_attn_linear_k_bias_to_fp16, weight = encoder_layers_5_self_attn_linear_k_weight_to_fp16_quantized, x = query_11_cast_fp16)[name = tensor<string, []>("linear_49_cast_fp16")];
            tensor<int32, [4]> var_1241 = const()[name = tensor<string, []>("op_1241"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> k_21_cast_fp16 = reshape(shape = var_1241, x = linear_49_cast_fp16)[name = tensor<string, []>("k_21_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_5_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_5_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(35959232))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36221440))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_5_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36222528)))];
            tensor<fp16, [1, 125, 512]> linear_50_cast_fp16 = linear(bias = encoder_layers_5_self_attn_linear_v_bias_to_fp16, weight = encoder_layers_5_self_attn_linear_v_weight_to_fp16_quantized, x = query_11_cast_fp16)[name = tensor<string, []>("linear_50_cast_fp16")];
            tensor<int32, [4]> var_1246 = const()[name = tensor<string, []>("op_1246"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> v_11_cast_fp16 = reshape(shape = var_1246, x = linear_50_cast_fp16)[name = tensor<string, []>("v_11_cast_fp16")];
            tensor<int32, [4]> value_11_perm_0 = const()[name = tensor<string, []>("value_11_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 64]> encoder_layers_5_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36223616)))];
            tensor<fp16, [1, 125, 8, 64]> var_1258_cast_fp16 = add(x = q_31_cast_fp16, y = encoder_layers_5_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_1258_cast_fp16")];
            tensor<fp16, [8, 64]> encoder_layers_5_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36224704)))];
            tensor<fp16, [1, 125, 8, 64]> var_1260_cast_fp16 = add(x = q_31_cast_fp16, y = encoder_layers_5_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_1260_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_11_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_11_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<bool, []> x_117_transpose_x_0 = const()[name = tensor<string, []>("x_117_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_117_transpose_y_0 = const()[name = tensor<string, []>("x_117_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 64, 249]> op_1262_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(3), name = tensor<string, []>("op_1262_to_fp16_quantized"), quantized_data = tensor<int8, [1, 8, 64, 249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36225792))), scale = tensor<fp16, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36353344))), zero_point = tensor<int8, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5281536)))];
            tensor<fp16, [1, 8, 125, 64]> q_with_bias_v_11_cast_fp16 = transpose(perm = q_with_bias_v_11_perm_0, x = var_1260_cast_fp16)[name = tensor<string, []>("transpose_185")];
            tensor<fp16, [1, 8, 125, 249]> x_117_cast_fp16 = matmul(transpose_x = x_117_transpose_x_0, transpose_y = x_117_transpose_y_0, x = q_with_bias_v_11_cast_fp16, y = op_1262_to_fp16_quantized)[name = tensor<string, []>("x_117_cast_fp16")];
            tensor<int32, [8]> x_119_pad_0 = const()[name = tensor<string, []>("x_119_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            tensor<string, []> x_119_mode_0 = const()[name = tensor<string, []>("x_119_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_120_to_fp16 = const()[name = tensor<string, []>("const_120_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 8, 125, 250]> x_119_cast_fp16 = pad(constant_val = const_120_to_fp16, mode = x_119_mode_0, pad = x_119_pad_0, x = x_117_cast_fp16)[name = tensor<string, []>("x_119_cast_fp16")];
            tensor<int32, [4]> var_1270 = const()[name = tensor<string, []>("op_1270"), val = tensor<int32, [4]>([1, 8, -1, 125])];
            tensor<fp16, [1, 8, 250, 125]> x_121_cast_fp16 = reshape(shape = var_1270, x = x_119_cast_fp16)[name = tensor<string, []>("x_121_cast_fp16")];
            tensor<int32, [4]> var_1274_begin_0 = const()[name = tensor<string, []>("op_1274_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_1274_end_0 = const()[name = tensor<string, []>("op_1274_end_0"), val = tensor<int32, [4]>([1, 8, 250, 125])];
            tensor<bool, [4]> var_1274_end_mask_0 = const()[name = tensor<string, []>("op_1274_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 249, 125]> var_1274_cast_fp16 = slice_by_index(begin = var_1274_begin_0, end = var_1274_end_0, end_mask = var_1274_end_mask_0, x = x_121_cast_fp16)[name = tensor<string, []>("op_1274_cast_fp16")];
            tensor<int32, [4]> var_1275 = const()[name = tensor<string, []>("op_1275"), val = tensor<int32, [4]>([1, 8, 125, 249])];
            tensor<fp16, [1, 8, 125, 249]> matrix_bd_21_cast_fp16 = reshape(shape = var_1275, x = var_1274_cast_fp16)[name = tensor<string, []>("matrix_bd_21_cast_fp16")];
            tensor<bool, []> matrix_ac_11_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_11_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matrix_ac_11_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_11_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_78_perm_0 = const()[name = tensor<string, []>("transpose_78_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_79_perm_0 = const()[name = tensor<string, []>("transpose_79_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 64, 125]> transpose_79 = transpose(perm = transpose_79_perm_0, x = k_21_cast_fp16)[name = tensor<string, []>("transpose_183")];
            tensor<fp16, [1, 8, 125, 64]> transpose_78 = transpose(perm = transpose_78_perm_0, x = var_1258_cast_fp16)[name = tensor<string, []>("transpose_184")];
            tensor<fp16, [1, 8, 125, 125]> matrix_ac_11_cast_fp16 = matmul(transpose_x = matrix_ac_11_transpose_x_0, transpose_y = matrix_ac_11_transpose_y_0, x = transpose_78, y = transpose_79)[name = tensor<string, []>("matrix_ac_11_cast_fp16")];
            tensor<int32, [4]> matrix_bd_23_begin_0 = const()[name = tensor<string, []>("matrix_bd_23_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_23_end_0 = const()[name = tensor<string, []>("matrix_bd_23_end_0"), val = tensor<int32, [4]>([1, 8, 125, 125])];
            tensor<bool, [4]> matrix_bd_23_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_23_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 125, 125]> matrix_bd_23_cast_fp16 = slice_by_index(begin = matrix_bd_23_begin_0, end = matrix_bd_23_end_0, end_mask = matrix_bd_23_end_mask_0, x = matrix_bd_21_cast_fp16)[name = tensor<string, []>("matrix_bd_23_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_1284_cast_fp16 = add(x = matrix_ac_11_cast_fp16, y = matrix_bd_23_cast_fp16)[name = tensor<string, []>("op_1284_cast_fp16")];
            tensor<fp16, []> _inversed_scores_21_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_21_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 8, 125, 125]> _inversed_scores_21_cast_fp16 = mul(x = var_1284_cast_fp16, y = _inversed_scores_21_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_21_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> scores_23_cast_fp16 = select(a = var_8_to_fp16, b = _inversed_scores_21_cast_fp16, cond = mask_11)[name = tensor<string, []>("scores_23_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_1290_cast_fp16 = softmax(axis = var_19, x = scores_23_cast_fp16)[name = tensor<string, []>("op_1290_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> input_293_cast_fp16 = select(a = var_7_to_fp16, b = var_1290_cast_fp16, cond = mask_11)[name = tensor<string, []>("input_293_cast_fp16")];
            tensor<bool, []> x_123_transpose_x_0 = const()[name = tensor<string, []>("x_123_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_123_transpose_y_0 = const()[name = tensor<string, []>("x_123_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 125, 64]> value_11_cast_fp16 = transpose(perm = value_11_perm_0, x = v_11_cast_fp16)[name = tensor<string, []>("transpose_182")];
            tensor<fp16, [1, 8, 125, 64]> x_123_cast_fp16 = matmul(transpose_x = x_123_transpose_x_0, transpose_y = x_123_transpose_y_0, x = input_293_cast_fp16, y = value_11_cast_fp16)[name = tensor<string, []>("x_123_cast_fp16")];
            tensor<int32, [4]> var_1294_perm_0 = const()[name = tensor<string, []>("op_1294_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1295 = const()[name = tensor<string, []>("op_1295"), val = tensor<int32, [3]>([1, -1, 512])];
            tensor<fp16, [1, 125, 8, 64]> var_1294_cast_fp16 = transpose(perm = var_1294_perm_0, x = x_123_cast_fp16)[name = tensor<string, []>("transpose_181")];
            tensor<fp16, [1, 125, 512]> input_295_cast_fp16 = reshape(shape = var_1295, x = var_1294_cast_fp16)[name = tensor<string, []>("input_295_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_5_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_5_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36353920))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36616128))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_5_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36617216)))];
            tensor<fp16, [1, 125, 512]> linear_52_cast_fp16 = linear(bias = encoder_layers_5_self_attn_linear_out_bias_to_fp16, weight = encoder_layers_5_self_attn_linear_out_weight_to_fp16_quantized, x = input_295_cast_fp16)[name = tensor<string, []>("linear_52_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_299_cast_fp16 = add(x = input_291_cast_fp16, y = linear_52_cast_fp16)[name = tensor<string, []>("input_299_cast_fp16")];
            tensor<int32, [1]> x_127_axes_0 = const()[name = tensor<string, []>("x_127_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_5_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_norm_conv_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36618304)))];
            tensor<fp16, [512]> encoder_layers_5_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_norm_conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36619392)))];
            tensor<fp16, [1, 125, 512]> x_127_cast_fp16 = layer_norm(axes = x_127_axes_0, beta = encoder_layers_5_norm_conv_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_5_norm_conv_weight_to_fp16, x = input_299_cast_fp16)[name = tensor<string, []>("x_127_cast_fp16")];
            tensor<int32, [3]> input_301_perm_0 = const()[name = tensor<string, []>("input_301_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> input_303_pad_type_0 = const()[name = tensor<string, []>("input_303_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> input_303_strides_0 = const()[name = tensor<string, []>("input_303_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_303_pad_0 = const()[name = tensor<string, []>("input_303_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_303_dilations_0 = const()[name = tensor<string, []>("input_303_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> input_303_groups_0 = const()[name = tensor<string, []>("input_303_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 512, 1]> encoder_layers_5_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_5_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1024, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(36620480))), scale = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37144832))), zero_point = tensor<int8, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6073344)))];
            tensor<fp16, [1024]> encoder_layers_5_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37146944)))];
            tensor<fp16, [1, 512, 125]> input_301_cast_fp16 = transpose(perm = input_301_perm_0, x = x_127_cast_fp16)[name = tensor<string, []>("transpose_180")];
            tensor<fp16, [1, 1024, 125]> input_303_cast_fp16 = conv(bias = encoder_layers_5_conv_pointwise_conv1_bias_to_fp16, dilations = input_303_dilations_0, groups = input_303_groups_0, pad = input_303_pad_0, pad_type = input_303_pad_type_0, strides = input_303_strides_0, weight = encoder_layers_5_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_301_cast_fp16)[name = tensor<string, []>("input_303_cast_fp16")];
            tensor<int32, []> x_129_split_num_splits_0 = const()[name = tensor<string, []>("x_129_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> x_129_split_axis_0 = const()[name = tensor<string, []>("x_129_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 512, 125]> x_129_split_cast_fp16_0, tensor<fp16, [1, 512, 125]> x_129_split_cast_fp16_1 = split(axis = x_129_split_axis_0, num_splits = x_129_split_num_splits_0, x = input_303_cast_fp16)[name = tensor<string, []>("x_129_split_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_129_split_1_sigmoid_cast_fp16 = sigmoid(x = x_129_split_cast_fp16_1)[name = tensor<string, []>("x_129_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_129_cast_fp16 = mul(x = x_129_split_cast_fp16_0, y = x_129_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_129_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_305_cast_fp16 = select(a = var_7_to_fp16, b = x_129_cast_fp16, cond = var_449)[name = tensor<string, []>("input_305_cast_fp16")];
            tensor<int32, [6]> input_307_pad_0 = const()[name = tensor<string, []>("input_307_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            tensor<string, []> input_307_mode_0 = const()[name = tensor<string, []>("input_307_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_123_to_fp16 = const()[name = tensor<string, []>("const_123_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 512, 133]> input_307_cast_fp16 = pad(constant_val = const_123_to_fp16, mode = input_307_mode_0, pad = input_307_pad_0, x = input_305_cast_fp16)[name = tensor<string, []>("input_307_cast_fp16")];
            tensor<string, []> input_309_pad_type_0 = const()[name = tensor<string, []>("input_309_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, []> input_309_groups_0 = const()[name = tensor<string, []>("input_309_groups_0"), val = tensor<int32, []>(512)];
            tensor<int32, [1]> input_309_strides_0 = const()[name = tensor<string, []>("input_309_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_309_pad_0 = const()[name = tensor<string, []>("input_309_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_309_dilations_0 = const()[name = tensor<string, []>("input_309_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 9]> const_247_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("const_247_to_fp16_quantized"), quantized_data = tensor<int8, [512, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37149056))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37153728))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> const_248_to_fp16 = const()[name = tensor<string, []>("const_248_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37154816)))];
            tensor<fp16, [1, 512, 125]> input_311_cast_fp16 = conv(bias = const_248_to_fp16, dilations = input_309_dilations_0, groups = input_309_groups_0, pad = input_309_pad_0, pad_type = input_309_pad_type_0, strides = input_309_strides_0, weight = const_247_to_fp16_quantized, x = input_307_cast_fp16)[name = tensor<string, []>("input_311_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_313_cast_fp16 = silu(x = input_311_cast_fp16)[name = tensor<string, []>("input_313_cast_fp16")];
            tensor<string, []> x_131_pad_type_0 = const()[name = tensor<string, []>("x_131_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> x_131_strides_0 = const()[name = tensor<string, []>("x_131_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_131_pad_0 = const()[name = tensor<string, []>("x_131_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_131_dilations_0 = const()[name = tensor<string, []>("x_131_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> x_131_groups_0 = const()[name = tensor<string, []>("x_131_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [512, 512, 1]> encoder_layers_5_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_5_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37155904))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37418112))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_5_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37419200)))];
            tensor<fp16, [1, 512, 125]> x_131_cast_fp16 = conv(bias = encoder_layers_5_conv_pointwise_conv2_bias_to_fp16, dilations = x_131_dilations_0, groups = x_131_groups_0, pad = x_131_pad_0, pad_type = x_131_pad_type_0, strides = x_131_strides_0, weight = encoder_layers_5_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_313_cast_fp16)[name = tensor<string, []>("x_131_cast_fp16")];
            tensor<int32, [3]> input_315_perm_0 = const()[name = tensor<string, []>("input_315_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 125, 512]> input_315_cast_fp16 = transpose(perm = input_315_perm_0, x = x_131_cast_fp16)[name = tensor<string, []>("transpose_179")];
            tensor<fp16, [1, 125, 512]> input_317_cast_fp16 = add(x = input_299_cast_fp16, y = input_315_cast_fp16)[name = tensor<string, []>("input_317_cast_fp16")];
            tensor<int32, [1]> input_319_axes_0 = const()[name = tensor<string, []>("input_319_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_5_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37420288)))];
            tensor<fp16, [512]> encoder_layers_5_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37421376)))];
            tensor<fp16, [1, 125, 512]> input_319_cast_fp16 = layer_norm(axes = input_319_axes_0, beta = encoder_layers_5_norm_feed_forward2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_5_norm_feed_forward2_weight_to_fp16, x = input_317_cast_fp16)[name = tensor<string, []>("input_319_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_5_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_5_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(37422464))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38471104))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_5_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38475264)))];
            tensor<fp16, [1, 125, 2048]> linear_53_cast_fp16 = linear(bias = encoder_layers_5_feed_forward2_linear1_bias_to_fp16, weight = encoder_layers_5_feed_forward2_linear1_weight_to_fp16_quantized, x = input_319_cast_fp16)[name = tensor<string, []>("linear_53_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_323_cast_fp16 = silu(x = linear_53_cast_fp16)[name = tensor<string, []>("input_323_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_5_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_5_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(38479424))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39528064))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_5_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39529152)))];
            tensor<fp16, [1, 125, 512]> linear_54_cast_fp16 = linear(bias = encoder_layers_5_feed_forward2_linear2_bias_to_fp16, weight = encoder_layers_5_feed_forward2_linear2_weight_to_fp16_quantized, x = input_323_cast_fp16)[name = tensor<string, []>("linear_54_cast_fp16")];
            tensor<fp16, []> var_1361_to_fp16 = const()[name = tensor<string, []>("op_1361_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_1362_cast_fp16 = mul(x = linear_54_cast_fp16, y = var_1361_to_fp16)[name = tensor<string, []>("op_1362_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_329_cast_fp16 = add(x = input_317_cast_fp16, y = var_1362_cast_fp16)[name = tensor<string, []>("input_329_cast_fp16")];
            tensor<int32, [1]> input_331_axes_0 = const()[name = tensor<string, []>("input_331_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_5_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_norm_out_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39530240)))];
            tensor<fp16, [512]> encoder_layers_5_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_5_norm_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39531328)))];
            tensor<fp16, [1, 125, 512]> input_331_cast_fp16 = layer_norm(axes = input_331_axes_0, beta = encoder_layers_5_norm_out_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_5_norm_out_weight_to_fp16, x = input_329_cast_fp16)[name = tensor<string, []>("input_331_cast_fp16")];
            tensor<int32, [1]> input_333_axes_0 = const()[name = tensor<string, []>("input_333_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_6_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39532416)))];
            tensor<fp16, [512]> encoder_layers_6_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39533504)))];
            tensor<fp16, [1, 125, 512]> input_333_cast_fp16 = layer_norm(axes = input_333_axes_0, beta = encoder_layers_6_norm_feed_forward1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_6_norm_feed_forward1_weight_to_fp16, x = input_331_cast_fp16)[name = tensor<string, []>("input_333_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_6_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_6_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(39534592))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40583232))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_6_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40587392)))];
            tensor<fp16, [1, 125, 2048]> linear_55_cast_fp16 = linear(bias = encoder_layers_6_feed_forward1_linear1_bias_to_fp16, weight = encoder_layers_6_feed_forward1_linear1_weight_to_fp16_quantized, x = input_333_cast_fp16)[name = tensor<string, []>("linear_55_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_337_cast_fp16 = silu(x = linear_55_cast_fp16)[name = tensor<string, []>("input_337_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_6_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_6_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(40591552))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41640192))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_6_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41641280)))];
            tensor<fp16, [1, 125, 512]> linear_56_cast_fp16 = linear(bias = encoder_layers_6_feed_forward1_linear2_bias_to_fp16, weight = encoder_layers_6_feed_forward1_linear2_weight_to_fp16_quantized, x = input_337_cast_fp16)[name = tensor<string, []>("linear_56_cast_fp16")];
            tensor<fp16, []> var_1392_to_fp16 = const()[name = tensor<string, []>("op_1392_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_1393_cast_fp16 = mul(x = linear_56_cast_fp16, y = var_1392_to_fp16)[name = tensor<string, []>("op_1393_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_343_cast_fp16 = add(x = input_331_cast_fp16, y = var_1393_cast_fp16)[name = tensor<string, []>("input_343_cast_fp16")];
            tensor<int32, [1]> query_13_axes_0 = const()[name = tensor<string, []>("query_13_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_6_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_norm_self_att_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41642368)))];
            tensor<fp16, [512]> encoder_layers_6_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_norm_self_att_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41643456)))];
            tensor<fp16, [1, 125, 512]> query_13_cast_fp16 = layer_norm(axes = query_13_axes_0, beta = encoder_layers_6_norm_self_att_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_6_norm_self_att_weight_to_fp16, x = input_343_cast_fp16)[name = tensor<string, []>("query_13_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_6_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_6_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41644544))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41906752))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_6_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41907840)))];
            tensor<fp16, [1, 125, 512]> linear_57_cast_fp16 = linear(bias = encoder_layers_6_self_attn_linear_q_bias_to_fp16, weight = encoder_layers_6_self_attn_linear_q_weight_to_fp16_quantized, x = query_13_cast_fp16)[name = tensor<string, []>("linear_57_cast_fp16")];
            tensor<int32, [4]> var_1410 = const()[name = tensor<string, []>("op_1410"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> q_37_cast_fp16 = reshape(shape = var_1410, x = linear_57_cast_fp16)[name = tensor<string, []>("q_37_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_6_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_6_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(41908928))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42171136))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_6_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42172224)))];
            tensor<fp16, [1, 125, 512]> linear_58_cast_fp16 = linear(bias = encoder_layers_6_self_attn_linear_k_bias_to_fp16, weight = encoder_layers_6_self_attn_linear_k_weight_to_fp16_quantized, x = query_13_cast_fp16)[name = tensor<string, []>("linear_58_cast_fp16")];
            tensor<int32, [4]> var_1415 = const()[name = tensor<string, []>("op_1415"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> k_25_cast_fp16 = reshape(shape = var_1415, x = linear_58_cast_fp16)[name = tensor<string, []>("k_25_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_6_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_6_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42173312))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42435520))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_6_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42436608)))];
            tensor<fp16, [1, 125, 512]> linear_59_cast_fp16 = linear(bias = encoder_layers_6_self_attn_linear_v_bias_to_fp16, weight = encoder_layers_6_self_attn_linear_v_weight_to_fp16_quantized, x = query_13_cast_fp16)[name = tensor<string, []>("linear_59_cast_fp16")];
            tensor<int32, [4]> var_1420 = const()[name = tensor<string, []>("op_1420"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> v_13_cast_fp16 = reshape(shape = var_1420, x = linear_59_cast_fp16)[name = tensor<string, []>("v_13_cast_fp16")];
            tensor<int32, [4]> value_13_perm_0 = const()[name = tensor<string, []>("value_13_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 64]> encoder_layers_6_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42437696)))];
            tensor<fp16, [1, 125, 8, 64]> var_1432_cast_fp16 = add(x = q_37_cast_fp16, y = encoder_layers_6_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_1432_cast_fp16")];
            tensor<fp16, [8, 64]> encoder_layers_6_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42438784)))];
            tensor<fp16, [1, 125, 8, 64]> var_1434_cast_fp16 = add(x = q_37_cast_fp16, y = encoder_layers_6_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_1434_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_13_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_13_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<bool, []> x_139_transpose_x_0 = const()[name = tensor<string, []>("x_139_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_139_transpose_y_0 = const()[name = tensor<string, []>("x_139_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 64, 249]> op_1436_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(3), name = tensor<string, []>("op_1436_to_fp16_quantized"), quantized_data = tensor<int8, [1, 8, 64, 249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42439872))), scale = tensor<fp16, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42567424))), zero_point = tensor<int8, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5281536)))];
            tensor<fp16, [1, 8, 125, 64]> q_with_bias_v_13_cast_fp16 = transpose(perm = q_with_bias_v_13_perm_0, x = var_1434_cast_fp16)[name = tensor<string, []>("transpose_178")];
            tensor<fp16, [1, 8, 125, 249]> x_139_cast_fp16 = matmul(transpose_x = x_139_transpose_x_0, transpose_y = x_139_transpose_y_0, x = q_with_bias_v_13_cast_fp16, y = op_1436_to_fp16_quantized)[name = tensor<string, []>("x_139_cast_fp16")];
            tensor<int32, [8]> x_141_pad_0 = const()[name = tensor<string, []>("x_141_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            tensor<string, []> x_141_mode_0 = const()[name = tensor<string, []>("x_141_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_130_to_fp16 = const()[name = tensor<string, []>("const_130_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 8, 125, 250]> x_141_cast_fp16 = pad(constant_val = const_130_to_fp16, mode = x_141_mode_0, pad = x_141_pad_0, x = x_139_cast_fp16)[name = tensor<string, []>("x_141_cast_fp16")];
            tensor<int32, [4]> var_1444 = const()[name = tensor<string, []>("op_1444"), val = tensor<int32, [4]>([1, 8, -1, 125])];
            tensor<fp16, [1, 8, 250, 125]> x_143_cast_fp16 = reshape(shape = var_1444, x = x_141_cast_fp16)[name = tensor<string, []>("x_143_cast_fp16")];
            tensor<int32, [4]> var_1448_begin_0 = const()[name = tensor<string, []>("op_1448_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_1448_end_0 = const()[name = tensor<string, []>("op_1448_end_0"), val = tensor<int32, [4]>([1, 8, 250, 125])];
            tensor<bool, [4]> var_1448_end_mask_0 = const()[name = tensor<string, []>("op_1448_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 249, 125]> var_1448_cast_fp16 = slice_by_index(begin = var_1448_begin_0, end = var_1448_end_0, end_mask = var_1448_end_mask_0, x = x_143_cast_fp16)[name = tensor<string, []>("op_1448_cast_fp16")];
            tensor<int32, [4]> var_1449 = const()[name = tensor<string, []>("op_1449"), val = tensor<int32, [4]>([1, 8, 125, 249])];
            tensor<fp16, [1, 8, 125, 249]> matrix_bd_25_cast_fp16 = reshape(shape = var_1449, x = var_1448_cast_fp16)[name = tensor<string, []>("matrix_bd_25_cast_fp16")];
            tensor<bool, []> matrix_ac_13_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_13_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matrix_ac_13_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_13_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_80_perm_0 = const()[name = tensor<string, []>("transpose_80_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_81_perm_0 = const()[name = tensor<string, []>("transpose_81_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 64, 125]> transpose_81 = transpose(perm = transpose_81_perm_0, x = k_25_cast_fp16)[name = tensor<string, []>("transpose_176")];
            tensor<fp16, [1, 8, 125, 64]> transpose_80 = transpose(perm = transpose_80_perm_0, x = var_1432_cast_fp16)[name = tensor<string, []>("transpose_177")];
            tensor<fp16, [1, 8, 125, 125]> matrix_ac_13_cast_fp16 = matmul(transpose_x = matrix_ac_13_transpose_x_0, transpose_y = matrix_ac_13_transpose_y_0, x = transpose_80, y = transpose_81)[name = tensor<string, []>("matrix_ac_13_cast_fp16")];
            tensor<int32, [4]> matrix_bd_27_begin_0 = const()[name = tensor<string, []>("matrix_bd_27_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_27_end_0 = const()[name = tensor<string, []>("matrix_bd_27_end_0"), val = tensor<int32, [4]>([1, 8, 125, 125])];
            tensor<bool, [4]> matrix_bd_27_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_27_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 125, 125]> matrix_bd_27_cast_fp16 = slice_by_index(begin = matrix_bd_27_begin_0, end = matrix_bd_27_end_0, end_mask = matrix_bd_27_end_mask_0, x = matrix_bd_25_cast_fp16)[name = tensor<string, []>("matrix_bd_27_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_1458_cast_fp16 = add(x = matrix_ac_13_cast_fp16, y = matrix_bd_27_cast_fp16)[name = tensor<string, []>("op_1458_cast_fp16")];
            tensor<fp16, []> _inversed_scores_25_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_25_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 8, 125, 125]> _inversed_scores_25_cast_fp16 = mul(x = var_1458_cast_fp16, y = _inversed_scores_25_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_25_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> scores_27_cast_fp16 = select(a = var_8_to_fp16, b = _inversed_scores_25_cast_fp16, cond = mask_11)[name = tensor<string, []>("scores_27_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_1464_cast_fp16 = softmax(axis = var_19, x = scores_27_cast_fp16)[name = tensor<string, []>("op_1464_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> input_345_cast_fp16 = select(a = var_7_to_fp16, b = var_1464_cast_fp16, cond = mask_11)[name = tensor<string, []>("input_345_cast_fp16")];
            tensor<bool, []> x_145_transpose_x_0 = const()[name = tensor<string, []>("x_145_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_145_transpose_y_0 = const()[name = tensor<string, []>("x_145_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 125, 64]> value_13_cast_fp16 = transpose(perm = value_13_perm_0, x = v_13_cast_fp16)[name = tensor<string, []>("transpose_175")];
            tensor<fp16, [1, 8, 125, 64]> x_145_cast_fp16 = matmul(transpose_x = x_145_transpose_x_0, transpose_y = x_145_transpose_y_0, x = input_345_cast_fp16, y = value_13_cast_fp16)[name = tensor<string, []>("x_145_cast_fp16")];
            tensor<int32, [4]> var_1468_perm_0 = const()[name = tensor<string, []>("op_1468_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1469 = const()[name = tensor<string, []>("op_1469"), val = tensor<int32, [3]>([1, -1, 512])];
            tensor<fp16, [1, 125, 8, 64]> var_1468_cast_fp16 = transpose(perm = var_1468_perm_0, x = x_145_cast_fp16)[name = tensor<string, []>("transpose_174")];
            tensor<fp16, [1, 125, 512]> input_347_cast_fp16 = reshape(shape = var_1469, x = var_1468_cast_fp16)[name = tensor<string, []>("input_347_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_6_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_6_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42568000))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42830208))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_6_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42831296)))];
            tensor<fp16, [1, 125, 512]> linear_61_cast_fp16 = linear(bias = encoder_layers_6_self_attn_linear_out_bias_to_fp16, weight = encoder_layers_6_self_attn_linear_out_weight_to_fp16_quantized, x = input_347_cast_fp16)[name = tensor<string, []>("linear_61_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_351_cast_fp16 = add(x = input_343_cast_fp16, y = linear_61_cast_fp16)[name = tensor<string, []>("input_351_cast_fp16")];
            tensor<int32, [1]> x_149_axes_0 = const()[name = tensor<string, []>("x_149_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_6_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_norm_conv_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42832384)))];
            tensor<fp16, [512]> encoder_layers_6_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_norm_conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42833472)))];
            tensor<fp16, [1, 125, 512]> x_149_cast_fp16 = layer_norm(axes = x_149_axes_0, beta = encoder_layers_6_norm_conv_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_6_norm_conv_weight_to_fp16, x = input_351_cast_fp16)[name = tensor<string, []>("x_149_cast_fp16")];
            tensor<int32, [3]> input_353_perm_0 = const()[name = tensor<string, []>("input_353_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> input_355_pad_type_0 = const()[name = tensor<string, []>("input_355_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> input_355_strides_0 = const()[name = tensor<string, []>("input_355_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_355_pad_0 = const()[name = tensor<string, []>("input_355_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_355_dilations_0 = const()[name = tensor<string, []>("input_355_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> input_355_groups_0 = const()[name = tensor<string, []>("input_355_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 512, 1]> encoder_layers_6_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_6_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1024, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(42834560))), scale = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43358912))), zero_point = tensor<int8, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6073344)))];
            tensor<fp16, [1024]> encoder_layers_6_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43361024)))];
            tensor<fp16, [1, 512, 125]> input_353_cast_fp16 = transpose(perm = input_353_perm_0, x = x_149_cast_fp16)[name = tensor<string, []>("transpose_173")];
            tensor<fp16, [1, 1024, 125]> input_355_cast_fp16 = conv(bias = encoder_layers_6_conv_pointwise_conv1_bias_to_fp16, dilations = input_355_dilations_0, groups = input_355_groups_0, pad = input_355_pad_0, pad_type = input_355_pad_type_0, strides = input_355_strides_0, weight = encoder_layers_6_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_353_cast_fp16)[name = tensor<string, []>("input_355_cast_fp16")];
            tensor<int32, []> x_151_split_num_splits_0 = const()[name = tensor<string, []>("x_151_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> x_151_split_axis_0 = const()[name = tensor<string, []>("x_151_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 512, 125]> x_151_split_cast_fp16_0, tensor<fp16, [1, 512, 125]> x_151_split_cast_fp16_1 = split(axis = x_151_split_axis_0, num_splits = x_151_split_num_splits_0, x = input_355_cast_fp16)[name = tensor<string, []>("x_151_split_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_151_split_1_sigmoid_cast_fp16 = sigmoid(x = x_151_split_cast_fp16_1)[name = tensor<string, []>("x_151_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_151_cast_fp16 = mul(x = x_151_split_cast_fp16_0, y = x_151_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_151_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_357_cast_fp16 = select(a = var_7_to_fp16, b = x_151_cast_fp16, cond = var_449)[name = tensor<string, []>("input_357_cast_fp16")];
            tensor<int32, [6]> input_359_pad_0 = const()[name = tensor<string, []>("input_359_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            tensor<string, []> input_359_mode_0 = const()[name = tensor<string, []>("input_359_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_133_to_fp16 = const()[name = tensor<string, []>("const_133_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 512, 133]> input_359_cast_fp16 = pad(constant_val = const_133_to_fp16, mode = input_359_mode_0, pad = input_359_pad_0, x = input_357_cast_fp16)[name = tensor<string, []>("input_359_cast_fp16")];
            tensor<string, []> input_361_pad_type_0 = const()[name = tensor<string, []>("input_361_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, []> input_361_groups_0 = const()[name = tensor<string, []>("input_361_groups_0"), val = tensor<int32, []>(512)];
            tensor<int32, [1]> input_361_strides_0 = const()[name = tensor<string, []>("input_361_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_361_pad_0 = const()[name = tensor<string, []>("input_361_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_361_dilations_0 = const()[name = tensor<string, []>("input_361_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 9]> const_249_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("const_249_to_fp16_quantized"), quantized_data = tensor<int8, [512, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43363136))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43367808))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> const_250_to_fp16 = const()[name = tensor<string, []>("const_250_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43368896)))];
            tensor<fp16, [1, 512, 125]> input_363_cast_fp16 = conv(bias = const_250_to_fp16, dilations = input_361_dilations_0, groups = input_361_groups_0, pad = input_361_pad_0, pad_type = input_361_pad_type_0, strides = input_361_strides_0, weight = const_249_to_fp16_quantized, x = input_359_cast_fp16)[name = tensor<string, []>("input_363_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_365_cast_fp16 = silu(x = input_363_cast_fp16)[name = tensor<string, []>("input_365_cast_fp16")];
            tensor<string, []> x_153_pad_type_0 = const()[name = tensor<string, []>("x_153_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> x_153_strides_0 = const()[name = tensor<string, []>("x_153_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_153_pad_0 = const()[name = tensor<string, []>("x_153_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_153_dilations_0 = const()[name = tensor<string, []>("x_153_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> x_153_groups_0 = const()[name = tensor<string, []>("x_153_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [512, 512, 1]> encoder_layers_6_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_6_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43369984))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43632192))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_6_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43633280)))];
            tensor<fp16, [1, 512, 125]> x_153_cast_fp16 = conv(bias = encoder_layers_6_conv_pointwise_conv2_bias_to_fp16, dilations = x_153_dilations_0, groups = x_153_groups_0, pad = x_153_pad_0, pad_type = x_153_pad_type_0, strides = x_153_strides_0, weight = encoder_layers_6_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_365_cast_fp16)[name = tensor<string, []>("x_153_cast_fp16")];
            tensor<int32, [3]> input_367_perm_0 = const()[name = tensor<string, []>("input_367_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 125, 512]> input_367_cast_fp16 = transpose(perm = input_367_perm_0, x = x_153_cast_fp16)[name = tensor<string, []>("transpose_172")];
            tensor<fp16, [1, 125, 512]> input_369_cast_fp16 = add(x = input_351_cast_fp16, y = input_367_cast_fp16)[name = tensor<string, []>("input_369_cast_fp16")];
            tensor<int32, [1]> input_371_axes_0 = const()[name = tensor<string, []>("input_371_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_6_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43634368)))];
            tensor<fp16, [512]> encoder_layers_6_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43635456)))];
            tensor<fp16, [1, 125, 512]> input_371_cast_fp16 = layer_norm(axes = input_371_axes_0, beta = encoder_layers_6_norm_feed_forward2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_6_norm_feed_forward2_weight_to_fp16, x = input_369_cast_fp16)[name = tensor<string, []>("input_371_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_6_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_6_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(43636544))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44685184))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_6_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44689344)))];
            tensor<fp16, [1, 125, 2048]> linear_62_cast_fp16 = linear(bias = encoder_layers_6_feed_forward2_linear1_bias_to_fp16, weight = encoder_layers_6_feed_forward2_linear1_weight_to_fp16_quantized, x = input_371_cast_fp16)[name = tensor<string, []>("linear_62_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_375_cast_fp16 = silu(x = linear_62_cast_fp16)[name = tensor<string, []>("input_375_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_6_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_6_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(44693504))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45742144))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_6_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45743232)))];
            tensor<fp16, [1, 125, 512]> linear_63_cast_fp16 = linear(bias = encoder_layers_6_feed_forward2_linear2_bias_to_fp16, weight = encoder_layers_6_feed_forward2_linear2_weight_to_fp16_quantized, x = input_375_cast_fp16)[name = tensor<string, []>("linear_63_cast_fp16")];
            tensor<fp16, []> var_1535_to_fp16 = const()[name = tensor<string, []>("op_1535_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_1536_cast_fp16 = mul(x = linear_63_cast_fp16, y = var_1535_to_fp16)[name = tensor<string, []>("op_1536_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_381_cast_fp16 = add(x = input_369_cast_fp16, y = var_1536_cast_fp16)[name = tensor<string, []>("input_381_cast_fp16")];
            tensor<int32, [1]> input_383_axes_0 = const()[name = tensor<string, []>("input_383_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_6_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_norm_out_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45744320)))];
            tensor<fp16, [512]> encoder_layers_6_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_6_norm_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45745408)))];
            tensor<fp16, [1, 125, 512]> input_383_cast_fp16 = layer_norm(axes = input_383_axes_0, beta = encoder_layers_6_norm_out_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_6_norm_out_weight_to_fp16, x = input_381_cast_fp16)[name = tensor<string, []>("input_383_cast_fp16")];
            tensor<int32, [1]> input_385_axes_0 = const()[name = tensor<string, []>("input_385_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_7_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45746496)))];
            tensor<fp16, [512]> encoder_layers_7_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45747584)))];
            tensor<fp16, [1, 125, 512]> input_385_cast_fp16 = layer_norm(axes = input_385_axes_0, beta = encoder_layers_7_norm_feed_forward1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_7_norm_feed_forward1_weight_to_fp16, x = input_383_cast_fp16)[name = tensor<string, []>("input_385_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_7_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_7_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(45748672))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46797312))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_7_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46801472)))];
            tensor<fp16, [1, 125, 2048]> linear_64_cast_fp16 = linear(bias = encoder_layers_7_feed_forward1_linear1_bias_to_fp16, weight = encoder_layers_7_feed_forward1_linear1_weight_to_fp16_quantized, x = input_385_cast_fp16)[name = tensor<string, []>("linear_64_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_389_cast_fp16 = silu(x = linear_64_cast_fp16)[name = tensor<string, []>("input_389_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_7_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_7_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(46805632))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(47854272))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_7_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(47855360)))];
            tensor<fp16, [1, 125, 512]> linear_65_cast_fp16 = linear(bias = encoder_layers_7_feed_forward1_linear2_bias_to_fp16, weight = encoder_layers_7_feed_forward1_linear2_weight_to_fp16_quantized, x = input_389_cast_fp16)[name = tensor<string, []>("linear_65_cast_fp16")];
            tensor<fp16, []> var_1566_to_fp16 = const()[name = tensor<string, []>("op_1566_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_1567_cast_fp16 = mul(x = linear_65_cast_fp16, y = var_1566_to_fp16)[name = tensor<string, []>("op_1567_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_395_cast_fp16 = add(x = input_383_cast_fp16, y = var_1567_cast_fp16)[name = tensor<string, []>("input_395_cast_fp16")];
            tensor<int32, [1]> query_15_axes_0 = const()[name = tensor<string, []>("query_15_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_7_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_norm_self_att_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(47856448)))];
            tensor<fp16, [512]> encoder_layers_7_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_norm_self_att_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(47857536)))];
            tensor<fp16, [1, 125, 512]> query_15_cast_fp16 = layer_norm(axes = query_15_axes_0, beta = encoder_layers_7_norm_self_att_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_7_norm_self_att_weight_to_fp16, x = input_395_cast_fp16)[name = tensor<string, []>("query_15_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_7_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_7_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(47858624))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48120832))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_7_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48121920)))];
            tensor<fp16, [1, 125, 512]> linear_66_cast_fp16 = linear(bias = encoder_layers_7_self_attn_linear_q_bias_to_fp16, weight = encoder_layers_7_self_attn_linear_q_weight_to_fp16_quantized, x = query_15_cast_fp16)[name = tensor<string, []>("linear_66_cast_fp16")];
            tensor<int32, [4]> var_1584 = const()[name = tensor<string, []>("op_1584"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> q_43_cast_fp16 = reshape(shape = var_1584, x = linear_66_cast_fp16)[name = tensor<string, []>("q_43_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_7_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_7_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48123008))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48385216))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_7_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48386304)))];
            tensor<fp16, [1, 125, 512]> linear_67_cast_fp16 = linear(bias = encoder_layers_7_self_attn_linear_k_bias_to_fp16, weight = encoder_layers_7_self_attn_linear_k_weight_to_fp16_quantized, x = query_15_cast_fp16)[name = tensor<string, []>("linear_67_cast_fp16")];
            tensor<int32, [4]> var_1589 = const()[name = tensor<string, []>("op_1589"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> k_29_cast_fp16 = reshape(shape = var_1589, x = linear_67_cast_fp16)[name = tensor<string, []>("k_29_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_7_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_7_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48387392))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48649600))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_7_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48650688)))];
            tensor<fp16, [1, 125, 512]> linear_68_cast_fp16 = linear(bias = encoder_layers_7_self_attn_linear_v_bias_to_fp16, weight = encoder_layers_7_self_attn_linear_v_weight_to_fp16_quantized, x = query_15_cast_fp16)[name = tensor<string, []>("linear_68_cast_fp16")];
            tensor<int32, [4]> var_1594 = const()[name = tensor<string, []>("op_1594"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> v_15_cast_fp16 = reshape(shape = var_1594, x = linear_68_cast_fp16)[name = tensor<string, []>("v_15_cast_fp16")];
            tensor<int32, [4]> value_15_perm_0 = const()[name = tensor<string, []>("value_15_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 64]> encoder_layers_7_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48651776)))];
            tensor<fp16, [1, 125, 8, 64]> var_1606_cast_fp16 = add(x = q_43_cast_fp16, y = encoder_layers_7_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_1606_cast_fp16")];
            tensor<fp16, [8, 64]> encoder_layers_7_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48652864)))];
            tensor<fp16, [1, 125, 8, 64]> var_1608_cast_fp16 = add(x = q_43_cast_fp16, y = encoder_layers_7_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_1608_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_15_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_15_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<bool, []> x_161_transpose_x_0 = const()[name = tensor<string, []>("x_161_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_161_transpose_y_0 = const()[name = tensor<string, []>("x_161_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 64, 249]> op_1610_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(3), name = tensor<string, []>("op_1610_to_fp16_quantized"), quantized_data = tensor<int8, [1, 8, 64, 249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48653952))), scale = tensor<fp16, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48781504))), zero_point = tensor<int8, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5281536)))];
            tensor<fp16, [1, 8, 125, 64]> q_with_bias_v_15_cast_fp16 = transpose(perm = q_with_bias_v_15_perm_0, x = var_1608_cast_fp16)[name = tensor<string, []>("transpose_171")];
            tensor<fp16, [1, 8, 125, 249]> x_161_cast_fp16 = matmul(transpose_x = x_161_transpose_x_0, transpose_y = x_161_transpose_y_0, x = q_with_bias_v_15_cast_fp16, y = op_1610_to_fp16_quantized)[name = tensor<string, []>("x_161_cast_fp16")];
            tensor<int32, [8]> x_163_pad_0 = const()[name = tensor<string, []>("x_163_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            tensor<string, []> x_163_mode_0 = const()[name = tensor<string, []>("x_163_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_140_to_fp16 = const()[name = tensor<string, []>("const_140_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 8, 125, 250]> x_163_cast_fp16 = pad(constant_val = const_140_to_fp16, mode = x_163_mode_0, pad = x_163_pad_0, x = x_161_cast_fp16)[name = tensor<string, []>("x_163_cast_fp16")];
            tensor<int32, [4]> var_1618 = const()[name = tensor<string, []>("op_1618"), val = tensor<int32, [4]>([1, 8, -1, 125])];
            tensor<fp16, [1, 8, 250, 125]> x_165_cast_fp16 = reshape(shape = var_1618, x = x_163_cast_fp16)[name = tensor<string, []>("x_165_cast_fp16")];
            tensor<int32, [4]> var_1622_begin_0 = const()[name = tensor<string, []>("op_1622_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_1622_end_0 = const()[name = tensor<string, []>("op_1622_end_0"), val = tensor<int32, [4]>([1, 8, 250, 125])];
            tensor<bool, [4]> var_1622_end_mask_0 = const()[name = tensor<string, []>("op_1622_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 249, 125]> var_1622_cast_fp16 = slice_by_index(begin = var_1622_begin_0, end = var_1622_end_0, end_mask = var_1622_end_mask_0, x = x_165_cast_fp16)[name = tensor<string, []>("op_1622_cast_fp16")];
            tensor<int32, [4]> var_1623 = const()[name = tensor<string, []>("op_1623"), val = tensor<int32, [4]>([1, 8, 125, 249])];
            tensor<fp16, [1, 8, 125, 249]> matrix_bd_29_cast_fp16 = reshape(shape = var_1623, x = var_1622_cast_fp16)[name = tensor<string, []>("matrix_bd_29_cast_fp16")];
            tensor<bool, []> matrix_ac_15_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_15_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matrix_ac_15_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_15_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_82_perm_0 = const()[name = tensor<string, []>("transpose_82_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_83_perm_0 = const()[name = tensor<string, []>("transpose_83_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 64, 125]> transpose_83 = transpose(perm = transpose_83_perm_0, x = k_29_cast_fp16)[name = tensor<string, []>("transpose_169")];
            tensor<fp16, [1, 8, 125, 64]> transpose_82 = transpose(perm = transpose_82_perm_0, x = var_1606_cast_fp16)[name = tensor<string, []>("transpose_170")];
            tensor<fp16, [1, 8, 125, 125]> matrix_ac_15_cast_fp16 = matmul(transpose_x = matrix_ac_15_transpose_x_0, transpose_y = matrix_ac_15_transpose_y_0, x = transpose_82, y = transpose_83)[name = tensor<string, []>("matrix_ac_15_cast_fp16")];
            tensor<int32, [4]> matrix_bd_31_begin_0 = const()[name = tensor<string, []>("matrix_bd_31_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_31_end_0 = const()[name = tensor<string, []>("matrix_bd_31_end_0"), val = tensor<int32, [4]>([1, 8, 125, 125])];
            tensor<bool, [4]> matrix_bd_31_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_31_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 125, 125]> matrix_bd_31_cast_fp16 = slice_by_index(begin = matrix_bd_31_begin_0, end = matrix_bd_31_end_0, end_mask = matrix_bd_31_end_mask_0, x = matrix_bd_29_cast_fp16)[name = tensor<string, []>("matrix_bd_31_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_1632_cast_fp16 = add(x = matrix_ac_15_cast_fp16, y = matrix_bd_31_cast_fp16)[name = tensor<string, []>("op_1632_cast_fp16")];
            tensor<fp16, []> _inversed_scores_29_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_29_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 8, 125, 125]> _inversed_scores_29_cast_fp16 = mul(x = var_1632_cast_fp16, y = _inversed_scores_29_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_29_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> scores_31_cast_fp16 = select(a = var_8_to_fp16, b = _inversed_scores_29_cast_fp16, cond = mask_11)[name = tensor<string, []>("scores_31_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_1638_cast_fp16 = softmax(axis = var_19, x = scores_31_cast_fp16)[name = tensor<string, []>("op_1638_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> input_397_cast_fp16 = select(a = var_7_to_fp16, b = var_1638_cast_fp16, cond = mask_11)[name = tensor<string, []>("input_397_cast_fp16")];
            tensor<bool, []> x_167_transpose_x_0 = const()[name = tensor<string, []>("x_167_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_167_transpose_y_0 = const()[name = tensor<string, []>("x_167_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 125, 64]> value_15_cast_fp16 = transpose(perm = value_15_perm_0, x = v_15_cast_fp16)[name = tensor<string, []>("transpose_168")];
            tensor<fp16, [1, 8, 125, 64]> x_167_cast_fp16 = matmul(transpose_x = x_167_transpose_x_0, transpose_y = x_167_transpose_y_0, x = input_397_cast_fp16, y = value_15_cast_fp16)[name = tensor<string, []>("x_167_cast_fp16")];
            tensor<int32, [4]> var_1642_perm_0 = const()[name = tensor<string, []>("op_1642_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1643 = const()[name = tensor<string, []>("op_1643"), val = tensor<int32, [3]>([1, -1, 512])];
            tensor<fp16, [1, 125, 8, 64]> var_1642_cast_fp16 = transpose(perm = var_1642_perm_0, x = x_167_cast_fp16)[name = tensor<string, []>("transpose_167")];
            tensor<fp16, [1, 125, 512]> input_399_cast_fp16 = reshape(shape = var_1643, x = var_1642_cast_fp16)[name = tensor<string, []>("input_399_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_7_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_7_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(48782080))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49044288))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_7_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49045376)))];
            tensor<fp16, [1, 125, 512]> linear_70_cast_fp16 = linear(bias = encoder_layers_7_self_attn_linear_out_bias_to_fp16, weight = encoder_layers_7_self_attn_linear_out_weight_to_fp16_quantized, x = input_399_cast_fp16)[name = tensor<string, []>("linear_70_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_403_cast_fp16 = add(x = input_395_cast_fp16, y = linear_70_cast_fp16)[name = tensor<string, []>("input_403_cast_fp16")];
            tensor<int32, [1]> x_171_axes_0 = const()[name = tensor<string, []>("x_171_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_7_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_norm_conv_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49046464)))];
            tensor<fp16, [512]> encoder_layers_7_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_norm_conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49047552)))];
            tensor<fp16, [1, 125, 512]> x_171_cast_fp16 = layer_norm(axes = x_171_axes_0, beta = encoder_layers_7_norm_conv_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_7_norm_conv_weight_to_fp16, x = input_403_cast_fp16)[name = tensor<string, []>("x_171_cast_fp16")];
            tensor<int32, [3]> input_405_perm_0 = const()[name = tensor<string, []>("input_405_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> input_407_pad_type_0 = const()[name = tensor<string, []>("input_407_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> input_407_strides_0 = const()[name = tensor<string, []>("input_407_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_407_pad_0 = const()[name = tensor<string, []>("input_407_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_407_dilations_0 = const()[name = tensor<string, []>("input_407_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> input_407_groups_0 = const()[name = tensor<string, []>("input_407_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 512, 1]> encoder_layers_7_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_7_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1024, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49048640))), scale = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49572992))), zero_point = tensor<int8, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6073344)))];
            tensor<fp16, [1024]> encoder_layers_7_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49575104)))];
            tensor<fp16, [1, 512, 125]> input_405_cast_fp16 = transpose(perm = input_405_perm_0, x = x_171_cast_fp16)[name = tensor<string, []>("transpose_166")];
            tensor<fp16, [1, 1024, 125]> input_407_cast_fp16 = conv(bias = encoder_layers_7_conv_pointwise_conv1_bias_to_fp16, dilations = input_407_dilations_0, groups = input_407_groups_0, pad = input_407_pad_0, pad_type = input_407_pad_type_0, strides = input_407_strides_0, weight = encoder_layers_7_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_405_cast_fp16)[name = tensor<string, []>("input_407_cast_fp16")];
            tensor<int32, []> x_173_split_num_splits_0 = const()[name = tensor<string, []>("x_173_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> x_173_split_axis_0 = const()[name = tensor<string, []>("x_173_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 512, 125]> x_173_split_cast_fp16_0, tensor<fp16, [1, 512, 125]> x_173_split_cast_fp16_1 = split(axis = x_173_split_axis_0, num_splits = x_173_split_num_splits_0, x = input_407_cast_fp16)[name = tensor<string, []>("x_173_split_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_173_split_1_sigmoid_cast_fp16 = sigmoid(x = x_173_split_cast_fp16_1)[name = tensor<string, []>("x_173_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_173_cast_fp16 = mul(x = x_173_split_cast_fp16_0, y = x_173_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_173_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_409_cast_fp16 = select(a = var_7_to_fp16, b = x_173_cast_fp16, cond = var_449)[name = tensor<string, []>("input_409_cast_fp16")];
            tensor<int32, [6]> input_411_pad_0 = const()[name = tensor<string, []>("input_411_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            tensor<string, []> input_411_mode_0 = const()[name = tensor<string, []>("input_411_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_143_to_fp16 = const()[name = tensor<string, []>("const_143_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 512, 133]> input_411_cast_fp16 = pad(constant_val = const_143_to_fp16, mode = input_411_mode_0, pad = input_411_pad_0, x = input_409_cast_fp16)[name = tensor<string, []>("input_411_cast_fp16")];
            tensor<string, []> input_413_pad_type_0 = const()[name = tensor<string, []>("input_413_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, []> input_413_groups_0 = const()[name = tensor<string, []>("input_413_groups_0"), val = tensor<int32, []>(512)];
            tensor<int32, [1]> input_413_strides_0 = const()[name = tensor<string, []>("input_413_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_413_pad_0 = const()[name = tensor<string, []>("input_413_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_413_dilations_0 = const()[name = tensor<string, []>("input_413_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 9]> const_251_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("const_251_to_fp16_quantized"), quantized_data = tensor<int8, [512, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49577216))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49581888))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> const_252_to_fp16 = const()[name = tensor<string, []>("const_252_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49582976)))];
            tensor<fp16, [1, 512, 125]> input_415_cast_fp16 = conv(bias = const_252_to_fp16, dilations = input_413_dilations_0, groups = input_413_groups_0, pad = input_413_pad_0, pad_type = input_413_pad_type_0, strides = input_413_strides_0, weight = const_251_to_fp16_quantized, x = input_411_cast_fp16)[name = tensor<string, []>("input_415_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_417_cast_fp16 = silu(x = input_415_cast_fp16)[name = tensor<string, []>("input_417_cast_fp16")];
            tensor<string, []> x_175_pad_type_0 = const()[name = tensor<string, []>("x_175_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> x_175_strides_0 = const()[name = tensor<string, []>("x_175_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_175_pad_0 = const()[name = tensor<string, []>("x_175_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_175_dilations_0 = const()[name = tensor<string, []>("x_175_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> x_175_groups_0 = const()[name = tensor<string, []>("x_175_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [512, 512, 1]> encoder_layers_7_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_7_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49584064))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49846272))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_7_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49847360)))];
            tensor<fp16, [1, 512, 125]> x_175_cast_fp16 = conv(bias = encoder_layers_7_conv_pointwise_conv2_bias_to_fp16, dilations = x_175_dilations_0, groups = x_175_groups_0, pad = x_175_pad_0, pad_type = x_175_pad_type_0, strides = x_175_strides_0, weight = encoder_layers_7_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_417_cast_fp16)[name = tensor<string, []>("x_175_cast_fp16")];
            tensor<int32, [3]> input_419_perm_0 = const()[name = tensor<string, []>("input_419_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 125, 512]> input_419_cast_fp16 = transpose(perm = input_419_perm_0, x = x_175_cast_fp16)[name = tensor<string, []>("transpose_165")];
            tensor<fp16, [1, 125, 512]> input_421_cast_fp16 = add(x = input_403_cast_fp16, y = input_419_cast_fp16)[name = tensor<string, []>("input_421_cast_fp16")];
            tensor<int32, [1]> input_423_axes_0 = const()[name = tensor<string, []>("input_423_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_7_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49848448)))];
            tensor<fp16, [512]> encoder_layers_7_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49849536)))];
            tensor<fp16, [1, 125, 512]> input_423_cast_fp16 = layer_norm(axes = input_423_axes_0, beta = encoder_layers_7_norm_feed_forward2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_7_norm_feed_forward2_weight_to_fp16, x = input_421_cast_fp16)[name = tensor<string, []>("input_423_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_7_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_7_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(49850624))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50899264))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_7_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50903424)))];
            tensor<fp16, [1, 125, 2048]> linear_71_cast_fp16 = linear(bias = encoder_layers_7_feed_forward2_linear1_bias_to_fp16, weight = encoder_layers_7_feed_forward2_linear1_weight_to_fp16_quantized, x = input_423_cast_fp16)[name = tensor<string, []>("linear_71_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_427_cast_fp16 = silu(x = linear_71_cast_fp16)[name = tensor<string, []>("input_427_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_7_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_7_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(50907584))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51956224))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_7_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51957312)))];
            tensor<fp16, [1, 125, 512]> linear_72_cast_fp16 = linear(bias = encoder_layers_7_feed_forward2_linear2_bias_to_fp16, weight = encoder_layers_7_feed_forward2_linear2_weight_to_fp16_quantized, x = input_427_cast_fp16)[name = tensor<string, []>("linear_72_cast_fp16")];
            tensor<fp16, []> var_1709_to_fp16 = const()[name = tensor<string, []>("op_1709_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_1710_cast_fp16 = mul(x = linear_72_cast_fp16, y = var_1709_to_fp16)[name = tensor<string, []>("op_1710_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_433_cast_fp16 = add(x = input_421_cast_fp16, y = var_1710_cast_fp16)[name = tensor<string, []>("input_433_cast_fp16")];
            tensor<int32, [1]> input_435_axes_0 = const()[name = tensor<string, []>("input_435_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_7_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_norm_out_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51958400)))];
            tensor<fp16, [512]> encoder_layers_7_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_7_norm_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51959488)))];
            tensor<fp16, [1, 125, 512]> input_435_cast_fp16 = layer_norm(axes = input_435_axes_0, beta = encoder_layers_7_norm_out_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_7_norm_out_weight_to_fp16, x = input_433_cast_fp16)[name = tensor<string, []>("input_435_cast_fp16")];
            tensor<int32, [1]> input_437_axes_0 = const()[name = tensor<string, []>("input_437_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_8_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51960576)))];
            tensor<fp16, [512]> encoder_layers_8_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51961664)))];
            tensor<fp16, [1, 125, 512]> input_437_cast_fp16 = layer_norm(axes = input_437_axes_0, beta = encoder_layers_8_norm_feed_forward1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_8_norm_feed_forward1_weight_to_fp16, x = input_435_cast_fp16)[name = tensor<string, []>("input_437_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_8_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_8_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(51962752))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53011392))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_8_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53015552)))];
            tensor<fp16, [1, 125, 2048]> linear_73_cast_fp16 = linear(bias = encoder_layers_8_feed_forward1_linear1_bias_to_fp16, weight = encoder_layers_8_feed_forward1_linear1_weight_to_fp16_quantized, x = input_437_cast_fp16)[name = tensor<string, []>("linear_73_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_441_cast_fp16 = silu(x = linear_73_cast_fp16)[name = tensor<string, []>("input_441_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_8_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_8_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(53019712))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54068352))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_8_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54069440)))];
            tensor<fp16, [1, 125, 512]> linear_74_cast_fp16 = linear(bias = encoder_layers_8_feed_forward1_linear2_bias_to_fp16, weight = encoder_layers_8_feed_forward1_linear2_weight_to_fp16_quantized, x = input_441_cast_fp16)[name = tensor<string, []>("linear_74_cast_fp16")];
            tensor<fp16, []> var_1740_to_fp16 = const()[name = tensor<string, []>("op_1740_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_1741_cast_fp16 = mul(x = linear_74_cast_fp16, y = var_1740_to_fp16)[name = tensor<string, []>("op_1741_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_447_cast_fp16 = add(x = input_435_cast_fp16, y = var_1741_cast_fp16)[name = tensor<string, []>("input_447_cast_fp16")];
            tensor<int32, [1]> query_17_axes_0 = const()[name = tensor<string, []>("query_17_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_8_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_norm_self_att_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54070528)))];
            tensor<fp16, [512]> encoder_layers_8_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_norm_self_att_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54071616)))];
            tensor<fp16, [1, 125, 512]> query_17_cast_fp16 = layer_norm(axes = query_17_axes_0, beta = encoder_layers_8_norm_self_att_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_8_norm_self_att_weight_to_fp16, x = input_447_cast_fp16)[name = tensor<string, []>("query_17_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_8_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_8_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54072704))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54334912))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_8_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54336000)))];
            tensor<fp16, [1, 125, 512]> linear_75_cast_fp16 = linear(bias = encoder_layers_8_self_attn_linear_q_bias_to_fp16, weight = encoder_layers_8_self_attn_linear_q_weight_to_fp16_quantized, x = query_17_cast_fp16)[name = tensor<string, []>("linear_75_cast_fp16")];
            tensor<int32, [4]> var_1758 = const()[name = tensor<string, []>("op_1758"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> q_49_cast_fp16 = reshape(shape = var_1758, x = linear_75_cast_fp16)[name = tensor<string, []>("q_49_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_8_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_8_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54337088))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54599296))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_8_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54600384)))];
            tensor<fp16, [1, 125, 512]> linear_76_cast_fp16 = linear(bias = encoder_layers_8_self_attn_linear_k_bias_to_fp16, weight = encoder_layers_8_self_attn_linear_k_weight_to_fp16_quantized, x = query_17_cast_fp16)[name = tensor<string, []>("linear_76_cast_fp16")];
            tensor<int32, [4]> var_1763 = const()[name = tensor<string, []>("op_1763"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> k_33_cast_fp16 = reshape(shape = var_1763, x = linear_76_cast_fp16)[name = tensor<string, []>("k_33_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_8_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_8_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54601472))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54863680))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_8_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54864768)))];
            tensor<fp16, [1, 125, 512]> linear_77_cast_fp16 = linear(bias = encoder_layers_8_self_attn_linear_v_bias_to_fp16, weight = encoder_layers_8_self_attn_linear_v_weight_to_fp16_quantized, x = query_17_cast_fp16)[name = tensor<string, []>("linear_77_cast_fp16")];
            tensor<int32, [4]> var_1768 = const()[name = tensor<string, []>("op_1768"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> v_17_cast_fp16 = reshape(shape = var_1768, x = linear_77_cast_fp16)[name = tensor<string, []>("v_17_cast_fp16")];
            tensor<int32, [4]> value_17_perm_0 = const()[name = tensor<string, []>("value_17_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 64]> encoder_layers_8_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54865856)))];
            tensor<fp16, [1, 125, 8, 64]> var_1780_cast_fp16 = add(x = q_49_cast_fp16, y = encoder_layers_8_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_1780_cast_fp16")];
            tensor<fp16, [8, 64]> encoder_layers_8_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54866944)))];
            tensor<fp16, [1, 125, 8, 64]> var_1782_cast_fp16 = add(x = q_49_cast_fp16, y = encoder_layers_8_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_1782_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_17_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_17_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<bool, []> x_183_transpose_x_0 = const()[name = tensor<string, []>("x_183_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_183_transpose_y_0 = const()[name = tensor<string, []>("x_183_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 64, 249]> op_1784_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(3), name = tensor<string, []>("op_1784_to_fp16_quantized"), quantized_data = tensor<int8, [1, 8, 64, 249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54868032))), scale = tensor<fp16, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54995584))), zero_point = tensor<int8, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5281536)))];
            tensor<fp16, [1, 8, 125, 64]> q_with_bias_v_17_cast_fp16 = transpose(perm = q_with_bias_v_17_perm_0, x = var_1782_cast_fp16)[name = tensor<string, []>("transpose_164")];
            tensor<fp16, [1, 8, 125, 249]> x_183_cast_fp16 = matmul(transpose_x = x_183_transpose_x_0, transpose_y = x_183_transpose_y_0, x = q_with_bias_v_17_cast_fp16, y = op_1784_to_fp16_quantized)[name = tensor<string, []>("x_183_cast_fp16")];
            tensor<int32, [8]> x_185_pad_0 = const()[name = tensor<string, []>("x_185_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            tensor<string, []> x_185_mode_0 = const()[name = tensor<string, []>("x_185_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_150_to_fp16 = const()[name = tensor<string, []>("const_150_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 8, 125, 250]> x_185_cast_fp16 = pad(constant_val = const_150_to_fp16, mode = x_185_mode_0, pad = x_185_pad_0, x = x_183_cast_fp16)[name = tensor<string, []>("x_185_cast_fp16")];
            tensor<int32, [4]> var_1792 = const()[name = tensor<string, []>("op_1792"), val = tensor<int32, [4]>([1, 8, -1, 125])];
            tensor<fp16, [1, 8, 250, 125]> x_187_cast_fp16 = reshape(shape = var_1792, x = x_185_cast_fp16)[name = tensor<string, []>("x_187_cast_fp16")];
            tensor<int32, [4]> var_1796_begin_0 = const()[name = tensor<string, []>("op_1796_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_1796_end_0 = const()[name = tensor<string, []>("op_1796_end_0"), val = tensor<int32, [4]>([1, 8, 250, 125])];
            tensor<bool, [4]> var_1796_end_mask_0 = const()[name = tensor<string, []>("op_1796_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 249, 125]> var_1796_cast_fp16 = slice_by_index(begin = var_1796_begin_0, end = var_1796_end_0, end_mask = var_1796_end_mask_0, x = x_187_cast_fp16)[name = tensor<string, []>("op_1796_cast_fp16")];
            tensor<int32, [4]> var_1797 = const()[name = tensor<string, []>("op_1797"), val = tensor<int32, [4]>([1, 8, 125, 249])];
            tensor<fp16, [1, 8, 125, 249]> matrix_bd_33_cast_fp16 = reshape(shape = var_1797, x = var_1796_cast_fp16)[name = tensor<string, []>("matrix_bd_33_cast_fp16")];
            tensor<bool, []> matrix_ac_17_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_17_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matrix_ac_17_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_17_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_84_perm_0 = const()[name = tensor<string, []>("transpose_84_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_85_perm_0 = const()[name = tensor<string, []>("transpose_85_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 64, 125]> transpose_85 = transpose(perm = transpose_85_perm_0, x = k_33_cast_fp16)[name = tensor<string, []>("transpose_162")];
            tensor<fp16, [1, 8, 125, 64]> transpose_84 = transpose(perm = transpose_84_perm_0, x = var_1780_cast_fp16)[name = tensor<string, []>("transpose_163")];
            tensor<fp16, [1, 8, 125, 125]> matrix_ac_17_cast_fp16 = matmul(transpose_x = matrix_ac_17_transpose_x_0, transpose_y = matrix_ac_17_transpose_y_0, x = transpose_84, y = transpose_85)[name = tensor<string, []>("matrix_ac_17_cast_fp16")];
            tensor<int32, [4]> matrix_bd_35_begin_0 = const()[name = tensor<string, []>("matrix_bd_35_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_35_end_0 = const()[name = tensor<string, []>("matrix_bd_35_end_0"), val = tensor<int32, [4]>([1, 8, 125, 125])];
            tensor<bool, [4]> matrix_bd_35_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_35_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 125, 125]> matrix_bd_35_cast_fp16 = slice_by_index(begin = matrix_bd_35_begin_0, end = matrix_bd_35_end_0, end_mask = matrix_bd_35_end_mask_0, x = matrix_bd_33_cast_fp16)[name = tensor<string, []>("matrix_bd_35_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_1806_cast_fp16 = add(x = matrix_ac_17_cast_fp16, y = matrix_bd_35_cast_fp16)[name = tensor<string, []>("op_1806_cast_fp16")];
            tensor<fp16, []> _inversed_scores_33_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_33_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 8, 125, 125]> _inversed_scores_33_cast_fp16 = mul(x = var_1806_cast_fp16, y = _inversed_scores_33_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_33_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> scores_35_cast_fp16 = select(a = var_8_to_fp16, b = _inversed_scores_33_cast_fp16, cond = mask_11)[name = tensor<string, []>("scores_35_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_1812_cast_fp16 = softmax(axis = var_19, x = scores_35_cast_fp16)[name = tensor<string, []>("op_1812_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> input_449_cast_fp16 = select(a = var_7_to_fp16, b = var_1812_cast_fp16, cond = mask_11)[name = tensor<string, []>("input_449_cast_fp16")];
            tensor<bool, []> x_189_transpose_x_0 = const()[name = tensor<string, []>("x_189_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_189_transpose_y_0 = const()[name = tensor<string, []>("x_189_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 125, 64]> value_17_cast_fp16 = transpose(perm = value_17_perm_0, x = v_17_cast_fp16)[name = tensor<string, []>("transpose_161")];
            tensor<fp16, [1, 8, 125, 64]> x_189_cast_fp16 = matmul(transpose_x = x_189_transpose_x_0, transpose_y = x_189_transpose_y_0, x = input_449_cast_fp16, y = value_17_cast_fp16)[name = tensor<string, []>("x_189_cast_fp16")];
            tensor<int32, [4]> var_1816_perm_0 = const()[name = tensor<string, []>("op_1816_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1817 = const()[name = tensor<string, []>("op_1817"), val = tensor<int32, [3]>([1, -1, 512])];
            tensor<fp16, [1, 125, 8, 64]> var_1816_cast_fp16 = transpose(perm = var_1816_perm_0, x = x_189_cast_fp16)[name = tensor<string, []>("transpose_160")];
            tensor<fp16, [1, 125, 512]> input_451_cast_fp16 = reshape(shape = var_1817, x = var_1816_cast_fp16)[name = tensor<string, []>("input_451_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_8_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_8_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(54996160))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55258368))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_8_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55259456)))];
            tensor<fp16, [1, 125, 512]> linear_79_cast_fp16 = linear(bias = encoder_layers_8_self_attn_linear_out_bias_to_fp16, weight = encoder_layers_8_self_attn_linear_out_weight_to_fp16_quantized, x = input_451_cast_fp16)[name = tensor<string, []>("linear_79_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_455_cast_fp16 = add(x = input_447_cast_fp16, y = linear_79_cast_fp16)[name = tensor<string, []>("input_455_cast_fp16")];
            tensor<int32, [1]> x_193_axes_0 = const()[name = tensor<string, []>("x_193_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_8_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_norm_conv_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55260544)))];
            tensor<fp16, [512]> encoder_layers_8_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_norm_conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55261632)))];
            tensor<fp16, [1, 125, 512]> x_193_cast_fp16 = layer_norm(axes = x_193_axes_0, beta = encoder_layers_8_norm_conv_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_8_norm_conv_weight_to_fp16, x = input_455_cast_fp16)[name = tensor<string, []>("x_193_cast_fp16")];
            tensor<int32, [3]> input_457_perm_0 = const()[name = tensor<string, []>("input_457_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> input_459_pad_type_0 = const()[name = tensor<string, []>("input_459_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> input_459_strides_0 = const()[name = tensor<string, []>("input_459_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_459_pad_0 = const()[name = tensor<string, []>("input_459_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_459_dilations_0 = const()[name = tensor<string, []>("input_459_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> input_459_groups_0 = const()[name = tensor<string, []>("input_459_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 512, 1]> encoder_layers_8_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_8_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1024, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55262720))), scale = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55787072))), zero_point = tensor<int8, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6073344)))];
            tensor<fp16, [1024]> encoder_layers_8_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55789184)))];
            tensor<fp16, [1, 512, 125]> input_457_cast_fp16 = transpose(perm = input_457_perm_0, x = x_193_cast_fp16)[name = tensor<string, []>("transpose_159")];
            tensor<fp16, [1, 1024, 125]> input_459_cast_fp16 = conv(bias = encoder_layers_8_conv_pointwise_conv1_bias_to_fp16, dilations = input_459_dilations_0, groups = input_459_groups_0, pad = input_459_pad_0, pad_type = input_459_pad_type_0, strides = input_459_strides_0, weight = encoder_layers_8_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_457_cast_fp16)[name = tensor<string, []>("input_459_cast_fp16")];
            tensor<int32, []> x_195_split_num_splits_0 = const()[name = tensor<string, []>("x_195_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> x_195_split_axis_0 = const()[name = tensor<string, []>("x_195_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 512, 125]> x_195_split_cast_fp16_0, tensor<fp16, [1, 512, 125]> x_195_split_cast_fp16_1 = split(axis = x_195_split_axis_0, num_splits = x_195_split_num_splits_0, x = input_459_cast_fp16)[name = tensor<string, []>("x_195_split_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_195_split_1_sigmoid_cast_fp16 = sigmoid(x = x_195_split_cast_fp16_1)[name = tensor<string, []>("x_195_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_195_cast_fp16 = mul(x = x_195_split_cast_fp16_0, y = x_195_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_195_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_461_cast_fp16 = select(a = var_7_to_fp16, b = x_195_cast_fp16, cond = var_449)[name = tensor<string, []>("input_461_cast_fp16")];
            tensor<int32, [6]> input_463_pad_0 = const()[name = tensor<string, []>("input_463_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            tensor<string, []> input_463_mode_0 = const()[name = tensor<string, []>("input_463_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_153_to_fp16 = const()[name = tensor<string, []>("const_153_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 512, 133]> input_463_cast_fp16 = pad(constant_val = const_153_to_fp16, mode = input_463_mode_0, pad = input_463_pad_0, x = input_461_cast_fp16)[name = tensor<string, []>("input_463_cast_fp16")];
            tensor<string, []> input_465_pad_type_0 = const()[name = tensor<string, []>("input_465_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, []> input_465_groups_0 = const()[name = tensor<string, []>("input_465_groups_0"), val = tensor<int32, []>(512)];
            tensor<int32, [1]> input_465_strides_0 = const()[name = tensor<string, []>("input_465_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_465_pad_0 = const()[name = tensor<string, []>("input_465_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_465_dilations_0 = const()[name = tensor<string, []>("input_465_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 9]> const_253_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("const_253_to_fp16_quantized"), quantized_data = tensor<int8, [512, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55791296))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55795968))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> const_254_to_fp16 = const()[name = tensor<string, []>("const_254_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55797056)))];
            tensor<fp16, [1, 512, 125]> input_467_cast_fp16 = conv(bias = const_254_to_fp16, dilations = input_465_dilations_0, groups = input_465_groups_0, pad = input_465_pad_0, pad_type = input_465_pad_type_0, strides = input_465_strides_0, weight = const_253_to_fp16_quantized, x = input_463_cast_fp16)[name = tensor<string, []>("input_467_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_469_cast_fp16 = silu(x = input_467_cast_fp16)[name = tensor<string, []>("input_469_cast_fp16")];
            tensor<string, []> x_197_pad_type_0 = const()[name = tensor<string, []>("x_197_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> x_197_strides_0 = const()[name = tensor<string, []>("x_197_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_197_pad_0 = const()[name = tensor<string, []>("x_197_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_197_dilations_0 = const()[name = tensor<string, []>("x_197_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> x_197_groups_0 = const()[name = tensor<string, []>("x_197_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [512, 512, 1]> encoder_layers_8_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_8_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(55798144))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56060352))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_8_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56061440)))];
            tensor<fp16, [1, 512, 125]> x_197_cast_fp16 = conv(bias = encoder_layers_8_conv_pointwise_conv2_bias_to_fp16, dilations = x_197_dilations_0, groups = x_197_groups_0, pad = x_197_pad_0, pad_type = x_197_pad_type_0, strides = x_197_strides_0, weight = encoder_layers_8_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_469_cast_fp16)[name = tensor<string, []>("x_197_cast_fp16")];
            tensor<int32, [3]> input_471_perm_0 = const()[name = tensor<string, []>("input_471_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 125, 512]> input_471_cast_fp16 = transpose(perm = input_471_perm_0, x = x_197_cast_fp16)[name = tensor<string, []>("transpose_158")];
            tensor<fp16, [1, 125, 512]> input_473_cast_fp16 = add(x = input_455_cast_fp16, y = input_471_cast_fp16)[name = tensor<string, []>("input_473_cast_fp16")];
            tensor<int32, [1]> input_475_axes_0 = const()[name = tensor<string, []>("input_475_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_8_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56062528)))];
            tensor<fp16, [512]> encoder_layers_8_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56063616)))];
            tensor<fp16, [1, 125, 512]> input_475_cast_fp16 = layer_norm(axes = input_475_axes_0, beta = encoder_layers_8_norm_feed_forward2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_8_norm_feed_forward2_weight_to_fp16, x = input_473_cast_fp16)[name = tensor<string, []>("input_475_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_8_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_8_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(56064704))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57113344))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_8_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57117504)))];
            tensor<fp16, [1, 125, 2048]> linear_80_cast_fp16 = linear(bias = encoder_layers_8_feed_forward2_linear1_bias_to_fp16, weight = encoder_layers_8_feed_forward2_linear1_weight_to_fp16_quantized, x = input_475_cast_fp16)[name = tensor<string, []>("linear_80_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_479_cast_fp16 = silu(x = linear_80_cast_fp16)[name = tensor<string, []>("input_479_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_8_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_8_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(57121664))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58170304))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_8_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58171392)))];
            tensor<fp16, [1, 125, 512]> linear_81_cast_fp16 = linear(bias = encoder_layers_8_feed_forward2_linear2_bias_to_fp16, weight = encoder_layers_8_feed_forward2_linear2_weight_to_fp16_quantized, x = input_479_cast_fp16)[name = tensor<string, []>("linear_81_cast_fp16")];
            tensor<fp16, []> var_1883_to_fp16 = const()[name = tensor<string, []>("op_1883_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_1884_cast_fp16 = mul(x = linear_81_cast_fp16, y = var_1883_to_fp16)[name = tensor<string, []>("op_1884_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_485_cast_fp16 = add(x = input_473_cast_fp16, y = var_1884_cast_fp16)[name = tensor<string, []>("input_485_cast_fp16")];
            tensor<int32, [1]> input_487_axes_0 = const()[name = tensor<string, []>("input_487_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_8_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_norm_out_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58172480)))];
            tensor<fp16, [512]> encoder_layers_8_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_8_norm_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58173568)))];
            tensor<fp16, [1, 125, 512]> input_487_cast_fp16 = layer_norm(axes = input_487_axes_0, beta = encoder_layers_8_norm_out_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_8_norm_out_weight_to_fp16, x = input_485_cast_fp16)[name = tensor<string, []>("input_487_cast_fp16")];
            tensor<int32, [1]> input_489_axes_0 = const()[name = tensor<string, []>("input_489_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_9_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58174656)))];
            tensor<fp16, [512]> encoder_layers_9_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58175744)))];
            tensor<fp16, [1, 125, 512]> input_489_cast_fp16 = layer_norm(axes = input_489_axes_0, beta = encoder_layers_9_norm_feed_forward1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_9_norm_feed_forward1_weight_to_fp16, x = input_487_cast_fp16)[name = tensor<string, []>("input_489_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_9_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_9_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(58176832))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59225472))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_9_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59229632)))];
            tensor<fp16, [1, 125, 2048]> linear_82_cast_fp16 = linear(bias = encoder_layers_9_feed_forward1_linear1_bias_to_fp16, weight = encoder_layers_9_feed_forward1_linear1_weight_to_fp16_quantized, x = input_489_cast_fp16)[name = tensor<string, []>("linear_82_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_493_cast_fp16 = silu(x = linear_82_cast_fp16)[name = tensor<string, []>("input_493_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_9_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_9_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(59233792))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60282432))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_9_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60283520)))];
            tensor<fp16, [1, 125, 512]> linear_83_cast_fp16 = linear(bias = encoder_layers_9_feed_forward1_linear2_bias_to_fp16, weight = encoder_layers_9_feed_forward1_linear2_weight_to_fp16_quantized, x = input_493_cast_fp16)[name = tensor<string, []>("linear_83_cast_fp16")];
            tensor<fp16, []> var_1914_to_fp16 = const()[name = tensor<string, []>("op_1914_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_1915_cast_fp16 = mul(x = linear_83_cast_fp16, y = var_1914_to_fp16)[name = tensor<string, []>("op_1915_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_499_cast_fp16 = add(x = input_487_cast_fp16, y = var_1915_cast_fp16)[name = tensor<string, []>("input_499_cast_fp16")];
            tensor<int32, [1]> query_19_axes_0 = const()[name = tensor<string, []>("query_19_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_9_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_norm_self_att_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60284608)))];
            tensor<fp16, [512]> encoder_layers_9_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_norm_self_att_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60285696)))];
            tensor<fp16, [1, 125, 512]> query_19_cast_fp16 = layer_norm(axes = query_19_axes_0, beta = encoder_layers_9_norm_self_att_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_9_norm_self_att_weight_to_fp16, x = input_499_cast_fp16)[name = tensor<string, []>("query_19_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_9_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_9_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60286784))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60548992))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_9_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60550080)))];
            tensor<fp16, [1, 125, 512]> linear_84_cast_fp16 = linear(bias = encoder_layers_9_self_attn_linear_q_bias_to_fp16, weight = encoder_layers_9_self_attn_linear_q_weight_to_fp16_quantized, x = query_19_cast_fp16)[name = tensor<string, []>("linear_84_cast_fp16")];
            tensor<int32, [4]> var_1932 = const()[name = tensor<string, []>("op_1932"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> q_55_cast_fp16 = reshape(shape = var_1932, x = linear_84_cast_fp16)[name = tensor<string, []>("q_55_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_9_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_9_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60551168))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60813376))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_9_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60814464)))];
            tensor<fp16, [1, 125, 512]> linear_85_cast_fp16 = linear(bias = encoder_layers_9_self_attn_linear_k_bias_to_fp16, weight = encoder_layers_9_self_attn_linear_k_weight_to_fp16_quantized, x = query_19_cast_fp16)[name = tensor<string, []>("linear_85_cast_fp16")];
            tensor<int32, [4]> var_1937 = const()[name = tensor<string, []>("op_1937"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> k_37_cast_fp16 = reshape(shape = var_1937, x = linear_85_cast_fp16)[name = tensor<string, []>("k_37_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_9_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_9_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(60815552))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61077760))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_9_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61078848)))];
            tensor<fp16, [1, 125, 512]> linear_86_cast_fp16 = linear(bias = encoder_layers_9_self_attn_linear_v_bias_to_fp16, weight = encoder_layers_9_self_attn_linear_v_weight_to_fp16_quantized, x = query_19_cast_fp16)[name = tensor<string, []>("linear_86_cast_fp16")];
            tensor<int32, [4]> var_1942 = const()[name = tensor<string, []>("op_1942"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> v_19_cast_fp16 = reshape(shape = var_1942, x = linear_86_cast_fp16)[name = tensor<string, []>("v_19_cast_fp16")];
            tensor<int32, [4]> value_19_perm_0 = const()[name = tensor<string, []>("value_19_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 64]> encoder_layers_9_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61079936)))];
            tensor<fp16, [1, 125, 8, 64]> var_1954_cast_fp16 = add(x = q_55_cast_fp16, y = encoder_layers_9_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_1954_cast_fp16")];
            tensor<fp16, [8, 64]> encoder_layers_9_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61081024)))];
            tensor<fp16, [1, 125, 8, 64]> var_1956_cast_fp16 = add(x = q_55_cast_fp16, y = encoder_layers_9_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_1956_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_19_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_19_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<bool, []> x_205_transpose_x_0 = const()[name = tensor<string, []>("x_205_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_205_transpose_y_0 = const()[name = tensor<string, []>("x_205_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 64, 249]> op_1958_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(3), name = tensor<string, []>("op_1958_to_fp16_quantized"), quantized_data = tensor<int8, [1, 8, 64, 249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61082112))), scale = tensor<fp16, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61209664))), zero_point = tensor<int8, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5281536)))];
            tensor<fp16, [1, 8, 125, 64]> q_with_bias_v_19_cast_fp16 = transpose(perm = q_with_bias_v_19_perm_0, x = var_1956_cast_fp16)[name = tensor<string, []>("transpose_157")];
            tensor<fp16, [1, 8, 125, 249]> x_205_cast_fp16 = matmul(transpose_x = x_205_transpose_x_0, transpose_y = x_205_transpose_y_0, x = q_with_bias_v_19_cast_fp16, y = op_1958_to_fp16_quantized)[name = tensor<string, []>("x_205_cast_fp16")];
            tensor<int32, [8]> x_207_pad_0 = const()[name = tensor<string, []>("x_207_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            tensor<string, []> x_207_mode_0 = const()[name = tensor<string, []>("x_207_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_160_to_fp16 = const()[name = tensor<string, []>("const_160_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 8, 125, 250]> x_207_cast_fp16 = pad(constant_val = const_160_to_fp16, mode = x_207_mode_0, pad = x_207_pad_0, x = x_205_cast_fp16)[name = tensor<string, []>("x_207_cast_fp16")];
            tensor<int32, [4]> var_1966 = const()[name = tensor<string, []>("op_1966"), val = tensor<int32, [4]>([1, 8, -1, 125])];
            tensor<fp16, [1, 8, 250, 125]> x_209_cast_fp16 = reshape(shape = var_1966, x = x_207_cast_fp16)[name = tensor<string, []>("x_209_cast_fp16")];
            tensor<int32, [4]> var_1970_begin_0 = const()[name = tensor<string, []>("op_1970_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_1970_end_0 = const()[name = tensor<string, []>("op_1970_end_0"), val = tensor<int32, [4]>([1, 8, 250, 125])];
            tensor<bool, [4]> var_1970_end_mask_0 = const()[name = tensor<string, []>("op_1970_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 249, 125]> var_1970_cast_fp16 = slice_by_index(begin = var_1970_begin_0, end = var_1970_end_0, end_mask = var_1970_end_mask_0, x = x_209_cast_fp16)[name = tensor<string, []>("op_1970_cast_fp16")];
            tensor<int32, [4]> var_1971 = const()[name = tensor<string, []>("op_1971"), val = tensor<int32, [4]>([1, 8, 125, 249])];
            tensor<fp16, [1, 8, 125, 249]> matrix_bd_37_cast_fp16 = reshape(shape = var_1971, x = var_1970_cast_fp16)[name = tensor<string, []>("matrix_bd_37_cast_fp16")];
            tensor<bool, []> matrix_ac_19_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_19_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matrix_ac_19_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_19_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_86_perm_0 = const()[name = tensor<string, []>("transpose_86_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_87_perm_0 = const()[name = tensor<string, []>("transpose_87_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 64, 125]> transpose_87 = transpose(perm = transpose_87_perm_0, x = k_37_cast_fp16)[name = tensor<string, []>("transpose_155")];
            tensor<fp16, [1, 8, 125, 64]> transpose_86 = transpose(perm = transpose_86_perm_0, x = var_1954_cast_fp16)[name = tensor<string, []>("transpose_156")];
            tensor<fp16, [1, 8, 125, 125]> matrix_ac_19_cast_fp16 = matmul(transpose_x = matrix_ac_19_transpose_x_0, transpose_y = matrix_ac_19_transpose_y_0, x = transpose_86, y = transpose_87)[name = tensor<string, []>("matrix_ac_19_cast_fp16")];
            tensor<int32, [4]> matrix_bd_39_begin_0 = const()[name = tensor<string, []>("matrix_bd_39_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_39_end_0 = const()[name = tensor<string, []>("matrix_bd_39_end_0"), val = tensor<int32, [4]>([1, 8, 125, 125])];
            tensor<bool, [4]> matrix_bd_39_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_39_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 125, 125]> matrix_bd_39_cast_fp16 = slice_by_index(begin = matrix_bd_39_begin_0, end = matrix_bd_39_end_0, end_mask = matrix_bd_39_end_mask_0, x = matrix_bd_37_cast_fp16)[name = tensor<string, []>("matrix_bd_39_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_1980_cast_fp16 = add(x = matrix_ac_19_cast_fp16, y = matrix_bd_39_cast_fp16)[name = tensor<string, []>("op_1980_cast_fp16")];
            tensor<fp16, []> _inversed_scores_37_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_37_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 8, 125, 125]> _inversed_scores_37_cast_fp16 = mul(x = var_1980_cast_fp16, y = _inversed_scores_37_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_37_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> scores_39_cast_fp16 = select(a = var_8_to_fp16, b = _inversed_scores_37_cast_fp16, cond = mask_11)[name = tensor<string, []>("scores_39_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_1986_cast_fp16 = softmax(axis = var_19, x = scores_39_cast_fp16)[name = tensor<string, []>("op_1986_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> input_501_cast_fp16 = select(a = var_7_to_fp16, b = var_1986_cast_fp16, cond = mask_11)[name = tensor<string, []>("input_501_cast_fp16")];
            tensor<bool, []> x_211_transpose_x_0 = const()[name = tensor<string, []>("x_211_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_211_transpose_y_0 = const()[name = tensor<string, []>("x_211_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 125, 64]> value_19_cast_fp16 = transpose(perm = value_19_perm_0, x = v_19_cast_fp16)[name = tensor<string, []>("transpose_154")];
            tensor<fp16, [1, 8, 125, 64]> x_211_cast_fp16 = matmul(transpose_x = x_211_transpose_x_0, transpose_y = x_211_transpose_y_0, x = input_501_cast_fp16, y = value_19_cast_fp16)[name = tensor<string, []>("x_211_cast_fp16")];
            tensor<int32, [4]> var_1990_perm_0 = const()[name = tensor<string, []>("op_1990_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_1991 = const()[name = tensor<string, []>("op_1991"), val = tensor<int32, [3]>([1, -1, 512])];
            tensor<fp16, [1, 125, 8, 64]> var_1990_cast_fp16 = transpose(perm = var_1990_perm_0, x = x_211_cast_fp16)[name = tensor<string, []>("transpose_153")];
            tensor<fp16, [1, 125, 512]> input_503_cast_fp16 = reshape(shape = var_1991, x = var_1990_cast_fp16)[name = tensor<string, []>("input_503_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_9_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_9_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61210240))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61472448))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_9_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61473536)))];
            tensor<fp16, [1, 125, 512]> linear_88_cast_fp16 = linear(bias = encoder_layers_9_self_attn_linear_out_bias_to_fp16, weight = encoder_layers_9_self_attn_linear_out_weight_to_fp16_quantized, x = input_503_cast_fp16)[name = tensor<string, []>("linear_88_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_507_cast_fp16 = add(x = input_499_cast_fp16, y = linear_88_cast_fp16)[name = tensor<string, []>("input_507_cast_fp16")];
            tensor<int32, [1]> x_215_axes_0 = const()[name = tensor<string, []>("x_215_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_9_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_norm_conv_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61474624)))];
            tensor<fp16, [512]> encoder_layers_9_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_norm_conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61475712)))];
            tensor<fp16, [1, 125, 512]> x_215_cast_fp16 = layer_norm(axes = x_215_axes_0, beta = encoder_layers_9_norm_conv_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_9_norm_conv_weight_to_fp16, x = input_507_cast_fp16)[name = tensor<string, []>("x_215_cast_fp16")];
            tensor<int32, [3]> input_509_perm_0 = const()[name = tensor<string, []>("input_509_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> input_511_pad_type_0 = const()[name = tensor<string, []>("input_511_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> input_511_strides_0 = const()[name = tensor<string, []>("input_511_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_511_pad_0 = const()[name = tensor<string, []>("input_511_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_511_dilations_0 = const()[name = tensor<string, []>("input_511_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> input_511_groups_0 = const()[name = tensor<string, []>("input_511_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 512, 1]> encoder_layers_9_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_9_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1024, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(61476800))), scale = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62001152))), zero_point = tensor<int8, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6073344)))];
            tensor<fp16, [1024]> encoder_layers_9_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62003264)))];
            tensor<fp16, [1, 512, 125]> input_509_cast_fp16 = transpose(perm = input_509_perm_0, x = x_215_cast_fp16)[name = tensor<string, []>("transpose_152")];
            tensor<fp16, [1, 1024, 125]> input_511_cast_fp16 = conv(bias = encoder_layers_9_conv_pointwise_conv1_bias_to_fp16, dilations = input_511_dilations_0, groups = input_511_groups_0, pad = input_511_pad_0, pad_type = input_511_pad_type_0, strides = input_511_strides_0, weight = encoder_layers_9_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_509_cast_fp16)[name = tensor<string, []>("input_511_cast_fp16")];
            tensor<int32, []> x_217_split_num_splits_0 = const()[name = tensor<string, []>("x_217_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> x_217_split_axis_0 = const()[name = tensor<string, []>("x_217_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 512, 125]> x_217_split_cast_fp16_0, tensor<fp16, [1, 512, 125]> x_217_split_cast_fp16_1 = split(axis = x_217_split_axis_0, num_splits = x_217_split_num_splits_0, x = input_511_cast_fp16)[name = tensor<string, []>("x_217_split_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_217_split_1_sigmoid_cast_fp16 = sigmoid(x = x_217_split_cast_fp16_1)[name = tensor<string, []>("x_217_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_217_cast_fp16 = mul(x = x_217_split_cast_fp16_0, y = x_217_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_217_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_513_cast_fp16 = select(a = var_7_to_fp16, b = x_217_cast_fp16, cond = var_449)[name = tensor<string, []>("input_513_cast_fp16")];
            tensor<int32, [6]> input_515_pad_0 = const()[name = tensor<string, []>("input_515_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            tensor<string, []> input_515_mode_0 = const()[name = tensor<string, []>("input_515_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_163_to_fp16 = const()[name = tensor<string, []>("const_163_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 512, 133]> input_515_cast_fp16 = pad(constant_val = const_163_to_fp16, mode = input_515_mode_0, pad = input_515_pad_0, x = input_513_cast_fp16)[name = tensor<string, []>("input_515_cast_fp16")];
            tensor<string, []> input_517_pad_type_0 = const()[name = tensor<string, []>("input_517_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, []> input_517_groups_0 = const()[name = tensor<string, []>("input_517_groups_0"), val = tensor<int32, []>(512)];
            tensor<int32, [1]> input_517_strides_0 = const()[name = tensor<string, []>("input_517_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_517_pad_0 = const()[name = tensor<string, []>("input_517_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_517_dilations_0 = const()[name = tensor<string, []>("input_517_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 9]> const_255_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("const_255_to_fp16_quantized"), quantized_data = tensor<int8, [512, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62005376))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62010048))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> const_256_to_fp16 = const()[name = tensor<string, []>("const_256_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62011136)))];
            tensor<fp16, [1, 512, 125]> input_519_cast_fp16 = conv(bias = const_256_to_fp16, dilations = input_517_dilations_0, groups = input_517_groups_0, pad = input_517_pad_0, pad_type = input_517_pad_type_0, strides = input_517_strides_0, weight = const_255_to_fp16_quantized, x = input_515_cast_fp16)[name = tensor<string, []>("input_519_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_521_cast_fp16 = silu(x = input_519_cast_fp16)[name = tensor<string, []>("input_521_cast_fp16")];
            tensor<string, []> x_219_pad_type_0 = const()[name = tensor<string, []>("x_219_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> x_219_strides_0 = const()[name = tensor<string, []>("x_219_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_219_pad_0 = const()[name = tensor<string, []>("x_219_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_219_dilations_0 = const()[name = tensor<string, []>("x_219_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> x_219_groups_0 = const()[name = tensor<string, []>("x_219_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [512, 512, 1]> encoder_layers_9_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_9_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62012224))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62274432))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_9_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62275520)))];
            tensor<fp16, [1, 512, 125]> x_219_cast_fp16 = conv(bias = encoder_layers_9_conv_pointwise_conv2_bias_to_fp16, dilations = x_219_dilations_0, groups = x_219_groups_0, pad = x_219_pad_0, pad_type = x_219_pad_type_0, strides = x_219_strides_0, weight = encoder_layers_9_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_521_cast_fp16)[name = tensor<string, []>("x_219_cast_fp16")];
            tensor<int32, [3]> input_523_perm_0 = const()[name = tensor<string, []>("input_523_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 125, 512]> input_523_cast_fp16 = transpose(perm = input_523_perm_0, x = x_219_cast_fp16)[name = tensor<string, []>("transpose_151")];
            tensor<fp16, [1, 125, 512]> input_525_cast_fp16 = add(x = input_507_cast_fp16, y = input_523_cast_fp16)[name = tensor<string, []>("input_525_cast_fp16")];
            tensor<int32, [1]> input_527_axes_0 = const()[name = tensor<string, []>("input_527_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_9_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62276608)))];
            tensor<fp16, [512]> encoder_layers_9_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62277696)))];
            tensor<fp16, [1, 125, 512]> input_527_cast_fp16 = layer_norm(axes = input_527_axes_0, beta = encoder_layers_9_norm_feed_forward2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_9_norm_feed_forward2_weight_to_fp16, x = input_525_cast_fp16)[name = tensor<string, []>("input_527_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_9_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_9_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(62278784))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63327424))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_9_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63331584)))];
            tensor<fp16, [1, 125, 2048]> linear_89_cast_fp16 = linear(bias = encoder_layers_9_feed_forward2_linear1_bias_to_fp16, weight = encoder_layers_9_feed_forward2_linear1_weight_to_fp16_quantized, x = input_527_cast_fp16)[name = tensor<string, []>("linear_89_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_531_cast_fp16 = silu(x = linear_89_cast_fp16)[name = tensor<string, []>("input_531_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_9_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_9_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(63335744))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64384384))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_9_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64385472)))];
            tensor<fp16, [1, 125, 512]> linear_90_cast_fp16 = linear(bias = encoder_layers_9_feed_forward2_linear2_bias_to_fp16, weight = encoder_layers_9_feed_forward2_linear2_weight_to_fp16_quantized, x = input_531_cast_fp16)[name = tensor<string, []>("linear_90_cast_fp16")];
            tensor<fp16, []> var_2057_to_fp16 = const()[name = tensor<string, []>("op_2057_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_2058_cast_fp16 = mul(x = linear_90_cast_fp16, y = var_2057_to_fp16)[name = tensor<string, []>("op_2058_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_537_cast_fp16 = add(x = input_525_cast_fp16, y = var_2058_cast_fp16)[name = tensor<string, []>("input_537_cast_fp16")];
            tensor<int32, [1]> input_539_axes_0 = const()[name = tensor<string, []>("input_539_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_9_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_norm_out_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64386560)))];
            tensor<fp16, [512]> encoder_layers_9_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_9_norm_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64387648)))];
            tensor<fp16, [1, 125, 512]> input_539_cast_fp16 = layer_norm(axes = input_539_axes_0, beta = encoder_layers_9_norm_out_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_9_norm_out_weight_to_fp16, x = input_537_cast_fp16)[name = tensor<string, []>("input_539_cast_fp16")];
            tensor<int32, [1]> input_541_axes_0 = const()[name = tensor<string, []>("input_541_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_10_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64388736)))];
            tensor<fp16, [512]> encoder_layers_10_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64389824)))];
            tensor<fp16, [1, 125, 512]> input_541_cast_fp16 = layer_norm(axes = input_541_axes_0, beta = encoder_layers_10_norm_feed_forward1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_10_norm_feed_forward1_weight_to_fp16, x = input_539_cast_fp16)[name = tensor<string, []>("input_541_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_10_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_10_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(64390912))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65439552))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_10_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65443712)))];
            tensor<fp16, [1, 125, 2048]> linear_91_cast_fp16 = linear(bias = encoder_layers_10_feed_forward1_linear1_bias_to_fp16, weight = encoder_layers_10_feed_forward1_linear1_weight_to_fp16_quantized, x = input_541_cast_fp16)[name = tensor<string, []>("linear_91_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_545_cast_fp16 = silu(x = linear_91_cast_fp16)[name = tensor<string, []>("input_545_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_10_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_10_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(65447872))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66496512))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_10_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66497600)))];
            tensor<fp16, [1, 125, 512]> linear_92_cast_fp16 = linear(bias = encoder_layers_10_feed_forward1_linear2_bias_to_fp16, weight = encoder_layers_10_feed_forward1_linear2_weight_to_fp16_quantized, x = input_545_cast_fp16)[name = tensor<string, []>("linear_92_cast_fp16")];
            tensor<fp16, []> var_2088_to_fp16 = const()[name = tensor<string, []>("op_2088_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_2089_cast_fp16 = mul(x = linear_92_cast_fp16, y = var_2088_to_fp16)[name = tensor<string, []>("op_2089_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_551_cast_fp16 = add(x = input_539_cast_fp16, y = var_2089_cast_fp16)[name = tensor<string, []>("input_551_cast_fp16")];
            tensor<int32, [1]> query_21_axes_0 = const()[name = tensor<string, []>("query_21_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_10_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_norm_self_att_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66498688)))];
            tensor<fp16, [512]> encoder_layers_10_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_norm_self_att_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66499776)))];
            tensor<fp16, [1, 125, 512]> query_21_cast_fp16 = layer_norm(axes = query_21_axes_0, beta = encoder_layers_10_norm_self_att_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_10_norm_self_att_weight_to_fp16, x = input_551_cast_fp16)[name = tensor<string, []>("query_21_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_10_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_10_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66500864))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66763072))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_10_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66764160)))];
            tensor<fp16, [1, 125, 512]> linear_93_cast_fp16 = linear(bias = encoder_layers_10_self_attn_linear_q_bias_to_fp16, weight = encoder_layers_10_self_attn_linear_q_weight_to_fp16_quantized, x = query_21_cast_fp16)[name = tensor<string, []>("linear_93_cast_fp16")];
            tensor<int32, [4]> var_2106 = const()[name = tensor<string, []>("op_2106"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> q_61_cast_fp16 = reshape(shape = var_2106, x = linear_93_cast_fp16)[name = tensor<string, []>("q_61_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_10_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_10_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(66765248))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67027456))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_10_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67028544)))];
            tensor<fp16, [1, 125, 512]> linear_94_cast_fp16 = linear(bias = encoder_layers_10_self_attn_linear_k_bias_to_fp16, weight = encoder_layers_10_self_attn_linear_k_weight_to_fp16_quantized, x = query_21_cast_fp16)[name = tensor<string, []>("linear_94_cast_fp16")];
            tensor<int32, [4]> var_2111 = const()[name = tensor<string, []>("op_2111"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> k_41_cast_fp16 = reshape(shape = var_2111, x = linear_94_cast_fp16)[name = tensor<string, []>("k_41_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_10_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_10_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67029632))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67291840))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_10_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67292928)))];
            tensor<fp16, [1, 125, 512]> linear_95_cast_fp16 = linear(bias = encoder_layers_10_self_attn_linear_v_bias_to_fp16, weight = encoder_layers_10_self_attn_linear_v_weight_to_fp16_quantized, x = query_21_cast_fp16)[name = tensor<string, []>("linear_95_cast_fp16")];
            tensor<int32, [4]> var_2116 = const()[name = tensor<string, []>("op_2116"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> v_21_cast_fp16 = reshape(shape = var_2116, x = linear_95_cast_fp16)[name = tensor<string, []>("v_21_cast_fp16")];
            tensor<int32, [4]> value_21_perm_0 = const()[name = tensor<string, []>("value_21_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 64]> encoder_layers_10_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67294016)))];
            tensor<fp16, [1, 125, 8, 64]> var_2128_cast_fp16 = add(x = q_61_cast_fp16, y = encoder_layers_10_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_2128_cast_fp16")];
            tensor<fp16, [8, 64]> encoder_layers_10_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67295104)))];
            tensor<fp16, [1, 125, 8, 64]> var_2130_cast_fp16 = add(x = q_61_cast_fp16, y = encoder_layers_10_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_2130_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_21_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_21_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<bool, []> x_227_transpose_x_0 = const()[name = tensor<string, []>("x_227_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_227_transpose_y_0 = const()[name = tensor<string, []>("x_227_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 64, 249]> op_2132_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(3), name = tensor<string, []>("op_2132_to_fp16_quantized"), quantized_data = tensor<int8, [1, 8, 64, 249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67296192))), scale = tensor<fp16, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67423744))), zero_point = tensor<int8, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5281536)))];
            tensor<fp16, [1, 8, 125, 64]> q_with_bias_v_21_cast_fp16 = transpose(perm = q_with_bias_v_21_perm_0, x = var_2130_cast_fp16)[name = tensor<string, []>("transpose_150")];
            tensor<fp16, [1, 8, 125, 249]> x_227_cast_fp16 = matmul(transpose_x = x_227_transpose_x_0, transpose_y = x_227_transpose_y_0, x = q_with_bias_v_21_cast_fp16, y = op_2132_to_fp16_quantized)[name = tensor<string, []>("x_227_cast_fp16")];
            tensor<int32, [8]> x_229_pad_0 = const()[name = tensor<string, []>("x_229_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            tensor<string, []> x_229_mode_0 = const()[name = tensor<string, []>("x_229_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_170_to_fp16 = const()[name = tensor<string, []>("const_170_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 8, 125, 250]> x_229_cast_fp16 = pad(constant_val = const_170_to_fp16, mode = x_229_mode_0, pad = x_229_pad_0, x = x_227_cast_fp16)[name = tensor<string, []>("x_229_cast_fp16")];
            tensor<int32, [4]> var_2140 = const()[name = tensor<string, []>("op_2140"), val = tensor<int32, [4]>([1, 8, -1, 125])];
            tensor<fp16, [1, 8, 250, 125]> x_231_cast_fp16 = reshape(shape = var_2140, x = x_229_cast_fp16)[name = tensor<string, []>("x_231_cast_fp16")];
            tensor<int32, [4]> var_2144_begin_0 = const()[name = tensor<string, []>("op_2144_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_2144_end_0 = const()[name = tensor<string, []>("op_2144_end_0"), val = tensor<int32, [4]>([1, 8, 250, 125])];
            tensor<bool, [4]> var_2144_end_mask_0 = const()[name = tensor<string, []>("op_2144_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 249, 125]> var_2144_cast_fp16 = slice_by_index(begin = var_2144_begin_0, end = var_2144_end_0, end_mask = var_2144_end_mask_0, x = x_231_cast_fp16)[name = tensor<string, []>("op_2144_cast_fp16")];
            tensor<int32, [4]> var_2145 = const()[name = tensor<string, []>("op_2145"), val = tensor<int32, [4]>([1, 8, 125, 249])];
            tensor<fp16, [1, 8, 125, 249]> matrix_bd_41_cast_fp16 = reshape(shape = var_2145, x = var_2144_cast_fp16)[name = tensor<string, []>("matrix_bd_41_cast_fp16")];
            tensor<bool, []> matrix_ac_21_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_21_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matrix_ac_21_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_21_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_88_perm_0 = const()[name = tensor<string, []>("transpose_88_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_89_perm_0 = const()[name = tensor<string, []>("transpose_89_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 64, 125]> transpose_89 = transpose(perm = transpose_89_perm_0, x = k_41_cast_fp16)[name = tensor<string, []>("transpose_148")];
            tensor<fp16, [1, 8, 125, 64]> transpose_88 = transpose(perm = transpose_88_perm_0, x = var_2128_cast_fp16)[name = tensor<string, []>("transpose_149")];
            tensor<fp16, [1, 8, 125, 125]> matrix_ac_21_cast_fp16 = matmul(transpose_x = matrix_ac_21_transpose_x_0, transpose_y = matrix_ac_21_transpose_y_0, x = transpose_88, y = transpose_89)[name = tensor<string, []>("matrix_ac_21_cast_fp16")];
            tensor<int32, [4]> matrix_bd_43_begin_0 = const()[name = tensor<string, []>("matrix_bd_43_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_43_end_0 = const()[name = tensor<string, []>("matrix_bd_43_end_0"), val = tensor<int32, [4]>([1, 8, 125, 125])];
            tensor<bool, [4]> matrix_bd_43_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_43_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 125, 125]> matrix_bd_43_cast_fp16 = slice_by_index(begin = matrix_bd_43_begin_0, end = matrix_bd_43_end_0, end_mask = matrix_bd_43_end_mask_0, x = matrix_bd_41_cast_fp16)[name = tensor<string, []>("matrix_bd_43_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_2154_cast_fp16 = add(x = matrix_ac_21_cast_fp16, y = matrix_bd_43_cast_fp16)[name = tensor<string, []>("op_2154_cast_fp16")];
            tensor<fp16, []> _inversed_scores_41_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_41_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 8, 125, 125]> _inversed_scores_41_cast_fp16 = mul(x = var_2154_cast_fp16, y = _inversed_scores_41_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_41_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> scores_43_cast_fp16 = select(a = var_8_to_fp16, b = _inversed_scores_41_cast_fp16, cond = mask_11)[name = tensor<string, []>("scores_43_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_2160_cast_fp16 = softmax(axis = var_19, x = scores_43_cast_fp16)[name = tensor<string, []>("op_2160_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> input_553_cast_fp16 = select(a = var_7_to_fp16, b = var_2160_cast_fp16, cond = mask_11)[name = tensor<string, []>("input_553_cast_fp16")];
            tensor<bool, []> x_233_transpose_x_0 = const()[name = tensor<string, []>("x_233_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_233_transpose_y_0 = const()[name = tensor<string, []>("x_233_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 125, 64]> value_21_cast_fp16 = transpose(perm = value_21_perm_0, x = v_21_cast_fp16)[name = tensor<string, []>("transpose_147")];
            tensor<fp16, [1, 8, 125, 64]> x_233_cast_fp16 = matmul(transpose_x = x_233_transpose_x_0, transpose_y = x_233_transpose_y_0, x = input_553_cast_fp16, y = value_21_cast_fp16)[name = tensor<string, []>("x_233_cast_fp16")];
            tensor<int32, [4]> var_2164_perm_0 = const()[name = tensor<string, []>("op_2164_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2165 = const()[name = tensor<string, []>("op_2165"), val = tensor<int32, [3]>([1, -1, 512])];
            tensor<fp16, [1, 125, 8, 64]> var_2164_cast_fp16 = transpose(perm = var_2164_perm_0, x = x_233_cast_fp16)[name = tensor<string, []>("transpose_146")];
            tensor<fp16, [1, 125, 512]> input_555_cast_fp16 = reshape(shape = var_2165, x = var_2164_cast_fp16)[name = tensor<string, []>("input_555_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_10_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_10_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67424320))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67686528))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_10_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67687616)))];
            tensor<fp16, [1, 125, 512]> linear_97_cast_fp16 = linear(bias = encoder_layers_10_self_attn_linear_out_bias_to_fp16, weight = encoder_layers_10_self_attn_linear_out_weight_to_fp16_quantized, x = input_555_cast_fp16)[name = tensor<string, []>("linear_97_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_559_cast_fp16 = add(x = input_551_cast_fp16, y = linear_97_cast_fp16)[name = tensor<string, []>("input_559_cast_fp16")];
            tensor<int32, [1]> x_237_axes_0 = const()[name = tensor<string, []>("x_237_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_10_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_norm_conv_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67688704)))];
            tensor<fp16, [512]> encoder_layers_10_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_norm_conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67689792)))];
            tensor<fp16, [1, 125, 512]> x_237_cast_fp16 = layer_norm(axes = x_237_axes_0, beta = encoder_layers_10_norm_conv_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_10_norm_conv_weight_to_fp16, x = input_559_cast_fp16)[name = tensor<string, []>("x_237_cast_fp16")];
            tensor<int32, [3]> input_561_perm_0 = const()[name = tensor<string, []>("input_561_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> input_563_pad_type_0 = const()[name = tensor<string, []>("input_563_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> input_563_strides_0 = const()[name = tensor<string, []>("input_563_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_563_pad_0 = const()[name = tensor<string, []>("input_563_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_563_dilations_0 = const()[name = tensor<string, []>("input_563_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> input_563_groups_0 = const()[name = tensor<string, []>("input_563_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 512, 1]> encoder_layers_10_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_10_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1024, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(67690880))), scale = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68215232))), zero_point = tensor<int8, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6073344)))];
            tensor<fp16, [1024]> encoder_layers_10_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68217344)))];
            tensor<fp16, [1, 512, 125]> input_561_cast_fp16 = transpose(perm = input_561_perm_0, x = x_237_cast_fp16)[name = tensor<string, []>("transpose_145")];
            tensor<fp16, [1, 1024, 125]> input_563_cast_fp16 = conv(bias = encoder_layers_10_conv_pointwise_conv1_bias_to_fp16, dilations = input_563_dilations_0, groups = input_563_groups_0, pad = input_563_pad_0, pad_type = input_563_pad_type_0, strides = input_563_strides_0, weight = encoder_layers_10_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_561_cast_fp16)[name = tensor<string, []>("input_563_cast_fp16")];
            tensor<int32, []> x_239_split_num_splits_0 = const()[name = tensor<string, []>("x_239_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> x_239_split_axis_0 = const()[name = tensor<string, []>("x_239_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 512, 125]> x_239_split_cast_fp16_0, tensor<fp16, [1, 512, 125]> x_239_split_cast_fp16_1 = split(axis = x_239_split_axis_0, num_splits = x_239_split_num_splits_0, x = input_563_cast_fp16)[name = tensor<string, []>("x_239_split_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_239_split_1_sigmoid_cast_fp16 = sigmoid(x = x_239_split_cast_fp16_1)[name = tensor<string, []>("x_239_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_239_cast_fp16 = mul(x = x_239_split_cast_fp16_0, y = x_239_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_239_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_565_cast_fp16 = select(a = var_7_to_fp16, b = x_239_cast_fp16, cond = var_449)[name = tensor<string, []>("input_565_cast_fp16")];
            tensor<int32, [6]> input_567_pad_0 = const()[name = tensor<string, []>("input_567_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            tensor<string, []> input_567_mode_0 = const()[name = tensor<string, []>("input_567_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_173_to_fp16 = const()[name = tensor<string, []>("const_173_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 512, 133]> input_567_cast_fp16 = pad(constant_val = const_173_to_fp16, mode = input_567_mode_0, pad = input_567_pad_0, x = input_565_cast_fp16)[name = tensor<string, []>("input_567_cast_fp16")];
            tensor<string, []> input_569_pad_type_0 = const()[name = tensor<string, []>("input_569_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, []> input_569_groups_0 = const()[name = tensor<string, []>("input_569_groups_0"), val = tensor<int32, []>(512)];
            tensor<int32, [1]> input_569_strides_0 = const()[name = tensor<string, []>("input_569_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_569_pad_0 = const()[name = tensor<string, []>("input_569_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_569_dilations_0 = const()[name = tensor<string, []>("input_569_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 9]> const_257_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("const_257_to_fp16_quantized"), quantized_data = tensor<int8, [512, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68219456))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68224128))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> const_258_to_fp16 = const()[name = tensor<string, []>("const_258_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68225216)))];
            tensor<fp16, [1, 512, 125]> input_571_cast_fp16 = conv(bias = const_258_to_fp16, dilations = input_569_dilations_0, groups = input_569_groups_0, pad = input_569_pad_0, pad_type = input_569_pad_type_0, strides = input_569_strides_0, weight = const_257_to_fp16_quantized, x = input_567_cast_fp16)[name = tensor<string, []>("input_571_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_573_cast_fp16 = silu(x = input_571_cast_fp16)[name = tensor<string, []>("input_573_cast_fp16")];
            tensor<string, []> x_241_pad_type_0 = const()[name = tensor<string, []>("x_241_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> x_241_strides_0 = const()[name = tensor<string, []>("x_241_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_241_pad_0 = const()[name = tensor<string, []>("x_241_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_241_dilations_0 = const()[name = tensor<string, []>("x_241_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> x_241_groups_0 = const()[name = tensor<string, []>("x_241_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [512, 512, 1]> encoder_layers_10_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_10_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68226304))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68488512))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_10_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68489600)))];
            tensor<fp16, [1, 512, 125]> x_241_cast_fp16 = conv(bias = encoder_layers_10_conv_pointwise_conv2_bias_to_fp16, dilations = x_241_dilations_0, groups = x_241_groups_0, pad = x_241_pad_0, pad_type = x_241_pad_type_0, strides = x_241_strides_0, weight = encoder_layers_10_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_573_cast_fp16)[name = tensor<string, []>("x_241_cast_fp16")];
            tensor<int32, [3]> input_575_perm_0 = const()[name = tensor<string, []>("input_575_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 125, 512]> input_575_cast_fp16 = transpose(perm = input_575_perm_0, x = x_241_cast_fp16)[name = tensor<string, []>("transpose_144")];
            tensor<fp16, [1, 125, 512]> input_577_cast_fp16 = add(x = input_559_cast_fp16, y = input_575_cast_fp16)[name = tensor<string, []>("input_577_cast_fp16")];
            tensor<int32, [1]> input_579_axes_0 = const()[name = tensor<string, []>("input_579_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_10_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68490688)))];
            tensor<fp16, [512]> encoder_layers_10_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68491776)))];
            tensor<fp16, [1, 125, 512]> input_579_cast_fp16 = layer_norm(axes = input_579_axes_0, beta = encoder_layers_10_norm_feed_forward2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_10_norm_feed_forward2_weight_to_fp16, x = input_577_cast_fp16)[name = tensor<string, []>("input_579_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_10_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_10_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(68492864))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(69541504))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_10_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(69545664)))];
            tensor<fp16, [1, 125, 2048]> linear_98_cast_fp16 = linear(bias = encoder_layers_10_feed_forward2_linear1_bias_to_fp16, weight = encoder_layers_10_feed_forward2_linear1_weight_to_fp16_quantized, x = input_579_cast_fp16)[name = tensor<string, []>("linear_98_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_583_cast_fp16 = silu(x = linear_98_cast_fp16)[name = tensor<string, []>("input_583_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_10_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_10_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(69549824))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70598464))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_10_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70599552)))];
            tensor<fp16, [1, 125, 512]> linear_99_cast_fp16 = linear(bias = encoder_layers_10_feed_forward2_linear2_bias_to_fp16, weight = encoder_layers_10_feed_forward2_linear2_weight_to_fp16_quantized, x = input_583_cast_fp16)[name = tensor<string, []>("linear_99_cast_fp16")];
            tensor<fp16, []> var_2231_to_fp16 = const()[name = tensor<string, []>("op_2231_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_2232_cast_fp16 = mul(x = linear_99_cast_fp16, y = var_2231_to_fp16)[name = tensor<string, []>("op_2232_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_589_cast_fp16 = add(x = input_577_cast_fp16, y = var_2232_cast_fp16)[name = tensor<string, []>("input_589_cast_fp16")];
            tensor<int32, [1]> input_591_axes_0 = const()[name = tensor<string, []>("input_591_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_10_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_norm_out_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70600640)))];
            tensor<fp16, [512]> encoder_layers_10_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_10_norm_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70601728)))];
            tensor<fp16, [1, 125, 512]> input_591_cast_fp16 = layer_norm(axes = input_591_axes_0, beta = encoder_layers_10_norm_out_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_10_norm_out_weight_to_fp16, x = input_589_cast_fp16)[name = tensor<string, []>("input_591_cast_fp16")];
            tensor<int32, [1]> input_593_axes_0 = const()[name = tensor<string, []>("input_593_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_11_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70602816)))];
            tensor<fp16, [512]> encoder_layers_11_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70603904)))];
            tensor<fp16, [1, 125, 512]> input_593_cast_fp16 = layer_norm(axes = input_593_axes_0, beta = encoder_layers_11_norm_feed_forward1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_11_norm_feed_forward1_weight_to_fp16, x = input_591_cast_fp16)[name = tensor<string, []>("input_593_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_11_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_11_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(70604992))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71653632))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_11_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71657792)))];
            tensor<fp16, [1, 125, 2048]> linear_100_cast_fp16 = linear(bias = encoder_layers_11_feed_forward1_linear1_bias_to_fp16, weight = encoder_layers_11_feed_forward1_linear1_weight_to_fp16_quantized, x = input_593_cast_fp16)[name = tensor<string, []>("linear_100_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_597_cast_fp16 = silu(x = linear_100_cast_fp16)[name = tensor<string, []>("input_597_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_11_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_11_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(71661952))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72710592))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_11_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72711680)))];
            tensor<fp16, [1, 125, 512]> linear_101_cast_fp16 = linear(bias = encoder_layers_11_feed_forward1_linear2_bias_to_fp16, weight = encoder_layers_11_feed_forward1_linear2_weight_to_fp16_quantized, x = input_597_cast_fp16)[name = tensor<string, []>("linear_101_cast_fp16")];
            tensor<fp16, []> var_2262_to_fp16 = const()[name = tensor<string, []>("op_2262_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_2263_cast_fp16 = mul(x = linear_101_cast_fp16, y = var_2262_to_fp16)[name = tensor<string, []>("op_2263_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_603_cast_fp16 = add(x = input_591_cast_fp16, y = var_2263_cast_fp16)[name = tensor<string, []>("input_603_cast_fp16")];
            tensor<int32, [1]> query_23_axes_0 = const()[name = tensor<string, []>("query_23_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_11_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_norm_self_att_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72712768)))];
            tensor<fp16, [512]> encoder_layers_11_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_norm_self_att_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72713856)))];
            tensor<fp16, [1, 125, 512]> query_23_cast_fp16 = layer_norm(axes = query_23_axes_0, beta = encoder_layers_11_norm_self_att_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_11_norm_self_att_weight_to_fp16, x = input_603_cast_fp16)[name = tensor<string, []>("query_23_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_11_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_11_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72714944))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72977152))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_11_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72978240)))];
            tensor<fp16, [1, 125, 512]> linear_102_cast_fp16 = linear(bias = encoder_layers_11_self_attn_linear_q_bias_to_fp16, weight = encoder_layers_11_self_attn_linear_q_weight_to_fp16_quantized, x = query_23_cast_fp16)[name = tensor<string, []>("linear_102_cast_fp16")];
            tensor<int32, [4]> var_2280 = const()[name = tensor<string, []>("op_2280"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> q_67_cast_fp16 = reshape(shape = var_2280, x = linear_102_cast_fp16)[name = tensor<string, []>("q_67_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_11_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_11_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(72979328))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73241536))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_11_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73242624)))];
            tensor<fp16, [1, 125, 512]> linear_103_cast_fp16 = linear(bias = encoder_layers_11_self_attn_linear_k_bias_to_fp16, weight = encoder_layers_11_self_attn_linear_k_weight_to_fp16_quantized, x = query_23_cast_fp16)[name = tensor<string, []>("linear_103_cast_fp16")];
            tensor<int32, [4]> var_2285 = const()[name = tensor<string, []>("op_2285"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> k_45_cast_fp16 = reshape(shape = var_2285, x = linear_103_cast_fp16)[name = tensor<string, []>("k_45_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_11_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_11_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73243712))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73505920))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_11_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73507008)))];
            tensor<fp16, [1, 125, 512]> linear_104_cast_fp16 = linear(bias = encoder_layers_11_self_attn_linear_v_bias_to_fp16, weight = encoder_layers_11_self_attn_linear_v_weight_to_fp16_quantized, x = query_23_cast_fp16)[name = tensor<string, []>("linear_104_cast_fp16")];
            tensor<int32, [4]> var_2290 = const()[name = tensor<string, []>("op_2290"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> v_23_cast_fp16 = reshape(shape = var_2290, x = linear_104_cast_fp16)[name = tensor<string, []>("v_23_cast_fp16")];
            tensor<int32, [4]> value_23_perm_0 = const()[name = tensor<string, []>("value_23_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 64]> encoder_layers_11_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73508096)))];
            tensor<fp16, [1, 125, 8, 64]> var_2302_cast_fp16 = add(x = q_67_cast_fp16, y = encoder_layers_11_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_2302_cast_fp16")];
            tensor<fp16, [8, 64]> encoder_layers_11_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73509184)))];
            tensor<fp16, [1, 125, 8, 64]> var_2304_cast_fp16 = add(x = q_67_cast_fp16, y = encoder_layers_11_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_2304_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_23_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_23_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<bool, []> x_249_transpose_x_0 = const()[name = tensor<string, []>("x_249_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_249_transpose_y_0 = const()[name = tensor<string, []>("x_249_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 64, 249]> op_2306_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(3), name = tensor<string, []>("op_2306_to_fp16_quantized"), quantized_data = tensor<int8, [1, 8, 64, 249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73510272))), scale = tensor<fp16, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73637824))), zero_point = tensor<int8, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5281536)))];
            tensor<fp16, [1, 8, 125, 64]> q_with_bias_v_23_cast_fp16 = transpose(perm = q_with_bias_v_23_perm_0, x = var_2304_cast_fp16)[name = tensor<string, []>("transpose_143")];
            tensor<fp16, [1, 8, 125, 249]> x_249_cast_fp16 = matmul(transpose_x = x_249_transpose_x_0, transpose_y = x_249_transpose_y_0, x = q_with_bias_v_23_cast_fp16, y = op_2306_to_fp16_quantized)[name = tensor<string, []>("x_249_cast_fp16")];
            tensor<int32, [8]> x_251_pad_0 = const()[name = tensor<string, []>("x_251_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            tensor<string, []> x_251_mode_0 = const()[name = tensor<string, []>("x_251_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_180_to_fp16 = const()[name = tensor<string, []>("const_180_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 8, 125, 250]> x_251_cast_fp16 = pad(constant_val = const_180_to_fp16, mode = x_251_mode_0, pad = x_251_pad_0, x = x_249_cast_fp16)[name = tensor<string, []>("x_251_cast_fp16")];
            tensor<int32, [4]> var_2314 = const()[name = tensor<string, []>("op_2314"), val = tensor<int32, [4]>([1, 8, -1, 125])];
            tensor<fp16, [1, 8, 250, 125]> x_253_cast_fp16 = reshape(shape = var_2314, x = x_251_cast_fp16)[name = tensor<string, []>("x_253_cast_fp16")];
            tensor<int32, [4]> var_2318_begin_0 = const()[name = tensor<string, []>("op_2318_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_2318_end_0 = const()[name = tensor<string, []>("op_2318_end_0"), val = tensor<int32, [4]>([1, 8, 250, 125])];
            tensor<bool, [4]> var_2318_end_mask_0 = const()[name = tensor<string, []>("op_2318_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 249, 125]> var_2318_cast_fp16 = slice_by_index(begin = var_2318_begin_0, end = var_2318_end_0, end_mask = var_2318_end_mask_0, x = x_253_cast_fp16)[name = tensor<string, []>("op_2318_cast_fp16")];
            tensor<int32, [4]> var_2319 = const()[name = tensor<string, []>("op_2319"), val = tensor<int32, [4]>([1, 8, 125, 249])];
            tensor<fp16, [1, 8, 125, 249]> matrix_bd_45_cast_fp16 = reshape(shape = var_2319, x = var_2318_cast_fp16)[name = tensor<string, []>("matrix_bd_45_cast_fp16")];
            tensor<bool, []> matrix_ac_23_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_23_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matrix_ac_23_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_23_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_90_perm_0 = const()[name = tensor<string, []>("transpose_90_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_91_perm_0 = const()[name = tensor<string, []>("transpose_91_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 64, 125]> transpose_91 = transpose(perm = transpose_91_perm_0, x = k_45_cast_fp16)[name = tensor<string, []>("transpose_141")];
            tensor<fp16, [1, 8, 125, 64]> transpose_90 = transpose(perm = transpose_90_perm_0, x = var_2302_cast_fp16)[name = tensor<string, []>("transpose_142")];
            tensor<fp16, [1, 8, 125, 125]> matrix_ac_23_cast_fp16 = matmul(transpose_x = matrix_ac_23_transpose_x_0, transpose_y = matrix_ac_23_transpose_y_0, x = transpose_90, y = transpose_91)[name = tensor<string, []>("matrix_ac_23_cast_fp16")];
            tensor<int32, [4]> matrix_bd_47_begin_0 = const()[name = tensor<string, []>("matrix_bd_47_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_47_end_0 = const()[name = tensor<string, []>("matrix_bd_47_end_0"), val = tensor<int32, [4]>([1, 8, 125, 125])];
            tensor<bool, [4]> matrix_bd_47_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_47_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 125, 125]> matrix_bd_47_cast_fp16 = slice_by_index(begin = matrix_bd_47_begin_0, end = matrix_bd_47_end_0, end_mask = matrix_bd_47_end_mask_0, x = matrix_bd_45_cast_fp16)[name = tensor<string, []>("matrix_bd_47_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_2328_cast_fp16 = add(x = matrix_ac_23_cast_fp16, y = matrix_bd_47_cast_fp16)[name = tensor<string, []>("op_2328_cast_fp16")];
            tensor<fp16, []> _inversed_scores_45_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_45_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 8, 125, 125]> _inversed_scores_45_cast_fp16 = mul(x = var_2328_cast_fp16, y = _inversed_scores_45_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_45_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> scores_47_cast_fp16 = select(a = var_8_to_fp16, b = _inversed_scores_45_cast_fp16, cond = mask_11)[name = tensor<string, []>("scores_47_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_2334_cast_fp16 = softmax(axis = var_19, x = scores_47_cast_fp16)[name = tensor<string, []>("op_2334_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> input_605_cast_fp16 = select(a = var_7_to_fp16, b = var_2334_cast_fp16, cond = mask_11)[name = tensor<string, []>("input_605_cast_fp16")];
            tensor<bool, []> x_255_transpose_x_0 = const()[name = tensor<string, []>("x_255_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_255_transpose_y_0 = const()[name = tensor<string, []>("x_255_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 125, 64]> value_23_cast_fp16 = transpose(perm = value_23_perm_0, x = v_23_cast_fp16)[name = tensor<string, []>("transpose_140")];
            tensor<fp16, [1, 8, 125, 64]> x_255_cast_fp16 = matmul(transpose_x = x_255_transpose_x_0, transpose_y = x_255_transpose_y_0, x = input_605_cast_fp16, y = value_23_cast_fp16)[name = tensor<string, []>("x_255_cast_fp16")];
            tensor<int32, [4]> var_2338_perm_0 = const()[name = tensor<string, []>("op_2338_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2339 = const()[name = tensor<string, []>("op_2339"), val = tensor<int32, [3]>([1, -1, 512])];
            tensor<fp16, [1, 125, 8, 64]> var_2338_cast_fp16 = transpose(perm = var_2338_perm_0, x = x_255_cast_fp16)[name = tensor<string, []>("transpose_139")];
            tensor<fp16, [1, 125, 512]> input_607_cast_fp16 = reshape(shape = var_2339, x = var_2338_cast_fp16)[name = tensor<string, []>("input_607_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_11_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_11_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73638400))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73900608))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_11_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73901696)))];
            tensor<fp16, [1, 125, 512]> linear_106_cast_fp16 = linear(bias = encoder_layers_11_self_attn_linear_out_bias_to_fp16, weight = encoder_layers_11_self_attn_linear_out_weight_to_fp16_quantized, x = input_607_cast_fp16)[name = tensor<string, []>("linear_106_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_611_cast_fp16 = add(x = input_603_cast_fp16, y = linear_106_cast_fp16)[name = tensor<string, []>("input_611_cast_fp16")];
            tensor<int32, [1]> x_259_axes_0 = const()[name = tensor<string, []>("x_259_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_11_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_norm_conv_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73902784)))];
            tensor<fp16, [512]> encoder_layers_11_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_norm_conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73903872)))];
            tensor<fp16, [1, 125, 512]> x_259_cast_fp16 = layer_norm(axes = x_259_axes_0, beta = encoder_layers_11_norm_conv_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_11_norm_conv_weight_to_fp16, x = input_611_cast_fp16)[name = tensor<string, []>("x_259_cast_fp16")];
            tensor<int32, [3]> input_613_perm_0 = const()[name = tensor<string, []>("input_613_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> input_615_pad_type_0 = const()[name = tensor<string, []>("input_615_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> input_615_strides_0 = const()[name = tensor<string, []>("input_615_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_615_pad_0 = const()[name = tensor<string, []>("input_615_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_615_dilations_0 = const()[name = tensor<string, []>("input_615_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> input_615_groups_0 = const()[name = tensor<string, []>("input_615_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 512, 1]> encoder_layers_11_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_11_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1024, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(73904960))), scale = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74429312))), zero_point = tensor<int8, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6073344)))];
            tensor<fp16, [1024]> encoder_layers_11_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74431424)))];
            tensor<fp16, [1, 512, 125]> input_613_cast_fp16 = transpose(perm = input_613_perm_0, x = x_259_cast_fp16)[name = tensor<string, []>("transpose_138")];
            tensor<fp16, [1, 1024, 125]> input_615_cast_fp16 = conv(bias = encoder_layers_11_conv_pointwise_conv1_bias_to_fp16, dilations = input_615_dilations_0, groups = input_615_groups_0, pad = input_615_pad_0, pad_type = input_615_pad_type_0, strides = input_615_strides_0, weight = encoder_layers_11_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_613_cast_fp16)[name = tensor<string, []>("input_615_cast_fp16")];
            tensor<int32, []> x_261_split_num_splits_0 = const()[name = tensor<string, []>("x_261_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> x_261_split_axis_0 = const()[name = tensor<string, []>("x_261_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 512, 125]> x_261_split_cast_fp16_0, tensor<fp16, [1, 512, 125]> x_261_split_cast_fp16_1 = split(axis = x_261_split_axis_0, num_splits = x_261_split_num_splits_0, x = input_615_cast_fp16)[name = tensor<string, []>("x_261_split_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_261_split_1_sigmoid_cast_fp16 = sigmoid(x = x_261_split_cast_fp16_1)[name = tensor<string, []>("x_261_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_261_cast_fp16 = mul(x = x_261_split_cast_fp16_0, y = x_261_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_261_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_617_cast_fp16 = select(a = var_7_to_fp16, b = x_261_cast_fp16, cond = var_449)[name = tensor<string, []>("input_617_cast_fp16")];
            tensor<int32, [6]> input_619_pad_0 = const()[name = tensor<string, []>("input_619_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            tensor<string, []> input_619_mode_0 = const()[name = tensor<string, []>("input_619_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_183_to_fp16 = const()[name = tensor<string, []>("const_183_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 512, 133]> input_619_cast_fp16 = pad(constant_val = const_183_to_fp16, mode = input_619_mode_0, pad = input_619_pad_0, x = input_617_cast_fp16)[name = tensor<string, []>("input_619_cast_fp16")];
            tensor<string, []> input_621_pad_type_0 = const()[name = tensor<string, []>("input_621_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, []> input_621_groups_0 = const()[name = tensor<string, []>("input_621_groups_0"), val = tensor<int32, []>(512)];
            tensor<int32, [1]> input_621_strides_0 = const()[name = tensor<string, []>("input_621_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_621_pad_0 = const()[name = tensor<string, []>("input_621_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_621_dilations_0 = const()[name = tensor<string, []>("input_621_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 9]> const_259_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("const_259_to_fp16_quantized"), quantized_data = tensor<int8, [512, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74433536))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74438208))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> const_260_to_fp16 = const()[name = tensor<string, []>("const_260_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74439296)))];
            tensor<fp16, [1, 512, 125]> input_623_cast_fp16 = conv(bias = const_260_to_fp16, dilations = input_621_dilations_0, groups = input_621_groups_0, pad = input_621_pad_0, pad_type = input_621_pad_type_0, strides = input_621_strides_0, weight = const_259_to_fp16_quantized, x = input_619_cast_fp16)[name = tensor<string, []>("input_623_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_625_cast_fp16 = silu(x = input_623_cast_fp16)[name = tensor<string, []>("input_625_cast_fp16")];
            tensor<string, []> x_263_pad_type_0 = const()[name = tensor<string, []>("x_263_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> x_263_strides_0 = const()[name = tensor<string, []>("x_263_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_263_pad_0 = const()[name = tensor<string, []>("x_263_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_263_dilations_0 = const()[name = tensor<string, []>("x_263_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> x_263_groups_0 = const()[name = tensor<string, []>("x_263_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [512, 512, 1]> encoder_layers_11_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_11_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74440384))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74702592))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_11_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74703680)))];
            tensor<fp16, [1, 512, 125]> x_263_cast_fp16 = conv(bias = encoder_layers_11_conv_pointwise_conv2_bias_to_fp16, dilations = x_263_dilations_0, groups = x_263_groups_0, pad = x_263_pad_0, pad_type = x_263_pad_type_0, strides = x_263_strides_0, weight = encoder_layers_11_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_625_cast_fp16)[name = tensor<string, []>("x_263_cast_fp16")];
            tensor<int32, [3]> input_627_perm_0 = const()[name = tensor<string, []>("input_627_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 125, 512]> input_627_cast_fp16 = transpose(perm = input_627_perm_0, x = x_263_cast_fp16)[name = tensor<string, []>("transpose_137")];
            tensor<fp16, [1, 125, 512]> input_629_cast_fp16 = add(x = input_611_cast_fp16, y = input_627_cast_fp16)[name = tensor<string, []>("input_629_cast_fp16")];
            tensor<int32, [1]> input_631_axes_0 = const()[name = tensor<string, []>("input_631_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_11_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74704768)))];
            tensor<fp16, [512]> encoder_layers_11_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74705856)))];
            tensor<fp16, [1, 125, 512]> input_631_cast_fp16 = layer_norm(axes = input_631_axes_0, beta = encoder_layers_11_norm_feed_forward2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_11_norm_feed_forward2_weight_to_fp16, x = input_629_cast_fp16)[name = tensor<string, []>("input_631_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_11_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_11_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(74706944))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75755584))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_11_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75759744)))];
            tensor<fp16, [1, 125, 2048]> linear_107_cast_fp16 = linear(bias = encoder_layers_11_feed_forward2_linear1_bias_to_fp16, weight = encoder_layers_11_feed_forward2_linear1_weight_to_fp16_quantized, x = input_631_cast_fp16)[name = tensor<string, []>("linear_107_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_635_cast_fp16 = silu(x = linear_107_cast_fp16)[name = tensor<string, []>("input_635_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_11_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_11_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(75763904))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76812544))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_11_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76813632)))];
            tensor<fp16, [1, 125, 512]> linear_108_cast_fp16 = linear(bias = encoder_layers_11_feed_forward2_linear2_bias_to_fp16, weight = encoder_layers_11_feed_forward2_linear2_weight_to_fp16_quantized, x = input_635_cast_fp16)[name = tensor<string, []>("linear_108_cast_fp16")];
            tensor<fp16, []> var_2405_to_fp16 = const()[name = tensor<string, []>("op_2405_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_2406_cast_fp16 = mul(x = linear_108_cast_fp16, y = var_2405_to_fp16)[name = tensor<string, []>("op_2406_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_641_cast_fp16 = add(x = input_629_cast_fp16, y = var_2406_cast_fp16)[name = tensor<string, []>("input_641_cast_fp16")];
            tensor<int32, [1]> input_643_axes_0 = const()[name = tensor<string, []>("input_643_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_11_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_norm_out_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76814720)))];
            tensor<fp16, [512]> encoder_layers_11_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_11_norm_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76815808)))];
            tensor<fp16, [1, 125, 512]> input_643_cast_fp16 = layer_norm(axes = input_643_axes_0, beta = encoder_layers_11_norm_out_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_11_norm_out_weight_to_fp16, x = input_641_cast_fp16)[name = tensor<string, []>("input_643_cast_fp16")];
            tensor<int32, [1]> input_645_axes_0 = const()[name = tensor<string, []>("input_645_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_12_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76816896)))];
            tensor<fp16, [512]> encoder_layers_12_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76817984)))];
            tensor<fp16, [1, 125, 512]> input_645_cast_fp16 = layer_norm(axes = input_645_axes_0, beta = encoder_layers_12_norm_feed_forward1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_12_norm_feed_forward1_weight_to_fp16, x = input_643_cast_fp16)[name = tensor<string, []>("input_645_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_12_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_12_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(76819072))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77867712))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_12_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77871872)))];
            tensor<fp16, [1, 125, 2048]> linear_109_cast_fp16 = linear(bias = encoder_layers_12_feed_forward1_linear1_bias_to_fp16, weight = encoder_layers_12_feed_forward1_linear1_weight_to_fp16_quantized, x = input_645_cast_fp16)[name = tensor<string, []>("linear_109_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_649_cast_fp16 = silu(x = linear_109_cast_fp16)[name = tensor<string, []>("input_649_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_12_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_12_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(77876032))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78924672))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_12_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78925760)))];
            tensor<fp16, [1, 125, 512]> linear_110_cast_fp16 = linear(bias = encoder_layers_12_feed_forward1_linear2_bias_to_fp16, weight = encoder_layers_12_feed_forward1_linear2_weight_to_fp16_quantized, x = input_649_cast_fp16)[name = tensor<string, []>("linear_110_cast_fp16")];
            tensor<fp16, []> var_2436_to_fp16 = const()[name = tensor<string, []>("op_2436_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_2437_cast_fp16 = mul(x = linear_110_cast_fp16, y = var_2436_to_fp16)[name = tensor<string, []>("op_2437_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_655_cast_fp16 = add(x = input_643_cast_fp16, y = var_2437_cast_fp16)[name = tensor<string, []>("input_655_cast_fp16")];
            tensor<int32, [1]> query_25_axes_0 = const()[name = tensor<string, []>("query_25_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_12_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_norm_self_att_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78926848)))];
            tensor<fp16, [512]> encoder_layers_12_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_norm_self_att_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78927936)))];
            tensor<fp16, [1, 125, 512]> query_25_cast_fp16 = layer_norm(axes = query_25_axes_0, beta = encoder_layers_12_norm_self_att_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_12_norm_self_att_weight_to_fp16, x = input_655_cast_fp16)[name = tensor<string, []>("query_25_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_12_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_12_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(78929024))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79191232))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_12_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79192320)))];
            tensor<fp16, [1, 125, 512]> linear_111_cast_fp16 = linear(bias = encoder_layers_12_self_attn_linear_q_bias_to_fp16, weight = encoder_layers_12_self_attn_linear_q_weight_to_fp16_quantized, x = query_25_cast_fp16)[name = tensor<string, []>("linear_111_cast_fp16")];
            tensor<int32, [4]> var_2454 = const()[name = tensor<string, []>("op_2454"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> q_73_cast_fp16 = reshape(shape = var_2454, x = linear_111_cast_fp16)[name = tensor<string, []>("q_73_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_12_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_12_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79193408))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79455616))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_12_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79456704)))];
            tensor<fp16, [1, 125, 512]> linear_112_cast_fp16 = linear(bias = encoder_layers_12_self_attn_linear_k_bias_to_fp16, weight = encoder_layers_12_self_attn_linear_k_weight_to_fp16_quantized, x = query_25_cast_fp16)[name = tensor<string, []>("linear_112_cast_fp16")];
            tensor<int32, [4]> var_2459 = const()[name = tensor<string, []>("op_2459"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> k_49_cast_fp16 = reshape(shape = var_2459, x = linear_112_cast_fp16)[name = tensor<string, []>("k_49_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_12_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_12_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79457792))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79720000))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_12_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79721088)))];
            tensor<fp16, [1, 125, 512]> linear_113_cast_fp16 = linear(bias = encoder_layers_12_self_attn_linear_v_bias_to_fp16, weight = encoder_layers_12_self_attn_linear_v_weight_to_fp16_quantized, x = query_25_cast_fp16)[name = tensor<string, []>("linear_113_cast_fp16")];
            tensor<int32, [4]> var_2464 = const()[name = tensor<string, []>("op_2464"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> v_25_cast_fp16 = reshape(shape = var_2464, x = linear_113_cast_fp16)[name = tensor<string, []>("v_25_cast_fp16")];
            tensor<int32, [4]> value_25_perm_0 = const()[name = tensor<string, []>("value_25_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 64]> encoder_layers_12_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79722176)))];
            tensor<fp16, [1, 125, 8, 64]> var_2476_cast_fp16 = add(x = q_73_cast_fp16, y = encoder_layers_12_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_2476_cast_fp16")];
            tensor<fp16, [8, 64]> encoder_layers_12_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79723264)))];
            tensor<fp16, [1, 125, 8, 64]> var_2478_cast_fp16 = add(x = q_73_cast_fp16, y = encoder_layers_12_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_2478_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_25_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_25_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<bool, []> x_271_transpose_x_0 = const()[name = tensor<string, []>("x_271_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_271_transpose_y_0 = const()[name = tensor<string, []>("x_271_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 64, 249]> op_2480_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(3), name = tensor<string, []>("op_2480_to_fp16_quantized"), quantized_data = tensor<int8, [1, 8, 64, 249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79724352))), scale = tensor<fp16, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79851904))), zero_point = tensor<int8, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5281536)))];
            tensor<fp16, [1, 8, 125, 64]> q_with_bias_v_25_cast_fp16 = transpose(perm = q_with_bias_v_25_perm_0, x = var_2478_cast_fp16)[name = tensor<string, []>("transpose_136")];
            tensor<fp16, [1, 8, 125, 249]> x_271_cast_fp16 = matmul(transpose_x = x_271_transpose_x_0, transpose_y = x_271_transpose_y_0, x = q_with_bias_v_25_cast_fp16, y = op_2480_to_fp16_quantized)[name = tensor<string, []>("x_271_cast_fp16")];
            tensor<int32, [8]> x_273_pad_0 = const()[name = tensor<string, []>("x_273_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            tensor<string, []> x_273_mode_0 = const()[name = tensor<string, []>("x_273_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_190_to_fp16 = const()[name = tensor<string, []>("const_190_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 8, 125, 250]> x_273_cast_fp16 = pad(constant_val = const_190_to_fp16, mode = x_273_mode_0, pad = x_273_pad_0, x = x_271_cast_fp16)[name = tensor<string, []>("x_273_cast_fp16")];
            tensor<int32, [4]> var_2488 = const()[name = tensor<string, []>("op_2488"), val = tensor<int32, [4]>([1, 8, -1, 125])];
            tensor<fp16, [1, 8, 250, 125]> x_275_cast_fp16 = reshape(shape = var_2488, x = x_273_cast_fp16)[name = tensor<string, []>("x_275_cast_fp16")];
            tensor<int32, [4]> var_2492_begin_0 = const()[name = tensor<string, []>("op_2492_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_2492_end_0 = const()[name = tensor<string, []>("op_2492_end_0"), val = tensor<int32, [4]>([1, 8, 250, 125])];
            tensor<bool, [4]> var_2492_end_mask_0 = const()[name = tensor<string, []>("op_2492_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 249, 125]> var_2492_cast_fp16 = slice_by_index(begin = var_2492_begin_0, end = var_2492_end_0, end_mask = var_2492_end_mask_0, x = x_275_cast_fp16)[name = tensor<string, []>("op_2492_cast_fp16")];
            tensor<int32, [4]> var_2493 = const()[name = tensor<string, []>("op_2493"), val = tensor<int32, [4]>([1, 8, 125, 249])];
            tensor<fp16, [1, 8, 125, 249]> matrix_bd_49_cast_fp16 = reshape(shape = var_2493, x = var_2492_cast_fp16)[name = tensor<string, []>("matrix_bd_49_cast_fp16")];
            tensor<bool, []> matrix_ac_25_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_25_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matrix_ac_25_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_25_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_92_perm_0 = const()[name = tensor<string, []>("transpose_92_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_93_perm_0 = const()[name = tensor<string, []>("transpose_93_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 64, 125]> transpose_93 = transpose(perm = transpose_93_perm_0, x = k_49_cast_fp16)[name = tensor<string, []>("transpose_134")];
            tensor<fp16, [1, 8, 125, 64]> transpose_92 = transpose(perm = transpose_92_perm_0, x = var_2476_cast_fp16)[name = tensor<string, []>("transpose_135")];
            tensor<fp16, [1, 8, 125, 125]> matrix_ac_25_cast_fp16 = matmul(transpose_x = matrix_ac_25_transpose_x_0, transpose_y = matrix_ac_25_transpose_y_0, x = transpose_92, y = transpose_93)[name = tensor<string, []>("matrix_ac_25_cast_fp16")];
            tensor<int32, [4]> matrix_bd_51_begin_0 = const()[name = tensor<string, []>("matrix_bd_51_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_51_end_0 = const()[name = tensor<string, []>("matrix_bd_51_end_0"), val = tensor<int32, [4]>([1, 8, 125, 125])];
            tensor<bool, [4]> matrix_bd_51_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_51_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 125, 125]> matrix_bd_51_cast_fp16 = slice_by_index(begin = matrix_bd_51_begin_0, end = matrix_bd_51_end_0, end_mask = matrix_bd_51_end_mask_0, x = matrix_bd_49_cast_fp16)[name = tensor<string, []>("matrix_bd_51_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_2502_cast_fp16 = add(x = matrix_ac_25_cast_fp16, y = matrix_bd_51_cast_fp16)[name = tensor<string, []>("op_2502_cast_fp16")];
            tensor<fp16, []> _inversed_scores_49_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_49_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 8, 125, 125]> _inversed_scores_49_cast_fp16 = mul(x = var_2502_cast_fp16, y = _inversed_scores_49_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_49_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> scores_51_cast_fp16 = select(a = var_8_to_fp16, b = _inversed_scores_49_cast_fp16, cond = mask_11)[name = tensor<string, []>("scores_51_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_2508_cast_fp16 = softmax(axis = var_19, x = scores_51_cast_fp16)[name = tensor<string, []>("op_2508_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> input_657_cast_fp16 = select(a = var_7_to_fp16, b = var_2508_cast_fp16, cond = mask_11)[name = tensor<string, []>("input_657_cast_fp16")];
            tensor<bool, []> x_277_transpose_x_0 = const()[name = tensor<string, []>("x_277_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_277_transpose_y_0 = const()[name = tensor<string, []>("x_277_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 125, 64]> value_25_cast_fp16 = transpose(perm = value_25_perm_0, x = v_25_cast_fp16)[name = tensor<string, []>("transpose_133")];
            tensor<fp16, [1, 8, 125, 64]> x_277_cast_fp16 = matmul(transpose_x = x_277_transpose_x_0, transpose_y = x_277_transpose_y_0, x = input_657_cast_fp16, y = value_25_cast_fp16)[name = tensor<string, []>("x_277_cast_fp16")];
            tensor<int32, [4]> var_2512_perm_0 = const()[name = tensor<string, []>("op_2512_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2513 = const()[name = tensor<string, []>("op_2513"), val = tensor<int32, [3]>([1, -1, 512])];
            tensor<fp16, [1, 125, 8, 64]> var_2512_cast_fp16 = transpose(perm = var_2512_perm_0, x = x_277_cast_fp16)[name = tensor<string, []>("transpose_132")];
            tensor<fp16, [1, 125, 512]> input_659_cast_fp16 = reshape(shape = var_2513, x = var_2512_cast_fp16)[name = tensor<string, []>("input_659_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_12_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_12_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(79852480))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80114688))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_12_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80115776)))];
            tensor<fp16, [1, 125, 512]> linear_115_cast_fp16 = linear(bias = encoder_layers_12_self_attn_linear_out_bias_to_fp16, weight = encoder_layers_12_self_attn_linear_out_weight_to_fp16_quantized, x = input_659_cast_fp16)[name = tensor<string, []>("linear_115_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_663_cast_fp16 = add(x = input_655_cast_fp16, y = linear_115_cast_fp16)[name = tensor<string, []>("input_663_cast_fp16")];
            tensor<int32, [1]> x_281_axes_0 = const()[name = tensor<string, []>("x_281_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_12_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_norm_conv_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80116864)))];
            tensor<fp16, [512]> encoder_layers_12_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_norm_conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80117952)))];
            tensor<fp16, [1, 125, 512]> x_281_cast_fp16 = layer_norm(axes = x_281_axes_0, beta = encoder_layers_12_norm_conv_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_12_norm_conv_weight_to_fp16, x = input_663_cast_fp16)[name = tensor<string, []>("x_281_cast_fp16")];
            tensor<int32, [3]> input_665_perm_0 = const()[name = tensor<string, []>("input_665_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> input_667_pad_type_0 = const()[name = tensor<string, []>("input_667_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> input_667_strides_0 = const()[name = tensor<string, []>("input_667_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_667_pad_0 = const()[name = tensor<string, []>("input_667_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_667_dilations_0 = const()[name = tensor<string, []>("input_667_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> input_667_groups_0 = const()[name = tensor<string, []>("input_667_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 512, 1]> encoder_layers_12_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_12_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1024, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80119040))), scale = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80643392))), zero_point = tensor<int8, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6073344)))];
            tensor<fp16, [1024]> encoder_layers_12_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80645504)))];
            tensor<fp16, [1, 512, 125]> input_665_cast_fp16 = transpose(perm = input_665_perm_0, x = x_281_cast_fp16)[name = tensor<string, []>("transpose_131")];
            tensor<fp16, [1, 1024, 125]> input_667_cast_fp16 = conv(bias = encoder_layers_12_conv_pointwise_conv1_bias_to_fp16, dilations = input_667_dilations_0, groups = input_667_groups_0, pad = input_667_pad_0, pad_type = input_667_pad_type_0, strides = input_667_strides_0, weight = encoder_layers_12_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_665_cast_fp16)[name = tensor<string, []>("input_667_cast_fp16")];
            tensor<int32, []> x_283_split_num_splits_0 = const()[name = tensor<string, []>("x_283_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> x_283_split_axis_0 = const()[name = tensor<string, []>("x_283_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 512, 125]> x_283_split_cast_fp16_0, tensor<fp16, [1, 512, 125]> x_283_split_cast_fp16_1 = split(axis = x_283_split_axis_0, num_splits = x_283_split_num_splits_0, x = input_667_cast_fp16)[name = tensor<string, []>("x_283_split_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_283_split_1_sigmoid_cast_fp16 = sigmoid(x = x_283_split_cast_fp16_1)[name = tensor<string, []>("x_283_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_283_cast_fp16 = mul(x = x_283_split_cast_fp16_0, y = x_283_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_283_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_669_cast_fp16 = select(a = var_7_to_fp16, b = x_283_cast_fp16, cond = var_449)[name = tensor<string, []>("input_669_cast_fp16")];
            tensor<int32, [6]> input_671_pad_0 = const()[name = tensor<string, []>("input_671_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            tensor<string, []> input_671_mode_0 = const()[name = tensor<string, []>("input_671_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_193_to_fp16 = const()[name = tensor<string, []>("const_193_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 512, 133]> input_671_cast_fp16 = pad(constant_val = const_193_to_fp16, mode = input_671_mode_0, pad = input_671_pad_0, x = input_669_cast_fp16)[name = tensor<string, []>("input_671_cast_fp16")];
            tensor<string, []> input_673_pad_type_0 = const()[name = tensor<string, []>("input_673_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, []> input_673_groups_0 = const()[name = tensor<string, []>("input_673_groups_0"), val = tensor<int32, []>(512)];
            tensor<int32, [1]> input_673_strides_0 = const()[name = tensor<string, []>("input_673_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_673_pad_0 = const()[name = tensor<string, []>("input_673_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_673_dilations_0 = const()[name = tensor<string, []>("input_673_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 9]> const_261_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("const_261_to_fp16_quantized"), quantized_data = tensor<int8, [512, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80647616))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80652288))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> const_262_to_fp16 = const()[name = tensor<string, []>("const_262_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80653376)))];
            tensor<fp16, [1, 512, 125]> input_675_cast_fp16 = conv(bias = const_262_to_fp16, dilations = input_673_dilations_0, groups = input_673_groups_0, pad = input_673_pad_0, pad_type = input_673_pad_type_0, strides = input_673_strides_0, weight = const_261_to_fp16_quantized, x = input_671_cast_fp16)[name = tensor<string, []>("input_675_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_677_cast_fp16 = silu(x = input_675_cast_fp16)[name = tensor<string, []>("input_677_cast_fp16")];
            tensor<string, []> x_285_pad_type_0 = const()[name = tensor<string, []>("x_285_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> x_285_strides_0 = const()[name = tensor<string, []>("x_285_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_285_pad_0 = const()[name = tensor<string, []>("x_285_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_285_dilations_0 = const()[name = tensor<string, []>("x_285_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> x_285_groups_0 = const()[name = tensor<string, []>("x_285_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [512, 512, 1]> encoder_layers_12_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_12_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80654464))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80916672))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_12_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80917760)))];
            tensor<fp16, [1, 512, 125]> x_285_cast_fp16 = conv(bias = encoder_layers_12_conv_pointwise_conv2_bias_to_fp16, dilations = x_285_dilations_0, groups = x_285_groups_0, pad = x_285_pad_0, pad_type = x_285_pad_type_0, strides = x_285_strides_0, weight = encoder_layers_12_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_677_cast_fp16)[name = tensor<string, []>("x_285_cast_fp16")];
            tensor<int32, [3]> input_679_perm_0 = const()[name = tensor<string, []>("input_679_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 125, 512]> input_679_cast_fp16 = transpose(perm = input_679_perm_0, x = x_285_cast_fp16)[name = tensor<string, []>("transpose_130")];
            tensor<fp16, [1, 125, 512]> input_681_cast_fp16 = add(x = input_663_cast_fp16, y = input_679_cast_fp16)[name = tensor<string, []>("input_681_cast_fp16")];
            tensor<int32, [1]> input_683_axes_0 = const()[name = tensor<string, []>("input_683_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_12_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80918848)))];
            tensor<fp16, [512]> encoder_layers_12_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80919936)))];
            tensor<fp16, [1, 125, 512]> input_683_cast_fp16 = layer_norm(axes = input_683_axes_0, beta = encoder_layers_12_norm_feed_forward2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_12_norm_feed_forward2_weight_to_fp16, x = input_681_cast_fp16)[name = tensor<string, []>("input_683_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_12_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_12_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(80921024))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81969664))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_12_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81973824)))];
            tensor<fp16, [1, 125, 2048]> linear_116_cast_fp16 = linear(bias = encoder_layers_12_feed_forward2_linear1_bias_to_fp16, weight = encoder_layers_12_feed_forward2_linear1_weight_to_fp16_quantized, x = input_683_cast_fp16)[name = tensor<string, []>("linear_116_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_687_cast_fp16 = silu(x = linear_116_cast_fp16)[name = tensor<string, []>("input_687_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_12_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_12_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(81977984))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83026624))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_12_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83027712)))];
            tensor<fp16, [1, 125, 512]> linear_117_cast_fp16 = linear(bias = encoder_layers_12_feed_forward2_linear2_bias_to_fp16, weight = encoder_layers_12_feed_forward2_linear2_weight_to_fp16_quantized, x = input_687_cast_fp16)[name = tensor<string, []>("linear_117_cast_fp16")];
            tensor<fp16, []> var_2579_to_fp16 = const()[name = tensor<string, []>("op_2579_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_2580_cast_fp16 = mul(x = linear_117_cast_fp16, y = var_2579_to_fp16)[name = tensor<string, []>("op_2580_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_693_cast_fp16 = add(x = input_681_cast_fp16, y = var_2580_cast_fp16)[name = tensor<string, []>("input_693_cast_fp16")];
            tensor<int32, [1]> input_695_axes_0 = const()[name = tensor<string, []>("input_695_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_12_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_norm_out_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83028800)))];
            tensor<fp16, [512]> encoder_layers_12_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_12_norm_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83029888)))];
            tensor<fp16, [1, 125, 512]> input_695_cast_fp16 = layer_norm(axes = input_695_axes_0, beta = encoder_layers_12_norm_out_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_12_norm_out_weight_to_fp16, x = input_693_cast_fp16)[name = tensor<string, []>("input_695_cast_fp16")];
            tensor<int32, [1]> input_697_axes_0 = const()[name = tensor<string, []>("input_697_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_13_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83030976)))];
            tensor<fp16, [512]> encoder_layers_13_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83032064)))];
            tensor<fp16, [1, 125, 512]> input_697_cast_fp16 = layer_norm(axes = input_697_axes_0, beta = encoder_layers_13_norm_feed_forward1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_13_norm_feed_forward1_weight_to_fp16, x = input_695_cast_fp16)[name = tensor<string, []>("input_697_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_13_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_13_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(83033152))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84081792))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_13_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84085952)))];
            tensor<fp16, [1, 125, 2048]> linear_118_cast_fp16 = linear(bias = encoder_layers_13_feed_forward1_linear1_bias_to_fp16, weight = encoder_layers_13_feed_forward1_linear1_weight_to_fp16_quantized, x = input_697_cast_fp16)[name = tensor<string, []>("linear_118_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_701_cast_fp16 = silu(x = linear_118_cast_fp16)[name = tensor<string, []>("input_701_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_13_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_13_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(84090112))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85138752))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_13_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85139840)))];
            tensor<fp16, [1, 125, 512]> linear_119_cast_fp16 = linear(bias = encoder_layers_13_feed_forward1_linear2_bias_to_fp16, weight = encoder_layers_13_feed_forward1_linear2_weight_to_fp16_quantized, x = input_701_cast_fp16)[name = tensor<string, []>("linear_119_cast_fp16")];
            tensor<fp16, []> var_2610_to_fp16 = const()[name = tensor<string, []>("op_2610_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_2611_cast_fp16 = mul(x = linear_119_cast_fp16, y = var_2610_to_fp16)[name = tensor<string, []>("op_2611_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_707_cast_fp16 = add(x = input_695_cast_fp16, y = var_2611_cast_fp16)[name = tensor<string, []>("input_707_cast_fp16")];
            tensor<int32, [1]> query_27_axes_0 = const()[name = tensor<string, []>("query_27_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_13_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_norm_self_att_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85140928)))];
            tensor<fp16, [512]> encoder_layers_13_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_norm_self_att_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85142016)))];
            tensor<fp16, [1, 125, 512]> query_27_cast_fp16 = layer_norm(axes = query_27_axes_0, beta = encoder_layers_13_norm_self_att_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_13_norm_self_att_weight_to_fp16, x = input_707_cast_fp16)[name = tensor<string, []>("query_27_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_13_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_13_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85143104))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85405312))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_13_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85406400)))];
            tensor<fp16, [1, 125, 512]> linear_120_cast_fp16 = linear(bias = encoder_layers_13_self_attn_linear_q_bias_to_fp16, weight = encoder_layers_13_self_attn_linear_q_weight_to_fp16_quantized, x = query_27_cast_fp16)[name = tensor<string, []>("linear_120_cast_fp16")];
            tensor<int32, [4]> var_2628 = const()[name = tensor<string, []>("op_2628"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> q_79_cast_fp16 = reshape(shape = var_2628, x = linear_120_cast_fp16)[name = tensor<string, []>("q_79_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_13_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_13_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85407488))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85669696))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_13_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85670784)))];
            tensor<fp16, [1, 125, 512]> linear_121_cast_fp16 = linear(bias = encoder_layers_13_self_attn_linear_k_bias_to_fp16, weight = encoder_layers_13_self_attn_linear_k_weight_to_fp16_quantized, x = query_27_cast_fp16)[name = tensor<string, []>("linear_121_cast_fp16")];
            tensor<int32, [4]> var_2633 = const()[name = tensor<string, []>("op_2633"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> k_53_cast_fp16 = reshape(shape = var_2633, x = linear_121_cast_fp16)[name = tensor<string, []>("k_53_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_13_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_13_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85671872))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85934080))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_13_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85935168)))];
            tensor<fp16, [1, 125, 512]> linear_122_cast_fp16 = linear(bias = encoder_layers_13_self_attn_linear_v_bias_to_fp16, weight = encoder_layers_13_self_attn_linear_v_weight_to_fp16_quantized, x = query_27_cast_fp16)[name = tensor<string, []>("linear_122_cast_fp16")];
            tensor<int32, [4]> var_2638 = const()[name = tensor<string, []>("op_2638"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> v_27_cast_fp16 = reshape(shape = var_2638, x = linear_122_cast_fp16)[name = tensor<string, []>("v_27_cast_fp16")];
            tensor<int32, [4]> value_27_perm_0 = const()[name = tensor<string, []>("value_27_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 64]> encoder_layers_13_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85936256)))];
            tensor<fp16, [1, 125, 8, 64]> var_2650_cast_fp16 = add(x = q_79_cast_fp16, y = encoder_layers_13_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_2650_cast_fp16")];
            tensor<fp16, [8, 64]> encoder_layers_13_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85937344)))];
            tensor<fp16, [1, 125, 8, 64]> var_2652_cast_fp16 = add(x = q_79_cast_fp16, y = encoder_layers_13_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_2652_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_27_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_27_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<bool, []> x_293_transpose_x_0 = const()[name = tensor<string, []>("x_293_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_293_transpose_y_0 = const()[name = tensor<string, []>("x_293_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 64, 249]> op_2654_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(3), name = tensor<string, []>("op_2654_to_fp16_quantized"), quantized_data = tensor<int8, [1, 8, 64, 249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(85938432))), scale = tensor<fp16, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86065984))), zero_point = tensor<int8, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5281536)))];
            tensor<fp16, [1, 8, 125, 64]> q_with_bias_v_27_cast_fp16 = transpose(perm = q_with_bias_v_27_perm_0, x = var_2652_cast_fp16)[name = tensor<string, []>("transpose_129")];
            tensor<fp16, [1, 8, 125, 249]> x_293_cast_fp16 = matmul(transpose_x = x_293_transpose_x_0, transpose_y = x_293_transpose_y_0, x = q_with_bias_v_27_cast_fp16, y = op_2654_to_fp16_quantized)[name = tensor<string, []>("x_293_cast_fp16")];
            tensor<int32, [8]> x_295_pad_0 = const()[name = tensor<string, []>("x_295_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            tensor<string, []> x_295_mode_0 = const()[name = tensor<string, []>("x_295_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_200_to_fp16 = const()[name = tensor<string, []>("const_200_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 8, 125, 250]> x_295_cast_fp16 = pad(constant_val = const_200_to_fp16, mode = x_295_mode_0, pad = x_295_pad_0, x = x_293_cast_fp16)[name = tensor<string, []>("x_295_cast_fp16")];
            tensor<int32, [4]> var_2662 = const()[name = tensor<string, []>("op_2662"), val = tensor<int32, [4]>([1, 8, -1, 125])];
            tensor<fp16, [1, 8, 250, 125]> x_297_cast_fp16 = reshape(shape = var_2662, x = x_295_cast_fp16)[name = tensor<string, []>("x_297_cast_fp16")];
            tensor<int32, [4]> var_2666_begin_0 = const()[name = tensor<string, []>("op_2666_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_2666_end_0 = const()[name = tensor<string, []>("op_2666_end_0"), val = tensor<int32, [4]>([1, 8, 250, 125])];
            tensor<bool, [4]> var_2666_end_mask_0 = const()[name = tensor<string, []>("op_2666_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 249, 125]> var_2666_cast_fp16 = slice_by_index(begin = var_2666_begin_0, end = var_2666_end_0, end_mask = var_2666_end_mask_0, x = x_297_cast_fp16)[name = tensor<string, []>("op_2666_cast_fp16")];
            tensor<int32, [4]> var_2667 = const()[name = tensor<string, []>("op_2667"), val = tensor<int32, [4]>([1, 8, 125, 249])];
            tensor<fp16, [1, 8, 125, 249]> matrix_bd_53_cast_fp16 = reshape(shape = var_2667, x = var_2666_cast_fp16)[name = tensor<string, []>("matrix_bd_53_cast_fp16")];
            tensor<bool, []> matrix_ac_27_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_27_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matrix_ac_27_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_27_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_94_perm_0 = const()[name = tensor<string, []>("transpose_94_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_95_perm_0 = const()[name = tensor<string, []>("transpose_95_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 64, 125]> transpose_95 = transpose(perm = transpose_95_perm_0, x = k_53_cast_fp16)[name = tensor<string, []>("transpose_127")];
            tensor<fp16, [1, 8, 125, 64]> transpose_94 = transpose(perm = transpose_94_perm_0, x = var_2650_cast_fp16)[name = tensor<string, []>("transpose_128")];
            tensor<fp16, [1, 8, 125, 125]> matrix_ac_27_cast_fp16 = matmul(transpose_x = matrix_ac_27_transpose_x_0, transpose_y = matrix_ac_27_transpose_y_0, x = transpose_94, y = transpose_95)[name = tensor<string, []>("matrix_ac_27_cast_fp16")];
            tensor<int32, [4]> matrix_bd_55_begin_0 = const()[name = tensor<string, []>("matrix_bd_55_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_55_end_0 = const()[name = tensor<string, []>("matrix_bd_55_end_0"), val = tensor<int32, [4]>([1, 8, 125, 125])];
            tensor<bool, [4]> matrix_bd_55_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_55_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 125, 125]> matrix_bd_55_cast_fp16 = slice_by_index(begin = matrix_bd_55_begin_0, end = matrix_bd_55_end_0, end_mask = matrix_bd_55_end_mask_0, x = matrix_bd_53_cast_fp16)[name = tensor<string, []>("matrix_bd_55_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_2676_cast_fp16 = add(x = matrix_ac_27_cast_fp16, y = matrix_bd_55_cast_fp16)[name = tensor<string, []>("op_2676_cast_fp16")];
            tensor<fp16, []> _inversed_scores_53_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_53_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 8, 125, 125]> _inversed_scores_53_cast_fp16 = mul(x = var_2676_cast_fp16, y = _inversed_scores_53_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_53_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> scores_55_cast_fp16 = select(a = var_8_to_fp16, b = _inversed_scores_53_cast_fp16, cond = mask_11)[name = tensor<string, []>("scores_55_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_2682_cast_fp16 = softmax(axis = var_19, x = scores_55_cast_fp16)[name = tensor<string, []>("op_2682_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> input_709_cast_fp16 = select(a = var_7_to_fp16, b = var_2682_cast_fp16, cond = mask_11)[name = tensor<string, []>("input_709_cast_fp16")];
            tensor<bool, []> x_299_transpose_x_0 = const()[name = tensor<string, []>("x_299_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_299_transpose_y_0 = const()[name = tensor<string, []>("x_299_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 125, 64]> value_27_cast_fp16 = transpose(perm = value_27_perm_0, x = v_27_cast_fp16)[name = tensor<string, []>("transpose_126")];
            tensor<fp16, [1, 8, 125, 64]> x_299_cast_fp16 = matmul(transpose_x = x_299_transpose_x_0, transpose_y = x_299_transpose_y_0, x = input_709_cast_fp16, y = value_27_cast_fp16)[name = tensor<string, []>("x_299_cast_fp16")];
            tensor<int32, [4]> var_2686_perm_0 = const()[name = tensor<string, []>("op_2686_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2687 = const()[name = tensor<string, []>("op_2687"), val = tensor<int32, [3]>([1, -1, 512])];
            tensor<fp16, [1, 125, 8, 64]> var_2686_cast_fp16 = transpose(perm = var_2686_perm_0, x = x_299_cast_fp16)[name = tensor<string, []>("transpose_125")];
            tensor<fp16, [1, 125, 512]> input_711_cast_fp16 = reshape(shape = var_2687, x = var_2686_cast_fp16)[name = tensor<string, []>("input_711_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_13_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_13_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86066560))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86328768))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_13_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86329856)))];
            tensor<fp16, [1, 125, 512]> linear_124_cast_fp16 = linear(bias = encoder_layers_13_self_attn_linear_out_bias_to_fp16, weight = encoder_layers_13_self_attn_linear_out_weight_to_fp16_quantized, x = input_711_cast_fp16)[name = tensor<string, []>("linear_124_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_715_cast_fp16 = add(x = input_707_cast_fp16, y = linear_124_cast_fp16)[name = tensor<string, []>("input_715_cast_fp16")];
            tensor<int32, [1]> x_303_axes_0 = const()[name = tensor<string, []>("x_303_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_13_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_norm_conv_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86330944)))];
            tensor<fp16, [512]> encoder_layers_13_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_norm_conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86332032)))];
            tensor<fp16, [1, 125, 512]> x_303_cast_fp16 = layer_norm(axes = x_303_axes_0, beta = encoder_layers_13_norm_conv_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_13_norm_conv_weight_to_fp16, x = input_715_cast_fp16)[name = tensor<string, []>("x_303_cast_fp16")];
            tensor<int32, [3]> input_717_perm_0 = const()[name = tensor<string, []>("input_717_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> input_719_pad_type_0 = const()[name = tensor<string, []>("input_719_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> input_719_strides_0 = const()[name = tensor<string, []>("input_719_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_719_pad_0 = const()[name = tensor<string, []>("input_719_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_719_dilations_0 = const()[name = tensor<string, []>("input_719_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> input_719_groups_0 = const()[name = tensor<string, []>("input_719_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 512, 1]> encoder_layers_13_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_13_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1024, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86333120))), scale = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86857472))), zero_point = tensor<int8, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6073344)))];
            tensor<fp16, [1024]> encoder_layers_13_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86859584)))];
            tensor<fp16, [1, 512, 125]> input_717_cast_fp16 = transpose(perm = input_717_perm_0, x = x_303_cast_fp16)[name = tensor<string, []>("transpose_124")];
            tensor<fp16, [1, 1024, 125]> input_719_cast_fp16 = conv(bias = encoder_layers_13_conv_pointwise_conv1_bias_to_fp16, dilations = input_719_dilations_0, groups = input_719_groups_0, pad = input_719_pad_0, pad_type = input_719_pad_type_0, strides = input_719_strides_0, weight = encoder_layers_13_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_717_cast_fp16)[name = tensor<string, []>("input_719_cast_fp16")];
            tensor<int32, []> x_305_split_num_splits_0 = const()[name = tensor<string, []>("x_305_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> x_305_split_axis_0 = const()[name = tensor<string, []>("x_305_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 512, 125]> x_305_split_cast_fp16_0, tensor<fp16, [1, 512, 125]> x_305_split_cast_fp16_1 = split(axis = x_305_split_axis_0, num_splits = x_305_split_num_splits_0, x = input_719_cast_fp16)[name = tensor<string, []>("x_305_split_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_305_split_1_sigmoid_cast_fp16 = sigmoid(x = x_305_split_cast_fp16_1)[name = tensor<string, []>("x_305_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_305_cast_fp16 = mul(x = x_305_split_cast_fp16_0, y = x_305_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_305_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_721_cast_fp16 = select(a = var_7_to_fp16, b = x_305_cast_fp16, cond = var_449)[name = tensor<string, []>("input_721_cast_fp16")];
            tensor<int32, [6]> input_723_pad_0 = const()[name = tensor<string, []>("input_723_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            tensor<string, []> input_723_mode_0 = const()[name = tensor<string, []>("input_723_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_203_to_fp16 = const()[name = tensor<string, []>("const_203_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 512, 133]> input_723_cast_fp16 = pad(constant_val = const_203_to_fp16, mode = input_723_mode_0, pad = input_723_pad_0, x = input_721_cast_fp16)[name = tensor<string, []>("input_723_cast_fp16")];
            tensor<string, []> input_725_pad_type_0 = const()[name = tensor<string, []>("input_725_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, []> input_725_groups_0 = const()[name = tensor<string, []>("input_725_groups_0"), val = tensor<int32, []>(512)];
            tensor<int32, [1]> input_725_strides_0 = const()[name = tensor<string, []>("input_725_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_725_pad_0 = const()[name = tensor<string, []>("input_725_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_725_dilations_0 = const()[name = tensor<string, []>("input_725_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 9]> const_263_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("const_263_to_fp16_quantized"), quantized_data = tensor<int8, [512, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86861696))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86866368))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> const_264_to_fp16 = const()[name = tensor<string, []>("const_264_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86867456)))];
            tensor<fp16, [1, 512, 125]> input_727_cast_fp16 = conv(bias = const_264_to_fp16, dilations = input_725_dilations_0, groups = input_725_groups_0, pad = input_725_pad_0, pad_type = input_725_pad_type_0, strides = input_725_strides_0, weight = const_263_to_fp16_quantized, x = input_723_cast_fp16)[name = tensor<string, []>("input_727_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_729_cast_fp16 = silu(x = input_727_cast_fp16)[name = tensor<string, []>("input_729_cast_fp16")];
            tensor<string, []> x_307_pad_type_0 = const()[name = tensor<string, []>("x_307_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> x_307_strides_0 = const()[name = tensor<string, []>("x_307_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_307_pad_0 = const()[name = tensor<string, []>("x_307_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_307_dilations_0 = const()[name = tensor<string, []>("x_307_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> x_307_groups_0 = const()[name = tensor<string, []>("x_307_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [512, 512, 1]> encoder_layers_13_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_13_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(86868544))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87130752))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_13_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87131840)))];
            tensor<fp16, [1, 512, 125]> x_307_cast_fp16 = conv(bias = encoder_layers_13_conv_pointwise_conv2_bias_to_fp16, dilations = x_307_dilations_0, groups = x_307_groups_0, pad = x_307_pad_0, pad_type = x_307_pad_type_0, strides = x_307_strides_0, weight = encoder_layers_13_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_729_cast_fp16)[name = tensor<string, []>("x_307_cast_fp16")];
            tensor<int32, [3]> input_731_perm_0 = const()[name = tensor<string, []>("input_731_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 125, 512]> input_731_cast_fp16 = transpose(perm = input_731_perm_0, x = x_307_cast_fp16)[name = tensor<string, []>("transpose_123")];
            tensor<fp16, [1, 125, 512]> input_733_cast_fp16 = add(x = input_715_cast_fp16, y = input_731_cast_fp16)[name = tensor<string, []>("input_733_cast_fp16")];
            tensor<int32, [1]> input_735_axes_0 = const()[name = tensor<string, []>("input_735_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_13_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87132928)))];
            tensor<fp16, [512]> encoder_layers_13_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87134016)))];
            tensor<fp16, [1, 125, 512]> input_735_cast_fp16 = layer_norm(axes = input_735_axes_0, beta = encoder_layers_13_norm_feed_forward2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_13_norm_feed_forward2_weight_to_fp16, x = input_733_cast_fp16)[name = tensor<string, []>("input_735_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_13_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_13_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(87135104))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88183744))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_13_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88187904)))];
            tensor<fp16, [1, 125, 2048]> linear_125_cast_fp16 = linear(bias = encoder_layers_13_feed_forward2_linear1_bias_to_fp16, weight = encoder_layers_13_feed_forward2_linear1_weight_to_fp16_quantized, x = input_735_cast_fp16)[name = tensor<string, []>("linear_125_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_739_cast_fp16 = silu(x = linear_125_cast_fp16)[name = tensor<string, []>("input_739_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_13_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_13_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(88192064))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89240704))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_13_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89241792)))];
            tensor<fp16, [1, 125, 512]> linear_126_cast_fp16 = linear(bias = encoder_layers_13_feed_forward2_linear2_bias_to_fp16, weight = encoder_layers_13_feed_forward2_linear2_weight_to_fp16_quantized, x = input_739_cast_fp16)[name = tensor<string, []>("linear_126_cast_fp16")];
            tensor<fp16, []> var_2753_to_fp16 = const()[name = tensor<string, []>("op_2753_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_2754_cast_fp16 = mul(x = linear_126_cast_fp16, y = var_2753_to_fp16)[name = tensor<string, []>("op_2754_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_745_cast_fp16 = add(x = input_733_cast_fp16, y = var_2754_cast_fp16)[name = tensor<string, []>("input_745_cast_fp16")];
            tensor<int32, [1]> input_747_axes_0 = const()[name = tensor<string, []>("input_747_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_13_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_norm_out_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89242880)))];
            tensor<fp16, [512]> encoder_layers_13_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_13_norm_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89243968)))];
            tensor<fp16, [1, 125, 512]> input_747_cast_fp16 = layer_norm(axes = input_747_axes_0, beta = encoder_layers_13_norm_out_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_13_norm_out_weight_to_fp16, x = input_745_cast_fp16)[name = tensor<string, []>("input_747_cast_fp16")];
            tensor<int32, [1]> input_749_axes_0 = const()[name = tensor<string, []>("input_749_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_14_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89245056)))];
            tensor<fp16, [512]> encoder_layers_14_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89246144)))];
            tensor<fp16, [1, 125, 512]> input_749_cast_fp16 = layer_norm(axes = input_749_axes_0, beta = encoder_layers_14_norm_feed_forward1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_14_norm_feed_forward1_weight_to_fp16, x = input_747_cast_fp16)[name = tensor<string, []>("input_749_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_14_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_14_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(89247232))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90295872))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_14_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90300032)))];
            tensor<fp16, [1, 125, 2048]> linear_127_cast_fp16 = linear(bias = encoder_layers_14_feed_forward1_linear1_bias_to_fp16, weight = encoder_layers_14_feed_forward1_linear1_weight_to_fp16_quantized, x = input_749_cast_fp16)[name = tensor<string, []>("linear_127_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_753_cast_fp16 = silu(x = linear_127_cast_fp16)[name = tensor<string, []>("input_753_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_14_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_14_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(90304192))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91352832))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_14_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91353920)))];
            tensor<fp16, [1, 125, 512]> linear_128_cast_fp16 = linear(bias = encoder_layers_14_feed_forward1_linear2_bias_to_fp16, weight = encoder_layers_14_feed_forward1_linear2_weight_to_fp16_quantized, x = input_753_cast_fp16)[name = tensor<string, []>("linear_128_cast_fp16")];
            tensor<fp16, []> var_2784_to_fp16 = const()[name = tensor<string, []>("op_2784_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_2785_cast_fp16 = mul(x = linear_128_cast_fp16, y = var_2784_to_fp16)[name = tensor<string, []>("op_2785_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_759_cast_fp16 = add(x = input_747_cast_fp16, y = var_2785_cast_fp16)[name = tensor<string, []>("input_759_cast_fp16")];
            tensor<int32, [1]> query_29_axes_0 = const()[name = tensor<string, []>("query_29_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_14_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_norm_self_att_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91355008)))];
            tensor<fp16, [512]> encoder_layers_14_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_norm_self_att_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91356096)))];
            tensor<fp16, [1, 125, 512]> query_29_cast_fp16 = layer_norm(axes = query_29_axes_0, beta = encoder_layers_14_norm_self_att_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_14_norm_self_att_weight_to_fp16, x = input_759_cast_fp16)[name = tensor<string, []>("query_29_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_14_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_14_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91357184))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91619392))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_14_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91620480)))];
            tensor<fp16, [1, 125, 512]> linear_129_cast_fp16 = linear(bias = encoder_layers_14_self_attn_linear_q_bias_to_fp16, weight = encoder_layers_14_self_attn_linear_q_weight_to_fp16_quantized, x = query_29_cast_fp16)[name = tensor<string, []>("linear_129_cast_fp16")];
            tensor<int32, [4]> var_2802 = const()[name = tensor<string, []>("op_2802"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> q_85_cast_fp16 = reshape(shape = var_2802, x = linear_129_cast_fp16)[name = tensor<string, []>("q_85_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_14_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_14_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91621568))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91883776))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_14_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91884864)))];
            tensor<fp16, [1, 125, 512]> linear_130_cast_fp16 = linear(bias = encoder_layers_14_self_attn_linear_k_bias_to_fp16, weight = encoder_layers_14_self_attn_linear_k_weight_to_fp16_quantized, x = query_29_cast_fp16)[name = tensor<string, []>("linear_130_cast_fp16")];
            tensor<int32, [4]> var_2807 = const()[name = tensor<string, []>("op_2807"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> k_57_cast_fp16 = reshape(shape = var_2807, x = linear_130_cast_fp16)[name = tensor<string, []>("k_57_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_14_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_14_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(91885952))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92148160))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_14_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92149248)))];
            tensor<fp16, [1, 125, 512]> linear_131_cast_fp16 = linear(bias = encoder_layers_14_self_attn_linear_v_bias_to_fp16, weight = encoder_layers_14_self_attn_linear_v_weight_to_fp16_quantized, x = query_29_cast_fp16)[name = tensor<string, []>("linear_131_cast_fp16")];
            tensor<int32, [4]> var_2812 = const()[name = tensor<string, []>("op_2812"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> v_29_cast_fp16 = reshape(shape = var_2812, x = linear_131_cast_fp16)[name = tensor<string, []>("v_29_cast_fp16")];
            tensor<int32, [4]> value_29_perm_0 = const()[name = tensor<string, []>("value_29_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 64]> encoder_layers_14_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92150336)))];
            tensor<fp16, [1, 125, 8, 64]> var_2824_cast_fp16 = add(x = q_85_cast_fp16, y = encoder_layers_14_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_2824_cast_fp16")];
            tensor<fp16, [8, 64]> encoder_layers_14_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92151424)))];
            tensor<fp16, [1, 125, 8, 64]> var_2826_cast_fp16 = add(x = q_85_cast_fp16, y = encoder_layers_14_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_2826_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_29_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_29_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<bool, []> x_315_transpose_x_0 = const()[name = tensor<string, []>("x_315_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_315_transpose_y_0 = const()[name = tensor<string, []>("x_315_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 64, 249]> op_2828_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(3), name = tensor<string, []>("op_2828_to_fp16_quantized"), quantized_data = tensor<int8, [1, 8, 64, 249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92152512))), scale = tensor<fp16, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92280064))), zero_point = tensor<int8, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5281536)))];
            tensor<fp16, [1, 8, 125, 64]> q_with_bias_v_29_cast_fp16 = transpose(perm = q_with_bias_v_29_perm_0, x = var_2826_cast_fp16)[name = tensor<string, []>("transpose_122")];
            tensor<fp16, [1, 8, 125, 249]> x_315_cast_fp16 = matmul(transpose_x = x_315_transpose_x_0, transpose_y = x_315_transpose_y_0, x = q_with_bias_v_29_cast_fp16, y = op_2828_to_fp16_quantized)[name = tensor<string, []>("x_315_cast_fp16")];
            tensor<int32, [8]> x_317_pad_0 = const()[name = tensor<string, []>("x_317_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            tensor<string, []> x_317_mode_0 = const()[name = tensor<string, []>("x_317_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_210_to_fp16 = const()[name = tensor<string, []>("const_210_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 8, 125, 250]> x_317_cast_fp16 = pad(constant_val = const_210_to_fp16, mode = x_317_mode_0, pad = x_317_pad_0, x = x_315_cast_fp16)[name = tensor<string, []>("x_317_cast_fp16")];
            tensor<int32, [4]> var_2836 = const()[name = tensor<string, []>("op_2836"), val = tensor<int32, [4]>([1, 8, -1, 125])];
            tensor<fp16, [1, 8, 250, 125]> x_319_cast_fp16 = reshape(shape = var_2836, x = x_317_cast_fp16)[name = tensor<string, []>("x_319_cast_fp16")];
            tensor<int32, [4]> var_2840_begin_0 = const()[name = tensor<string, []>("op_2840_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_2840_end_0 = const()[name = tensor<string, []>("op_2840_end_0"), val = tensor<int32, [4]>([1, 8, 250, 125])];
            tensor<bool, [4]> var_2840_end_mask_0 = const()[name = tensor<string, []>("op_2840_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 249, 125]> var_2840_cast_fp16 = slice_by_index(begin = var_2840_begin_0, end = var_2840_end_0, end_mask = var_2840_end_mask_0, x = x_319_cast_fp16)[name = tensor<string, []>("op_2840_cast_fp16")];
            tensor<int32, [4]> var_2841 = const()[name = tensor<string, []>("op_2841"), val = tensor<int32, [4]>([1, 8, 125, 249])];
            tensor<fp16, [1, 8, 125, 249]> matrix_bd_57_cast_fp16 = reshape(shape = var_2841, x = var_2840_cast_fp16)[name = tensor<string, []>("matrix_bd_57_cast_fp16")];
            tensor<bool, []> matrix_ac_29_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_29_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matrix_ac_29_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_29_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_96_perm_0 = const()[name = tensor<string, []>("transpose_96_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_97_perm_0 = const()[name = tensor<string, []>("transpose_97_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 64, 125]> transpose_97 = transpose(perm = transpose_97_perm_0, x = k_57_cast_fp16)[name = tensor<string, []>("transpose_120")];
            tensor<fp16, [1, 8, 125, 64]> transpose_96 = transpose(perm = transpose_96_perm_0, x = var_2824_cast_fp16)[name = tensor<string, []>("transpose_121")];
            tensor<fp16, [1, 8, 125, 125]> matrix_ac_29_cast_fp16 = matmul(transpose_x = matrix_ac_29_transpose_x_0, transpose_y = matrix_ac_29_transpose_y_0, x = transpose_96, y = transpose_97)[name = tensor<string, []>("matrix_ac_29_cast_fp16")];
            tensor<int32, [4]> matrix_bd_59_begin_0 = const()[name = tensor<string, []>("matrix_bd_59_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_59_end_0 = const()[name = tensor<string, []>("matrix_bd_59_end_0"), val = tensor<int32, [4]>([1, 8, 125, 125])];
            tensor<bool, [4]> matrix_bd_59_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_59_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 125, 125]> matrix_bd_59_cast_fp16 = slice_by_index(begin = matrix_bd_59_begin_0, end = matrix_bd_59_end_0, end_mask = matrix_bd_59_end_mask_0, x = matrix_bd_57_cast_fp16)[name = tensor<string, []>("matrix_bd_59_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_2850_cast_fp16 = add(x = matrix_ac_29_cast_fp16, y = matrix_bd_59_cast_fp16)[name = tensor<string, []>("op_2850_cast_fp16")];
            tensor<fp16, []> _inversed_scores_57_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_57_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 8, 125, 125]> _inversed_scores_57_cast_fp16 = mul(x = var_2850_cast_fp16, y = _inversed_scores_57_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_57_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> scores_59_cast_fp16 = select(a = var_8_to_fp16, b = _inversed_scores_57_cast_fp16, cond = mask_11)[name = tensor<string, []>("scores_59_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_2856_cast_fp16 = softmax(axis = var_19, x = scores_59_cast_fp16)[name = tensor<string, []>("op_2856_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> input_761_cast_fp16 = select(a = var_7_to_fp16, b = var_2856_cast_fp16, cond = mask_11)[name = tensor<string, []>("input_761_cast_fp16")];
            tensor<bool, []> x_321_transpose_x_0 = const()[name = tensor<string, []>("x_321_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_321_transpose_y_0 = const()[name = tensor<string, []>("x_321_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 125, 64]> value_29_cast_fp16 = transpose(perm = value_29_perm_0, x = v_29_cast_fp16)[name = tensor<string, []>("transpose_119")];
            tensor<fp16, [1, 8, 125, 64]> x_321_cast_fp16 = matmul(transpose_x = x_321_transpose_x_0, transpose_y = x_321_transpose_y_0, x = input_761_cast_fp16, y = value_29_cast_fp16)[name = tensor<string, []>("x_321_cast_fp16")];
            tensor<int32, [4]> var_2860_perm_0 = const()[name = tensor<string, []>("op_2860_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_2861 = const()[name = tensor<string, []>("op_2861"), val = tensor<int32, [3]>([1, -1, 512])];
            tensor<fp16, [1, 125, 8, 64]> var_2860_cast_fp16 = transpose(perm = var_2860_perm_0, x = x_321_cast_fp16)[name = tensor<string, []>("transpose_118")];
            tensor<fp16, [1, 125, 512]> input_763_cast_fp16 = reshape(shape = var_2861, x = var_2860_cast_fp16)[name = tensor<string, []>("input_763_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_14_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_14_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92280640))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92542848))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_14_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92543936)))];
            tensor<fp16, [1, 125, 512]> linear_133_cast_fp16 = linear(bias = encoder_layers_14_self_attn_linear_out_bias_to_fp16, weight = encoder_layers_14_self_attn_linear_out_weight_to_fp16_quantized, x = input_763_cast_fp16)[name = tensor<string, []>("linear_133_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_767_cast_fp16 = add(x = input_759_cast_fp16, y = linear_133_cast_fp16)[name = tensor<string, []>("input_767_cast_fp16")];
            tensor<int32, [1]> x_325_axes_0 = const()[name = tensor<string, []>("x_325_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_14_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_norm_conv_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92545024)))];
            tensor<fp16, [512]> encoder_layers_14_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_norm_conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92546112)))];
            tensor<fp16, [1, 125, 512]> x_325_cast_fp16 = layer_norm(axes = x_325_axes_0, beta = encoder_layers_14_norm_conv_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_14_norm_conv_weight_to_fp16, x = input_767_cast_fp16)[name = tensor<string, []>("x_325_cast_fp16")];
            tensor<int32, [3]> input_769_perm_0 = const()[name = tensor<string, []>("input_769_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> input_771_pad_type_0 = const()[name = tensor<string, []>("input_771_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> input_771_strides_0 = const()[name = tensor<string, []>("input_771_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_771_pad_0 = const()[name = tensor<string, []>("input_771_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_771_dilations_0 = const()[name = tensor<string, []>("input_771_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> input_771_groups_0 = const()[name = tensor<string, []>("input_771_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 512, 1]> encoder_layers_14_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_14_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1024, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(92547200))), scale = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93071552))), zero_point = tensor<int8, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6073344)))];
            tensor<fp16, [1024]> encoder_layers_14_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93073664)))];
            tensor<fp16, [1, 512, 125]> input_769_cast_fp16 = transpose(perm = input_769_perm_0, x = x_325_cast_fp16)[name = tensor<string, []>("transpose_117")];
            tensor<fp16, [1, 1024, 125]> input_771_cast_fp16 = conv(bias = encoder_layers_14_conv_pointwise_conv1_bias_to_fp16, dilations = input_771_dilations_0, groups = input_771_groups_0, pad = input_771_pad_0, pad_type = input_771_pad_type_0, strides = input_771_strides_0, weight = encoder_layers_14_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_769_cast_fp16)[name = tensor<string, []>("input_771_cast_fp16")];
            tensor<int32, []> x_327_split_num_splits_0 = const()[name = tensor<string, []>("x_327_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> x_327_split_axis_0 = const()[name = tensor<string, []>("x_327_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 512, 125]> x_327_split_cast_fp16_0, tensor<fp16, [1, 512, 125]> x_327_split_cast_fp16_1 = split(axis = x_327_split_axis_0, num_splits = x_327_split_num_splits_0, x = input_771_cast_fp16)[name = tensor<string, []>("x_327_split_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_327_split_1_sigmoid_cast_fp16 = sigmoid(x = x_327_split_cast_fp16_1)[name = tensor<string, []>("x_327_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_327_cast_fp16 = mul(x = x_327_split_cast_fp16_0, y = x_327_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_327_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_773_cast_fp16 = select(a = var_7_to_fp16, b = x_327_cast_fp16, cond = var_449)[name = tensor<string, []>("input_773_cast_fp16")];
            tensor<int32, [6]> input_775_pad_0 = const()[name = tensor<string, []>("input_775_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            tensor<string, []> input_775_mode_0 = const()[name = tensor<string, []>("input_775_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_213_to_fp16 = const()[name = tensor<string, []>("const_213_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 512, 133]> input_775_cast_fp16 = pad(constant_val = const_213_to_fp16, mode = input_775_mode_0, pad = input_775_pad_0, x = input_773_cast_fp16)[name = tensor<string, []>("input_775_cast_fp16")];
            tensor<string, []> input_777_pad_type_0 = const()[name = tensor<string, []>("input_777_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, []> input_777_groups_0 = const()[name = tensor<string, []>("input_777_groups_0"), val = tensor<int32, []>(512)];
            tensor<int32, [1]> input_777_strides_0 = const()[name = tensor<string, []>("input_777_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_777_pad_0 = const()[name = tensor<string, []>("input_777_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_777_dilations_0 = const()[name = tensor<string, []>("input_777_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 9]> const_265_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("const_265_to_fp16_quantized"), quantized_data = tensor<int8, [512, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93075776))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93080448))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> const_266_to_fp16 = const()[name = tensor<string, []>("const_266_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93081536)))];
            tensor<fp16, [1, 512, 125]> input_779_cast_fp16 = conv(bias = const_266_to_fp16, dilations = input_777_dilations_0, groups = input_777_groups_0, pad = input_777_pad_0, pad_type = input_777_pad_type_0, strides = input_777_strides_0, weight = const_265_to_fp16_quantized, x = input_775_cast_fp16)[name = tensor<string, []>("input_779_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_781_cast_fp16 = silu(x = input_779_cast_fp16)[name = tensor<string, []>("input_781_cast_fp16")];
            tensor<string, []> x_329_pad_type_0 = const()[name = tensor<string, []>("x_329_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> x_329_strides_0 = const()[name = tensor<string, []>("x_329_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_329_pad_0 = const()[name = tensor<string, []>("x_329_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_329_dilations_0 = const()[name = tensor<string, []>("x_329_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> x_329_groups_0 = const()[name = tensor<string, []>("x_329_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [512, 512, 1]> encoder_layers_14_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_14_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93082624))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93344832))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_14_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93345920)))];
            tensor<fp16, [1, 512, 125]> x_329_cast_fp16 = conv(bias = encoder_layers_14_conv_pointwise_conv2_bias_to_fp16, dilations = x_329_dilations_0, groups = x_329_groups_0, pad = x_329_pad_0, pad_type = x_329_pad_type_0, strides = x_329_strides_0, weight = encoder_layers_14_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_781_cast_fp16)[name = tensor<string, []>("x_329_cast_fp16")];
            tensor<int32, [3]> input_783_perm_0 = const()[name = tensor<string, []>("input_783_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 125, 512]> input_783_cast_fp16 = transpose(perm = input_783_perm_0, x = x_329_cast_fp16)[name = tensor<string, []>("transpose_116")];
            tensor<fp16, [1, 125, 512]> input_785_cast_fp16 = add(x = input_767_cast_fp16, y = input_783_cast_fp16)[name = tensor<string, []>("input_785_cast_fp16")];
            tensor<int32, [1]> input_787_axes_0 = const()[name = tensor<string, []>("input_787_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_14_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93347008)))];
            tensor<fp16, [512]> encoder_layers_14_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93348096)))];
            tensor<fp16, [1, 125, 512]> input_787_cast_fp16 = layer_norm(axes = input_787_axes_0, beta = encoder_layers_14_norm_feed_forward2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_14_norm_feed_forward2_weight_to_fp16, x = input_785_cast_fp16)[name = tensor<string, []>("input_787_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_14_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_14_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(93349184))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94397824))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_14_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94401984)))];
            tensor<fp16, [1, 125, 2048]> linear_134_cast_fp16 = linear(bias = encoder_layers_14_feed_forward2_linear1_bias_to_fp16, weight = encoder_layers_14_feed_forward2_linear1_weight_to_fp16_quantized, x = input_787_cast_fp16)[name = tensor<string, []>("linear_134_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_791_cast_fp16 = silu(x = linear_134_cast_fp16)[name = tensor<string, []>("input_791_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_14_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_14_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(94406144))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95454784))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_14_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95455872)))];
            tensor<fp16, [1, 125, 512]> linear_135_cast_fp16 = linear(bias = encoder_layers_14_feed_forward2_linear2_bias_to_fp16, weight = encoder_layers_14_feed_forward2_linear2_weight_to_fp16_quantized, x = input_791_cast_fp16)[name = tensor<string, []>("linear_135_cast_fp16")];
            tensor<fp16, []> var_2927_to_fp16 = const()[name = tensor<string, []>("op_2927_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_2928_cast_fp16 = mul(x = linear_135_cast_fp16, y = var_2927_to_fp16)[name = tensor<string, []>("op_2928_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_797_cast_fp16 = add(x = input_785_cast_fp16, y = var_2928_cast_fp16)[name = tensor<string, []>("input_797_cast_fp16")];
            tensor<int32, [1]> input_799_axes_0 = const()[name = tensor<string, []>("input_799_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_14_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_norm_out_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95456960)))];
            tensor<fp16, [512]> encoder_layers_14_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_14_norm_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95458048)))];
            tensor<fp16, [1, 125, 512]> input_799_cast_fp16 = layer_norm(axes = input_799_axes_0, beta = encoder_layers_14_norm_out_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_14_norm_out_weight_to_fp16, x = input_797_cast_fp16)[name = tensor<string, []>("input_799_cast_fp16")];
            tensor<int32, [1]> input_801_axes_0 = const()[name = tensor<string, []>("input_801_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_15_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95459136)))];
            tensor<fp16, [512]> encoder_layers_15_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95460224)))];
            tensor<fp16, [1, 125, 512]> input_801_cast_fp16 = layer_norm(axes = input_801_axes_0, beta = encoder_layers_15_norm_feed_forward1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_15_norm_feed_forward1_weight_to_fp16, x = input_799_cast_fp16)[name = tensor<string, []>("input_801_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_15_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_15_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(95461312))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96509952))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_15_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96514112)))];
            tensor<fp16, [1, 125, 2048]> linear_136_cast_fp16 = linear(bias = encoder_layers_15_feed_forward1_linear1_bias_to_fp16, weight = encoder_layers_15_feed_forward1_linear1_weight_to_fp16_quantized, x = input_801_cast_fp16)[name = tensor<string, []>("linear_136_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_805_cast_fp16 = silu(x = linear_136_cast_fp16)[name = tensor<string, []>("input_805_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_15_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_15_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(96518272))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97566912))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_15_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97568000)))];
            tensor<fp16, [1, 125, 512]> linear_137_cast_fp16 = linear(bias = encoder_layers_15_feed_forward1_linear2_bias_to_fp16, weight = encoder_layers_15_feed_forward1_linear2_weight_to_fp16_quantized, x = input_805_cast_fp16)[name = tensor<string, []>("linear_137_cast_fp16")];
            tensor<fp16, []> var_2958_to_fp16 = const()[name = tensor<string, []>("op_2958_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_2959_cast_fp16 = mul(x = linear_137_cast_fp16, y = var_2958_to_fp16)[name = tensor<string, []>("op_2959_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_811_cast_fp16 = add(x = input_799_cast_fp16, y = var_2959_cast_fp16)[name = tensor<string, []>("input_811_cast_fp16")];
            tensor<int32, [1]> query_31_axes_0 = const()[name = tensor<string, []>("query_31_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_15_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_norm_self_att_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97569088)))];
            tensor<fp16, [512]> encoder_layers_15_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_norm_self_att_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97570176)))];
            tensor<fp16, [1, 125, 512]> query_31_cast_fp16 = layer_norm(axes = query_31_axes_0, beta = encoder_layers_15_norm_self_att_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_15_norm_self_att_weight_to_fp16, x = input_811_cast_fp16)[name = tensor<string, []>("query_31_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_15_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_15_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97571264))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97833472))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_15_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97834560)))];
            tensor<fp16, [1, 125, 512]> linear_138_cast_fp16 = linear(bias = encoder_layers_15_self_attn_linear_q_bias_to_fp16, weight = encoder_layers_15_self_attn_linear_q_weight_to_fp16_quantized, x = query_31_cast_fp16)[name = tensor<string, []>("linear_138_cast_fp16")];
            tensor<int32, [4]> var_2976 = const()[name = tensor<string, []>("op_2976"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> q_91_cast_fp16 = reshape(shape = var_2976, x = linear_138_cast_fp16)[name = tensor<string, []>("q_91_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_15_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_15_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(97835648))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98097856))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_15_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98098944)))];
            tensor<fp16, [1, 125, 512]> linear_139_cast_fp16 = linear(bias = encoder_layers_15_self_attn_linear_k_bias_to_fp16, weight = encoder_layers_15_self_attn_linear_k_weight_to_fp16_quantized, x = query_31_cast_fp16)[name = tensor<string, []>("linear_139_cast_fp16")];
            tensor<int32, [4]> var_2981 = const()[name = tensor<string, []>("op_2981"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> k_61_cast_fp16 = reshape(shape = var_2981, x = linear_139_cast_fp16)[name = tensor<string, []>("k_61_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_15_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_15_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98100032))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98362240))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_15_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98363328)))];
            tensor<fp16, [1, 125, 512]> linear_140_cast_fp16 = linear(bias = encoder_layers_15_self_attn_linear_v_bias_to_fp16, weight = encoder_layers_15_self_attn_linear_v_weight_to_fp16_quantized, x = query_31_cast_fp16)[name = tensor<string, []>("linear_140_cast_fp16")];
            tensor<int32, [4]> var_2986 = const()[name = tensor<string, []>("op_2986"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> v_31_cast_fp16 = reshape(shape = var_2986, x = linear_140_cast_fp16)[name = tensor<string, []>("v_31_cast_fp16")];
            tensor<int32, [4]> value_31_perm_0 = const()[name = tensor<string, []>("value_31_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 64]> encoder_layers_15_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98364416)))];
            tensor<fp16, [1, 125, 8, 64]> var_2998_cast_fp16 = add(x = q_91_cast_fp16, y = encoder_layers_15_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_2998_cast_fp16")];
            tensor<fp16, [8, 64]> encoder_layers_15_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98365504)))];
            tensor<fp16, [1, 125, 8, 64]> var_3000_cast_fp16 = add(x = q_91_cast_fp16, y = encoder_layers_15_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_3000_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_31_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_31_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<bool, []> x_337_transpose_x_0 = const()[name = tensor<string, []>("x_337_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_337_transpose_y_0 = const()[name = tensor<string, []>("x_337_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 64, 249]> op_3002_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(3), name = tensor<string, []>("op_3002_to_fp16_quantized"), quantized_data = tensor<int8, [1, 8, 64, 249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98366592))), scale = tensor<fp16, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98494144))), zero_point = tensor<int8, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5281536)))];
            tensor<fp16, [1, 8, 125, 64]> q_with_bias_v_31_cast_fp16 = transpose(perm = q_with_bias_v_31_perm_0, x = var_3000_cast_fp16)[name = tensor<string, []>("transpose_115")];
            tensor<fp16, [1, 8, 125, 249]> x_337_cast_fp16 = matmul(transpose_x = x_337_transpose_x_0, transpose_y = x_337_transpose_y_0, x = q_with_bias_v_31_cast_fp16, y = op_3002_to_fp16_quantized)[name = tensor<string, []>("x_337_cast_fp16")];
            tensor<int32, [8]> x_339_pad_0 = const()[name = tensor<string, []>("x_339_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            tensor<string, []> x_339_mode_0 = const()[name = tensor<string, []>("x_339_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_220_to_fp16 = const()[name = tensor<string, []>("const_220_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 8, 125, 250]> x_339_cast_fp16 = pad(constant_val = const_220_to_fp16, mode = x_339_mode_0, pad = x_339_pad_0, x = x_337_cast_fp16)[name = tensor<string, []>("x_339_cast_fp16")];
            tensor<int32, [4]> var_3010 = const()[name = tensor<string, []>("op_3010"), val = tensor<int32, [4]>([1, 8, -1, 125])];
            tensor<fp16, [1, 8, 250, 125]> x_341_cast_fp16 = reshape(shape = var_3010, x = x_339_cast_fp16)[name = tensor<string, []>("x_341_cast_fp16")];
            tensor<int32, [4]> var_3014_begin_0 = const()[name = tensor<string, []>("op_3014_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_3014_end_0 = const()[name = tensor<string, []>("op_3014_end_0"), val = tensor<int32, [4]>([1, 8, 250, 125])];
            tensor<bool, [4]> var_3014_end_mask_0 = const()[name = tensor<string, []>("op_3014_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 249, 125]> var_3014_cast_fp16 = slice_by_index(begin = var_3014_begin_0, end = var_3014_end_0, end_mask = var_3014_end_mask_0, x = x_341_cast_fp16)[name = tensor<string, []>("op_3014_cast_fp16")];
            tensor<int32, [4]> var_3015 = const()[name = tensor<string, []>("op_3015"), val = tensor<int32, [4]>([1, 8, 125, 249])];
            tensor<fp16, [1, 8, 125, 249]> matrix_bd_61_cast_fp16 = reshape(shape = var_3015, x = var_3014_cast_fp16)[name = tensor<string, []>("matrix_bd_61_cast_fp16")];
            tensor<bool, []> matrix_ac_31_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_31_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matrix_ac_31_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_31_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_98_perm_0 = const()[name = tensor<string, []>("transpose_98_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_99_perm_0 = const()[name = tensor<string, []>("transpose_99_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 64, 125]> transpose_99 = transpose(perm = transpose_99_perm_0, x = k_61_cast_fp16)[name = tensor<string, []>("transpose_113")];
            tensor<fp16, [1, 8, 125, 64]> transpose_98 = transpose(perm = transpose_98_perm_0, x = var_2998_cast_fp16)[name = tensor<string, []>("transpose_114")];
            tensor<fp16, [1, 8, 125, 125]> matrix_ac_31_cast_fp16 = matmul(transpose_x = matrix_ac_31_transpose_x_0, transpose_y = matrix_ac_31_transpose_y_0, x = transpose_98, y = transpose_99)[name = tensor<string, []>("matrix_ac_31_cast_fp16")];
            tensor<int32, [4]> matrix_bd_63_begin_0 = const()[name = tensor<string, []>("matrix_bd_63_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_63_end_0 = const()[name = tensor<string, []>("matrix_bd_63_end_0"), val = tensor<int32, [4]>([1, 8, 125, 125])];
            tensor<bool, [4]> matrix_bd_63_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_63_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 125, 125]> matrix_bd_63_cast_fp16 = slice_by_index(begin = matrix_bd_63_begin_0, end = matrix_bd_63_end_0, end_mask = matrix_bd_63_end_mask_0, x = matrix_bd_61_cast_fp16)[name = tensor<string, []>("matrix_bd_63_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_3024_cast_fp16 = add(x = matrix_ac_31_cast_fp16, y = matrix_bd_63_cast_fp16)[name = tensor<string, []>("op_3024_cast_fp16")];
            tensor<fp16, []> _inversed_scores_61_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_61_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 8, 125, 125]> _inversed_scores_61_cast_fp16 = mul(x = var_3024_cast_fp16, y = _inversed_scores_61_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_61_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> scores_63_cast_fp16 = select(a = var_8_to_fp16, b = _inversed_scores_61_cast_fp16, cond = mask_11)[name = tensor<string, []>("scores_63_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_3030_cast_fp16 = softmax(axis = var_19, x = scores_63_cast_fp16)[name = tensor<string, []>("op_3030_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> input_813_cast_fp16 = select(a = var_7_to_fp16, b = var_3030_cast_fp16, cond = mask_11)[name = tensor<string, []>("input_813_cast_fp16")];
            tensor<bool, []> x_343_transpose_x_0 = const()[name = tensor<string, []>("x_343_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_343_transpose_y_0 = const()[name = tensor<string, []>("x_343_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 125, 64]> value_31_cast_fp16 = transpose(perm = value_31_perm_0, x = v_31_cast_fp16)[name = tensor<string, []>("transpose_112")];
            tensor<fp16, [1, 8, 125, 64]> x_343_cast_fp16 = matmul(transpose_x = x_343_transpose_x_0, transpose_y = x_343_transpose_y_0, x = input_813_cast_fp16, y = value_31_cast_fp16)[name = tensor<string, []>("x_343_cast_fp16")];
            tensor<int32, [4]> var_3034_perm_0 = const()[name = tensor<string, []>("op_3034_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3035 = const()[name = tensor<string, []>("op_3035"), val = tensor<int32, [3]>([1, -1, 512])];
            tensor<fp16, [1, 125, 8, 64]> var_3034_cast_fp16 = transpose(perm = var_3034_perm_0, x = x_343_cast_fp16)[name = tensor<string, []>("transpose_111")];
            tensor<fp16, [1, 125, 512]> input_815_cast_fp16 = reshape(shape = var_3035, x = var_3034_cast_fp16)[name = tensor<string, []>("input_815_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_15_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_15_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98494720))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98756928))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_15_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98758016)))];
            tensor<fp16, [1, 125, 512]> linear_142_cast_fp16 = linear(bias = encoder_layers_15_self_attn_linear_out_bias_to_fp16, weight = encoder_layers_15_self_attn_linear_out_weight_to_fp16_quantized, x = input_815_cast_fp16)[name = tensor<string, []>("linear_142_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_819_cast_fp16 = add(x = input_811_cast_fp16, y = linear_142_cast_fp16)[name = tensor<string, []>("input_819_cast_fp16")];
            tensor<int32, [1]> x_347_axes_0 = const()[name = tensor<string, []>("x_347_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_15_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_norm_conv_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98759104)))];
            tensor<fp16, [512]> encoder_layers_15_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_norm_conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98760192)))];
            tensor<fp16, [1, 125, 512]> x_347_cast_fp16 = layer_norm(axes = x_347_axes_0, beta = encoder_layers_15_norm_conv_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_15_norm_conv_weight_to_fp16, x = input_819_cast_fp16)[name = tensor<string, []>("x_347_cast_fp16")];
            tensor<int32, [3]> input_821_perm_0 = const()[name = tensor<string, []>("input_821_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> input_823_pad_type_0 = const()[name = tensor<string, []>("input_823_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> input_823_strides_0 = const()[name = tensor<string, []>("input_823_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_823_pad_0 = const()[name = tensor<string, []>("input_823_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_823_dilations_0 = const()[name = tensor<string, []>("input_823_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> input_823_groups_0 = const()[name = tensor<string, []>("input_823_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 512, 1]> encoder_layers_15_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_15_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1024, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(98761280))), scale = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99285632))), zero_point = tensor<int8, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6073344)))];
            tensor<fp16, [1024]> encoder_layers_15_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99287744)))];
            tensor<fp16, [1, 512, 125]> input_821_cast_fp16 = transpose(perm = input_821_perm_0, x = x_347_cast_fp16)[name = tensor<string, []>("transpose_110")];
            tensor<fp16, [1, 1024, 125]> input_823_cast_fp16 = conv(bias = encoder_layers_15_conv_pointwise_conv1_bias_to_fp16, dilations = input_823_dilations_0, groups = input_823_groups_0, pad = input_823_pad_0, pad_type = input_823_pad_type_0, strides = input_823_strides_0, weight = encoder_layers_15_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_821_cast_fp16)[name = tensor<string, []>("input_823_cast_fp16")];
            tensor<int32, []> x_349_split_num_splits_0 = const()[name = tensor<string, []>("x_349_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> x_349_split_axis_0 = const()[name = tensor<string, []>("x_349_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 512, 125]> x_349_split_cast_fp16_0, tensor<fp16, [1, 512, 125]> x_349_split_cast_fp16_1 = split(axis = x_349_split_axis_0, num_splits = x_349_split_num_splits_0, x = input_823_cast_fp16)[name = tensor<string, []>("x_349_split_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_349_split_1_sigmoid_cast_fp16 = sigmoid(x = x_349_split_cast_fp16_1)[name = tensor<string, []>("x_349_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_349_cast_fp16 = mul(x = x_349_split_cast_fp16_0, y = x_349_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_349_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_825_cast_fp16 = select(a = var_7_to_fp16, b = x_349_cast_fp16, cond = var_449)[name = tensor<string, []>("input_825_cast_fp16")];
            tensor<int32, [6]> input_827_pad_0 = const()[name = tensor<string, []>("input_827_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            tensor<string, []> input_827_mode_0 = const()[name = tensor<string, []>("input_827_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_223_to_fp16 = const()[name = tensor<string, []>("const_223_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 512, 133]> input_827_cast_fp16 = pad(constant_val = const_223_to_fp16, mode = input_827_mode_0, pad = input_827_pad_0, x = input_825_cast_fp16)[name = tensor<string, []>("input_827_cast_fp16")];
            tensor<string, []> input_829_pad_type_0 = const()[name = tensor<string, []>("input_829_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, []> input_829_groups_0 = const()[name = tensor<string, []>("input_829_groups_0"), val = tensor<int32, []>(512)];
            tensor<int32, [1]> input_829_strides_0 = const()[name = tensor<string, []>("input_829_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_829_pad_0 = const()[name = tensor<string, []>("input_829_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_829_dilations_0 = const()[name = tensor<string, []>("input_829_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 9]> const_267_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("const_267_to_fp16_quantized"), quantized_data = tensor<int8, [512, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99289856))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99294528))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> const_268_to_fp16 = const()[name = tensor<string, []>("const_268_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99295616)))];
            tensor<fp16, [1, 512, 125]> input_831_cast_fp16 = conv(bias = const_268_to_fp16, dilations = input_829_dilations_0, groups = input_829_groups_0, pad = input_829_pad_0, pad_type = input_829_pad_type_0, strides = input_829_strides_0, weight = const_267_to_fp16_quantized, x = input_827_cast_fp16)[name = tensor<string, []>("input_831_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_833_cast_fp16 = silu(x = input_831_cast_fp16)[name = tensor<string, []>("input_833_cast_fp16")];
            tensor<string, []> x_351_pad_type_0 = const()[name = tensor<string, []>("x_351_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> x_351_strides_0 = const()[name = tensor<string, []>("x_351_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_351_pad_0 = const()[name = tensor<string, []>("x_351_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_351_dilations_0 = const()[name = tensor<string, []>("x_351_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> x_351_groups_0 = const()[name = tensor<string, []>("x_351_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [512, 512, 1]> encoder_layers_15_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_15_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99296704))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99558912))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_15_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99560000)))];
            tensor<fp16, [1, 512, 125]> x_351_cast_fp16 = conv(bias = encoder_layers_15_conv_pointwise_conv2_bias_to_fp16, dilations = x_351_dilations_0, groups = x_351_groups_0, pad = x_351_pad_0, pad_type = x_351_pad_type_0, strides = x_351_strides_0, weight = encoder_layers_15_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_833_cast_fp16)[name = tensor<string, []>("x_351_cast_fp16")];
            tensor<int32, [3]> input_835_perm_0 = const()[name = tensor<string, []>("input_835_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 125, 512]> input_835_cast_fp16 = transpose(perm = input_835_perm_0, x = x_351_cast_fp16)[name = tensor<string, []>("transpose_109")];
            tensor<fp16, [1, 125, 512]> input_837_cast_fp16 = add(x = input_819_cast_fp16, y = input_835_cast_fp16)[name = tensor<string, []>("input_837_cast_fp16")];
            tensor<int32, [1]> input_839_axes_0 = const()[name = tensor<string, []>("input_839_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_15_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99561088)))];
            tensor<fp16, [512]> encoder_layers_15_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99562176)))];
            tensor<fp16, [1, 125, 512]> input_839_cast_fp16 = layer_norm(axes = input_839_axes_0, beta = encoder_layers_15_norm_feed_forward2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_15_norm_feed_forward2_weight_to_fp16, x = input_837_cast_fp16)[name = tensor<string, []>("input_839_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_15_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_15_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(99563264))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100611904))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_15_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100616064)))];
            tensor<fp16, [1, 125, 2048]> linear_143_cast_fp16 = linear(bias = encoder_layers_15_feed_forward2_linear1_bias_to_fp16, weight = encoder_layers_15_feed_forward2_linear1_weight_to_fp16_quantized, x = input_839_cast_fp16)[name = tensor<string, []>("linear_143_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_843_cast_fp16 = silu(x = linear_143_cast_fp16)[name = tensor<string, []>("input_843_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_15_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_15_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(100620224))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101668864))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_15_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101669952)))];
            tensor<fp16, [1, 125, 512]> linear_144_cast_fp16 = linear(bias = encoder_layers_15_feed_forward2_linear2_bias_to_fp16, weight = encoder_layers_15_feed_forward2_linear2_weight_to_fp16_quantized, x = input_843_cast_fp16)[name = tensor<string, []>("linear_144_cast_fp16")];
            tensor<fp16, []> var_3101_to_fp16 = const()[name = tensor<string, []>("op_3101_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_3102_cast_fp16 = mul(x = linear_144_cast_fp16, y = var_3101_to_fp16)[name = tensor<string, []>("op_3102_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_849_cast_fp16 = add(x = input_837_cast_fp16, y = var_3102_cast_fp16)[name = tensor<string, []>("input_849_cast_fp16")];
            tensor<int32, [1]> input_851_axes_0 = const()[name = tensor<string, []>("input_851_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_15_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_norm_out_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101671040)))];
            tensor<fp16, [512]> encoder_layers_15_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_15_norm_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101672128)))];
            tensor<fp16, [1, 125, 512]> input_851_cast_fp16 = layer_norm(axes = input_851_axes_0, beta = encoder_layers_15_norm_out_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_15_norm_out_weight_to_fp16, x = input_849_cast_fp16)[name = tensor<string, []>("input_851_cast_fp16")];
            tensor<int32, [1]> input_853_axes_0 = const()[name = tensor<string, []>("input_853_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_16_norm_feed_forward1_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_norm_feed_forward1_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101673216)))];
            tensor<fp16, [512]> encoder_layers_16_norm_feed_forward1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_norm_feed_forward1_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101674304)))];
            tensor<fp16, [1, 125, 512]> input_853_cast_fp16 = layer_norm(axes = input_853_axes_0, beta = encoder_layers_16_norm_feed_forward1_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_16_norm_feed_forward1_weight_to_fp16, x = input_851_cast_fp16)[name = tensor<string, []>("input_853_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_16_feed_forward1_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_16_feed_forward1_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(101675392))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102724032))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_16_feed_forward1_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_feed_forward1_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102728192)))];
            tensor<fp16, [1, 125, 2048]> linear_145_cast_fp16 = linear(bias = encoder_layers_16_feed_forward1_linear1_bias_to_fp16, weight = encoder_layers_16_feed_forward1_linear1_weight_to_fp16_quantized, x = input_853_cast_fp16)[name = tensor<string, []>("linear_145_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_857_cast_fp16 = silu(x = linear_145_cast_fp16)[name = tensor<string, []>("input_857_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_16_feed_forward1_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_16_feed_forward1_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(102732352))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103780992))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_16_feed_forward1_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_feed_forward1_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103782080)))];
            tensor<fp16, [1, 125, 512]> linear_146_cast_fp16 = linear(bias = encoder_layers_16_feed_forward1_linear2_bias_to_fp16, weight = encoder_layers_16_feed_forward1_linear2_weight_to_fp16_quantized, x = input_857_cast_fp16)[name = tensor<string, []>("linear_146_cast_fp16")];
            tensor<fp16, []> var_3132_to_fp16 = const()[name = tensor<string, []>("op_3132_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_3133_cast_fp16 = mul(x = linear_146_cast_fp16, y = var_3132_to_fp16)[name = tensor<string, []>("op_3133_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_863_cast_fp16 = add(x = input_851_cast_fp16, y = var_3133_cast_fp16)[name = tensor<string, []>("input_863_cast_fp16")];
            tensor<int32, [1]> query_axes_0 = const()[name = tensor<string, []>("query_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_16_norm_self_att_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_norm_self_att_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103783168)))];
            tensor<fp16, [512]> encoder_layers_16_norm_self_att_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_norm_self_att_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103784256)))];
            tensor<fp16, [1, 125, 512]> query_cast_fp16 = layer_norm(axes = query_axes_0, beta = encoder_layers_16_norm_self_att_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_16_norm_self_att_weight_to_fp16, x = input_863_cast_fp16)[name = tensor<string, []>("query_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_16_self_attn_linear_q_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_16_self_attn_linear_q_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(103785344))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104047552))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_16_self_attn_linear_q_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_self_attn_linear_q_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104048640)))];
            tensor<fp16, [1, 125, 512]> linear_147_cast_fp16 = linear(bias = encoder_layers_16_self_attn_linear_q_bias_to_fp16, weight = encoder_layers_16_self_attn_linear_q_weight_to_fp16_quantized, x = query_cast_fp16)[name = tensor<string, []>("linear_147_cast_fp16")];
            tensor<int32, [4]> var_3150 = const()[name = tensor<string, []>("op_3150"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> q_97_cast_fp16 = reshape(shape = var_3150, x = linear_147_cast_fp16)[name = tensor<string, []>("q_97_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_16_self_attn_linear_k_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_16_self_attn_linear_k_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104049728))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104311936))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_16_self_attn_linear_k_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_self_attn_linear_k_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104313024)))];
            tensor<fp16, [1, 125, 512]> linear_148_cast_fp16 = linear(bias = encoder_layers_16_self_attn_linear_k_bias_to_fp16, weight = encoder_layers_16_self_attn_linear_k_weight_to_fp16_quantized, x = query_cast_fp16)[name = tensor<string, []>("linear_148_cast_fp16")];
            tensor<int32, [4]> var_3155 = const()[name = tensor<string, []>("op_3155"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> k_65_cast_fp16 = reshape(shape = var_3155, x = linear_148_cast_fp16)[name = tensor<string, []>("k_65_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_16_self_attn_linear_v_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_16_self_attn_linear_v_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104314112))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104576320))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_16_self_attn_linear_v_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_self_attn_linear_v_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104577408)))];
            tensor<fp16, [1, 125, 512]> linear_149_cast_fp16 = linear(bias = encoder_layers_16_self_attn_linear_v_bias_to_fp16, weight = encoder_layers_16_self_attn_linear_v_weight_to_fp16_quantized, x = query_cast_fp16)[name = tensor<string, []>("linear_149_cast_fp16")];
            tensor<int32, [4]> var_3160 = const()[name = tensor<string, []>("op_3160"), val = tensor<int32, [4]>([1, -1, 8, 64])];
            tensor<fp16, [1, 125, 8, 64]> v_cast_fp16 = reshape(shape = var_3160, x = linear_149_cast_fp16)[name = tensor<string, []>("v_cast_fp16")];
            tensor<int32, [4]> value_perm_0 = const()[name = tensor<string, []>("value_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<fp16, [8, 64]> encoder_layers_16_self_attn_pos_bias_u_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_self_attn_pos_bias_u_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104578496)))];
            tensor<fp16, [1, 125, 8, 64]> var_3172_cast_fp16 = add(x = q_97_cast_fp16, y = encoder_layers_16_self_attn_pos_bias_u_to_fp16)[name = tensor<string, []>("op_3172_cast_fp16")];
            tensor<fp16, [8, 64]> encoder_layers_16_self_attn_pos_bias_v_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_self_attn_pos_bias_v_to_fp16"), val = tensor<fp16, [8, 64]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104579584)))];
            tensor<fp16, [1, 125, 8, 64]> var_3174_cast_fp16 = add(x = q_97_cast_fp16, y = encoder_layers_16_self_attn_pos_bias_v_to_fp16)[name = tensor<string, []>("op_3174_cast_fp16")];
            tensor<int32, [4]> q_with_bias_v_perm_0 = const()[name = tensor<string, []>("q_with_bias_v_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<bool, []> x_359_transpose_x_0 = const()[name = tensor<string, []>("x_359_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_359_transpose_y_0 = const()[name = tensor<string, []>("x_359_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 64, 249]> op_3176_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(3), name = tensor<string, []>("op_3176_to_fp16_quantized"), quantized_data = tensor<int8, [1, 8, 64, 249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104580672))), scale = tensor<fp16, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104708224))), zero_point = tensor<int8, [249]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(5281536)))];
            tensor<fp16, [1, 8, 125, 64]> q_with_bias_v_cast_fp16 = transpose(perm = q_with_bias_v_perm_0, x = var_3174_cast_fp16)[name = tensor<string, []>("transpose_108")];
            tensor<fp16, [1, 8, 125, 249]> x_359_cast_fp16 = matmul(transpose_x = x_359_transpose_x_0, transpose_y = x_359_transpose_y_0, x = q_with_bias_v_cast_fp16, y = op_3176_to_fp16_quantized)[name = tensor<string, []>("x_359_cast_fp16")];
            tensor<int32, [8]> x_361_pad_0 = const()[name = tensor<string, []>("x_361_pad_0"), val = tensor<int32, [8]>([0, 0, 0, 0, 0, 0, 1, 0])];
            tensor<string, []> x_361_mode_0 = const()[name = tensor<string, []>("x_361_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_230_to_fp16 = const()[name = tensor<string, []>("const_230_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 8, 125, 250]> x_361_cast_fp16 = pad(constant_val = const_230_to_fp16, mode = x_361_mode_0, pad = x_361_pad_0, x = x_359_cast_fp16)[name = tensor<string, []>("x_361_cast_fp16")];
            tensor<int32, [4]> var_3184 = const()[name = tensor<string, []>("op_3184"), val = tensor<int32, [4]>([1, 8, -1, 125])];
            tensor<fp16, [1, 8, 250, 125]> x_363_cast_fp16 = reshape(shape = var_3184, x = x_361_cast_fp16)[name = tensor<string, []>("x_363_cast_fp16")];
            tensor<int32, [4]> var_3188_begin_0 = const()[name = tensor<string, []>("op_3188_begin_0"), val = tensor<int32, [4]>([0, 0, 1, 0])];
            tensor<int32, [4]> var_3188_end_0 = const()[name = tensor<string, []>("op_3188_end_0"), val = tensor<int32, [4]>([1, 8, 250, 125])];
            tensor<bool, [4]> var_3188_end_mask_0 = const()[name = tensor<string, []>("op_3188_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 8, 249, 125]> var_3188_cast_fp16 = slice_by_index(begin = var_3188_begin_0, end = var_3188_end_0, end_mask = var_3188_end_mask_0, x = x_363_cast_fp16)[name = tensor<string, []>("op_3188_cast_fp16")];
            tensor<int32, [4]> var_3189 = const()[name = tensor<string, []>("op_3189"), val = tensor<int32, [4]>([1, 8, 125, 249])];
            tensor<fp16, [1, 8, 125, 249]> matrix_bd_65_cast_fp16 = reshape(shape = var_3189, x = var_3188_cast_fp16)[name = tensor<string, []>("matrix_bd_65_cast_fp16")];
            tensor<bool, []> matrix_ac_transpose_x_0 = const()[name = tensor<string, []>("matrix_ac_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> matrix_ac_transpose_y_0 = const()[name = tensor<string, []>("matrix_ac_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<int32, [4]> transpose_100_perm_0 = const()[name = tensor<string, []>("transpose_100_perm_0"), val = tensor<int32, [4]>([0, 2, -3, -1])];
            tensor<int32, [4]> transpose_101_perm_0 = const()[name = tensor<string, []>("transpose_101_perm_0"), val = tensor<int32, [4]>([0, 2, -1, -3])];
            tensor<fp16, [1, 8, 64, 125]> transpose_101 = transpose(perm = transpose_101_perm_0, x = k_65_cast_fp16)[name = tensor<string, []>("transpose_106")];
            tensor<fp16, [1, 8, 125, 64]> transpose_100 = transpose(perm = transpose_100_perm_0, x = var_3172_cast_fp16)[name = tensor<string, []>("transpose_107")];
            tensor<fp16, [1, 8, 125, 125]> matrix_ac_cast_fp16 = matmul(transpose_x = matrix_ac_transpose_x_0, transpose_y = matrix_ac_transpose_y_0, x = transpose_100, y = transpose_101)[name = tensor<string, []>("matrix_ac_cast_fp16")];
            tensor<int32, [4]> matrix_bd_begin_0 = const()[name = tensor<string, []>("matrix_bd_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> matrix_bd_end_0 = const()[name = tensor<string, []>("matrix_bd_end_0"), val = tensor<int32, [4]>([1, 8, 125, 125])];
            tensor<bool, [4]> matrix_bd_end_mask_0 = const()[name = tensor<string, []>("matrix_bd_end_mask_0"), val = tensor<bool, [4]>([true, true, true, false])];
            tensor<fp16, [1, 8, 125, 125]> matrix_bd_cast_fp16 = slice_by_index(begin = matrix_bd_begin_0, end = matrix_bd_end_0, end_mask = matrix_bd_end_mask_0, x = matrix_bd_65_cast_fp16)[name = tensor<string, []>("matrix_bd_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_3198_cast_fp16 = add(x = matrix_ac_cast_fp16, y = matrix_bd_cast_fp16)[name = tensor<string, []>("op_3198_cast_fp16")];
            tensor<fp16, []> _inversed_scores_65_y_0_to_fp16 = const()[name = tensor<string, []>("_inversed_scores_65_y_0_to_fp16"), val = tensor<fp16, []>(0x1p-3)];
            tensor<fp16, [1, 8, 125, 125]> _inversed_scores_65_cast_fp16 = mul(x = var_3198_cast_fp16, y = _inversed_scores_65_y_0_to_fp16)[name = tensor<string, []>("_inversed_scores_65_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> scores_cast_fp16 = select(a = var_8_to_fp16, b = _inversed_scores_65_cast_fp16, cond = mask_11)[name = tensor<string, []>("scores_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> var_3204_cast_fp16 = softmax(axis = var_19, x = scores_cast_fp16)[name = tensor<string, []>("op_3204_cast_fp16")];
            tensor<fp16, [1, 8, 125, 125]> input_865_cast_fp16 = select(a = var_7_to_fp16, b = var_3204_cast_fp16, cond = mask_11)[name = tensor<string, []>("input_865_cast_fp16")];
            tensor<bool, []> x_365_transpose_x_0 = const()[name = tensor<string, []>("x_365_transpose_x_0"), val = tensor<bool, []>(false)];
            tensor<bool, []> x_365_transpose_y_0 = const()[name = tensor<string, []>("x_365_transpose_y_0"), val = tensor<bool, []>(false)];
            tensor<fp16, [1, 8, 125, 64]> value_cast_fp16 = transpose(perm = value_perm_0, x = v_cast_fp16)[name = tensor<string, []>("transpose_105")];
            tensor<fp16, [1, 8, 125, 64]> x_365_cast_fp16 = matmul(transpose_x = x_365_transpose_x_0, transpose_y = x_365_transpose_y_0, x = input_865_cast_fp16, y = value_cast_fp16)[name = tensor<string, []>("x_365_cast_fp16")];
            tensor<int32, [4]> var_3208_perm_0 = const()[name = tensor<string, []>("op_3208_perm_0"), val = tensor<int32, [4]>([0, 2, 1, 3])];
            tensor<int32, [3]> var_3209 = const()[name = tensor<string, []>("op_3209"), val = tensor<int32, [3]>([1, -1, 512])];
            tensor<fp16, [1, 125, 8, 64]> var_3208_cast_fp16 = transpose(perm = var_3208_perm_0, x = x_365_cast_fp16)[name = tensor<string, []>("transpose_104")];
            tensor<fp16, [1, 125, 512]> input_867_cast_fp16 = reshape(shape = var_3209, x = var_3208_cast_fp16)[name = tensor<string, []>("input_867_cast_fp16")];
            tensor<fp16, [512, 512]> encoder_layers_16_self_attn_linear_out_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_16_self_attn_linear_out_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104708800))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104971008))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_16_self_attn_linear_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_self_attn_linear_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104972096)))];
            tensor<fp16, [1, 125, 512]> linear_151_cast_fp16 = linear(bias = encoder_layers_16_self_attn_linear_out_bias_to_fp16, weight = encoder_layers_16_self_attn_linear_out_weight_to_fp16_quantized, x = input_867_cast_fp16)[name = tensor<string, []>("linear_151_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_871_cast_fp16 = add(x = input_863_cast_fp16, y = linear_151_cast_fp16)[name = tensor<string, []>("input_871_cast_fp16")];
            tensor<int32, [1]> x_369_axes_0 = const()[name = tensor<string, []>("x_369_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_16_norm_conv_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_norm_conv_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104973184)))];
            tensor<fp16, [512]> encoder_layers_16_norm_conv_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_norm_conv_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104974272)))];
            tensor<fp16, [1, 125, 512]> x_369_cast_fp16 = layer_norm(axes = x_369_axes_0, beta = encoder_layers_16_norm_conv_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_16_norm_conv_weight_to_fp16, x = input_871_cast_fp16)[name = tensor<string, []>("x_369_cast_fp16")];
            tensor<int32, [3]> input_873_perm_0 = const()[name = tensor<string, []>("input_873_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<string, []> input_875_pad_type_0 = const()[name = tensor<string, []>("input_875_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> input_875_strides_0 = const()[name = tensor<string, []>("input_875_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_875_pad_0 = const()[name = tensor<string, []>("input_875_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_875_dilations_0 = const()[name = tensor<string, []>("input_875_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> input_875_groups_0 = const()[name = tensor<string, []>("input_875_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1024, 512, 1]> encoder_layers_16_conv_pointwise_conv1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_16_conv_pointwise_conv1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1024, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(104975360))), scale = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105499712))), zero_point = tensor<int8, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6073344)))];
            tensor<fp16, [1024]> encoder_layers_16_conv_pointwise_conv1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_conv_pointwise_conv1_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105501824)))];
            tensor<fp16, [1, 512, 125]> input_873_cast_fp16 = transpose(perm = input_873_perm_0, x = x_369_cast_fp16)[name = tensor<string, []>("transpose_103")];
            tensor<fp16, [1, 1024, 125]> input_875_cast_fp16 = conv(bias = encoder_layers_16_conv_pointwise_conv1_bias_to_fp16, dilations = input_875_dilations_0, groups = input_875_groups_0, pad = input_875_pad_0, pad_type = input_875_pad_type_0, strides = input_875_strides_0, weight = encoder_layers_16_conv_pointwise_conv1_weight_to_fp16_quantized, x = input_873_cast_fp16)[name = tensor<string, []>("input_875_cast_fp16")];
            tensor<int32, []> x_371_split_num_splits_0 = const()[name = tensor<string, []>("x_371_split_num_splits_0"), val = tensor<int32, []>(2)];
            tensor<int32, []> x_371_split_axis_0 = const()[name = tensor<string, []>("x_371_split_axis_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [1, 512, 125]> x_371_split_cast_fp16_0, tensor<fp16, [1, 512, 125]> x_371_split_cast_fp16_1 = split(axis = x_371_split_axis_0, num_splits = x_371_split_num_splits_0, x = input_875_cast_fp16)[name = tensor<string, []>("x_371_split_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_371_split_1_sigmoid_cast_fp16 = sigmoid(x = x_371_split_cast_fp16_1)[name = tensor<string, []>("x_371_split_1_sigmoid_cast_fp16")];
            tensor<fp16, [1, 512, 125]> x_371_cast_fp16 = mul(x = x_371_split_cast_fp16_0, y = x_371_split_1_sigmoid_cast_fp16)[name = tensor<string, []>("x_371_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_877_cast_fp16 = select(a = var_7_to_fp16, b = x_371_cast_fp16, cond = var_449)[name = tensor<string, []>("input_877_cast_fp16")];
            tensor<int32, [6]> input_879_pad_0 = const()[name = tensor<string, []>("input_879_pad_0"), val = tensor<int32, [6]>([0, 0, 0, 0, 4, 4])];
            tensor<string, []> input_879_mode_0 = const()[name = tensor<string, []>("input_879_mode_0"), val = tensor<string, []>("constant")];
            tensor<fp16, []> const_233_to_fp16 = const()[name = tensor<string, []>("const_233_to_fp16"), val = tensor<fp16, []>(0x0p+0)];
            tensor<fp16, [1, 512, 133]> input_879_cast_fp16 = pad(constant_val = const_233_to_fp16, mode = input_879_mode_0, pad = input_879_pad_0, x = input_877_cast_fp16)[name = tensor<string, []>("input_879_cast_fp16")];
            tensor<string, []> input_881_pad_type_0 = const()[name = tensor<string, []>("input_881_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, []> input_881_groups_0 = const()[name = tensor<string, []>("input_881_groups_0"), val = tensor<int32, []>(512)];
            tensor<int32, [1]> input_881_strides_0 = const()[name = tensor<string, []>("input_881_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> input_881_pad_0 = const()[name = tensor<string, []>("input_881_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> input_881_dilations_0 = const()[name = tensor<string, []>("input_881_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<fp16, [512, 1, 9]> const_269_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("const_269_to_fp16_quantized"), quantized_data = tensor<int8, [512, 1, 9]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105503936))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105508608))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> const_270_to_fp16 = const()[name = tensor<string, []>("const_270_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105509696)))];
            tensor<fp16, [1, 512, 125]> input_883_cast_fp16 = conv(bias = const_270_to_fp16, dilations = input_881_dilations_0, groups = input_881_groups_0, pad = input_881_pad_0, pad_type = input_881_pad_type_0, strides = input_881_strides_0, weight = const_269_to_fp16_quantized, x = input_879_cast_fp16)[name = tensor<string, []>("input_883_cast_fp16")];
            tensor<fp16, [1, 512, 125]> input_885_cast_fp16 = silu(x = input_883_cast_fp16)[name = tensor<string, []>("input_885_cast_fp16")];
            tensor<string, []> x_373_pad_type_0 = const()[name = tensor<string, []>("x_373_pad_type_0"), val = tensor<string, []>("valid")];
            tensor<int32, [1]> x_373_strides_0 = const()[name = tensor<string, []>("x_373_strides_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, [2]> x_373_pad_0 = const()[name = tensor<string, []>("x_373_pad_0"), val = tensor<int32, [2]>([0, 0])];
            tensor<int32, [1]> x_373_dilations_0 = const()[name = tensor<string, []>("x_373_dilations_0"), val = tensor<int32, [1]>([1])];
            tensor<int32, []> x_373_groups_0 = const()[name = tensor<string, []>("x_373_groups_0"), val = tensor<int32, []>(1)];
            tensor<fp16, [512, 512, 1]> encoder_layers_16_conv_pointwise_conv2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_16_conv_pointwise_conv2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 512, 1]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105510784))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105772992))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_16_conv_pointwise_conv2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_conv_pointwise_conv2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105774080)))];
            tensor<fp16, [1, 512, 125]> x_373_cast_fp16 = conv(bias = encoder_layers_16_conv_pointwise_conv2_bias_to_fp16, dilations = x_373_dilations_0, groups = x_373_groups_0, pad = x_373_pad_0, pad_type = x_373_pad_type_0, strides = x_373_strides_0, weight = encoder_layers_16_conv_pointwise_conv2_weight_to_fp16_quantized, x = input_885_cast_fp16)[name = tensor<string, []>("x_373_cast_fp16")];
            tensor<int32, [3]> input_887_perm_0 = const()[name = tensor<string, []>("input_887_perm_0"), val = tensor<int32, [3]>([0, 2, 1])];
            tensor<fp16, [1, 125, 512]> input_887_cast_fp16 = transpose(perm = input_887_perm_0, x = x_373_cast_fp16)[name = tensor<string, []>("transpose_102")];
            tensor<fp16, [1, 125, 512]> input_889_cast_fp16 = add(x = input_871_cast_fp16, y = input_887_cast_fp16)[name = tensor<string, []>("input_889_cast_fp16")];
            tensor<int32, [1]> input_891_axes_0 = const()[name = tensor<string, []>("input_891_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_16_norm_feed_forward2_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_norm_feed_forward2_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105775168)))];
            tensor<fp16, [512]> encoder_layers_16_norm_feed_forward2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_norm_feed_forward2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105776256)))];
            tensor<fp16, [1, 125, 512]> input_891_cast_fp16 = layer_norm(axes = input_891_axes_0, beta = encoder_layers_16_norm_feed_forward2_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_16_norm_feed_forward2_weight_to_fp16, x = input_889_cast_fp16)[name = tensor<string, []>("input_891_cast_fp16")];
            tensor<fp16, [2048, 512]> encoder_layers_16_feed_forward2_linear1_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_16_feed_forward2_linear1_weight_to_fp16_quantized"), quantized_data = tensor<int8, [2048, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(105777344))), scale = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106825984))), zero_point = tensor<int8, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(3295232)))];
            tensor<fp16, [2048]> encoder_layers_16_feed_forward2_linear1_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_feed_forward2_linear1_bias_to_fp16"), val = tensor<fp16, [2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106830144)))];
            tensor<fp16, [1, 125, 2048]> linear_152_cast_fp16 = linear(bias = encoder_layers_16_feed_forward2_linear1_bias_to_fp16, weight = encoder_layers_16_feed_forward2_linear1_weight_to_fp16_quantized, x = input_891_cast_fp16)[name = tensor<string, []>("linear_152_cast_fp16")];
            tensor<fp16, [1, 125, 2048]> input_895_cast_fp16 = silu(x = linear_152_cast_fp16)[name = tensor<string, []>("input_895_cast_fp16")];
            tensor<fp16, [512, 2048]> encoder_layers_16_feed_forward2_linear2_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("encoder_layers_16_feed_forward2_linear2_weight_to_fp16_quantized"), quantized_data = tensor<int8, [512, 2048]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(106834304))), scale = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107882944))), zero_point = tensor<int8, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(2241664)))];
            tensor<fp16, [512]> encoder_layers_16_feed_forward2_linear2_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_feed_forward2_linear2_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107884032)))];
            tensor<fp16, [1, 125, 512]> linear_153_cast_fp16 = linear(bias = encoder_layers_16_feed_forward2_linear2_bias_to_fp16, weight = encoder_layers_16_feed_forward2_linear2_weight_to_fp16_quantized, x = input_895_cast_fp16)[name = tensor<string, []>("linear_153_cast_fp16")];
            tensor<fp16, []> var_3275_to_fp16 = const()[name = tensor<string, []>("op_3275_to_fp16"), val = tensor<fp16, []>(0x1p-1)];
            tensor<fp16, [1, 125, 512]> var_3276_cast_fp16 = mul(x = linear_153_cast_fp16, y = var_3275_to_fp16)[name = tensor<string, []>("op_3276_cast_fp16")];
            tensor<fp16, [1, 125, 512]> input_901_cast_fp16 = add(x = input_889_cast_fp16, y = var_3276_cast_fp16)[name = tensor<string, []>("input_901_cast_fp16")];
            tensor<int32, [1]> audio_signal_axes_0 = const()[name = tensor<string, []>("audio_signal_axes_0"), val = tensor<int32, [1]>([-1])];
            tensor<fp16, [512]> encoder_layers_16_norm_out_weight_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_norm_out_weight_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107885120)))];
            tensor<fp16, [512]> encoder_layers_16_norm_out_bias_to_fp16 = const()[name = tensor<string, []>("encoder_layers_16_norm_out_bias_to_fp16"), val = tensor<fp16, [512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107886208)))];
            tensor<fp16, [1, 125, 512]> audio_signal_cast_fp16 = layer_norm(axes = audio_signal_axes_0, beta = encoder_layers_16_norm_out_bias_to_fp16, epsilon = var_5_to_fp16, gamma = encoder_layers_16_norm_out_weight_to_fp16, x = input_901_cast_fp16)[name = tensor<string, []>("audio_signal_cast_fp16")];
            tensor<fp16, [1024, 512]> proj_weight_to_fp16_quantized = constexpr_affine_dequantize()[axis = tensor<int32, []>(0), name = tensor<string, []>("proj_weight_to_fp16_quantized"), quantized_data = tensor<int8, [1024, 512]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(107887296))), scale = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108411648))), zero_point = tensor<int8, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(6073344)))];
            tensor<fp16, [1024]> proj_bias_to_fp16 = const()[name = tensor<string, []>("proj_bias_to_fp16"), val = tensor<fp16, [1024]>(BLOBFILE(path = tensor<string, []>("@model_path/weights/weight.bin"), offset = tensor<uint64, []>(108413760)))];
            tensor<fp16, [1, 125, 1024]> encoder_embeddings = linear(bias = proj_bias_to_fp16, weight = proj_weight_to_fp16_quantized, x = audio_signal_cast_fp16)[name = tensor<string, []>("linear_154_cast_fp16")];
            tensor<string, []> cast_159_dtype_0 = const()[name = tensor<string, []>("cast_159_dtype_0"), val = tensor<string, []>("fp16")];
            tensor<fp16, [1, 125]> encoder_mask = cast(dtype = cast_159_dtype_0, x = time_mask)[name = tensor<string, []>("cast_0")];
        } -> (encoder_embeddings, encoder_mask);
}