Hugging Face's logo

SolsticeAI
/
voz

Automatic Speech Recognition
LiteRT
ONNX
speech
speech-recognition
transcription
word-timestamps
on-device
core-ml
apple-neural-engine
multilingual
Model card Files
xet
 Community
voz
File size: 99,410 Bytes
e69441b
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
program(1.3)
[buildInfo = dict<string, string>({{"coremlc-component-MIL", "3600.16.1"}, {"coremlc-version", "3600.25.1"}})]
{
    func decoder_1<ios18>(tensor<fp16, [1, 1280, 1, 1]> c_in, tensor<fp16, [1, 640, 1, 1]> embed, tensor<fp16, [1, 640, 1, 8]> enc_step, tensor<fp16, [1, 1280, 1, 1]> h_in) {
            int32 var_14 = const()[name = string("op_14"), val = int32(1)];
            tensor<int32, [4]> input_1_begin_0 = const()[name = string("input_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> input_1_end_0 = const()[name = string("input_1_end_0"), val = tensor<int32, [4]>([1, 640, 1, 1])];
            tensor<bool, [4]> input_1_end_mask_0 = const()[name = string("input_1_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 640, 1, 1]> input_1_cast_fp16 = slice_by_index(begin = input_1_begin_0, end = input_1_end_0, end_mask = input_1_end_mask_0, x = h_in)[name = string("input_1_cast_fp16")];
            tensor<int32, [4]> c_1_begin_0 = const()[name = string("c_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> c_1_end_0 = const()[name = string("c_1_end_0"), val = tensor<int32, [4]>([1, 640, 1, 1])];
            tensor<bool, [4]> c_1_end_mask_0 = const()[name = string("c_1_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 640, 1, 1]> c_1_cast_fp16 = slice_by_index(begin = c_1_begin_0, end = c_1_end_0, end_mask = c_1_end_mask_0, x = c_in)[name = string("c_1_cast_fp16")];
            string var_34_pad_type_0 = const()[name = string("op_34_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_34_strides_0 = const()[name = string("op_34_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_34_pad_0 = const()[name = string("op_34_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_34_dilations_0 = const()[name = string("op_34_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_34_groups_0 = const()[name = string("op_34_groups_0"), val = int32(1)];
            tensor<fp16, [2560, 640, 1, 1]> decoder_cells_0_ih_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2560, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64))), scale = tensor<fp16, [2560, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1638528))))[name = string("decoder_cells_0_ih_weight_to_fp16_quantized")];
            tensor<fp16, [2560]> decoder_cells_0_ih_bias_to_fp16 = const()[name = string("decoder_cells_0_ih_bias_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1643712)))];
            tensor<fp16, [1, 2560, 1, 1]> var_34_cast_fp16 = conv(bias = decoder_cells_0_ih_bias_to_fp16, dilations = var_34_dilations_0, groups = var_34_groups_0, pad = var_34_pad_0, pad_type = var_34_pad_type_0, strides = var_34_strides_0, weight = decoder_cells_0_ih_weight_to_fp16_quantized, x = embed)[name = string("op_34_cast_fp16")];
            string var_40_pad_type_0 = const()[name = string("op_40_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_40_strides_0 = const()[name = string("op_40_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_40_pad_0 = const()[name = string("op_40_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_40_dilations_0 = const()[name = string("op_40_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_40_groups_0 = const()[name = string("op_40_groups_0"), val = int32(1)];
            tensor<fp16, [2560, 640, 1, 1]> decoder_cells_0_hh_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2560, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1648896))), scale = tensor<fp16, [2560, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3287360))))[name = string("decoder_cells_0_hh_weight_to_fp16_quantized")];
            tensor<fp16, [1, 2560, 1, 1]> var_40_cast_fp16 = conv(dilations = var_40_dilations_0, groups = var_40_groups_0, pad = var_40_pad_0, pad_type = var_40_pad_type_0, strides = var_40_strides_0, weight = decoder_cells_0_hh_weight_to_fp16_quantized, x = input_1_cast_fp16)[name = string("op_40_cast_fp16")];
            tensor<fp16, [1, 2560, 1, 1]> gates_1_cast_fp16 = add(x = var_34_cast_fp16, y = var_40_cast_fp16)[name = string("gates_1_cast_fp16")];
            tensor<int32, [4]> var_43_begin_0 = const()[name = string("op_43_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_43_end_0 = const()[name = string("op_43_end_0"), val = tensor<int32, [4]>([1, 640, 1, 1])];
            tensor<bool, [4]> var_43_end_mask_0 = const()[name = string("op_43_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 640, 1, 1]> var_43_cast_fp16 = slice_by_index(begin = var_43_begin_0, end = var_43_end_0, end_mask = var_43_end_mask_0, x = gates_1_cast_fp16)[name = string("op_43_cast_fp16")];
            tensor<fp16, [1, 640, 1, 1]> i_1_cast_fp16 = sigmoid(x = var_43_cast_fp16)[name = string("i_1_cast_fp16")];
            tensor<int32, [4]> var_46_begin_0 = const()[name = string("op_46_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_46_end_0 = const()[name = string("op_46_end_0"), val = tensor<int32, [4]>([1, 1280, 1, 1])];
            tensor<bool, [4]> var_46_end_mask_0 = const()[name = string("op_46_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 640, 1, 1]> var_46_cast_fp16 = slice_by_index(begin = var_46_begin_0, end = var_46_end_0, end_mask = var_46_end_mask_0, x = gates_1_cast_fp16)[name = string("op_46_cast_fp16")];
            tensor<fp16, [1, 640, 1, 1]> f_1_cast_fp16 = sigmoid(x = var_46_cast_fp16)[name = string("f_1_cast_fp16")];
            tensor<int32, [4]> var_49_begin_0 = const()[name = string("op_49_begin_0"), val = tensor<int32, [4]>([0, 1280, 0, 0])];
            tensor<int32, [4]> var_49_end_0 = const()[name = string("op_49_end_0"), val = tensor<int32, [4]>([1, 1920, 1, 1])];
            tensor<bool, [4]> var_49_end_mask_0 = const()[name = string("op_49_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 640, 1, 1]> var_49_cast_fp16 = slice_by_index(begin = var_49_begin_0, end = var_49_end_0, end_mask = var_49_end_mask_0, x = gates_1_cast_fp16)[name = string("op_49_cast_fp16")];
            tensor<fp16, [1, 640, 1, 1]> g_1_cast_fp16 = tanh(x = var_49_cast_fp16)[name = string("g_1_cast_fp16")];
            tensor<int32, [4]> var_52_begin_0 = const()[name = string("op_52_begin_0"), val = tensor<int32, [4]>([0, 1920, 0, 0])];
            tensor<int32, [4]> var_52_end_0 = const()[name = string("op_52_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> var_52_end_mask_0 = const()[name = string("op_52_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 640, 1, 1]> var_52_cast_fp16 = slice_by_index(begin = var_52_begin_0, end = var_52_end_0, end_mask = var_52_end_mask_0, x = gates_1_cast_fp16)[name = string("op_52_cast_fp16")];
            tensor<fp16, [1, 640, 1, 1]> o_1_cast_fp16 = sigmoid(x = var_52_cast_fp16)[name = string("o_1_cast_fp16")];
            tensor<fp16, [1, 640, 1, 1]> var_54_cast_fp16 = mul(x = f_1_cast_fp16, y = c_1_cast_fp16)[name = string("op_54_cast_fp16")];
            tensor<fp16, [1, 640, 1, 1]> var_55_cast_fp16 = mul(x = i_1_cast_fp16, y = g_1_cast_fp16)[name = string("op_55_cast_fp16")];
            tensor<fp16, [1, 640, 1, 1]> c_new_1_cast_fp16 = add(x = var_54_cast_fp16, y = var_55_cast_fp16)[name = string("c_new_1_cast_fp16")];
            tensor<fp16, [1, 640, 1, 1]> var_57_cast_fp16 = tanh(x = c_new_1_cast_fp16)[name = string("op_57_cast_fp16")];
            tensor<fp16, [1, 640, 1, 1]> input_3_cast_fp16 = mul(x = o_1_cast_fp16, y = var_57_cast_fp16)[name = string("input_3_cast_fp16")];
            tensor<int32, [4]> input_5_begin_0 = const()[name = string("input_5_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> input_5_end_0 = const()[name = string("input_5_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> input_5_end_mask_0 = const()[name = string("input_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 640, 1, 1]> input_5_cast_fp16 = slice_by_index(begin = input_5_begin_0, end = input_5_end_0, end_mask = input_5_end_mask_0, x = h_in)[name = string("input_5_cast_fp16")];
            tensor<int32, [4]> c_begin_0 = const()[name = string("c_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> c_end_0 = const()[name = string("c_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> c_end_mask_0 = const()[name = string("c_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 640, 1, 1]> c_cast_fp16 = slice_by_index(begin = c_begin_0, end = c_end_0, end_mask = c_end_mask_0, x = c_in)[name = string("c_cast_fp16")];
            string var_74_pad_type_0 = const()[name = string("op_74_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_74_strides_0 = const()[name = string("op_74_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_74_pad_0 = const()[name = string("op_74_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_74_dilations_0 = const()[name = string("op_74_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_74_groups_0 = const()[name = string("op_74_groups_0"), val = int32(1)];
            tensor<fp16, [2560, 640, 1, 1]> decoder_cells_1_ih_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2560, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3292544))), scale = tensor<fp16, [2560, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4931008))))[name = string("decoder_cells_1_ih_weight_to_fp16_quantized")];
            tensor<fp16, [2560]> decoder_cells_1_ih_bias_to_fp16 = const()[name = string("decoder_cells_1_ih_bias_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4936192)))];
            tensor<fp16, [1, 2560, 1, 1]> var_74_cast_fp16 = conv(bias = decoder_cells_1_ih_bias_to_fp16, dilations = var_74_dilations_0, groups = var_74_groups_0, pad = var_74_pad_0, pad_type = var_74_pad_type_0, strides = var_74_strides_0, weight = decoder_cells_1_ih_weight_to_fp16_quantized, x = input_3_cast_fp16)[name = string("op_74_cast_fp16")];
            string var_80_pad_type_0 = const()[name = string("op_80_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_80_strides_0 = const()[name = string("op_80_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_80_pad_0 = const()[name = string("op_80_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_80_dilations_0 = const()[name = string("op_80_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_80_groups_0 = const()[name = string("op_80_groups_0"), val = int32(1)];
            tensor<fp16, [2560, 640, 1, 1]> decoder_cells_1_hh_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2560, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4941376))), scale = tensor<fp16, [2560, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6579840))))[name = string("decoder_cells_1_hh_weight_to_fp16_quantized")];
            tensor<fp16, [1, 2560, 1, 1]> var_80_cast_fp16 = conv(dilations = var_80_dilations_0, groups = var_80_groups_0, pad = var_80_pad_0, pad_type = var_80_pad_type_0, strides = var_80_strides_0, weight = decoder_cells_1_hh_weight_to_fp16_quantized, x = input_5_cast_fp16)[name = string("op_80_cast_fp16")];
            tensor<fp16, [1, 2560, 1, 1]> gates_cast_fp16 = add(x = var_74_cast_fp16, y = var_80_cast_fp16)[name = string("gates_cast_fp16")];
            tensor<int32, [4]> var_83_begin_0 = const()[name = string("op_83_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_83_end_0 = const()[name = string("op_83_end_0"), val = tensor<int32, [4]>([1, 640, 1, 1])];
            tensor<bool, [4]> var_83_end_mask_0 = const()[name = string("op_83_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 640, 1, 1]> var_83_cast_fp16 = slice_by_index(begin = var_83_begin_0, end = var_83_end_0, end_mask = var_83_end_mask_0, x = gates_cast_fp16)[name = string("op_83_cast_fp16")];
            tensor<fp16, [1, 640, 1, 1]> i_cast_fp16 = sigmoid(x = var_83_cast_fp16)[name = string("i_cast_fp16")];
            tensor<int32, [4]> var_86_begin_0 = const()[name = string("op_86_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_86_end_0 = const()[name = string("op_86_end_0"), val = tensor<int32, [4]>([1, 1280, 1, 1])];
            tensor<bool, [4]> var_86_end_mask_0 = const()[name = string("op_86_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 640, 1, 1]> var_86_cast_fp16 = slice_by_index(begin = var_86_begin_0, end = var_86_end_0, end_mask = var_86_end_mask_0, x = gates_cast_fp16)[name = string("op_86_cast_fp16")];
            tensor<fp16, [1, 640, 1, 1]> f_cast_fp16 = sigmoid(x = var_86_cast_fp16)[name = string("f_cast_fp16")];
            tensor<int32, [4]> var_89_begin_0 = const()[name = string("op_89_begin_0"), val = tensor<int32, [4]>([0, 1280, 0, 0])];
            tensor<int32, [4]> var_89_end_0 = const()[name = string("op_89_end_0"), val = tensor<int32, [4]>([1, 1920, 1, 1])];
            tensor<bool, [4]> var_89_end_mask_0 = const()[name = string("op_89_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [1, 640, 1, 1]> var_89_cast_fp16 = slice_by_index(begin = var_89_begin_0, end = var_89_end_0, end_mask = var_89_end_mask_0, x = gates_cast_fp16)[name = string("op_89_cast_fp16")];
            tensor<fp16, [1, 640, 1, 1]> g_cast_fp16 = tanh(x = var_89_cast_fp16)[name = string("g_cast_fp16")];
            tensor<int32, [4]> var_92_begin_0 = const()[name = string("op_92_begin_0"), val = tensor<int32, [4]>([0, 1920, 0, 0])];
            tensor<int32, [4]> var_92_end_0 = const()[name = string("op_92_end_0"), val = tensor<int32, [4]>([1, 1, 1, 1])];
            tensor<bool, [4]> var_92_end_mask_0 = const()[name = string("op_92_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [1, 640, 1, 1]> var_92_cast_fp16 = slice_by_index(begin = var_92_begin_0, end = var_92_end_0, end_mask = var_92_end_mask_0, x = gates_cast_fp16)[name = string("op_92_cast_fp16")];
            tensor<fp16, [1, 640, 1, 1]> o_cast_fp16 = sigmoid(x = var_92_cast_fp16)[name = string("o_cast_fp16")];
            tensor<fp16, [1, 640, 1, 1]> var_94_cast_fp16 = mul(x = f_cast_fp16, y = c_cast_fp16)[name = string("op_94_cast_fp16")];
            tensor<fp16, [1, 640, 1, 1]> var_95_cast_fp16 = mul(x = i_cast_fp16, y = g_cast_fp16)[name = string("op_95_cast_fp16")];
            tensor<fp16, [1, 640, 1, 1]> c_new_cast_fp16 = add(x = var_94_cast_fp16, y = var_95_cast_fp16)[name = string("c_new_cast_fp16")];
            tensor<fp16, [1, 640, 1, 1]> var_97_cast_fp16 = tanh(x = c_new_cast_fp16)[name = string("op_97_cast_fp16")];
            tensor<fp16, [1, 640, 1, 1]> input_7_cast_fp16 = mul(x = o_cast_fp16, y = var_97_cast_fp16)[name = string("input_7_cast_fp16")];
            string pred_pad_type_0 = const()[name = string("pred_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> pred_strides_0 = const()[name = string("pred_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> pred_pad_0 = const()[name = string("pred_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> pred_dilations_0 = const()[name = string("pred_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 pred_groups_0 = const()[name = string("pred_groups_0"), val = int32(1)];
            tensor<fp16, [640, 640, 1, 1]> decoder_joint_pred_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [640, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6585024))), scale = tensor<fp16, [640, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6994688))))[name = string("decoder_joint_pred_weight_to_fp16_quantized")];
            tensor<fp16, [640]> decoder_joint_pred_bias_to_fp16 = const()[name = string("decoder_joint_pred_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6996032)))];
            tensor<fp16, [1, 640, 1, 1]> pred_cast_fp16 = conv(bias = decoder_joint_pred_bias_to_fp16, dilations = pred_dilations_0, groups = pred_groups_0, pad = pred_pad_0, pad_type = pred_pad_type_0, strides = pred_strides_0, weight = decoder_joint_pred_weight_to_fp16_quantized, x = input_7_cast_fp16)[name = string("pred_cast_fp16")];
            bool var_110_interleave_0 = const()[name = string("op_110_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 1280, 1, 1]> h_out = concat(axis = var_14, interleave = var_110_interleave_0, values = (input_3_cast_fp16, input_7_cast_fp16))[name = string("op_110_cast_fp16")];
            bool var_112_interleave_0 = const()[name = string("op_112_interleave_0"), val = bool(false)];
            tensor<fp16, [1, 1280, 1, 1]> c_out = concat(axis = var_14, interleave = var_112_interleave_0, values = (c_new_1_cast_fp16, c_new_cast_fp16))[name = string("op_112_cast_fp16")];
            tensor<fp16, [1, 640, 1, 8]> var_122_cast_fp16 = add(x = enc_step, y = pred_cast_fp16)[name = string("op_122_cast_fp16")];
            tensor<fp16, [1, 640, 1, 8]> input_cast_fp16 = relu(x = var_122_cast_fp16)[name = string("input_cast_fp16")];
            string x_pad_type_0 = const()[name = string("x_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> x_strides_0 = const()[name = string("x_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_pad_0 = const()[name = string("x_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_dilations_0 = const()[name = string("x_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 x_groups_0 = const()[name = string("x_groups_0"), val = int32(1)];
            tensor<fp16, [8198, 640, 1, 1]> joint_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [8198, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6997376))), scale = tensor<fp16, [8198, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12244160))))[name = string("joint_out_weight_to_fp16_quantized")];
            tensor<fp16, [8198]> joint_out_bias_to_fp16 = const()[name = string("joint_out_bias_to_fp16"), val = tensor<fp16, [8198]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12260672)))];
            tensor<fp16, [1, 8198, 1, 8]> logits = conv(bias = joint_out_bias_to_fp16, dilations = x_dilations_0, groups = x_groups_0, pad = x_pad_0, pad_type = x_pad_type_0, strides = x_strides_0, weight = joint_out_weight_to_fp16_quantized, x = input_cast_fp16)[name = string("x_cast_fp16")];
        } -> (logits, h_out, c_out);
    func decoder_2<ios18>(tensor<fp16, [2, 1280, 1, 1]> c_in, tensor<fp16, [2, 640, 1, 1]> embed, tensor<fp16, [2, 640, 1, 8]> enc_step, tensor<fp16, [2, 1280, 1, 1]> h_in) {
            int32 var_14 = const()[name = string("op_14"), val = int32(1)];
            tensor<int32, [4]> input_1_begin_0 = const()[name = string("input_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> input_1_end_0 = const()[name = string("input_1_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1])];
            tensor<bool, [4]> input_1_end_mask_0 = const()[name = string("input_1_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 640, 1, 1]> input_1_cast_fp16 = slice_by_index(begin = input_1_begin_0, end = input_1_end_0, end_mask = input_1_end_mask_0, x = h_in)[name = string("input_1_cast_fp16")];
            tensor<int32, [4]> c_1_begin_0 = const()[name = string("c_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> c_1_end_0 = const()[name = string("c_1_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1])];
            tensor<bool, [4]> c_1_end_mask_0 = const()[name = string("c_1_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 640, 1, 1]> c_1_cast_fp16 = slice_by_index(begin = c_1_begin_0, end = c_1_end_0, end_mask = c_1_end_mask_0, x = c_in)[name = string("c_1_cast_fp16")];
            string var_34_pad_type_0 = const()[name = string("op_34_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_34_strides_0 = const()[name = string("op_34_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_34_pad_0 = const()[name = string("op_34_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_34_dilations_0 = const()[name = string("op_34_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_34_groups_0 = const()[name = string("op_34_groups_0"), val = int32(1)];
            tensor<fp16, [2560, 640, 1, 1]> decoder_cells_0_ih_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2560, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64))), scale = tensor<fp16, [2560, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1638528))))[name = string("decoder_cells_0_ih_weight_to_fp16_quantized")];
            tensor<fp16, [2560]> decoder_cells_0_ih_bias_to_fp16 = const()[name = string("decoder_cells_0_ih_bias_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1643712)))];
            tensor<fp16, [2, 2560, 1, 1]> var_34_cast_fp16 = conv(bias = decoder_cells_0_ih_bias_to_fp16, dilations = var_34_dilations_0, groups = var_34_groups_0, pad = var_34_pad_0, pad_type = var_34_pad_type_0, strides = var_34_strides_0, weight = decoder_cells_0_ih_weight_to_fp16_quantized, x = embed)[name = string("op_34_cast_fp16")];
            string var_40_pad_type_0 = const()[name = string("op_40_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_40_strides_0 = const()[name = string("op_40_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_40_pad_0 = const()[name = string("op_40_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_40_dilations_0 = const()[name = string("op_40_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_40_groups_0 = const()[name = string("op_40_groups_0"), val = int32(1)];
            tensor<fp16, [2560, 640, 1, 1]> decoder_cells_0_hh_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2560, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1648896))), scale = tensor<fp16, [2560, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3287360))))[name = string("decoder_cells_0_hh_weight_to_fp16_quantized")];
            tensor<fp16, [2, 2560, 1, 1]> var_40_cast_fp16 = conv(dilations = var_40_dilations_0, groups = var_40_groups_0, pad = var_40_pad_0, pad_type = var_40_pad_type_0, strides = var_40_strides_0, weight = decoder_cells_0_hh_weight_to_fp16_quantized, x = input_1_cast_fp16)[name = string("op_40_cast_fp16")];
            tensor<fp16, [2, 2560, 1, 1]> gates_1_cast_fp16 = add(x = var_34_cast_fp16, y = var_40_cast_fp16)[name = string("gates_1_cast_fp16")];
            tensor<int32, [4]> var_43_begin_0 = const()[name = string("op_43_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_43_end_0 = const()[name = string("op_43_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1])];
            tensor<bool, [4]> var_43_end_mask_0 = const()[name = string("op_43_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 640, 1, 1]> var_43_cast_fp16 = slice_by_index(begin = var_43_begin_0, end = var_43_end_0, end_mask = var_43_end_mask_0, x = gates_1_cast_fp16)[name = string("op_43_cast_fp16")];
            tensor<fp16, [2, 640, 1, 1]> i_1_cast_fp16 = sigmoid(x = var_43_cast_fp16)[name = string("i_1_cast_fp16")];
            tensor<int32, [4]> var_46_begin_0 = const()[name = string("op_46_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_46_end_0 = const()[name = string("op_46_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1])];
            tensor<bool, [4]> var_46_end_mask_0 = const()[name = string("op_46_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 640, 1, 1]> var_46_cast_fp16 = slice_by_index(begin = var_46_begin_0, end = var_46_end_0, end_mask = var_46_end_mask_0, x = gates_1_cast_fp16)[name = string("op_46_cast_fp16")];
            tensor<fp16, [2, 640, 1, 1]> f_1_cast_fp16 = sigmoid(x = var_46_cast_fp16)[name = string("f_1_cast_fp16")];
            tensor<int32, [4]> var_49_begin_0 = const()[name = string("op_49_begin_0"), val = tensor<int32, [4]>([0, 1280, 0, 0])];
            tensor<int32, [4]> var_49_end_0 = const()[name = string("op_49_end_0"), val = tensor<int32, [4]>([2, 1920, 1, 1])];
            tensor<bool, [4]> var_49_end_mask_0 = const()[name = string("op_49_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 640, 1, 1]> var_49_cast_fp16 = slice_by_index(begin = var_49_begin_0, end = var_49_end_0, end_mask = var_49_end_mask_0, x = gates_1_cast_fp16)[name = string("op_49_cast_fp16")];
            tensor<fp16, [2, 640, 1, 1]> g_1_cast_fp16 = tanh(x = var_49_cast_fp16)[name = string("g_1_cast_fp16")];
            tensor<int32, [4]> var_52_begin_0 = const()[name = string("op_52_begin_0"), val = tensor<int32, [4]>([0, 1920, 0, 0])];
            tensor<int32, [4]> var_52_end_0 = const()[name = string("op_52_end_0"), val = tensor<int32, [4]>([2, 1, 1, 1])];
            tensor<bool, [4]> var_52_end_mask_0 = const()[name = string("op_52_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [2, 640, 1, 1]> var_52_cast_fp16 = slice_by_index(begin = var_52_begin_0, end = var_52_end_0, end_mask = var_52_end_mask_0, x = gates_1_cast_fp16)[name = string("op_52_cast_fp16")];
            tensor<fp16, [2, 640, 1, 1]> o_1_cast_fp16 = sigmoid(x = var_52_cast_fp16)[name = string("o_1_cast_fp16")];
            tensor<fp16, [2, 640, 1, 1]> var_54_cast_fp16 = mul(x = f_1_cast_fp16, y = c_1_cast_fp16)[name = string("op_54_cast_fp16")];
            tensor<fp16, [2, 640, 1, 1]> var_55_cast_fp16 = mul(x = i_1_cast_fp16, y = g_1_cast_fp16)[name = string("op_55_cast_fp16")];
            tensor<fp16, [2, 640, 1, 1]> c_new_1_cast_fp16 = add(x = var_54_cast_fp16, y = var_55_cast_fp16)[name = string("c_new_1_cast_fp16")];
            tensor<fp16, [2, 640, 1, 1]> var_57_cast_fp16 = tanh(x = c_new_1_cast_fp16)[name = string("op_57_cast_fp16")];
            tensor<fp16, [2, 640, 1, 1]> input_3_cast_fp16 = mul(x = o_1_cast_fp16, y = var_57_cast_fp16)[name = string("input_3_cast_fp16")];
            tensor<int32, [4]> input_5_begin_0 = const()[name = string("input_5_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> input_5_end_0 = const()[name = string("input_5_end_0"), val = tensor<int32, [4]>([2, 1, 1, 1])];
            tensor<bool, [4]> input_5_end_mask_0 = const()[name = string("input_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [2, 640, 1, 1]> input_5_cast_fp16 = slice_by_index(begin = input_5_begin_0, end = input_5_end_0, end_mask = input_5_end_mask_0, x = h_in)[name = string("input_5_cast_fp16")];
            tensor<int32, [4]> c_begin_0 = const()[name = string("c_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> c_end_0 = const()[name = string("c_end_0"), val = tensor<int32, [4]>([2, 1, 1, 1])];
            tensor<bool, [4]> c_end_mask_0 = const()[name = string("c_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [2, 640, 1, 1]> c_cast_fp16 = slice_by_index(begin = c_begin_0, end = c_end_0, end_mask = c_end_mask_0, x = c_in)[name = string("c_cast_fp16")];
            string var_74_pad_type_0 = const()[name = string("op_74_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_74_strides_0 = const()[name = string("op_74_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_74_pad_0 = const()[name = string("op_74_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_74_dilations_0 = const()[name = string("op_74_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_74_groups_0 = const()[name = string("op_74_groups_0"), val = int32(1)];
            tensor<fp16, [2560, 640, 1, 1]> decoder_cells_1_ih_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2560, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3292544))), scale = tensor<fp16, [2560, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4931008))))[name = string("decoder_cells_1_ih_weight_to_fp16_quantized")];
            tensor<fp16, [2560]> decoder_cells_1_ih_bias_to_fp16 = const()[name = string("decoder_cells_1_ih_bias_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4936192)))];
            tensor<fp16, [2, 2560, 1, 1]> var_74_cast_fp16 = conv(bias = decoder_cells_1_ih_bias_to_fp16, dilations = var_74_dilations_0, groups = var_74_groups_0, pad = var_74_pad_0, pad_type = var_74_pad_type_0, strides = var_74_strides_0, weight = decoder_cells_1_ih_weight_to_fp16_quantized, x = input_3_cast_fp16)[name = string("op_74_cast_fp16")];
            string var_80_pad_type_0 = const()[name = string("op_80_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_80_strides_0 = const()[name = string("op_80_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_80_pad_0 = const()[name = string("op_80_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_80_dilations_0 = const()[name = string("op_80_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_80_groups_0 = const()[name = string("op_80_groups_0"), val = int32(1)];
            tensor<fp16, [2560, 640, 1, 1]> decoder_cells_1_hh_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2560, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4941376))), scale = tensor<fp16, [2560, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6579840))))[name = string("decoder_cells_1_hh_weight_to_fp16_quantized")];
            tensor<fp16, [2, 2560, 1, 1]> var_80_cast_fp16 = conv(dilations = var_80_dilations_0, groups = var_80_groups_0, pad = var_80_pad_0, pad_type = var_80_pad_type_0, strides = var_80_strides_0, weight = decoder_cells_1_hh_weight_to_fp16_quantized, x = input_5_cast_fp16)[name = string("op_80_cast_fp16")];
            tensor<fp16, [2, 2560, 1, 1]> gates_cast_fp16 = add(x = var_74_cast_fp16, y = var_80_cast_fp16)[name = string("gates_cast_fp16")];
            tensor<int32, [4]> var_83_begin_0 = const()[name = string("op_83_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_83_end_0 = const()[name = string("op_83_end_0"), val = tensor<int32, [4]>([2, 640, 1, 1])];
            tensor<bool, [4]> var_83_end_mask_0 = const()[name = string("op_83_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 640, 1, 1]> var_83_cast_fp16 = slice_by_index(begin = var_83_begin_0, end = var_83_end_0, end_mask = var_83_end_mask_0, x = gates_cast_fp16)[name = string("op_83_cast_fp16")];
            tensor<fp16, [2, 640, 1, 1]> i_cast_fp16 = sigmoid(x = var_83_cast_fp16)[name = string("i_cast_fp16")];
            tensor<int32, [4]> var_86_begin_0 = const()[name = string("op_86_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_86_end_0 = const()[name = string("op_86_end_0"), val = tensor<int32, [4]>([2, 1280, 1, 1])];
            tensor<bool, [4]> var_86_end_mask_0 = const()[name = string("op_86_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 640, 1, 1]> var_86_cast_fp16 = slice_by_index(begin = var_86_begin_0, end = var_86_end_0, end_mask = var_86_end_mask_0, x = gates_cast_fp16)[name = string("op_86_cast_fp16")];
            tensor<fp16, [2, 640, 1, 1]> f_cast_fp16 = sigmoid(x = var_86_cast_fp16)[name = string("f_cast_fp16")];
            tensor<int32, [4]> var_89_begin_0 = const()[name = string("op_89_begin_0"), val = tensor<int32, [4]>([0, 1280, 0, 0])];
            tensor<int32, [4]> var_89_end_0 = const()[name = string("op_89_end_0"), val = tensor<int32, [4]>([2, 1920, 1, 1])];
            tensor<bool, [4]> var_89_end_mask_0 = const()[name = string("op_89_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [2, 640, 1, 1]> var_89_cast_fp16 = slice_by_index(begin = var_89_begin_0, end = var_89_end_0, end_mask = var_89_end_mask_0, x = gates_cast_fp16)[name = string("op_89_cast_fp16")];
            tensor<fp16, [2, 640, 1, 1]> g_cast_fp16 = tanh(x = var_89_cast_fp16)[name = string("g_cast_fp16")];
            tensor<int32, [4]> var_92_begin_0 = const()[name = string("op_92_begin_0"), val = tensor<int32, [4]>([0, 1920, 0, 0])];
            tensor<int32, [4]> var_92_end_0 = const()[name = string("op_92_end_0"), val = tensor<int32, [4]>([2, 1, 1, 1])];
            tensor<bool, [4]> var_92_end_mask_0 = const()[name = string("op_92_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [2, 640, 1, 1]> var_92_cast_fp16 = slice_by_index(begin = var_92_begin_0, end = var_92_end_0, end_mask = var_92_end_mask_0, x = gates_cast_fp16)[name = string("op_92_cast_fp16")];
            tensor<fp16, [2, 640, 1, 1]> o_cast_fp16 = sigmoid(x = var_92_cast_fp16)[name = string("o_cast_fp16")];
            tensor<fp16, [2, 640, 1, 1]> var_94_cast_fp16 = mul(x = f_cast_fp16, y = c_cast_fp16)[name = string("op_94_cast_fp16")];
            tensor<fp16, [2, 640, 1, 1]> var_95_cast_fp16 = mul(x = i_cast_fp16, y = g_cast_fp16)[name = string("op_95_cast_fp16")];
            tensor<fp16, [2, 640, 1, 1]> c_new_cast_fp16 = add(x = var_94_cast_fp16, y = var_95_cast_fp16)[name = string("c_new_cast_fp16")];
            tensor<fp16, [2, 640, 1, 1]> var_97_cast_fp16 = tanh(x = c_new_cast_fp16)[name = string("op_97_cast_fp16")];
            tensor<fp16, [2, 640, 1, 1]> input_7_cast_fp16 = mul(x = o_cast_fp16, y = var_97_cast_fp16)[name = string("input_7_cast_fp16")];
            string pred_pad_type_0 = const()[name = string("pred_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> pred_strides_0 = const()[name = string("pred_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> pred_pad_0 = const()[name = string("pred_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> pred_dilations_0 = const()[name = string("pred_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 pred_groups_0 = const()[name = string("pred_groups_0"), val = int32(1)];
            tensor<fp16, [640, 640, 1, 1]> decoder_joint_pred_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [640, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6585024))), scale = tensor<fp16, [640, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6994688))))[name = string("decoder_joint_pred_weight_to_fp16_quantized")];
            tensor<fp16, [640]> decoder_joint_pred_bias_to_fp16 = const()[name = string("decoder_joint_pred_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6996032)))];
            tensor<fp16, [2, 640, 1, 1]> pred_cast_fp16 = conv(bias = decoder_joint_pred_bias_to_fp16, dilations = pred_dilations_0, groups = pred_groups_0, pad = pred_pad_0, pad_type = pred_pad_type_0, strides = pred_strides_0, weight = decoder_joint_pred_weight_to_fp16_quantized, x = input_7_cast_fp16)[name = string("pred_cast_fp16")];
            bool var_110_interleave_0 = const()[name = string("op_110_interleave_0"), val = bool(false)];
            tensor<fp16, [2, 1280, 1, 1]> h_out = concat(axis = var_14, interleave = var_110_interleave_0, values = (input_3_cast_fp16, input_7_cast_fp16))[name = string("op_110_cast_fp16")];
            bool var_112_interleave_0 = const()[name = string("op_112_interleave_0"), val = bool(false)];
            tensor<fp16, [2, 1280, 1, 1]> c_out = concat(axis = var_14, interleave = var_112_interleave_0, values = (c_new_1_cast_fp16, c_new_cast_fp16))[name = string("op_112_cast_fp16")];
            tensor<fp16, [2, 640, 1, 8]> var_122_cast_fp16 = add(x = enc_step, y = pred_cast_fp16)[name = string("op_122_cast_fp16")];
            tensor<fp16, [2, 640, 1, 8]> input_cast_fp16 = relu(x = var_122_cast_fp16)[name = string("input_cast_fp16")];
            string x_pad_type_0 = const()[name = string("x_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> x_strides_0 = const()[name = string("x_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_pad_0 = const()[name = string("x_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_dilations_0 = const()[name = string("x_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 x_groups_0 = const()[name = string("x_groups_0"), val = int32(1)];
            tensor<fp16, [8198, 640, 1, 1]> joint_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [8198, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6997376))), scale = tensor<fp16, [8198, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12244160))))[name = string("joint_out_weight_to_fp16_quantized")];
            tensor<fp16, [8198]> joint_out_bias_to_fp16 = const()[name = string("joint_out_bias_to_fp16"), val = tensor<fp16, [8198]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12260672)))];
            tensor<fp16, [2, 8198, 1, 8]> logits = conv(bias = joint_out_bias_to_fp16, dilations = x_dilations_0, groups = x_groups_0, pad = x_pad_0, pad_type = x_pad_type_0, strides = x_strides_0, weight = joint_out_weight_to_fp16_quantized, x = input_cast_fp16)[name = string("x_cast_fp16")];
        } -> (logits, h_out, c_out);
    func decoder_4<ios18>(tensor<fp16, [4, 1280, 1, 1]> c_in, tensor<fp16, [4, 640, 1, 1]> embed, tensor<fp16, [4, 640, 1, 8]> enc_step, tensor<fp16, [4, 1280, 1, 1]> h_in) {
            int32 var_14 = const()[name = string("op_14"), val = int32(1)];
            tensor<int32, [4]> input_1_begin_0 = const()[name = string("input_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> input_1_end_0 = const()[name = string("input_1_end_0"), val = tensor<int32, [4]>([4, 640, 1, 1])];
            tensor<bool, [4]> input_1_end_mask_0 = const()[name = string("input_1_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [4, 640, 1, 1]> input_1_cast_fp16 = slice_by_index(begin = input_1_begin_0, end = input_1_end_0, end_mask = input_1_end_mask_0, x = h_in)[name = string("input_1_cast_fp16")];
            tensor<int32, [4]> c_1_begin_0 = const()[name = string("c_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> c_1_end_0 = const()[name = string("c_1_end_0"), val = tensor<int32, [4]>([4, 640, 1, 1])];
            tensor<bool, [4]> c_1_end_mask_0 = const()[name = string("c_1_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [4, 640, 1, 1]> c_1_cast_fp16 = slice_by_index(begin = c_1_begin_0, end = c_1_end_0, end_mask = c_1_end_mask_0, x = c_in)[name = string("c_1_cast_fp16")];
            string var_34_pad_type_0 = const()[name = string("op_34_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_34_strides_0 = const()[name = string("op_34_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_34_pad_0 = const()[name = string("op_34_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_34_dilations_0 = const()[name = string("op_34_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_34_groups_0 = const()[name = string("op_34_groups_0"), val = int32(1)];
            tensor<fp16, [2560, 640, 1, 1]> decoder_cells_0_ih_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2560, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64))), scale = tensor<fp16, [2560, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1638528))))[name = string("decoder_cells_0_ih_weight_to_fp16_quantized")];
            tensor<fp16, [2560]> decoder_cells_0_ih_bias_to_fp16 = const()[name = string("decoder_cells_0_ih_bias_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1643712)))];
            tensor<fp16, [4, 2560, 1, 1]> var_34_cast_fp16 = conv(bias = decoder_cells_0_ih_bias_to_fp16, dilations = var_34_dilations_0, groups = var_34_groups_0, pad = var_34_pad_0, pad_type = var_34_pad_type_0, strides = var_34_strides_0, weight = decoder_cells_0_ih_weight_to_fp16_quantized, x = embed)[name = string("op_34_cast_fp16")];
            string var_40_pad_type_0 = const()[name = string("op_40_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_40_strides_0 = const()[name = string("op_40_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_40_pad_0 = const()[name = string("op_40_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_40_dilations_0 = const()[name = string("op_40_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_40_groups_0 = const()[name = string("op_40_groups_0"), val = int32(1)];
            tensor<fp16, [2560, 640, 1, 1]> decoder_cells_0_hh_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2560, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1648896))), scale = tensor<fp16, [2560, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3287360))))[name = string("decoder_cells_0_hh_weight_to_fp16_quantized")];
            tensor<fp16, [4, 2560, 1, 1]> var_40_cast_fp16 = conv(dilations = var_40_dilations_0, groups = var_40_groups_0, pad = var_40_pad_0, pad_type = var_40_pad_type_0, strides = var_40_strides_0, weight = decoder_cells_0_hh_weight_to_fp16_quantized, x = input_1_cast_fp16)[name = string("op_40_cast_fp16")];
            tensor<fp16, [4, 2560, 1, 1]> gates_1_cast_fp16 = add(x = var_34_cast_fp16, y = var_40_cast_fp16)[name = string("gates_1_cast_fp16")];
            tensor<int32, [4]> var_43_begin_0 = const()[name = string("op_43_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_43_end_0 = const()[name = string("op_43_end_0"), val = tensor<int32, [4]>([4, 640, 1, 1])];
            tensor<bool, [4]> var_43_end_mask_0 = const()[name = string("op_43_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [4, 640, 1, 1]> var_43_cast_fp16 = slice_by_index(begin = var_43_begin_0, end = var_43_end_0, end_mask = var_43_end_mask_0, x = gates_1_cast_fp16)[name = string("op_43_cast_fp16")];
            tensor<fp16, [4, 640, 1, 1]> i_1_cast_fp16 = sigmoid(x = var_43_cast_fp16)[name = string("i_1_cast_fp16")];
            tensor<int32, [4]> var_46_begin_0 = const()[name = string("op_46_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_46_end_0 = const()[name = string("op_46_end_0"), val = tensor<int32, [4]>([4, 1280, 1, 1])];
            tensor<bool, [4]> var_46_end_mask_0 = const()[name = string("op_46_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [4, 640, 1, 1]> var_46_cast_fp16 = slice_by_index(begin = var_46_begin_0, end = var_46_end_0, end_mask = var_46_end_mask_0, x = gates_1_cast_fp16)[name = string("op_46_cast_fp16")];
            tensor<fp16, [4, 640, 1, 1]> f_1_cast_fp16 = sigmoid(x = var_46_cast_fp16)[name = string("f_1_cast_fp16")];
            tensor<int32, [4]> var_49_begin_0 = const()[name = string("op_49_begin_0"), val = tensor<int32, [4]>([0, 1280, 0, 0])];
            tensor<int32, [4]> var_49_end_0 = const()[name = string("op_49_end_0"), val = tensor<int32, [4]>([4, 1920, 1, 1])];
            tensor<bool, [4]> var_49_end_mask_0 = const()[name = string("op_49_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [4, 640, 1, 1]> var_49_cast_fp16 = slice_by_index(begin = var_49_begin_0, end = var_49_end_0, end_mask = var_49_end_mask_0, x = gates_1_cast_fp16)[name = string("op_49_cast_fp16")];
            tensor<fp16, [4, 640, 1, 1]> g_1_cast_fp16 = tanh(x = var_49_cast_fp16)[name = string("g_1_cast_fp16")];
            tensor<int32, [4]> var_52_begin_0 = const()[name = string("op_52_begin_0"), val = tensor<int32, [4]>([0, 1920, 0, 0])];
            tensor<int32, [4]> var_52_end_0 = const()[name = string("op_52_end_0"), val = tensor<int32, [4]>([4, 1, 1, 1])];
            tensor<bool, [4]> var_52_end_mask_0 = const()[name = string("op_52_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [4, 640, 1, 1]> var_52_cast_fp16 = slice_by_index(begin = var_52_begin_0, end = var_52_end_0, end_mask = var_52_end_mask_0, x = gates_1_cast_fp16)[name = string("op_52_cast_fp16")];
            tensor<fp16, [4, 640, 1, 1]> o_1_cast_fp16 = sigmoid(x = var_52_cast_fp16)[name = string("o_1_cast_fp16")];
            tensor<fp16, [4, 640, 1, 1]> var_54_cast_fp16 = mul(x = f_1_cast_fp16, y = c_1_cast_fp16)[name = string("op_54_cast_fp16")];
            tensor<fp16, [4, 640, 1, 1]> var_55_cast_fp16 = mul(x = i_1_cast_fp16, y = g_1_cast_fp16)[name = string("op_55_cast_fp16")];
            tensor<fp16, [4, 640, 1, 1]> c_new_1_cast_fp16 = add(x = var_54_cast_fp16, y = var_55_cast_fp16)[name = string("c_new_1_cast_fp16")];
            tensor<fp16, [4, 640, 1, 1]> var_57_cast_fp16 = tanh(x = c_new_1_cast_fp16)[name = string("op_57_cast_fp16")];
            tensor<fp16, [4, 640, 1, 1]> input_3_cast_fp16 = mul(x = o_1_cast_fp16, y = var_57_cast_fp16)[name = string("input_3_cast_fp16")];
            tensor<int32, [4]> input_5_begin_0 = const()[name = string("input_5_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> input_5_end_0 = const()[name = string("input_5_end_0"), val = tensor<int32, [4]>([4, 1, 1, 1])];
            tensor<bool, [4]> input_5_end_mask_0 = const()[name = string("input_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [4, 640, 1, 1]> input_5_cast_fp16 = slice_by_index(begin = input_5_begin_0, end = input_5_end_0, end_mask = input_5_end_mask_0, x = h_in)[name = string("input_5_cast_fp16")];
            tensor<int32, [4]> c_begin_0 = const()[name = string("c_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> c_end_0 = const()[name = string("c_end_0"), val = tensor<int32, [4]>([4, 1, 1, 1])];
            tensor<bool, [4]> c_end_mask_0 = const()[name = string("c_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [4, 640, 1, 1]> c_cast_fp16 = slice_by_index(begin = c_begin_0, end = c_end_0, end_mask = c_end_mask_0, x = c_in)[name = string("c_cast_fp16")];
            string var_74_pad_type_0 = const()[name = string("op_74_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_74_strides_0 = const()[name = string("op_74_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_74_pad_0 = const()[name = string("op_74_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_74_dilations_0 = const()[name = string("op_74_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_74_groups_0 = const()[name = string("op_74_groups_0"), val = int32(1)];
            tensor<fp16, [2560, 640, 1, 1]> decoder_cells_1_ih_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2560, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3292544))), scale = tensor<fp16, [2560, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4931008))))[name = string("decoder_cells_1_ih_weight_to_fp16_quantized")];
            tensor<fp16, [2560]> decoder_cells_1_ih_bias_to_fp16 = const()[name = string("decoder_cells_1_ih_bias_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4936192)))];
            tensor<fp16, [4, 2560, 1, 1]> var_74_cast_fp16 = conv(bias = decoder_cells_1_ih_bias_to_fp16, dilations = var_74_dilations_0, groups = var_74_groups_0, pad = var_74_pad_0, pad_type = var_74_pad_type_0, strides = var_74_strides_0, weight = decoder_cells_1_ih_weight_to_fp16_quantized, x = input_3_cast_fp16)[name = string("op_74_cast_fp16")];
            string var_80_pad_type_0 = const()[name = string("op_80_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_80_strides_0 = const()[name = string("op_80_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_80_pad_0 = const()[name = string("op_80_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_80_dilations_0 = const()[name = string("op_80_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_80_groups_0 = const()[name = string("op_80_groups_0"), val = int32(1)];
            tensor<fp16, [2560, 640, 1, 1]> decoder_cells_1_hh_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2560, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4941376))), scale = tensor<fp16, [2560, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6579840))))[name = string("decoder_cells_1_hh_weight_to_fp16_quantized")];
            tensor<fp16, [4, 2560, 1, 1]> var_80_cast_fp16 = conv(dilations = var_80_dilations_0, groups = var_80_groups_0, pad = var_80_pad_0, pad_type = var_80_pad_type_0, strides = var_80_strides_0, weight = decoder_cells_1_hh_weight_to_fp16_quantized, x = input_5_cast_fp16)[name = string("op_80_cast_fp16")];
            tensor<fp16, [4, 2560, 1, 1]> gates_cast_fp16 = add(x = var_74_cast_fp16, y = var_80_cast_fp16)[name = string("gates_cast_fp16")];
            tensor<int32, [4]> var_83_begin_0 = const()[name = string("op_83_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_83_end_0 = const()[name = string("op_83_end_0"), val = tensor<int32, [4]>([4, 640, 1, 1])];
            tensor<bool, [4]> var_83_end_mask_0 = const()[name = string("op_83_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [4, 640, 1, 1]> var_83_cast_fp16 = slice_by_index(begin = var_83_begin_0, end = var_83_end_0, end_mask = var_83_end_mask_0, x = gates_cast_fp16)[name = string("op_83_cast_fp16")];
            tensor<fp16, [4, 640, 1, 1]> i_cast_fp16 = sigmoid(x = var_83_cast_fp16)[name = string("i_cast_fp16")];
            tensor<int32, [4]> var_86_begin_0 = const()[name = string("op_86_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_86_end_0 = const()[name = string("op_86_end_0"), val = tensor<int32, [4]>([4, 1280, 1, 1])];
            tensor<bool, [4]> var_86_end_mask_0 = const()[name = string("op_86_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [4, 640, 1, 1]> var_86_cast_fp16 = slice_by_index(begin = var_86_begin_0, end = var_86_end_0, end_mask = var_86_end_mask_0, x = gates_cast_fp16)[name = string("op_86_cast_fp16")];
            tensor<fp16, [4, 640, 1, 1]> f_cast_fp16 = sigmoid(x = var_86_cast_fp16)[name = string("f_cast_fp16")];
            tensor<int32, [4]> var_89_begin_0 = const()[name = string("op_89_begin_0"), val = tensor<int32, [4]>([0, 1280, 0, 0])];
            tensor<int32, [4]> var_89_end_0 = const()[name = string("op_89_end_0"), val = tensor<int32, [4]>([4, 1920, 1, 1])];
            tensor<bool, [4]> var_89_end_mask_0 = const()[name = string("op_89_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [4, 640, 1, 1]> var_89_cast_fp16 = slice_by_index(begin = var_89_begin_0, end = var_89_end_0, end_mask = var_89_end_mask_0, x = gates_cast_fp16)[name = string("op_89_cast_fp16")];
            tensor<fp16, [4, 640, 1, 1]> g_cast_fp16 = tanh(x = var_89_cast_fp16)[name = string("g_cast_fp16")];
            tensor<int32, [4]> var_92_begin_0 = const()[name = string("op_92_begin_0"), val = tensor<int32, [4]>([0, 1920, 0, 0])];
            tensor<int32, [4]> var_92_end_0 = const()[name = string("op_92_end_0"), val = tensor<int32, [4]>([4, 1, 1, 1])];
            tensor<bool, [4]> var_92_end_mask_0 = const()[name = string("op_92_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [4, 640, 1, 1]> var_92_cast_fp16 = slice_by_index(begin = var_92_begin_0, end = var_92_end_0, end_mask = var_92_end_mask_0, x = gates_cast_fp16)[name = string("op_92_cast_fp16")];
            tensor<fp16, [4, 640, 1, 1]> o_cast_fp16 = sigmoid(x = var_92_cast_fp16)[name = string("o_cast_fp16")];
            tensor<fp16, [4, 640, 1, 1]> var_94_cast_fp16 = mul(x = f_cast_fp16, y = c_cast_fp16)[name = string("op_94_cast_fp16")];
            tensor<fp16, [4, 640, 1, 1]> var_95_cast_fp16 = mul(x = i_cast_fp16, y = g_cast_fp16)[name = string("op_95_cast_fp16")];
            tensor<fp16, [4, 640, 1, 1]> c_new_cast_fp16 = add(x = var_94_cast_fp16, y = var_95_cast_fp16)[name = string("c_new_cast_fp16")];
            tensor<fp16, [4, 640, 1, 1]> var_97_cast_fp16 = tanh(x = c_new_cast_fp16)[name = string("op_97_cast_fp16")];
            tensor<fp16, [4, 640, 1, 1]> input_7_cast_fp16 = mul(x = o_cast_fp16, y = var_97_cast_fp16)[name = string("input_7_cast_fp16")];
            string pred_pad_type_0 = const()[name = string("pred_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> pred_strides_0 = const()[name = string("pred_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> pred_pad_0 = const()[name = string("pred_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> pred_dilations_0 = const()[name = string("pred_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 pred_groups_0 = const()[name = string("pred_groups_0"), val = int32(1)];
            tensor<fp16, [640, 640, 1, 1]> decoder_joint_pred_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [640, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6585024))), scale = tensor<fp16, [640, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6994688))))[name = string("decoder_joint_pred_weight_to_fp16_quantized")];
            tensor<fp16, [640]> decoder_joint_pred_bias_to_fp16 = const()[name = string("decoder_joint_pred_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6996032)))];
            tensor<fp16, [4, 640, 1, 1]> pred_cast_fp16 = conv(bias = decoder_joint_pred_bias_to_fp16, dilations = pred_dilations_0, groups = pred_groups_0, pad = pred_pad_0, pad_type = pred_pad_type_0, strides = pred_strides_0, weight = decoder_joint_pred_weight_to_fp16_quantized, x = input_7_cast_fp16)[name = string("pred_cast_fp16")];
            bool var_110_interleave_0 = const()[name = string("op_110_interleave_0"), val = bool(false)];
            tensor<fp16, [4, 1280, 1, 1]> h_out = concat(axis = var_14, interleave = var_110_interleave_0, values = (input_3_cast_fp16, input_7_cast_fp16))[name = string("op_110_cast_fp16")];
            bool var_112_interleave_0 = const()[name = string("op_112_interleave_0"), val = bool(false)];
            tensor<fp16, [4, 1280, 1, 1]> c_out = concat(axis = var_14, interleave = var_112_interleave_0, values = (c_new_1_cast_fp16, c_new_cast_fp16))[name = string("op_112_cast_fp16")];
            tensor<fp16, [4, 640, 1, 8]> var_122_cast_fp16 = add(x = enc_step, y = pred_cast_fp16)[name = string("op_122_cast_fp16")];
            tensor<fp16, [4, 640, 1, 8]> input_cast_fp16 = relu(x = var_122_cast_fp16)[name = string("input_cast_fp16")];
            string x_pad_type_0 = const()[name = string("x_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> x_strides_0 = const()[name = string("x_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_pad_0 = const()[name = string("x_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_dilations_0 = const()[name = string("x_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 x_groups_0 = const()[name = string("x_groups_0"), val = int32(1)];
            tensor<fp16, [8198, 640, 1, 1]> joint_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [8198, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6997376))), scale = tensor<fp16, [8198, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12244160))))[name = string("joint_out_weight_to_fp16_quantized")];
            tensor<fp16, [8198]> joint_out_bias_to_fp16 = const()[name = string("joint_out_bias_to_fp16"), val = tensor<fp16, [8198]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12260672)))];
            tensor<fp16, [4, 8198, 1, 8]> logits = conv(bias = joint_out_bias_to_fp16, dilations = x_dilations_0, groups = x_groups_0, pad = x_pad_0, pad_type = x_pad_type_0, strides = x_strides_0, weight = joint_out_weight_to_fp16_quantized, x = input_cast_fp16)[name = string("x_cast_fp16")];
        } -> (logits, h_out, c_out);
    func decoder_8<ios18>(tensor<fp16, [8, 1280, 1, 1]> c_in, tensor<fp16, [8, 640, 1, 1]> embed, tensor<fp16, [8, 640, 1, 8]> enc_step, tensor<fp16, [8, 1280, 1, 1]> h_in) {
            int32 var_14 = const()[name = string("op_14"), val = int32(1)];
            tensor<int32, [4]> input_1_begin_0 = const()[name = string("input_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> input_1_end_0 = const()[name = string("input_1_end_0"), val = tensor<int32, [4]>([8, 640, 1, 1])];
            tensor<bool, [4]> input_1_end_mask_0 = const()[name = string("input_1_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [8, 640, 1, 1]> input_1_cast_fp16 = slice_by_index(begin = input_1_begin_0, end = input_1_end_0, end_mask = input_1_end_mask_0, x = h_in)[name = string("input_1_cast_fp16")];
            tensor<int32, [4]> c_1_begin_0 = const()[name = string("c_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> c_1_end_0 = const()[name = string("c_1_end_0"), val = tensor<int32, [4]>([8, 640, 1, 1])];
            tensor<bool, [4]> c_1_end_mask_0 = const()[name = string("c_1_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [8, 640, 1, 1]> c_1_cast_fp16 = slice_by_index(begin = c_1_begin_0, end = c_1_end_0, end_mask = c_1_end_mask_0, x = c_in)[name = string("c_1_cast_fp16")];
            string var_34_pad_type_0 = const()[name = string("op_34_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_34_strides_0 = const()[name = string("op_34_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_34_pad_0 = const()[name = string("op_34_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_34_dilations_0 = const()[name = string("op_34_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_34_groups_0 = const()[name = string("op_34_groups_0"), val = int32(1)];
            tensor<fp16, [2560, 640, 1, 1]> decoder_cells_0_ih_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2560, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64))), scale = tensor<fp16, [2560, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1638528))))[name = string("decoder_cells_0_ih_weight_to_fp16_quantized")];
            tensor<fp16, [2560]> decoder_cells_0_ih_bias_to_fp16 = const()[name = string("decoder_cells_0_ih_bias_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1643712)))];
            tensor<fp16, [8, 2560, 1, 1]> var_34_cast_fp16 = conv(bias = decoder_cells_0_ih_bias_to_fp16, dilations = var_34_dilations_0, groups = var_34_groups_0, pad = var_34_pad_0, pad_type = var_34_pad_type_0, strides = var_34_strides_0, weight = decoder_cells_0_ih_weight_to_fp16_quantized, x = embed)[name = string("op_34_cast_fp16")];
            string var_40_pad_type_0 = const()[name = string("op_40_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_40_strides_0 = const()[name = string("op_40_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_40_pad_0 = const()[name = string("op_40_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_40_dilations_0 = const()[name = string("op_40_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_40_groups_0 = const()[name = string("op_40_groups_0"), val = int32(1)];
            tensor<fp16, [2560, 640, 1, 1]> decoder_cells_0_hh_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2560, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1648896))), scale = tensor<fp16, [2560, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3287360))))[name = string("decoder_cells_0_hh_weight_to_fp16_quantized")];
            tensor<fp16, [8, 2560, 1, 1]> var_40_cast_fp16 = conv(dilations = var_40_dilations_0, groups = var_40_groups_0, pad = var_40_pad_0, pad_type = var_40_pad_type_0, strides = var_40_strides_0, weight = decoder_cells_0_hh_weight_to_fp16_quantized, x = input_1_cast_fp16)[name = string("op_40_cast_fp16")];
            tensor<fp16, [8, 2560, 1, 1]> gates_1_cast_fp16 = add(x = var_34_cast_fp16, y = var_40_cast_fp16)[name = string("gates_1_cast_fp16")];
            tensor<int32, [4]> var_43_begin_0 = const()[name = string("op_43_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_43_end_0 = const()[name = string("op_43_end_0"), val = tensor<int32, [4]>([8, 640, 1, 1])];
            tensor<bool, [4]> var_43_end_mask_0 = const()[name = string("op_43_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [8, 640, 1, 1]> var_43_cast_fp16 = slice_by_index(begin = var_43_begin_0, end = var_43_end_0, end_mask = var_43_end_mask_0, x = gates_1_cast_fp16)[name = string("op_43_cast_fp16")];
            tensor<fp16, [8, 640, 1, 1]> i_1_cast_fp16 = sigmoid(x = var_43_cast_fp16)[name = string("i_1_cast_fp16")];
            tensor<int32, [4]> var_46_begin_0 = const()[name = string("op_46_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_46_end_0 = const()[name = string("op_46_end_0"), val = tensor<int32, [4]>([8, 1280, 1, 1])];
            tensor<bool, [4]> var_46_end_mask_0 = const()[name = string("op_46_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [8, 640, 1, 1]> var_46_cast_fp16 = slice_by_index(begin = var_46_begin_0, end = var_46_end_0, end_mask = var_46_end_mask_0, x = gates_1_cast_fp16)[name = string("op_46_cast_fp16")];
            tensor<fp16, [8, 640, 1, 1]> f_1_cast_fp16 = sigmoid(x = var_46_cast_fp16)[name = string("f_1_cast_fp16")];
            tensor<int32, [4]> var_49_begin_0 = const()[name = string("op_49_begin_0"), val = tensor<int32, [4]>([0, 1280, 0, 0])];
            tensor<int32, [4]> var_49_end_0 = const()[name = string("op_49_end_0"), val = tensor<int32, [4]>([8, 1920, 1, 1])];
            tensor<bool, [4]> var_49_end_mask_0 = const()[name = string("op_49_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [8, 640, 1, 1]> var_49_cast_fp16 = slice_by_index(begin = var_49_begin_0, end = var_49_end_0, end_mask = var_49_end_mask_0, x = gates_1_cast_fp16)[name = string("op_49_cast_fp16")];
            tensor<fp16, [8, 640, 1, 1]> g_1_cast_fp16 = tanh(x = var_49_cast_fp16)[name = string("g_1_cast_fp16")];
            tensor<int32, [4]> var_52_begin_0 = const()[name = string("op_52_begin_0"), val = tensor<int32, [4]>([0, 1920, 0, 0])];
            tensor<int32, [4]> var_52_end_0 = const()[name = string("op_52_end_0"), val = tensor<int32, [4]>([8, 1, 1, 1])];
            tensor<bool, [4]> var_52_end_mask_0 = const()[name = string("op_52_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [8, 640, 1, 1]> var_52_cast_fp16 = slice_by_index(begin = var_52_begin_0, end = var_52_end_0, end_mask = var_52_end_mask_0, x = gates_1_cast_fp16)[name = string("op_52_cast_fp16")];
            tensor<fp16, [8, 640, 1, 1]> o_1_cast_fp16 = sigmoid(x = var_52_cast_fp16)[name = string("o_1_cast_fp16")];
            tensor<fp16, [8, 640, 1, 1]> var_54_cast_fp16 = mul(x = f_1_cast_fp16, y = c_1_cast_fp16)[name = string("op_54_cast_fp16")];
            tensor<fp16, [8, 640, 1, 1]> var_55_cast_fp16 = mul(x = i_1_cast_fp16, y = g_1_cast_fp16)[name = string("op_55_cast_fp16")];
            tensor<fp16, [8, 640, 1, 1]> c_new_1_cast_fp16 = add(x = var_54_cast_fp16, y = var_55_cast_fp16)[name = string("c_new_1_cast_fp16")];
            tensor<fp16, [8, 640, 1, 1]> var_57_cast_fp16 = tanh(x = c_new_1_cast_fp16)[name = string("op_57_cast_fp16")];
            tensor<fp16, [8, 640, 1, 1]> input_3_cast_fp16 = mul(x = o_1_cast_fp16, y = var_57_cast_fp16)[name = string("input_3_cast_fp16")];
            tensor<int32, [4]> input_5_begin_0 = const()[name = string("input_5_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> input_5_end_0 = const()[name = string("input_5_end_0"), val = tensor<int32, [4]>([8, 1, 1, 1])];
            tensor<bool, [4]> input_5_end_mask_0 = const()[name = string("input_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [8, 640, 1, 1]> input_5_cast_fp16 = slice_by_index(begin = input_5_begin_0, end = input_5_end_0, end_mask = input_5_end_mask_0, x = h_in)[name = string("input_5_cast_fp16")];
            tensor<int32, [4]> c_begin_0 = const()[name = string("c_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> c_end_0 = const()[name = string("c_end_0"), val = tensor<int32, [4]>([8, 1, 1, 1])];
            tensor<bool, [4]> c_end_mask_0 = const()[name = string("c_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [8, 640, 1, 1]> c_cast_fp16 = slice_by_index(begin = c_begin_0, end = c_end_0, end_mask = c_end_mask_0, x = c_in)[name = string("c_cast_fp16")];
            string var_74_pad_type_0 = const()[name = string("op_74_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_74_strides_0 = const()[name = string("op_74_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_74_pad_0 = const()[name = string("op_74_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_74_dilations_0 = const()[name = string("op_74_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_74_groups_0 = const()[name = string("op_74_groups_0"), val = int32(1)];
            tensor<fp16, [2560, 640, 1, 1]> decoder_cells_1_ih_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2560, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3292544))), scale = tensor<fp16, [2560, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4931008))))[name = string("decoder_cells_1_ih_weight_to_fp16_quantized")];
            tensor<fp16, [2560]> decoder_cells_1_ih_bias_to_fp16 = const()[name = string("decoder_cells_1_ih_bias_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4936192)))];
            tensor<fp16, [8, 2560, 1, 1]> var_74_cast_fp16 = conv(bias = decoder_cells_1_ih_bias_to_fp16, dilations = var_74_dilations_0, groups = var_74_groups_0, pad = var_74_pad_0, pad_type = var_74_pad_type_0, strides = var_74_strides_0, weight = decoder_cells_1_ih_weight_to_fp16_quantized, x = input_3_cast_fp16)[name = string("op_74_cast_fp16")];
            string var_80_pad_type_0 = const()[name = string("op_80_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_80_strides_0 = const()[name = string("op_80_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_80_pad_0 = const()[name = string("op_80_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_80_dilations_0 = const()[name = string("op_80_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_80_groups_0 = const()[name = string("op_80_groups_0"), val = int32(1)];
            tensor<fp16, [2560, 640, 1, 1]> decoder_cells_1_hh_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2560, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4941376))), scale = tensor<fp16, [2560, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6579840))))[name = string("decoder_cells_1_hh_weight_to_fp16_quantized")];
            tensor<fp16, [8, 2560, 1, 1]> var_80_cast_fp16 = conv(dilations = var_80_dilations_0, groups = var_80_groups_0, pad = var_80_pad_0, pad_type = var_80_pad_type_0, strides = var_80_strides_0, weight = decoder_cells_1_hh_weight_to_fp16_quantized, x = input_5_cast_fp16)[name = string("op_80_cast_fp16")];
            tensor<fp16, [8, 2560, 1, 1]> gates_cast_fp16 = add(x = var_74_cast_fp16, y = var_80_cast_fp16)[name = string("gates_cast_fp16")];
            tensor<int32, [4]> var_83_begin_0 = const()[name = string("op_83_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_83_end_0 = const()[name = string("op_83_end_0"), val = tensor<int32, [4]>([8, 640, 1, 1])];
            tensor<bool, [4]> var_83_end_mask_0 = const()[name = string("op_83_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [8, 640, 1, 1]> var_83_cast_fp16 = slice_by_index(begin = var_83_begin_0, end = var_83_end_0, end_mask = var_83_end_mask_0, x = gates_cast_fp16)[name = string("op_83_cast_fp16")];
            tensor<fp16, [8, 640, 1, 1]> i_cast_fp16 = sigmoid(x = var_83_cast_fp16)[name = string("i_cast_fp16")];
            tensor<int32, [4]> var_86_begin_0 = const()[name = string("op_86_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_86_end_0 = const()[name = string("op_86_end_0"), val = tensor<int32, [4]>([8, 1280, 1, 1])];
            tensor<bool, [4]> var_86_end_mask_0 = const()[name = string("op_86_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [8, 640, 1, 1]> var_86_cast_fp16 = slice_by_index(begin = var_86_begin_0, end = var_86_end_0, end_mask = var_86_end_mask_0, x = gates_cast_fp16)[name = string("op_86_cast_fp16")];
            tensor<fp16, [8, 640, 1, 1]> f_cast_fp16 = sigmoid(x = var_86_cast_fp16)[name = string("f_cast_fp16")];
            tensor<int32, [4]> var_89_begin_0 = const()[name = string("op_89_begin_0"), val = tensor<int32, [4]>([0, 1280, 0, 0])];
            tensor<int32, [4]> var_89_end_0 = const()[name = string("op_89_end_0"), val = tensor<int32, [4]>([8, 1920, 1, 1])];
            tensor<bool, [4]> var_89_end_mask_0 = const()[name = string("op_89_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [8, 640, 1, 1]> var_89_cast_fp16 = slice_by_index(begin = var_89_begin_0, end = var_89_end_0, end_mask = var_89_end_mask_0, x = gates_cast_fp16)[name = string("op_89_cast_fp16")];
            tensor<fp16, [8, 640, 1, 1]> g_cast_fp16 = tanh(x = var_89_cast_fp16)[name = string("g_cast_fp16")];
            tensor<int32, [4]> var_92_begin_0 = const()[name = string("op_92_begin_0"), val = tensor<int32, [4]>([0, 1920, 0, 0])];
            tensor<int32, [4]> var_92_end_0 = const()[name = string("op_92_end_0"), val = tensor<int32, [4]>([8, 1, 1, 1])];
            tensor<bool, [4]> var_92_end_mask_0 = const()[name = string("op_92_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [8, 640, 1, 1]> var_92_cast_fp16 = slice_by_index(begin = var_92_begin_0, end = var_92_end_0, end_mask = var_92_end_mask_0, x = gates_cast_fp16)[name = string("op_92_cast_fp16")];
            tensor<fp16, [8, 640, 1, 1]> o_cast_fp16 = sigmoid(x = var_92_cast_fp16)[name = string("o_cast_fp16")];
            tensor<fp16, [8, 640, 1, 1]> var_94_cast_fp16 = mul(x = f_cast_fp16, y = c_cast_fp16)[name = string("op_94_cast_fp16")];
            tensor<fp16, [8, 640, 1, 1]> var_95_cast_fp16 = mul(x = i_cast_fp16, y = g_cast_fp16)[name = string("op_95_cast_fp16")];
            tensor<fp16, [8, 640, 1, 1]> c_new_cast_fp16 = add(x = var_94_cast_fp16, y = var_95_cast_fp16)[name = string("c_new_cast_fp16")];
            tensor<fp16, [8, 640, 1, 1]> var_97_cast_fp16 = tanh(x = c_new_cast_fp16)[name = string("op_97_cast_fp16")];
            tensor<fp16, [8, 640, 1, 1]> input_7_cast_fp16 = mul(x = o_cast_fp16, y = var_97_cast_fp16)[name = string("input_7_cast_fp16")];
            string pred_pad_type_0 = const()[name = string("pred_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> pred_strides_0 = const()[name = string("pred_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> pred_pad_0 = const()[name = string("pred_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> pred_dilations_0 = const()[name = string("pred_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 pred_groups_0 = const()[name = string("pred_groups_0"), val = int32(1)];
            tensor<fp16, [640, 640, 1, 1]> decoder_joint_pred_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [640, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6585024))), scale = tensor<fp16, [640, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6994688))))[name = string("decoder_joint_pred_weight_to_fp16_quantized")];
            tensor<fp16, [640]> decoder_joint_pred_bias_to_fp16 = const()[name = string("decoder_joint_pred_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6996032)))];
            tensor<fp16, [8, 640, 1, 1]> pred_cast_fp16 = conv(bias = decoder_joint_pred_bias_to_fp16, dilations = pred_dilations_0, groups = pred_groups_0, pad = pred_pad_0, pad_type = pred_pad_type_0, strides = pred_strides_0, weight = decoder_joint_pred_weight_to_fp16_quantized, x = input_7_cast_fp16)[name = string("pred_cast_fp16")];
            bool var_110_interleave_0 = const()[name = string("op_110_interleave_0"), val = bool(false)];
            tensor<fp16, [8, 1280, 1, 1]> h_out = concat(axis = var_14, interleave = var_110_interleave_0, values = (input_3_cast_fp16, input_7_cast_fp16))[name = string("op_110_cast_fp16")];
            bool var_112_interleave_0 = const()[name = string("op_112_interleave_0"), val = bool(false)];
            tensor<fp16, [8, 1280, 1, 1]> c_out = concat(axis = var_14, interleave = var_112_interleave_0, values = (c_new_1_cast_fp16, c_new_cast_fp16))[name = string("op_112_cast_fp16")];
            tensor<fp16, [8, 640, 1, 8]> var_122_cast_fp16 = add(x = enc_step, y = pred_cast_fp16)[name = string("op_122_cast_fp16")];
            tensor<fp16, [8, 640, 1, 8]> input_cast_fp16 = relu(x = var_122_cast_fp16)[name = string("input_cast_fp16")];
            string x_pad_type_0 = const()[name = string("x_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> x_strides_0 = const()[name = string("x_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_pad_0 = const()[name = string("x_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_dilations_0 = const()[name = string("x_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 x_groups_0 = const()[name = string("x_groups_0"), val = int32(1)];
            tensor<fp16, [8198, 640, 1, 1]> joint_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [8198, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6997376))), scale = tensor<fp16, [8198, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12244160))))[name = string("joint_out_weight_to_fp16_quantized")];
            tensor<fp16, [8198]> joint_out_bias_to_fp16 = const()[name = string("joint_out_bias_to_fp16"), val = tensor<fp16, [8198]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12260672)))];
            tensor<fp16, [8, 8198, 1, 8]> logits = conv(bias = joint_out_bias_to_fp16, dilations = x_dilations_0, groups = x_groups_0, pad = x_pad_0, pad_type = x_pad_type_0, strides = x_strides_0, weight = joint_out_weight_to_fp16_quantized, x = input_cast_fp16)[name = string("x_cast_fp16")];
        } -> (logits, h_out, c_out);
    func main<ios18>(tensor<fp16, [16, 1280, 1, 1]> c_in, tensor<fp16, [16, 640, 1, 1]> embed, tensor<fp16, [16, 640, 1, 8]> enc_step, tensor<fp16, [16, 1280, 1, 1]> h_in) {
            int32 var_14 = const()[name = string("op_14"), val = int32(1)];
            tensor<int32, [4]> input_1_begin_0 = const()[name = string("input_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> input_1_end_0 = const()[name = string("input_1_end_0"), val = tensor<int32, [4]>([16, 640, 1, 1])];
            tensor<bool, [4]> input_1_end_mask_0 = const()[name = string("input_1_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [16, 640, 1, 1]> input_1_cast_fp16 = slice_by_index(begin = input_1_begin_0, end = input_1_end_0, end_mask = input_1_end_mask_0, x = h_in)[name = string("input_1_cast_fp16")];
            tensor<int32, [4]> c_1_begin_0 = const()[name = string("c_1_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> c_1_end_0 = const()[name = string("c_1_end_0"), val = tensor<int32, [4]>([16, 640, 1, 1])];
            tensor<bool, [4]> c_1_end_mask_0 = const()[name = string("c_1_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [16, 640, 1, 1]> c_1_cast_fp16 = slice_by_index(begin = c_1_begin_0, end = c_1_end_0, end_mask = c_1_end_mask_0, x = c_in)[name = string("c_1_cast_fp16")];
            string var_34_pad_type_0 = const()[name = string("op_34_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_34_strides_0 = const()[name = string("op_34_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_34_pad_0 = const()[name = string("op_34_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_34_dilations_0 = const()[name = string("op_34_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_34_groups_0 = const()[name = string("op_34_groups_0"), val = int32(1)];
            tensor<fp16, [2560, 640, 1, 1]> decoder_cells_0_ih_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2560, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(64))), scale = tensor<fp16, [2560, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1638528))))[name = string("decoder_cells_0_ih_weight_to_fp16_quantized")];
            tensor<fp16, [2560]> decoder_cells_0_ih_bias_to_fp16 = const()[name = string("decoder_cells_0_ih_bias_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1643712)))];
            tensor<fp16, [16, 2560, 1, 1]> var_34_cast_fp16 = conv(bias = decoder_cells_0_ih_bias_to_fp16, dilations = var_34_dilations_0, groups = var_34_groups_0, pad = var_34_pad_0, pad_type = var_34_pad_type_0, strides = var_34_strides_0, weight = decoder_cells_0_ih_weight_to_fp16_quantized, x = embed)[name = string("op_34_cast_fp16")];
            string var_40_pad_type_0 = const()[name = string("op_40_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_40_strides_0 = const()[name = string("op_40_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_40_pad_0 = const()[name = string("op_40_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_40_dilations_0 = const()[name = string("op_40_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_40_groups_0 = const()[name = string("op_40_groups_0"), val = int32(1)];
            tensor<fp16, [2560, 640, 1, 1]> decoder_cells_0_hh_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2560, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(1648896))), scale = tensor<fp16, [2560, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3287360))))[name = string("decoder_cells_0_hh_weight_to_fp16_quantized")];
            tensor<fp16, [16, 2560, 1, 1]> var_40_cast_fp16 = conv(dilations = var_40_dilations_0, groups = var_40_groups_0, pad = var_40_pad_0, pad_type = var_40_pad_type_0, strides = var_40_strides_0, weight = decoder_cells_0_hh_weight_to_fp16_quantized, x = input_1_cast_fp16)[name = string("op_40_cast_fp16")];
            tensor<fp16, [16, 2560, 1, 1]> gates_1_cast_fp16 = add(x = var_34_cast_fp16, y = var_40_cast_fp16)[name = string("gates_1_cast_fp16")];
            tensor<int32, [4]> var_43_begin_0 = const()[name = string("op_43_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_43_end_0 = const()[name = string("op_43_end_0"), val = tensor<int32, [4]>([16, 640, 1, 1])];
            tensor<bool, [4]> var_43_end_mask_0 = const()[name = string("op_43_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [16, 640, 1, 1]> var_43_cast_fp16 = slice_by_index(begin = var_43_begin_0, end = var_43_end_0, end_mask = var_43_end_mask_0, x = gates_1_cast_fp16)[name = string("op_43_cast_fp16")];
            tensor<fp16, [16, 640, 1, 1]> i_1_cast_fp16 = sigmoid(x = var_43_cast_fp16)[name = string("i_1_cast_fp16")];
            tensor<int32, [4]> var_46_begin_0 = const()[name = string("op_46_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_46_end_0 = const()[name = string("op_46_end_0"), val = tensor<int32, [4]>([16, 1280, 1, 1])];
            tensor<bool, [4]> var_46_end_mask_0 = const()[name = string("op_46_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [16, 640, 1, 1]> var_46_cast_fp16 = slice_by_index(begin = var_46_begin_0, end = var_46_end_0, end_mask = var_46_end_mask_0, x = gates_1_cast_fp16)[name = string("op_46_cast_fp16")];
            tensor<fp16, [16, 640, 1, 1]> f_1_cast_fp16 = sigmoid(x = var_46_cast_fp16)[name = string("f_1_cast_fp16")];
            tensor<int32, [4]> var_49_begin_0 = const()[name = string("op_49_begin_0"), val = tensor<int32, [4]>([0, 1280, 0, 0])];
            tensor<int32, [4]> var_49_end_0 = const()[name = string("op_49_end_0"), val = tensor<int32, [4]>([16, 1920, 1, 1])];
            tensor<bool, [4]> var_49_end_mask_0 = const()[name = string("op_49_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [16, 640, 1, 1]> var_49_cast_fp16 = slice_by_index(begin = var_49_begin_0, end = var_49_end_0, end_mask = var_49_end_mask_0, x = gates_1_cast_fp16)[name = string("op_49_cast_fp16")];
            tensor<fp16, [16, 640, 1, 1]> g_1_cast_fp16 = tanh(x = var_49_cast_fp16)[name = string("g_1_cast_fp16")];
            tensor<int32, [4]> var_52_begin_0 = const()[name = string("op_52_begin_0"), val = tensor<int32, [4]>([0, 1920, 0, 0])];
            tensor<int32, [4]> var_52_end_0 = const()[name = string("op_52_end_0"), val = tensor<int32, [4]>([16, 1, 1, 1])];
            tensor<bool, [4]> var_52_end_mask_0 = const()[name = string("op_52_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [16, 640, 1, 1]> var_52_cast_fp16 = slice_by_index(begin = var_52_begin_0, end = var_52_end_0, end_mask = var_52_end_mask_0, x = gates_1_cast_fp16)[name = string("op_52_cast_fp16")];
            tensor<fp16, [16, 640, 1, 1]> o_1_cast_fp16 = sigmoid(x = var_52_cast_fp16)[name = string("o_1_cast_fp16")];
            tensor<fp16, [16, 640, 1, 1]> var_54_cast_fp16 = mul(x = f_1_cast_fp16, y = c_1_cast_fp16)[name = string("op_54_cast_fp16")];
            tensor<fp16, [16, 640, 1, 1]> var_55_cast_fp16 = mul(x = i_1_cast_fp16, y = g_1_cast_fp16)[name = string("op_55_cast_fp16")];
            tensor<fp16, [16, 640, 1, 1]> c_new_1_cast_fp16 = add(x = var_54_cast_fp16, y = var_55_cast_fp16)[name = string("c_new_1_cast_fp16")];
            tensor<fp16, [16, 640, 1, 1]> var_57_cast_fp16 = tanh(x = c_new_1_cast_fp16)[name = string("op_57_cast_fp16")];
            tensor<fp16, [16, 640, 1, 1]> input_3_cast_fp16 = mul(x = o_1_cast_fp16, y = var_57_cast_fp16)[name = string("input_3_cast_fp16")];
            tensor<int32, [4]> input_5_begin_0 = const()[name = string("input_5_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> input_5_end_0 = const()[name = string("input_5_end_0"), val = tensor<int32, [4]>([16, 1, 1, 1])];
            tensor<bool, [4]> input_5_end_mask_0 = const()[name = string("input_5_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [16, 640, 1, 1]> input_5_cast_fp16 = slice_by_index(begin = input_5_begin_0, end = input_5_end_0, end_mask = input_5_end_mask_0, x = h_in)[name = string("input_5_cast_fp16")];
            tensor<int32, [4]> c_begin_0 = const()[name = string("c_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> c_end_0 = const()[name = string("c_end_0"), val = tensor<int32, [4]>([16, 1, 1, 1])];
            tensor<bool, [4]> c_end_mask_0 = const()[name = string("c_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [16, 640, 1, 1]> c_cast_fp16 = slice_by_index(begin = c_begin_0, end = c_end_0, end_mask = c_end_mask_0, x = c_in)[name = string("c_cast_fp16")];
            string var_74_pad_type_0 = const()[name = string("op_74_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_74_strides_0 = const()[name = string("op_74_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_74_pad_0 = const()[name = string("op_74_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_74_dilations_0 = const()[name = string("op_74_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_74_groups_0 = const()[name = string("op_74_groups_0"), val = int32(1)];
            tensor<fp16, [2560, 640, 1, 1]> decoder_cells_1_ih_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2560, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(3292544))), scale = tensor<fp16, [2560, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4931008))))[name = string("decoder_cells_1_ih_weight_to_fp16_quantized")];
            tensor<fp16, [2560]> decoder_cells_1_ih_bias_to_fp16 = const()[name = string("decoder_cells_1_ih_bias_to_fp16"), val = tensor<fp16, [2560]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4936192)))];
            tensor<fp16, [16, 2560, 1, 1]> var_74_cast_fp16 = conv(bias = decoder_cells_1_ih_bias_to_fp16, dilations = var_74_dilations_0, groups = var_74_groups_0, pad = var_74_pad_0, pad_type = var_74_pad_type_0, strides = var_74_strides_0, weight = decoder_cells_1_ih_weight_to_fp16_quantized, x = input_3_cast_fp16)[name = string("op_74_cast_fp16")];
            string var_80_pad_type_0 = const()[name = string("op_80_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> var_80_strides_0 = const()[name = string("op_80_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> var_80_pad_0 = const()[name = string("op_80_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> var_80_dilations_0 = const()[name = string("op_80_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 var_80_groups_0 = const()[name = string("op_80_groups_0"), val = int32(1)];
            tensor<fp16, [2560, 640, 1, 1]> decoder_cells_1_hh_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [2560, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(4941376))), scale = tensor<fp16, [2560, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6579840))))[name = string("decoder_cells_1_hh_weight_to_fp16_quantized")];
            tensor<fp16, [16, 2560, 1, 1]> var_80_cast_fp16 = conv(dilations = var_80_dilations_0, groups = var_80_groups_0, pad = var_80_pad_0, pad_type = var_80_pad_type_0, strides = var_80_strides_0, weight = decoder_cells_1_hh_weight_to_fp16_quantized, x = input_5_cast_fp16)[name = string("op_80_cast_fp16")];
            tensor<fp16, [16, 2560, 1, 1]> gates_cast_fp16 = add(x = var_74_cast_fp16, y = var_80_cast_fp16)[name = string("gates_cast_fp16")];
            tensor<int32, [4]> var_83_begin_0 = const()[name = string("op_83_begin_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [4]> var_83_end_0 = const()[name = string("op_83_end_0"), val = tensor<int32, [4]>([16, 640, 1, 1])];
            tensor<bool, [4]> var_83_end_mask_0 = const()[name = string("op_83_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [16, 640, 1, 1]> var_83_cast_fp16 = slice_by_index(begin = var_83_begin_0, end = var_83_end_0, end_mask = var_83_end_mask_0, x = gates_cast_fp16)[name = string("op_83_cast_fp16")];
            tensor<fp16, [16, 640, 1, 1]> i_cast_fp16 = sigmoid(x = var_83_cast_fp16)[name = string("i_cast_fp16")];
            tensor<int32, [4]> var_86_begin_0 = const()[name = string("op_86_begin_0"), val = tensor<int32, [4]>([0, 640, 0, 0])];
            tensor<int32, [4]> var_86_end_0 = const()[name = string("op_86_end_0"), val = tensor<int32, [4]>([16, 1280, 1, 1])];
            tensor<bool, [4]> var_86_end_mask_0 = const()[name = string("op_86_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [16, 640, 1, 1]> var_86_cast_fp16 = slice_by_index(begin = var_86_begin_0, end = var_86_end_0, end_mask = var_86_end_mask_0, x = gates_cast_fp16)[name = string("op_86_cast_fp16")];
            tensor<fp16, [16, 640, 1, 1]> f_cast_fp16 = sigmoid(x = var_86_cast_fp16)[name = string("f_cast_fp16")];
            tensor<int32, [4]> var_89_begin_0 = const()[name = string("op_89_begin_0"), val = tensor<int32, [4]>([0, 1280, 0, 0])];
            tensor<int32, [4]> var_89_end_0 = const()[name = string("op_89_end_0"), val = tensor<int32, [4]>([16, 1920, 1, 1])];
            tensor<bool, [4]> var_89_end_mask_0 = const()[name = string("op_89_end_mask_0"), val = tensor<bool, [4]>([true, false, true, true])];
            tensor<fp16, [16, 640, 1, 1]> var_89_cast_fp16 = slice_by_index(begin = var_89_begin_0, end = var_89_end_0, end_mask = var_89_end_mask_0, x = gates_cast_fp16)[name = string("op_89_cast_fp16")];
            tensor<fp16, [16, 640, 1, 1]> g_cast_fp16 = tanh(x = var_89_cast_fp16)[name = string("g_cast_fp16")];
            tensor<int32, [4]> var_92_begin_0 = const()[name = string("op_92_begin_0"), val = tensor<int32, [4]>([0, 1920, 0, 0])];
            tensor<int32, [4]> var_92_end_0 = const()[name = string("op_92_end_0"), val = tensor<int32, [4]>([16, 1, 1, 1])];
            tensor<bool, [4]> var_92_end_mask_0 = const()[name = string("op_92_end_mask_0"), val = tensor<bool, [4]>([true, true, true, true])];
            tensor<fp16, [16, 640, 1, 1]> var_92_cast_fp16 = slice_by_index(begin = var_92_begin_0, end = var_92_end_0, end_mask = var_92_end_mask_0, x = gates_cast_fp16)[name = string("op_92_cast_fp16")];
            tensor<fp16, [16, 640, 1, 1]> o_cast_fp16 = sigmoid(x = var_92_cast_fp16)[name = string("o_cast_fp16")];
            tensor<fp16, [16, 640, 1, 1]> var_94_cast_fp16 = mul(x = f_cast_fp16, y = c_cast_fp16)[name = string("op_94_cast_fp16")];
            tensor<fp16, [16, 640, 1, 1]> var_95_cast_fp16 = mul(x = i_cast_fp16, y = g_cast_fp16)[name = string("op_95_cast_fp16")];
            tensor<fp16, [16, 640, 1, 1]> c_new_cast_fp16 = add(x = var_94_cast_fp16, y = var_95_cast_fp16)[name = string("c_new_cast_fp16")];
            tensor<fp16, [16, 640, 1, 1]> var_97_cast_fp16 = tanh(x = c_new_cast_fp16)[name = string("op_97_cast_fp16")];
            tensor<fp16, [16, 640, 1, 1]> input_7_cast_fp16 = mul(x = o_cast_fp16, y = var_97_cast_fp16)[name = string("input_7_cast_fp16")];
            string pred_pad_type_0 = const()[name = string("pred_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> pred_strides_0 = const()[name = string("pred_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> pred_pad_0 = const()[name = string("pred_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> pred_dilations_0 = const()[name = string("pred_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 pred_groups_0 = const()[name = string("pred_groups_0"), val = int32(1)];
            tensor<fp16, [640, 640, 1, 1]> decoder_joint_pred_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [640, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6585024))), scale = tensor<fp16, [640, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6994688))))[name = string("decoder_joint_pred_weight_to_fp16_quantized")];
            tensor<fp16, [640]> decoder_joint_pred_bias_to_fp16 = const()[name = string("decoder_joint_pred_bias_to_fp16"), val = tensor<fp16, [640]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6996032)))];
            tensor<fp16, [16, 640, 1, 1]> pred_cast_fp16 = conv(bias = decoder_joint_pred_bias_to_fp16, dilations = pred_dilations_0, groups = pred_groups_0, pad = pred_pad_0, pad_type = pred_pad_type_0, strides = pred_strides_0, weight = decoder_joint_pred_weight_to_fp16_quantized, x = input_7_cast_fp16)[name = string("pred_cast_fp16")];
            bool var_110_interleave_0 = const()[name = string("op_110_interleave_0"), val = bool(false)];
            tensor<fp16, [16, 1280, 1, 1]> h_out = concat(axis = var_14, interleave = var_110_interleave_0, values = (input_3_cast_fp16, input_7_cast_fp16))[name = string("op_110_cast_fp16")];
            bool var_112_interleave_0 = const()[name = string("op_112_interleave_0"), val = bool(false)];
            tensor<fp16, [16, 1280, 1, 1]> c_out = concat(axis = var_14, interleave = var_112_interleave_0, values = (c_new_1_cast_fp16, c_new_cast_fp16))[name = string("op_112_cast_fp16")];
            tensor<fp16, [16, 640, 1, 8]> var_122_cast_fp16 = add(x = enc_step, y = pred_cast_fp16)[name = string("op_122_cast_fp16")];
            tensor<fp16, [16, 640, 1, 8]> input_cast_fp16 = relu(x = var_122_cast_fp16)[name = string("input_cast_fp16")];
            string x_pad_type_0 = const()[name = string("x_pad_type_0"), val = string("valid")];
            tensor<int32, [2]> x_strides_0 = const()[name = string("x_strides_0"), val = tensor<int32, [2]>([1, 1])];
            tensor<int32, [4]> x_pad_0 = const()[name = string("x_pad_0"), val = tensor<int32, [4]>([0, 0, 0, 0])];
            tensor<int32, [2]> x_dilations_0 = const()[name = string("x_dilations_0"), val = tensor<int32, [2]>([1, 1])];
            int32 x_groups_0 = const()[name = string("x_groups_0"), val = int32(1)];
            tensor<fp16, [8198, 640, 1, 1]> joint_out_weight_to_fp16_quantized = constexpr_blockwise_shift_scale(data = tensor<int8, [8198, 640, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(6997376))), scale = tensor<fp16, [8198, 1, 1, 1]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12244160))))[name = string("joint_out_weight_to_fp16_quantized")];
            tensor<fp16, [8198]> joint_out_bias_to_fp16 = const()[name = string("joint_out_bias_to_fp16"), val = tensor<fp16, [8198]>(BLOBFILE(path = string("@model_path/weights/weight.bin"), offset = uint64(12260672)))];
            tensor<fp16, [16, 8198, 1, 8]> logits = conv(bias = joint_out_bias_to_fp16, dilations = x_dilations_0, groups = x_groups_0, pad = x_pad_0, pad_type = x_pad_type_0, strides = x_strides_0, weight = joint_out_weight_to_fp16_quantized, x = input_cast_fp16)[name = string("x_cast_fp16")];
        } -> (logits, h_out, c_out);
}