File size: 42,889 Bytes
032e687
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
# Copyright (c) OpenMMLab. All rights reserved.
import math
import os.path as osp
import warnings
from collections import OrderedDict
from typing import List, Optional
import torch.nn.functional as F
import torch
import torch.nn as nn
from accelerate import init_empty_weights
from mmengine import print_log
from mmengine.config import Config, ConfigDict
from mmengine.model import BaseModel
from peft import get_peft_model, prepare_model_for_kbit_training
from transformers import (AddedToken, AutoConfig, CLIPImageProcessor, PreTrainedModel,
                          CLIPVisionModel, LlamaForCausalLM,
                          LlamaTokenizerFast, LlavaConfig,
                          LlavaForConditionalGeneration, LlavaProcessor)
from transformers.integrations import is_deepspeed_zero3_enabled

from xtuner.registry import BUILDER
from xtuner.model.modules.dispatch import SUPPORT_FLASH1, SUPPORT_FLASH2
from xtuner.model.utils import (LoadWoInit, find_all_linear_names,
                                get_peft_model_state_dict, guess_load_checkpoint,
                                make_inputs_require_grad, traverse_dict)
from xtuner.utils import IGNORE_INDEX, IMAGE_TOKEN_INDEX
from xtuner.tools.utils import get_stop_criteria, is_cn_string
from transformers import GenerationConfig
from xtuner.utils import (DEFAULT_IMAGE_TOKEN, IMAGE_TOKEN_INDEX,
                          PROMPT_TEMPLATE)


def convert_state_dict_to_hf(state_dict, mapping):
    new_state_dict = {}
    for key, value in state_dict.items():
        if key.endswith('.inv_freq'):
            continue
        for key_to_modify, new_key in mapping.items():
            if key_to_modify in key:
                key = key.replace(key_to_modify, new_key)
        new_state_dict[key] = value
    return new_state_dict


class SingleLLaVAModelSFT(BaseModel):

    def __init__(self,
                 llm,
                 visual_encoder=None,
                 tokenizer=None,
                 freeze_llm=False,
                 freeze_visual_encoder=False,
                 visual_select_layer=-2,
                 pretrained_pth=None,
                 projector_depth=0,
                 llm_lora=None,
                 visual_encoder_lora=None,
                 use_activation_checkpointing=True,
                 max_position_embeddings=None,
                 add_cls_token=False,
                 template=None,
                 ):
        super().__init__()

        if tokenizer is not None:
            self.tokenizer = tokenizer
            tokenizer_type = self.tokenizer['type']
            del self.tokenizer['type']
            self.tokenizer = tokenizer_type(**self.tokenizer)

        self.freeze_llm = freeze_llm
        self.freeze_visual_encoder = freeze_visual_encoder
        with LoadWoInit():
            if isinstance(llm, dict):
                llm = self._dispatch_lm_model_cfg(llm, max_position_embeddings)

            self.llm = self._build_from_cfg_or_module(llm)

            if visual_encoder is not None:
                self.visual_encoder = self._build_from_cfg_or_module(
                    visual_encoder)
            else:
                self.visual_encoder = None

        self.llm.config.use_cache = False

        # dispatch_modules(self.llm)

        self.projector_depth = projector_depth
        # projector_config = ProjectorConfig(
        #     visual_hidden_size=self.visual_encoder.config.hidden_size,
        #     llm_hidden_size=self.llm.config.hidden_size,
        #     depth=self.projector_depth)
        # self.projector = ProjectorModel(projector_config).to(
        #     self.visual_encoder.dtype)
        self.projector = None

        if self.freeze_llm:
            self.llm.requires_grad_(False)
        if self.freeze_visual_encoder:
            if self.visual_encoder is not None:
                self.visual_encoder.requires_grad_(False)

        if use_activation_checkpointing:
            # For backward compatibility
            if hasattr(self.llm, 'enable_input_require_grads'):
                self.llm.enable_input_require_grads()
            else:
                self.llm.get_input_embeddings().register_forward_hook(
                    make_inputs_require_grad)

            # enable gradient (activation) checkpointing for memory efficiency
            self.gradient_checkpointing_enable()

        self.use_llm_lora = llm_lora is not None
        self.use_visual_encoder_lora = visual_encoder_lora is not None

        if self.use_llm_lora:
            self._prepare_llm_for_lora(llm_lora, use_activation_checkpointing)
        if self.use_visual_encoder_lora:
            self._prepare_visual_encoder_for_lora(
                visual_encoder_lora, use_activation_checkpointing)

        if pretrained_pth is not None:
            pretrained_state_dict = guess_load_checkpoint(pretrained_pth)

            self.load_state_dict(pretrained_state_dict, strict=False)
            print_log(f'Load pretrained weight from {pretrained_pth}',
                      'current')

        self.visual_select_layer = visual_select_layer

        self._is_init = True

        self.is_first_iter = True

        self.add_cls_token = add_cls_token

        self.template = template

    def _parse_lora_config(self, lora_config):
        if isinstance(lora_config, dict) or isinstance(
                lora_config, Config) or isinstance(lora_config, ConfigDict):
            lora_config = BUILDER.build(lora_config)
        return lora_config

    def _prepare_llm_for_lora(self,
                              lora_config,
                              use_activation_checkpointing=True):
        lora_config = self._parse_lora_config(lora_config)
        self.llm = prepare_model_for_kbit_training(
            self.llm, use_activation_checkpointing)
        if lora_config.target_modules is None:
            modules = find_all_linear_names(self.llm)
            lora_config.target_modules = modules
        self.llm = get_peft_model(self.llm, lora_config)

    def _prepare_visual_encoder_for_lora(self,
                                         lora_config,
                                         use_activation_checkpointing=True):
        lora_config = self._parse_lora_config(lora_config)
        if lora_config.target_modules is None:
            modules = find_all_linear_names(self.visual_encoder)
            lora_config.target_modules = modules
        self.visual_encoder = get_peft_model(self.visual_encoder, lora_config)

    def gradient_checkpointing_enable(self):
        self.activation_checkpointing_enable()

    def activation_checkpointing_enable(self):
        self.llm.gradient_checkpointing_enable()
        if self.visual_encoder is not None:
            self.visual_encoder.gradient_checkpointing_enable()
        if self.projector is not None:
            self.projector.gradient_checkpointing_enable()

    def gradient_checkpointing_disable(self):
        self.activation_checkpointing_disable()

    def activation_checkpointing_disable(self):
        self.llm.gradient_checkpointing_disable()
        if self.visual_encoder is not None:
            self.visual_encoder.gradient_checkpointing_disable()
        if self.projector is not None:
            self.projector.gradient_checkpointing_disable()

    def init_weights(self):
        pass

    def state_dict(self, *args, **kwargs):
        state_dict = super().state_dict(*args, **kwargs)
        to_return = OrderedDict()
        # Step 1. visual_encoder
        if self.use_visual_encoder_lora:
            to_return.update(
                get_peft_model_state_dict(
                    self.visual_encoder, state_dict=state_dict))
        elif not self.freeze_visual_encoder:
            to_return.update({
                k: v
                for k, v in state_dict.items() if 'visual_encoder.' in k
            })
        # Step 2. LLM
        if self.use_llm_lora:
            to_return.update(
                get_peft_model_state_dict(self.llm, state_dict=state_dict))
        elif not self.freeze_llm:
            to_return.update(
                {k: v
                 for k, v in state_dict.items() if 'llm.' in k})
        # Step 3. Projector
        to_return.update(
            {k: v
             for k, v in state_dict.items() if 'projector.' in k})
        return to_return

    @staticmethod
    def _prepare_for_long_context_training(cfg, llm_cfg,
                                           max_position_embeddings):

        orig_rope_scaling = getattr(llm_cfg, 'rope_scaling', None)
        if orig_rope_scaling is None:
            orig_rope_scaling = {'factor': 1}

        orig_rope_scaling_factor = orig_rope_scaling[
            'factor'] if 'factor' in orig_rope_scaling.keys() else 1
        orig_ctx_len = getattr(llm_cfg, 'max_position_embeddings', None)
        if orig_ctx_len:
            orig_ctx_len *= orig_rope_scaling_factor
            if max_position_embeddings > orig_ctx_len:
                scaling_factor = float(
                    math.ceil(max_position_embeddings / orig_ctx_len))
                llm_cfg.rope_scaling = {
                    'type': 'linear',
                    'factor': scaling_factor
                }

        # hardcode for internlm2
        llm_cfg.attn_implementation = 'flash_attention_2'
        cfg.config = llm_cfg

        return cfg, llm_cfg

    @staticmethod
    def _prepare_for_flash_attn(cfg, llm_cfg):
        cls_name = type(llm_cfg).__name__
        SUPPORT_SDPA_ATTN = ('LlamaConfig', 'GemmaConfig', 'MistralConfig',
                             'MixtralConfig', 'Qwen2Config', 'Qwen2MoeConfig',
                             'Starcoder2Config', 'Starcoder2Config',
                             'Phi3Config')
        SUPPORT_FLASH_ATTN2 = ('InternLM2Config', 'LlamaConfig', 'GemmaConfig',
                               'MistralConfig', 'MixtralConfig', 'Qwen2Config',
                               'Qwen2MoeConfig', 'Starcoder2Config',
                               'Starcoder2Config', 'Phi3Config')

        torch_dtype = torch.bfloat16 if (
                torch.cuda.is_available() and torch.cuda.is_bf16_supported()) \
            else torch.float16

        if getattr(cfg, 'attn_implementation', None) is not None:
            # Flash Attention 2.0 only supports torch.float16 and
            # torch.bfloat16 dtypes
            if cfg.attn_implementation == 'flash_attention_2':
                cfg.torch_dtype = torch_dtype
        elif SUPPORT_FLASH2 and cls_name in SUPPORT_FLASH_ATTN2:
            cfg.torch_dtype = torch_dtype
            cfg.attn_implementation = 'flash_attention_2'
        elif SUPPORT_FLASH1 and cls_name in SUPPORT_SDPA_ATTN:
            cfg.attn_implementation = 'sdpa'

        return cfg, llm_cfg

    @staticmethod
    def _prepare_for_qlora_zero3(cfg):
        if (not is_deepspeed_zero3_enabled()) or (not hasattr(
                cfg, 'quantization_config')):
            return cfg

        torch_dtype = torch.bfloat16 if (
                torch.cuda.is_available() and torch.cuda.is_bf16_supported()) \
            else torch.float16

        cfg.torch_dtype = torch_dtype
        quantization_config = cfg.quantization_config
        quantization_config.bnb_4bit_compute_dtype = torch_dtype
        quantization_config.bnb_4bit_quant_storage = torch_dtype

        return cfg

    def _dispatch_lm_model_cfg(self, cfg, max_position_embeddings=None):
        cfg = self._prepare_for_qlora_zero3(cfg)
        pretrained_model_name_or_path = cfg.pretrained_model_name_or_path
        llm_cfg = AutoConfig.from_pretrained(
            pretrained_model_name_or_path, trust_remote_code=True)
        cfg, llm_cfg = self._prepare_for_flash_attn(cfg, llm_cfg)
        if max_position_embeddings is not None:
            cfg, llm_cfg = self._prepare_for_long_context_training(
                cfg, llm_cfg, max_position_embeddings)
        return cfg

    def _build_from_cfg_or_module(self, cfg_or_mod):
        if isinstance(cfg_or_mod, nn.Module):
            return cfg_or_mod
        elif isinstance(cfg_or_mod, dict):
            traverse_dict(cfg_or_mod)
            return BUILDER.build(cfg_or_mod)
        else:
            raise NotImplementedError

    def forward(self, data, data_samples=None, mode='loss'):
        if self.is_first_iter:
            # hardcode for qlora DeepSpeed ZeRO3, put buffers and QuantState to
            # device
            # Only required in `LLaVAModel` .
            # We do not need this in `SupervisedFinetune` .
            self.to(data['input_ids'].device)
            self.is_first_iter = False

        # if 'pixel_values' in data:
        #     # no visual encoder
        #     # visual_outputs = self.visual_encoder(
        #     #     data['pixel_values'].to(self.visual_encoder.dtype),
        #     #     output_hidden_states=True)
        #     # pixel_values = self.projector(
        #     #     visual_outputs.hidden_states[self.visual_select_layer][:, 1:])
        #     # data['pixel_values'] = pixel_values
        #     # only merge the image and text tokens.
        #     data = prepare_inputs_labels_for_multimodal_solo(llm=self.llm, tokenizer=self.tokenizer, **data,
        #                                                      add_CLS=self.add_cls_token)
        data = prepare_inputs_labels_for_multimodal_solo(llm=self.llm, tokenizer=self.tokenizer, **data,
                                                         add_CLS=self.add_cls_token)

        if mode == 'loss':
            loss = self.compute_loss(data, data_samples)
            if torch.isnan(loss["loss"]):
                print("loss nan here")
            return loss
        elif mode == 'predict':
            return self.predict(data, data_samples)
        elif mode == 'tensor':
            return self._forward(data, data_samples)
        else:
            raise NotImplementedError

    def _forward(self, data, data_samples=None):

        outputs = self.llm(**data)

        return outputs

    def predict(self, data, data_samples=None):
        outputs = self.llm(**data)
        logits_dict = [{'logits': logits} for logits in outputs.logits]
        return logits_dict

    def compute_loss(self, data, data_samples=None):
        outputs = self.llm(**data)
        loss_dict = {'loss': outputs.loss}
        return loss_dict

    def __getattr__(self, name: str):
        # return super().__getattr__(name)
        try:
            return super().__getattr__(name)
        except AttributeError:
            return getattr(self.llm, name)

    def to_hf(self,
              cfg,
              save_dir,
              fp32=False,
              save_pretrained_kwargs={},
              save_format='xtuner',
              **kwargs):
        if save_format == 'xtuner':
            self.to_xtuner_llava(cfg, save_dir, fp32, save_pretrained_kwargs)
        elif save_format == 'huggingface':
            self.to_huggingface_llava(cfg, save_dir, fp32,
                                      save_pretrained_kwargs)
        elif save_format == 'official':
            self.to_official_llava(cfg, save_dir, fp32, save_pretrained_kwargs)
        else:
            raise NotImplementedError

    def to_xtuner_llava(self,
                        cfg,
                        save_dir,
                        fp32=False,
                        save_pretrained_kwargs={}):
        # LLM
        self.llm.config.use_cache = True
        if not fp32:
            print_log('Convert LLM to float16', 'current')
            self.llm.half()
        if self.use_llm_lora:
            llm_path = osp.join(save_dir, 'llm_adapter')
            print_log(f'Saving LLM adapter to {llm_path}', 'current')
            self.llm.save_pretrained(llm_path, **save_pretrained_kwargs)
        elif not self.freeze_llm:
            llm_path = save_dir
            print_log(f'Saving LLM tokenizer to {llm_path}', 'current')
            tokenizer = BUILDER.build(cfg.tokenizer)
            tokenizer.save_pretrained(llm_path, **save_pretrained_kwargs)
            print_log(f'Saving LLM to {llm_path}', 'current')
            self.llm.save_pretrained(llm_path, **save_pretrained_kwargs)
        self.llm.config.use_cache = False

        # Visual Encoder
        if self.use_visual_encoder_lora:
            visual_encoder_path = osp.join(save_dir, 'visual_encoder_adapter')
            print_log(
                f'Saving visual_encoder adapter to {visual_encoder_path}',
                'current')
            self.visual_encoder.save_pretrained(visual_encoder_path,
                                                **save_pretrained_kwargs)
        elif not self.freeze_visual_encoder:
            visual_encoder_path = osp.join(save_dir, 'visual_encoder')
            print_log(
                'Saving visual_encoder image_processor to'
                f'{visual_encoder_path}', 'current')
            image_processor = BUILDER.build(cfg.image_processor)
            image_processor.save_pretrained(visual_encoder_path,
                                            **save_pretrained_kwargs)
            print_log(f'Saving visual_encoder to {visual_encoder_path}',
                      'current')
            self.visual_encoder.save_pretrained(visual_encoder_path,
                                                **save_pretrained_kwargs)

        # Projector
        projector_path = osp.join(save_dir, 'projector')
        print_log(f'Saving projector to {projector_path}', 'current')
        self.projector.save_pretrained(projector_path,
                                       **save_pretrained_kwargs)

    def to_huggingface_llava(self,
                             cfg,
                             save_dir,
                             fp32=False,
                             save_pretrained_kwargs={}):

        LLM_MAPPING = {
            'model': 'language_model.model',
            'lm_head': 'language_model.lm_head',
        }
        VIT_MAPPING = {
            'vision_model': 'vision_tower.vision_model',
        }
        PROJECTOR_MAPPING = {
            'model.0': 'multi_modal_projector.linear_1',
            'model.2': 'multi_modal_projector.linear_2',
        }

        assert getattr(self.llm, 'hf_quantizer', None) is None, \
            'This conversion format does not support quantized LLM.'

        # get state_dict
        llm = self.llm
        if self.use_llm_lora:
            llm = self.llm.merge_and_unload()
        llm.config.use_cache = True
        if not fp32:
            print_log('Convert LLM to float16', 'current')
            llm.half()

        assert isinstance(llm, LlamaForCausalLM), \
            'This conversion format only supports LlamaForCausalLM.'
        llm_state_dict = llm.state_dict()
        llm_state_dict = convert_state_dict_to_hf(llm_state_dict, LLM_MAPPING)

        need_visual_encoder = (not self.freeze_visual_encoder
                               or self.use_visual_encoder_lora)
        visual_encoder = self.visual_encoder
        if self.use_visual_encoder_lora:
            visual_encoder = self.visual_encoder.merge_and_unload()
        assert isinstance(visual_encoder, CLIPVisionModel), \
            'This conversion format only supports CLIPVisionModel.'
        if need_visual_encoder:
            visual_encoder_state_dict = visual_encoder.state_dict()
            visual_encoder_state_dict = convert_state_dict_to_hf(
                visual_encoder_state_dict, VIT_MAPPING)
        else:
            visual_encoder_state_dict = {}

        projector_state_dict = self.projector.state_dict()
        projector_state_dict = convert_state_dict_to_hf(
            projector_state_dict, PROJECTOR_MAPPING)

        state_dict = {
            **projector_state_dict,
            **llm_state_dict,
            **visual_encoder_state_dict
        }

        # init model
        text_config = llm.config
        vision_config = visual_encoder.config
        config = LlavaConfig(
            text_config=text_config,
            vision_config=vision_config,
            attn_implementation='eager')

        with init_empty_weights():
            with warnings.catch_warnings():
                warnings.filterwarnings(
                    'ignore', message='.*non-meta.*', category=UserWarning)
                model = LlavaForConditionalGeneration(config)
        model.load_state_dict(state_dict, strict=True, assign=True)

        # processor
        cfg.tokenizer.type = LlamaTokenizerFast.from_pretrained
        tokenizer = BUILDER.build(cfg.tokenizer)

        tokenizer.add_tokens(
            AddedToken(DEFAULT_IMAGE_TOKEN, special=True, normalized=False),
            special_tokens=True)
        tokenizer.add_special_tokens({'pad_token': '<pad>'})

        image_processor = BUILDER.build(cfg.image_processor)
        assert isinstance(image_processor, CLIPImageProcessor), \
            'This conversion format only supports CLIPImageProcessor.'

        processor = LlavaProcessor(
            tokenizer=tokenizer, image_processor=image_processor)

        # Pad to 64 for performance reasons
        pad_shape = 64

        pre_expansion_embeddings = \
            model.language_model.model.embed_tokens.weight.data
        mu = torch.mean(pre_expansion_embeddings, dim=0).float()
        n = pre_expansion_embeddings.size()[0]
        sigma = ((pre_expansion_embeddings - mu).T
                 @ (pre_expansion_embeddings - mu)) / n
        dist = torch.distributions.multivariate_normal.MultivariateNormal(
            mu, covariance_matrix=1e-5 * sigma)

        # We add an image token so we need to resize the model
        ori_vocab_size = config.text_config.vocab_size
        tokenizer_vocab_size = tokenizer.encode('<pad>')[-1]
        added_token = tokenizer_vocab_size - ori_vocab_size

        if added_token > 0:
            model.resize_token_embeddings(ori_vocab_size + added_token,
                                          pad_shape)
            model.language_model.model.embed_tokens.weight.data[
            ori_vocab_size:] = torch.stack(
                tuple(
                    dist.sample()
                    for _ in range(model.language_model.model.embed_tokens.
                                   weight.data[ori_vocab_size:].shape[0])),
                dim=0,
            )
            model.language_model.lm_head.weight.data[
            ori_vocab_size:] = torch.stack(
                tuple(dist.sample()
                      for _ in range(model.language_model.lm_head.weight.
                                     data[ori_vocab_size:].shape[0])),
                dim=0,
            )
        model.config.image_token_index = tokenizer.encode(
            DEFAULT_IMAGE_TOKEN)[-1]
        model.config.pad_token_id = tokenizer.encode('<pad>')[-1]

        # save
        print_log(f'Saving to {save_dir}', 'current')
        model.save_pretrained(save_dir, **save_pretrained_kwargs)
        processor.save_pretrained(save_dir, **save_pretrained_kwargs)

    def to_official_llava(self,
                          cfg,
                          save_dir,
                          fp32=False,
                          save_pretrained_kwargs={}):

        VIT_MAPPING = {
            'vision_model': 'model.vision_tower.vision_tower.vision_model',
        }
        PROJECTOR_MAPPING = {
            'model.0': 'model.mm_projector.0',
            'model.2': 'model.mm_projector.2',
        }

        try:
            from llava.model import LlavaConfig, LlavaLlamaForCausalLM
        except ImportError:
            raise ImportError(
                'Please install llava with '
                '`pip install git+https://github.com/haotian-liu/LLaVA.git '
                '--no-deps`.')

        assert getattr(self.llm, 'hf_quantizer', None) is None, \
            'This conversion format does not support quantized LLM.'

        # get state_dict
        llm = self.llm
        if self.use_llm_lora:
            llm = self.llm.merge_and_unload()
        llm.config.use_cache = True
        if not fp32:
            print_log('Convert LLM to float16', 'current')
            llm.half()

        assert isinstance(llm, LlamaForCausalLM), \
            'This conversion format only supports LlamaForCausalLM.'
        llm_state_dict = llm.state_dict()

        need_visual_encoder = (not self.freeze_visual_encoder
                               or self.use_visual_encoder_lora)
        visual_encoder = self.visual_encoder
        if self.use_visual_encoder_lora:
            visual_encoder = self.visual_encoder.merge_and_unload()
        assert isinstance(visual_encoder, CLIPVisionModel), \
            'This conversion format only supports CLIPVisionModel.'
        if need_visual_encoder:
            visual_encoder_state_dict = visual_encoder.state_dict()
            visual_encoder_state_dict = convert_state_dict_to_hf(
                visual_encoder_state_dict, VIT_MAPPING)
        else:
            visual_encoder_state_dict = {}

        projector_state_dict = self.projector.state_dict()
        projector_state_dict = convert_state_dict_to_hf(
            projector_state_dict, PROJECTOR_MAPPING)

        state_dict = {
            **projector_state_dict,
            **llm_state_dict,
            **visual_encoder_state_dict
        }

        # init model
        tokenizer = BUILDER.build(cfg.tokenizer)
        image_processor = BUILDER.build(cfg.image_processor)
        assert isinstance(image_processor, CLIPImageProcessor), \
            'This conversion format only supports CLIPImageProcessor.'

        llava_config_dict = llm.config.__dict__.copy()
        llava_config_dict.update(
            dict(
                image_aspect_ratio='pad',
                mm_hidden_size=visual_encoder.config.hidden_size,
                mm_projector_type=f'mlp{self.projector_depth}x_gelu',
                mm_use_im_patch_token=False,
                mm_use_im_start_end=False,
                mm_vision_select_feature='patch',
                mm_vision_select_layer=self.visual_select_layer,
                mm_vision_tower=visual_encoder.config.name_or_path,
                unfreeze_mm_vision_tower=need_visual_encoder,
                model_type='llava',
                use_cache=True,
                use_mm_proj=True))

        llava_config = LlavaConfig(**llava_config_dict)

        with init_empty_weights():
            with warnings.catch_warnings():
                warnings.filterwarnings(
                    'ignore', message='.*non-meta.*', category=UserWarning)
                model = LlavaLlamaForCausalLM(llava_config)

        model.load_state_dict(state_dict, strict=True, assign=True)

        # save
        print_log(f'Saving to {save_dir}', 'current')

        model.save_pretrained(save_dir, **save_pretrained_kwargs)
        image_processor.save_pretrained(save_dir, **save_pretrained_kwargs)
        tokenizer.save_pretrained(save_dir, **save_pretrained_kwargs)

    def preparing_for_generation(self, metainfo):
        # set stop criteria and generation configs for model
        assert hasattr(self, 'tokenizer'), "The Model does not have the tokenizer!!!"
        self.bot_name = 'BOT'
        # template = PROMPT_TEMPLATE['mistral']
        # self.template = template
        stop_words = []
        stop_words += self.template.get('STOP_WORDS', [])
        stop_criteria = get_stop_criteria(
            tokenizer=self.tokenizer, stop_words=stop_words)
        self.stop_criteria = stop_criteria

        default_generation_kwargs = dict(
            # keep the max tokens
            max_new_tokens=2048,
            do_sample=False,
            eos_token_id=self.tokenizer.eos_token_id,
            pad_token_id=(
                self.tokenizer.pad_token_id
                if self.tokenizer.pad_token_id is not None
                else self.tokenizer.eos_token_id
            ),
        )
        default_generation_kwargs.update(metainfo.get('generation_kwargs', {}))
        self.gen_config = GenerationConfig(**default_generation_kwargs)
        self.init_prediction_config = True
        return

    def predict_forward(
            self, pixel_values, text_prompts, **kwargs):
        # pixel_values: image tensor
        # text_prompts: question without template
        text_prompts = text_prompts.replace('<image>\n', '').strip()
        # print("text_prompt: ", text_prompts)
        assert self.init_prediction_config, "Please set prediction configs using self.preparing_for_generation()"
        # add template for text
        input_text = ''
        input_text += self.template['INSTRUCTION'].format(
            input=text_prompts, round=1, bot_name=self.bot_name)
        # input_text = '<image>' + input_text
        cur_encode = self.tokenizer.encode(input_text)

        # chunk_encode = []
        # for idx, chunk in enumerate(input_text.split(DEFAULT_IMAGE_TOKEN)):
        #     # if idx == 0:
        #     #     cur_encode = self.tokenizer.encode(chunk)
        #     # else:
        #     #     cur_encode = self.tokenizer.encode(chunk, add_special_tokens=False)
        #     cur_encode = self.tokenizer.encode(chunk)
        #     chunk_encode.append(cur_encode)
        #
        # assert len(chunk_encode) == 2
        ids = [IMAGE_TOKEN_INDEX]
        ids.extend(cur_encode)
        # for idx, cur_chunk_encode in enumerate(chunk_encode):
        #     ids.extend(cur_chunk_encode)
        #     if idx != len(chunk_encode) - 1:
        #         ids.append(IMAGE_TOKEN_INDEX)
        ids = torch.tensor(ids).cuda().unsqueeze(0)

        pixel_values = pixel_values.cuda().unsqueeze(0)
        # print(torch.max(pixel_values), '   ', torch.min(pixel_values))
        h, w = pixel_values.shape[-2:]
        if max(h, w) > 1024:
            if h > w:
                h_new = 1024
                w_new = int(w * h_new / h)
                w_new = pad_32(w_new)
            else:
                w_new = 1024
                h_new = int(h * w_new / w)
                h_new = pad_32(h_new)
        else:
            h_new = pad_32(h)
            w_new = pad_32(w)
        dtype = pixel_values.dtype
        pixel_values = F.interpolate(pixel_values.to(torch.float32),
                                     size=(h_new, w_new), mode='bilinear',
                                     align_corners=False).to(dtype)

        mm_inputs = prepare_inputs_labels_for_multimodal_solo(
            llm=self.llm,
            tokenizer=self.tokenizer,
            input_ids=ids,
            pixel_values=pixel_values)

        if 'input_ids' in mm_inputs.keys() and mm_inputs['input_ids'] is not None:
            inp_length = mm_inputs['input_ids'].shape[1]
        else:
            inp_length = 0
        # print(inp_length)
        generate_output = self.llm.generate(
            **mm_inputs,
            generation_config=self.gen_config,
            streamer=None,
            bos_token_id=self.tokenizer.bos_token_id,
            stopping_criteria=self.stop_criteria,
            output_hidden_states=False,
            return_dict_in_generate=True
        )
        # predict = self.tokenizer.decode(
        #     generate_output.sequences[0], skip_special_tokens=False)
        # print(predict)
        # print('\n\n\n', '--------------------------------')
        predict = self.tokenizer.decode(
            generate_output.sequences[0][inp_length:], skip_special_tokens=True).strip()
        print(predict)
        return {'prediction': predict}


def prepare_inputs_labels_for_multimodal_solo(
        llm: PreTrainedModel,
        tokenizer=None,
        input_ids: torch.LongTensor = None,
        position_ids: Optional[torch.LongTensor] = None,
        attention_mask: Optional[torch.Tensor] = None,
        past_key_values: Optional[List[torch.FloatTensor]] = None,
        labels: Optional[torch.LongTensor] = None,
        pixel_values: Optional[torch.FloatTensor] = None,
        add_CLS: bool = False,
):  # (1, 3, 1024, 1024)

    ori_input_ids = input_ids

    # no image data
    if pixel_values is None:
        return {
            'input_ids': input_ids,
            'position_ids': position_ids,
            'attention_mask': attention_mask,
            'past_key_values': past_key_values,
            'inputs_embeds': None,
            'labels': labels
        }
    # image data
    _labels = labels  # (B, N)
    _position_ids = position_ids
    _attention_mask = attention_mask

    vision_patch_indices = []
    vision_patches = []
    visual_tokens = []

    patch_size = 32
    NON_VISION_TOKEN = -1
    if isinstance(pixel_values, torch.Tensor):
        assert pixel_values.shape[0] == 1
        pixel_values = pixel_values[0]
        patches = pixel_values.unfold(1, patch_size, patch_size) \
            .unfold(2, patch_size, patch_size)
        patches = patches.permute(1, 2, 0, 3, 4).contiguous()  # -> (N_H_PATCHES, N_W_PATCHES, C, PATCH_H, PATCH_W)

        n_rows, n_cols = patches.shape[:2]
        n_patches = n_rows * n_cols
        patches = patches.view(n_patches, -1)

        img_tokens = ["<vision>"]
        cur_patch_indices = [NON_VISION_TOKEN]
        for row_idx in range(n_rows):
            for col_idx in range(n_cols):
                if row_idx != 0 and col_idx == 0:  # when new row starts
                    img_tokens.append(f"<vrow_sep>")
                    cur_patch_indices.append(NON_VISION_TOKEN)
                img_tokens.append(f"<vpatch>")
                cur_patch_indices.append(row_idx * n_cols + col_idx)
        # note we use </vision> for consistency.
        img_tokens.append("</vision>")
        cur_patch_indices.append(NON_VISION_TOKEN)

        if add_CLS:
            # add cls token to align pretrain
            img_tokens.append("<|vis_cls|>")
            cur_patch_indices.append(NON_VISION_TOKEN)

        cur_tokens = torch.Tensor(tokenizer.convert_tokens_to_ids(img_tokens, ))
        assert len(cur_tokens) == len(cur_patch_indices), f"{len(cur_tokens)} != {len(cur_patch_indices)}"

        vision_patch_indices.append(torch.Tensor(cur_patch_indices).to(ori_input_ids))
        vision_patches.append(patches.to(pixel_values.dtype))
        visual_tokens.append(cur_tokens)

    else:
        for pixel_value in pixel_values:
            per_image_patches = pixel_value.unfold(1, patch_size, patch_size) \
                .unfold(2, patch_size, patch_size)
            per_image_patches = per_image_patches.permute(1, 2, 0, 3,
                                                          4).contiguous()  # -> (N_H_PATCHES, N_W_PATCHES, C, PATCH_H, PATCH_W)
            n_rows, n_cols = per_image_patches.shape[:2]
            n_patches = n_rows * n_cols
            per_image_patches = per_image_patches.view(n_patches, -1)

            img_tokens = ["<vision>"]
            cur_patch_indices = [NON_VISION_TOKEN]
            for row_idx in range(n_rows):
                for col_idx in range(n_cols):
                    if row_idx != 0 and col_idx == 0:  # when new row starts
                        img_tokens.append(f"<vrow_sep>")
                        cur_patch_indices.append(NON_VISION_TOKEN)
                    img_tokens.append(f"<vpatch>")
                    cur_patch_indices.append(row_idx * n_cols + col_idx)

            # note we use </vision>
            img_tokens.append("</vision>")
            cur_patch_indices.append(NON_VISION_TOKEN)

            if add_CLS:
                # add cls token to align pretrain
                img_tokens.append("<|vis_cls|>")
                cur_patch_indices.append(NON_VISION_TOKEN)

            cur_tokens = torch.Tensor(tokenizer.convert_tokens_to_ids(img_tokens, ))
            assert len(cur_tokens) == len(cur_patch_indices), f"{len(cur_tokens)} != {len(cur_patch_indices)}"

            vision_patch_indices.append(torch.Tensor(cur_patch_indices).to(ori_input_ids))
            vision_patches.append(per_image_patches.to(pixel_value.dtype))
            visual_tokens.append(cur_tokens)

    # for support multi batch
    prefix_num = 0
    for i in range(len(vision_patch_indices)):
        vision_patch_indices[i] = vision_patch_indices[i] + prefix_num
        prefix_num += len(vision_patches[i])
    vision_patches = torch.cat(vision_patches, dim=0)

    if attention_mask is None:
        attention_mask = torch.ones_like(input_ids, dtype=torch.bool)
    else:
        attention_mask = attention_mask.bool()
    if position_ids is None:
        position_ids = torch.arange(
            0, input_ids.shape[1], dtype=torch.long, device=input_ids.device)
    if labels is None:
        labels = torch.full_like(input_ids, IGNORE_INDEX)

    # remove the padding using attention_mask -- TODO: double check
    input_ids = [
        cur_input_ids[cur_attention_mask]
        for cur_input_ids, cur_attention_mask in zip(input_ids, attention_mask)
    ]
    labels = [
        cur_labels[cur_attention_mask]
        for cur_labels, cur_attention_mask in zip(labels, attention_mask)
    ]

    new_inputs_ids = []
    new_vision_ids = []
    new_labels = []
    cur_image_idx = 0

    for batch_idx, cur_input_ids in enumerate(input_ids):
        num_images = (cur_input_ids == IMAGE_TOKEN_INDEX).sum()
        if num_images == 0:
            new_inputs_ids.append(cur_input_ids)
            new_labels.append(labels[batch_idx])
            new_vision_ids.append(cur_input_ids * 0 + NON_VISION_TOKEN)
            cur_image_idx += 1
            continue

        need_replace = cur_input_ids == IMAGE_TOKEN_INDEX
        num_replace = need_replace.sum()

        image_token_indices = [-1] + torch.where(
            need_replace)[0].tolist() + [
                                  cur_input_ids.shape[0]
                              ]
        cur_input_ids_noim = []
        cur_labels = labels[batch_idx]
        cur_labels_noim = []
        for i in range(len(image_token_indices) - 1):
            cur_input_ids_noim.append(cur_input_ids[image_token_indices[i] +
                                                    1:image_token_indices[i +
                                                                          1]])
            cur_labels_noim.append(cur_labels[image_token_indices[i] +
                                              1:image_token_indices[i + 1]])
        cur_new_inputs_ids = []
        cur_new_labels = []
        cur_new_vision_ids = []

        for i in range(num_replace + 1):
            cur_new_inputs_ids.append(cur_input_ids_noim[i])
            cur_new_vision_ids.append(cur_input_ids_noim[i] * 0 + NON_VISION_TOKEN)
            cur_new_labels.append(cur_labels_noim[i])
            if i < num_replace:
                # image
                cur_vision_tokens = visual_tokens[cur_image_idx].to(ori_input_ids)
                cur_new_inputs_ids.append(cur_vision_tokens)
                cur_new_vision_ids.append(vision_patch_indices[cur_image_idx])
                cur_new_labels.append(
                    torch.full((cur_vision_tokens.shape[0],),
                               IGNORE_INDEX,
                               device=cur_labels.device,
                               dtype=cur_labels.dtype))
                cur_image_idx += 1

        cur_new_inputs_ids = torch.cat(cur_new_inputs_ids)
        cur_new_vision_ids = torch.cat(cur_new_vision_ids)
        cur_new_labels = torch.cat(cur_new_labels)

        new_inputs_ids.append(cur_new_inputs_ids)
        new_vision_ids.append(cur_new_vision_ids)
        new_labels.append(cur_new_labels)

    # Combine them
    max_len = max(x.shape[0] for x in new_inputs_ids)
    batch_size = len(new_inputs_ids)

    new_inputs_ids_padded = []
    new_vision_ids_padded = []
    new_labels_padded = torch.full((batch_size, max_len),
                                   IGNORE_INDEX,
                                   dtype=new_labels[0].dtype,
                                   device=new_labels[0].device)
    attention_mask = torch.zeros((batch_size, max_len),
                                 dtype=attention_mask.dtype,
                                 device=attention_mask.device)
    position_ids = torch.zeros((batch_size, max_len),
                               dtype=position_ids.dtype,
                               device=position_ids.device)

    for i, (cur_new_id,
            cur_new_labels) in enumerate(zip(new_inputs_ids, new_labels)):
        # print(i, new_vision_ids)
        cur_vision_id = new_vision_ids[i]
        cur_len = cur_new_id.shape[0]
        new_inputs_ids_padded.append(
            torch.cat((cur_new_id,
                       torch.zeros((max_len - cur_len,),
                                   dtype=cur_new_id.dtype,
                                   device=cur_new_id.device)),
                      dim=0))
        new_vision_ids_padded.append(
            torch.cat((cur_vision_id,
                       torch.zeros((max_len - cur_len,),
                                   dtype=cur_new_id.dtype,
                                   device=cur_new_id.device) + NON_VISION_TOKEN),
                      dim=0))
        if cur_len > 0:
            new_labels_padded[i, :cur_len] = cur_new_labels
            attention_mask[i, :cur_len] = True
            position_ids[i, :cur_len] = torch.arange(
                0,
                cur_len,
                dtype=position_ids.dtype,
                device=position_ids.device)

    new_inputs_ids = torch.stack(new_inputs_ids_padded, dim=0)
    new_vision_ids = torch.stack(new_vision_ids_padded, dim=0)

    if _labels is None:
        new_labels = None
    else:
        new_labels = new_labels_padded

    if _attention_mask is None:
        attention_mask = None
    else:
        attention_mask = attention_mask.to(dtype=_attention_mask.dtype)

    if _position_ids is None:
        position_ids = None

    vpatch_id = tokenizer.encode("<vpatch>", add_special_tokens=False)[0]
    vpatch_indices = new_inputs_ids.clone().detach()
    vpatch_indices[vpatch_indices != vpatch_id] = NON_VISION_TOKEN
    if vision_patches is not None:
        assert vision_patches.size(0) == (vpatch_indices == vpatch_id).sum().item(), \
            f"number of vision patches is the the same as indicated in indices: {vision_patches.size(0)} vs {(vpatch_indices == vpatch_id).sum().item()}."
    vpatch_indices[vpatch_indices == vpatch_id] = torch.arange((vpatch_indices == vpatch_id).sum(),
                                                               device=vpatch_indices.device)

    return {
        'input_ids': new_inputs_ids,
        'position_ids': position_ids,
        'attention_mask': attention_mask,
        'past_key_values': past_key_values,
        # 'inputs_embeds': None,
        'labels': new_labels,
        'vision_patch_indices': vpatch_indices,  # new_vision_ids,
        'vision_patches': vision_patches,  # only 1 image
    }


def pad_32(val):
    if val % 32 == 0:
        return val
    else:
        return (val // 32 + 1) * 32