File size: 33,651 Bytes
e63f3e2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# Adopted from https://github.com/lm-sys/FastChat. Below is the original copyright:
# Adopted from tatsu-lab@stanford_alpaca. Below is the original copyright:
#    Copyright 2023 Rohan Taori, Ishaan Gulrajani, Tianyi Zhang, Yann Dubois, Xuechen Li
#
#    Licensed under the Apache License, Version 2.0 (the "License");
#    you may not use this file except in compliance with the License.
#    You may obtain a copy of the License at
#
#        http://www.apache.org/licenses/LICENSE-2.0
#
#    Unless required by applicable law or agreed to in writing, software
#    distributed under the License is distributed on an "AS IS" BASIS,
#    WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
#    See the License for the specific language governing permissions and
#    limitations under the License.

import os
import copy
from dataclasses import dataclass, field
import json
import logging
import pathlib
from typing import Dict, Optional, Sequence, List

from PIL import ImageFile
ImageFile.LOAD_TRUNCATED_IMAGES = True 

import torch

import transformers
from transformers.models.clip.image_processing_clip import CLIPImageProcessor

from torch.utils.data import Dataset
from q_align.train.mplug_owl2_trainer import MPLUGOwl2Trainer
from q_align.constants import IGNORE_INDEX, IMAGE_TOKEN_INDEX, DEFAULT_IMAGE_TOKEN

from q_align import conversation as conversation_lib
from q_align.model import *
from q_align.mm_utils import tokenizer_image_token

from PIL import Image
from icecream import ic

local_rank = None


def rank0_print(*args):
    if local_rank == 0:
        print(*args)


@dataclass
class ModelArguments:
    model_name_or_path: Optional[str] = field(default="facebook/opt-125m")
    version: Optional[str] = field(default="v0")
    freeze_backbone: bool = field(default=False)

@dataclass
class DataArguments:
    data_path: str = field(default=None,
                           metadata={"help": "Path to the training data."})
    lazy_preprocess: bool = False
    is_multimodal: bool = False
    image_folder: Optional[str] = field(default=None)
    image_aspect_ratio: str = 'square'
    image_grid_pinpoints: Optional[str] = field(default=None)


@dataclass
class TrainingArguments(transformers.TrainingArguments):
    cache_dir: Optional[str] = field(default=None)
    optim: str = field(default="adamw_torch")
    remove_unused_columns: bool = field(default=False)
    
    tune_visual_abstractor: bool = field(default=True)
    freeze_vision_model: bool = field(default=True)

    model_max_length: int = field(
        default=512,
        metadata={
            "help":
            "Maximum sequence length. Sequences will be right padded (and possibly truncated)."
        },
    )
    double_quant: bool = field(
        default=True,
        metadata={"help": "Compress the quantization statistics through double quantization."}
    )
    quant_type: str = field(
        default="nf4",
        metadata={"help": "Quantization data type to use. Should be one of `fp4` or `nf4`."}
    )
    bits: int = field(
        default=16,
        metadata={"help": "How many bits to use."}
    )
    lora_enable: bool = False
    lora_r: int = 64
    lora_alpha: int = 16
    lora_dropout: float = 0.05
    lora_weight_path: str = ""
    lora_bias: str = "none"
    visual_abstractor_lr: Optional[float] = None
    group_by_modality_length: bool = field(default=False)


def maybe_zero_3(param, ignore_status=False, name=None):
    from deepspeed import zero
    from deepspeed.runtime.zero.partition_parameters import ZeroParamStatus
    if hasattr(param, "ds_id"):
        if param.ds_status == ZeroParamStatus.NOT_AVAILABLE:
            if not ignore_status:
                logging.warning(f"{name}: param.ds_status != ZeroParamStatus.NOT_AVAILABLE: {param.ds_status}")
        with zero.GatheredParameters([param]):
            param = param.data.detach().cpu().clone()
    else:
        param = param.detach().cpu().clone()
    return param


# Borrowed from peft.utils.get_peft_model_state_dict
def get_peft_state_maybe_zero_3(named_params, bias):
    if bias == "none":
        to_return = {k: t for k, t in named_params if "lora_" in k}
    elif bias == "all":
        to_return = {k: t for k, t in named_params if "lora_" in k or "bias" in k}
    elif bias == "lora_only":
        to_return = {}
        maybe_lora_bias = {}
        lora_bias_names = set()
        for k, t in named_params:
            if "lora_" in k:
                to_return[k] = t
                bias_name = k.split("lora_")[0] + "bias"
                lora_bias_names.add(bias_name)
            elif "bias" in k:
                maybe_lora_bias[k] = t
        for k, t in maybe_lora_bias:
            if bias_name in lora_bias_names:
                to_return[bias_name] = t
    else:
        raise NotImplementedError
    to_return = {k: maybe_zero_3(v, ignore_status=True) for k, v in to_return.items()}
    return to_return


def get_peft_state_non_lora_maybe_zero_3(named_params, require_grad_only=True):
    to_return = {k: t for k, t in named_params if "lora_" not in k}
    if require_grad_only:
        to_return = {k: t for k, t in to_return.items() if t.requires_grad}
    to_return = {k: maybe_zero_3(v, ignore_status=True).cpu() for k, v in to_return.items()}
    return to_return


def get_mm_adapter_state_maybe_zero_3(named_params, keys_to_match):
    to_return = {k: t for k, t in named_params if any(key_match in k for key_match in keys_to_match)}
    to_return = {k: maybe_zero_3(v, ignore_status=True).cpu() for k, v in to_return.items()}
    return to_return


def find_all_linear_names(model):
    cls = torch.nn.Linear
    lora_module_names = set()
    multimodal_keywords = ['vision_model', 'visual_abstractor']
    for name, module in model.named_modules():
        if any(mm_keyword in name for mm_keyword in multimodal_keywords):
            continue
        if isinstance(module, cls):
            lora_module_names.add(name)

    if 'lm_head' in lora_module_names: # needed for 16-bit
        lora_module_names.remove('lm_head')
    return list(lora_module_names)


def safe_save_model_for_hf_trainer(trainer: transformers.Trainer,
                                   output_dir: str):
    """Collects the state dict and dump to disk."""

    if trainer.deepspeed:
        torch.cuda.synchronize()
        trainer.save_model(output_dir)
        return

    state_dict = trainer.model.state_dict()
    if trainer.args.should_save:
        cpu_state_dict = {
            key: value.cpu()
            for key, value in state_dict.items()
        }
        del state_dict
        trainer._save(output_dir, state_dict=cpu_state_dict)  # noqa


def smart_tokenizer_and_embedding_resize(
    special_tokens_dict: Dict,
    tokenizer: transformers.PreTrainedTokenizer,
    model: transformers.PreTrainedModel,
):
    """Resize tokenizer and embedding.

    Note: This is the unoptimized version that may make your embedding size not be divisible by 64.
    """
    num_new_tokens = tokenizer.add_special_tokens(special_tokens_dict)
    model.resize_token_embeddings(len(tokenizer))

    if num_new_tokens > 0:
        input_embeddings = model.get_input_embeddings().weight.data
        output_embeddings = model.get_output_embeddings().weight.data

        input_embeddings_avg = input_embeddings[:-num_new_tokens].mean(
            dim=0, keepdim=True)
        output_embeddings_avg = output_embeddings[:-num_new_tokens].mean(
            dim=0, keepdim=True)

        input_embeddings[-num_new_tokens:] = input_embeddings_avg
        output_embeddings[-num_new_tokens:] = output_embeddings_avg


def _tokenize_fn(strings: Sequence[str],
                 tokenizer: transformers.PreTrainedTokenizer) -> Dict:
    """Tokenize a list of strings."""
    tokenized_list = [
        tokenizer(
            text,
            return_tensors="pt",
            padding="longest",
            max_length=tokenizer.model_max_length,
            truncation=True,
        ) for text in strings
    ]
    input_ids = labels = [
        tokenized.input_ids[0] for tokenized in tokenized_list
    ]
    input_ids_lens = labels_lens = [
        tokenized.input_ids.ne(tokenizer.pad_token_id).sum().item()
        for tokenized in tokenized_list
    ]
    return dict(
        input_ids=input_ids,
        labels=labels,
        input_ids_lens=input_ids_lens,
        labels_lens=labels_lens,
    )


def _mask_targets(target, tokenized_lens, speakers):
    # cur_idx = 0
    cur_idx = tokenized_lens[0]
    tokenized_lens = tokenized_lens[1:]
    target[:cur_idx] = IGNORE_INDEX
    for tokenized_len, speaker in zip(tokenized_lens, speakers):
        if speaker == "human":
            target[cur_idx+2:cur_idx + tokenized_len] = IGNORE_INDEX
        cur_idx += tokenized_len


def _add_speaker_and_signal(header, source, get_conversation=True):
    """Add speaker and start/end signal on each round."""
    BEGIN_SIGNAL = "### "
    END_SIGNAL = "\n"
    conversation = header
    for sentence in source:
        from_str = sentence["from"]
        if from_str.lower() == "human":
            from_str = conversation_lib.default_conversation.roles[0]
        elif from_str.lower() == "gpt":
            from_str = conversation_lib.default_conversation.roles[1]
        else:
            from_str = 'unknown'
        sentence["value"] = (BEGIN_SIGNAL + from_str + ": " +
                             sentence["value"] + END_SIGNAL)
        if get_conversation:
            conversation += sentence["value"]
    conversation += BEGIN_SIGNAL
    return conversation


def preprocess_multimodal(
    sources: Sequence[str],
    data_args: DataArguments
) -> Dict:
    is_multimodal = data_args.is_multimodal
    if not is_multimodal:
        return sources

    for source in sources:
        for sentence in source:
            if DEFAULT_IMAGE_TOKEN in sentence['value']:
                sentence['value'] = sentence['value'].replace(DEFAULT_IMAGE_TOKEN, '').strip()
                sentence['value'] = DEFAULT_IMAGE_TOKEN + '\n' + sentence['value']
                sentence['value'] = sentence['value'].strip()
               
            replace_token = DEFAULT_IMAGE_TOKEN
            sentence["value"] = sentence["value"].replace(DEFAULT_IMAGE_TOKEN, replace_token)
        
    return sources


def preprocess_v1(
    sources,
    tokenizer: transformers.PreTrainedTokenizer,
    has_image: bool = False
) -> Dict:
    conv = conversation_lib.default_conversation.copy()
    roles = {"human": conv.roles[0], "gpt": conv.roles[1]}

    # Apply prompt templates
    conversations = []
    for i, source in enumerate(sources):
        if roles[source[0]["from"]] != conv.roles[0]:
            # Skip the first one if it is not from human
            source = source[1:]

        conv.messages = []
        for j, sentence in enumerate(source):
            role = roles[sentence["from"]]
            assert role == conv.roles[j % 2], f"{i}"
            conv.append_message(role, sentence["value"])
        conversations.append(conv.get_prompt())

    # Tokenize conversations

    if has_image:
        input_ids = torch.stack([tokenizer_image_token(prompt, tokenizer, return_tensors='pt') for prompt in conversations], dim=0)
    else:
        input_ids = tokenizer(
            conversations,
            return_tensors="pt",
            padding="longest",
            max_length=tokenizer.model_max_length,
            truncation=True,
        ).input_ids

    targets = input_ids.clone()

    assert conv.sep_style == conversation_lib.SeparatorStyle.TWO or conv.sep_style == conversation_lib.SeparatorStyle.TWO_NO_SYS

    # Mask targets
    sep = conv.sep + conv.roles[1] + ": "
    for conversation, target in zip(conversations, targets):
        total_len = int(target.ne(tokenizer.pad_token_id).sum())

        rounds = conversation.split(conv.sep2)
        cur_len = 1
        target[:cur_len] = IGNORE_INDEX
        for i, rou in enumerate(rounds):
            if rou == "":
                break

            parts = rou.split(sep)
            if len(parts) != 2:
                break
            parts[0] += sep

            if has_image:
                round_len = len(tokenizer_image_token(rou, tokenizer))
                instruction_len = len(tokenizer_image_token(parts[0], tokenizer)) - 2
            else:
                round_len = len(tokenizer(rou).input_ids)
                instruction_len = len(tokenizer(parts[0]).input_ids) - 2

            target[cur_len : cur_len + instruction_len] = IGNORE_INDEX

            cur_len += round_len
        target[cur_len:] = IGNORE_INDEX

        if cur_len < tokenizer.model_max_length:
            if cur_len != total_len:
                target[:] = IGNORE_INDEX
                print(
                    f"WARNING: tokenization mismatch: {cur_len} vs. {total_len}."
                    f" (ignored)"
                )

    return dict(
        input_ids=input_ids,
        labels=targets,
    )


def preprocess_plain(
    sources: Sequence[str],
    tokenizer: transformers.PreTrainedTokenizer,
) -> Dict:
    # add end signal and concatenate together
    conversations = []
    for source in sources:
        assert len(source) == 2
        assert DEFAULT_IMAGE_TOKEN in source[0]['value']
        source[0]['value'] = DEFAULT_IMAGE_TOKEN
        conversation = source[0]['value'] + source[1]['value'] + conversation_lib.default_conversation.sep
        conversations.append(conversation)
    # tokenize conversations
    input_ids = [tokenizer_image_token(prompt, tokenizer, return_tensors='pt') for prompt in conversations]
    targets = copy.deepcopy(input_ids)
    for target, source in zip(targets, sources):
        tokenized_len = len(tokenizer_image_token(source[0]['value'], tokenizer))
        target[:tokenized_len] = IGNORE_INDEX

    return dict(input_ids=input_ids, labels=targets)


def preprocess(
    sources: Sequence[str],
    tokenizer: transformers.PreTrainedTokenizer,
    has_image: bool = False
) -> Dict:
    """
    Given a list of sources, each is a conversation list. This transform:
    1. Add signal '### ' at the beginning each sentence, with end signal '\n';
    2. Concatenate conversations together;
    3. Tokenize the concatenated conversation;
    4. Make a deepcopy as the target. Mask human words with IGNORE_INDEX.
    """
    if conversation_lib.default_conversation.sep_style == conversation_lib.SeparatorStyle.PLAIN:
        return preprocess_plain(sources, tokenizer)
    if conversation_lib.default_conversation.version.startswith("v1"):
        return preprocess_v1(sources, tokenizer, has_image=has_image)
    # add end signal and concatenate together
    conversations = []
    for source in sources:
        header = f"{conversation_lib.default_conversation.system}\n\n"
        conversation = _add_speaker_and_signal(header, source)
        conversations.append(conversation)
    # tokenize conversations
    def get_tokenize_len(prompts):
        return [len(tokenizer_image_token(prompt, tokenizer)) for prompt in prompts]
    if has_image:
        input_ids = [tokenizer_image_token(prompt, tokenizer, return_tensors='pt') for prompt in conversations]
    else:
        conversations_tokenized = _tokenize_fn(conversations, tokenizer)
        input_ids = conversations_tokenized["input_ids"]

    targets = copy.deepcopy(input_ids)
    for target, source in zip(targets, sources):
        if has_image:
            tokenized_lens = get_tokenize_len([header] + [s["value"] for s in source])
        else:
            tokenized_lens = _tokenize_fn([header] + [s["value"] for s in source], tokenizer)["input_ids_lens"]
        speakers = [sentence["from"] for sentence in source]
        _mask_targets(target, tokenized_lens, speakers)

    return dict(input_ids=input_ids, labels=targets)


def load_video(video_file):
    from decord import VideoReader
    vr = VideoReader(video_file)

    # Get video frame rate
    fps = vr.get_avg_fps()

    # Calculate frame indices for 1fps
    frame_indices = [int(fps * i) for i in range(int(len(vr) / fps))]
    frames = vr.get_batch(frame_indices).asnumpy()
    return [Image.fromarray(frames[i]) for i in range(int(len(vr) / fps))]

def expand2square(pil_img, background_color):
    width, height = pil_img.size
    if width == height:
        return pil_img
    elif width > height:
        result = Image.new(pil_img.mode, (width, width), background_color)
        result.paste(pil_img, (0, (width - height) // 2))
        return result
    else:
        result = Image.new(pil_img.mode, (height, height), background_color)
        result.paste(pil_img, ((height - width) // 2, 0))
        return result

class LazySupervisedDataset(Dataset):
    """Dataset for supervised fine-tuning."""

    def __init__(self, data_path: str,
                 tokenizer: transformers.PreTrainedTokenizer,
                 data_args: DataArguments):
        super(LazySupervisedDataset, self).__init__()
        list_data_dict = json.load(open(data_path, "r"))

        rank0_print("Formatting inputs...Skip in lazy mode")
        self.tokenizer = tokenizer
        self.list_data_dict = list_data_dict
        self.data_args = data_args

    def __len__(self):
        return len(self.list_data_dict)

    @property
    def lengths(self):
        length_list = []
        for sample in self.list_data_dict:
            img_tokens = 128 if 'image' in sample else 0
            length_list.append(sum(len(conv['value'].split()) for conv in sample['conversations']) + img_tokens)
        return length_list


    @property
    def modality_lengths(self):
        length_list = []
        for sample in self.list_data_dict:
            cur_len = sum(len(conv['value'].split()) for conv in sample['conversations'])
            cur_len = cur_len if 'image' in sample else -cur_len
            length_list.append(cur_len)
        return length_list

#     def __getitem__(self, i) -> Dict[str, torch.Tensor]:
#         sources = self.list_data_dict[i]
#         if isinstance(i, int):
#             sources = [sources]
#         assert len(sources) == 1, "Don't know why it is wrapped to a list"  # FIXME
#         if 'image' in sources[0]:
#             image_file = self.list_data_dict[i]['image']
#             image_folder = self.data_args.image_folder
#             processor = self.data_args.image_processor
#             image = Image.open(os.path.join(image_folder, image_file)).convert('RGB')
#             if self.data_args.image_aspect_ratio == 'pad':
#                 def expand2square(pil_img, background_color):
#                     width, height = pil_img.size
#                     if width == height:
#                         return pil_img
#                     elif width > height:
#                         result = Image.new(pil_img.mode, (width, width), background_color)
#                         result.paste(pil_img, (0, (width - height) // 2))
#                         return result
#                     else:
#                         result = Image.new(pil_img.mode, (height, height), background_color)
#                         result.paste(pil_img, ((height - width) // 2, 0))
#                         return result
#                 image = expand2square(image, tuple(int(x*255) for x in processor.image_mean))
#                 image = processor.preprocess(image, return_tensors='pt')['pixel_values'][0]
#             else:
#                 image = processor.preprocess(image, return_tensors='pt')['pixel_values'][0]
#             sources = preprocess_multimodal(
#                 copy.deepcopy([e["conversations"] for e in sources]),
#                 self.data_args)
#         else:
#             sources = copy.deepcopy([e["conversations"] for e in sources])
#         data_dict = preprocess(
#             sources,
#             self.tokenizer,
#             has_image=('image' in self.list_data_dict[i]))
#         if isinstance(i, int):
#             data_dict = dict(input_ids=data_dict["input_ids"][0],
#                              labels=data_dict["labels"][0])

#         # image exist in the data
#         if 'image' in self.list_data_dict[i]:
#             data_dict['image'] = image
#         elif self.data_args.is_multimodal:
#             # image does not exist in the data, but the model is multimodal
#             crop_size = self.data_args.image_processor.crop_size
#             data_dict['image'] = torch.zeros(3, crop_size['height'], crop_size['width'])
#         return data_dict
    
    def next_rand(self):
        import random
        return random.randint(0,len(self)-1)

    def __getitem__(self, i) -> Dict[str, torch.Tensor]:
        while True:
            sources = self.list_data_dict[i]
            if isinstance(i, int):
                sources = [sources]
            assert len(sources) == 1, "Don't know why it is wrapped to a list"  # FIXME
            if 'image' in sources[0]:
                image_file = self.list_data_dict[i]['image']
                    
                image_folder = self.data_args.image_folder
                processor = self.data_args.image_processor
                from pathlib import Path
                #if not Path(os.path.join(image_folder, image_file)).exists():
                #    i = self.next_rand()
                #    continue
                if isinstance(image_file, list):
                    # Multiple Images as Input
                    try:
                        image = [Image.open(os.path.join(image_folder, imfile)).convert('RGB') for imfile in image_file]
                    except Exception as ex:
                        print(ex)
                        i = self.next_rand()
                        continue
                    if self.data_args.image_aspect_ratio == 'pad':
                        image = [expand2square(img, tuple(int(x*255) for x in processor.image_mean)) for img in image]
                        image = processor.preprocess(image, return_tensors='pt')['pixel_values']
                    else:
                        image = processor.preprocess(image, return_tensors='pt')['pixel_values']
                elif os.path.join(image_folder, image_file).endswith("mp4"):
                    # Video as Input
                    image = load_video(os.path.join(image_folder, image_file))
                    if self.data_args.image_aspect_ratio == 'pad':
                        image = [expand2square(img, tuple(int(x*255) for x in processor.image_mean)) for img in image]
                        image = processor.preprocess(image, return_tensors='pt')['pixel_values']
                    else:
                        image = processor.preprocess(image, return_tensors='pt')['pixel_values']
                else:
                    try:
                        image = Image.open(os.path.join(image_folder, image_file)).convert('RGB')
                    except Exception as ex:
                        print(ex)
                        i = self.next_rand()
                        continue
                    if self.data_args.image_aspect_ratio == 'pad':
                        image = expand2square(image, tuple(int(x*255) for x in processor.image_mean))
                        image = processor.preprocess(image, return_tensors='pt')['pixel_values'][0]
                    else:
                        image = processor.preprocess(image, return_tensors='pt')['pixel_values'][0]
                sources = preprocess_multimodal(
                    copy.deepcopy([e["conversations"] for e in sources]),
                    self.data_args)
            else:
                
                sources = copy.deepcopy([e["conversations"] for e in sources])
            data_dict = preprocess(
                sources,
                self.tokenizer,
                has_image=('image' in self.list_data_dict[i]))
            if isinstance(i, int):
                data_dict = dict(input_ids=data_dict["input_ids"][0],
                                labels=data_dict["labels"][0])

            # image exist in the data
            if 'image' in self.list_data_dict[i]:
                data_dict['image'] = image
            elif self.data_args.is_multimodal:
                # image does not exist in the data, but the model is multimodal
                crop_size = self.data_args.image_processor.crop_size
                data_dict['image'] = torch.zeros(3, crop_size['height'], crop_size['width'])
            return data_dict


@dataclass
class DataCollatorForSupervisedDataset(object):
    """Collate examples for supervised fine-tuning."""

    tokenizer: transformers.PreTrainedTokenizer

    def __call__(self, instances: Sequence[Dict]) -> Dict[str, torch.Tensor]:
        input_ids, labels = tuple([instance[key] for instance in instances]
                                  for key in ("input_ids", "labels"))
        input_ids = torch.nn.utils.rnn.pad_sequence(
            input_ids,
            batch_first=True,
            padding_value=self.tokenizer.pad_token_id)
        labels = torch.nn.utils.rnn.pad_sequence(labels,
                                                 batch_first=True,
                                                 padding_value=IGNORE_INDEX)
        input_ids = input_ids[:, :self.tokenizer.model_max_length]
        labels = labels[:, :self.tokenizer.model_max_length]
        batch = dict(
            input_ids=input_ids,
            labels=labels,
            attention_mask=input_ids.ne(self.tokenizer.pad_token_id),
        )

        if 'image' in instances[0]:
            images = [instance['image'] for instance in instances]
            if all(x is not None and x.shape == images[0].shape for x in images):
                batch['images'] = torch.stack(images)
            else:
                batch['images'] = images

        return batch


def make_supervised_data_module(tokenizer: transformers.PreTrainedTokenizer,
                                data_args) -> Dict:
    """Make dataset and collator for supervised fine-tuning."""
    train_dataset = LazySupervisedDataset(tokenizer=tokenizer,
                                data_path=data_args.data_path,
                                data_args=data_args)
    data_collator = DataCollatorForSupervisedDataset(tokenizer=tokenizer)
    return dict(train_dataset=train_dataset,
                eval_dataset=None,
                data_collator=data_collator)


def train():
    global local_rank

    parser = transformers.HfArgumentParser(
        (ModelArguments, DataArguments, TrainingArguments))
    model_args, data_args, training_args = parser.parse_args_into_dataclasses()
    local_rank = training_args.local_rank
    compute_dtype = (torch.float16 if training_args.fp16 else (torch.bfloat16 if training_args.bf16 else torch.float32))

    bnb_model_from_pretrained_args = {}
    if training_args.bits in [4, 8]:
        from transformers import BitsAndBytesConfig
        bnb_model_from_pretrained_args.update(dict(
            device_map={"": training_args.device},
            load_in_4bit=training_args.bits == 4,
            load_in_8bit=training_args.bits == 8,
            quantization_config=BitsAndBytesConfig(
                load_in_4bit=training_args.bits == 4,
                load_in_8bit=training_args.bits == 8,
                llm_int8_threshold=6.0,
                llm_int8_has_fp16_weight=False,
                bnb_4bit_compute_dtype=compute_dtype,
                bnb_4bit_use_double_quant=training_args.double_quant,
                bnb_4bit_quant_type=training_args.quant_type # {'fp4', 'nf4'}
            )
        ))

    model = MPLUGOwl2LlamaForCausalLM.from_pretrained(
        model_args.model_name_or_path,
        cache_dir=training_args.cache_dir,
        **bnb_model_from_pretrained_args
    )
    model.config.use_cache = False

    if model_args.freeze_backbone:
        model.model.requires_grad_(False)

    if training_args.bits in [4, 8]:
        from peft import prepare_model_for_kbit_training
        model.config.torch_dtype=(torch.float32 if training_args.fp16 else (torch.bfloat16 if training_args.bf16 else torch.float32))
        model = prepare_model_for_kbit_training(model, use_gradient_checkpointing=training_args.gradient_checkpointing)

    if training_args.gradient_checkpointing:
        if hasattr(model, "enable_input_require_grads"):
            model.enable_input_require_grads()
        else:
            def make_inputs_require_grad(module, input, output):
                output.requires_grad_(True)
            model.get_input_embeddings().register_forward_hook(make_inputs_require_grad)

    if training_args.lora_enable:
        from peft import LoraConfig, get_peft_model
        lora_config = LoraConfig(
            r=training_args.lora_r,
            lora_alpha=training_args.lora_alpha,
            target_modules=find_all_linear_names(model),
            lora_dropout=training_args.lora_dropout,
            bias=training_args.lora_bias,
            task_type="CAUSAL_LM",
        )
        if training_args.bits == 16:
            if training_args.bf16:
                model.to(torch.bfloat16)
            if training_args.fp16:
                model.to(torch.float16)
        rank0_print("Adding LoRA adapters...")
        model = get_peft_model(model, lora_config)

    tokenizer = transformers.AutoTokenizer.from_pretrained(
        model_args.model_name_or_path,
        cache_dir=training_args.cache_dir,
        model_max_length=training_args.model_max_length,
        padding_side="right",
        use_fast=False,
    )

    
    tokenizer.pad_token = tokenizer.unk_token
    if model_args.version in conversation_lib.conv_templates:
        conversation_lib.default_conversation = conversation_lib.conv_templates[model_args.version]
    else:
        conversation_lib.default_conversation = conversation_lib.conv_templates["vicuna_v1"]
    
    if not training_args.freeze_vision_model and training_args.bits in [4, 8]:
        model.get_model().vision_model.to(dtype=compute_dtype, device=training_args.device)
    else:
        vision_tower = model.get_model().vision_model
        vision_tower.to(dtype=torch.bfloat16 if training_args.bf16 else torch.float16, device=training_args.device)
    
    if training_args.tune_visual_abstractor and training_args.bits in [4, 8]:
        model.get_model().visual_abstractor.to(dtype=compute_dtype, device=training_args.device)
    else:
        visual_abstractor = model.get_model().visual_abstractor
        visual_abstractor.to(dtype=torch.bfloat16 if training_args.bf16 else torch.float16, device=training_args.device)
    
    data_args.image_processor = CLIPImageProcessor.from_pretrained(model_args.model_name_or_path)
    data_args.is_multimodal = True

    model.config.image_aspect_ratio = data_args.image_aspect_ratio
    model.config.image_grid_pinpoints = data_args.image_grid_pinpoints
    model.config.tune_visual_abstractor = model_args.tune_visual_abstractor = training_args.tune_visual_abstractor
    ic(training_args.tune_visual_abstractor)
    model.requires_grad_(True)
    if training_args.tune_visual_abstractor:
        # model.requires_grad_(False)
        for p in model.get_model().visual_abstractor.parameters():
            p.requires_grad = True
            
    model.config.freeze_vision_model = training_args.freeze_vision_model
    ic(training_args.freeze_vision_model)
    if training_args.freeze_vision_model:
        for p in model.get_model().vision_model.parameters():
            p.requires_grad = False
            
    model.config.visual_abstractor_lr = training_args.visual_abstractor_lr


    if training_args.bits in [4, 8]:
        from peft.tuners.lora import LoraLayer
        for name, module in model.named_modules():
            if isinstance(module, LoraLayer):
                if training_args.bf16:
                    module = module.to(torch.bfloat16)
            if 'norm' in name:
                module = module.to(torch.float32)
            if 'lm_head' in name or 'embed_tokens' in name:
                if hasattr(module, 'weight'):
                    if training_args.bf16 and module.weight.dtype == torch.float32:
                        module = module.to(torch.bfloat16)

    data_module = make_supervised_data_module(tokenizer=tokenizer,
                                              data_args=data_args)
    trainer = MPLUGOwl2Trainer(model=model,
                    tokenizer=tokenizer,
                    args=training_args,
                    **data_module)

    # if list(pathlib.Path(training_args.output_dir).glob("checkpoint-*")):
    #     trainer.train(resume_from_checkpoint=True)
    # else:
    #     trainer.train()
    
    # TODO I dont like auto resume << REMOVE IT AND UNCOMMENT THE ABOVE CODE
    trainer.train()

    trainer.save_state()

    model.config.use_cache = True

    if training_args.lora_enable:
        state_dict = get_peft_state_maybe_zero_3(
            model.named_parameters(), training_args.lora_bias
        )
        non_lora_state_dict = get_peft_state_non_lora_maybe_zero_3(
            model.named_parameters()
        )
        if training_args.local_rank == 0 or training_args.local_rank == -1:
            model.config.save_pretrained(training_args.output_dir)
            model.save_pretrained(training_args.output_dir, state_dict=state_dict)
            torch.save(non_lora_state_dict, os.path.join(training_args.output_dir, 'non_lora_trainables.bin'))
    else:
        safe_save_model_for_hf_trainer(trainer=trainer,
                                       output_dir=training_args.output_dir)


if __name__ == "__main__":
    train()