File size: 29,396 Bytes
32d2edb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from typing import List, Dict, Optional, Tuple
from PIL import Image, ImageOps, ImageDraw, ImageFont
import torch
import torch.nn as nn
from torchvision import transforms
from transformers import TextStreamer
from transformers.tokenization_utils import PreTrainedTokenizer as T
from abc import ABC
import re
import numpy as np


def load_image(image_path):
    try:
        image = Image.open(image_path)
        corrected_image = ImageOps.exif_transpose(image)

        return corrected_image
        
    except Exception as e:
        print(f"error: {e}")

        return None


def re_match(text):
    pattern = r'(<\|ref\|>(.*?)<\|/ref\|><\|det\|>(.*?)<\|/det\|>)'
    matches = re.findall(pattern, text, re.DOTALL)

    # pattern1 = r'<\|ref\|>.*?<\|/ref\|>\n'
    # new_text1 = re.sub(pattern1, '', text, flags=re.DOTALL)

    mathes_image = []
    mathes_other = []
    for a_match in matches:
        if '<|ref|>image<|/ref|>' in a_match[0]:
            mathes_image.append(a_match[0])
        else:
            mathes_other.append(a_match[0])
    return matches, mathes_image, mathes_other


def extract_coordinates_and_label(ref_text, image_width, image_height):

    try:
        label_type = ref_text[1]
        cor_list = eval(ref_text[2])
    except Exception as e:
        print(e)
        return None

    return (label_type, cor_list)


def draw_bounding_boxes(image, refs, ouput_path):

    image_width, image_height = image.size
    
    img_draw = image.copy()
    draw = ImageDraw.Draw(img_draw)

    overlay = Image.new('RGBA', img_draw.size, (0, 0, 0, 0))
    draw2 = ImageDraw.Draw(overlay)
    
    font = ImageFont.load_default()

    img_idx = 0
    
    for i, ref in enumerate(refs):
        try:
            result = extract_coordinates_and_label(ref, image_width, image_height)
            if result:
                label_type, points_list = result
                
                color = (np.random.randint(0, 200), np.random.randint(0, 200), np.random.randint(0, 255))

                color_a = color + (20, )
                for points in points_list:
                    x1, y1, x2, y2 = points

                    x1 = int(x1 / 999 * image_width)
                    y1 = int(y1 / 999 * image_height)

                    x2 = int(x2 / 999 * image_width)
                    y2 = int(y2 / 999 * image_height)

                    if label_type == 'image':
                        try:
                            cropped = image.crop((x1, y1, x2, y2))
                            cropped.save(f"{ouput_path}/images/{img_idx}.jpg")
                        except Exception as e:
                            print(e)
                            pass
                        img_idx += 1
                        
                    try:
                        if label_type == 'title':
                            draw.rectangle([x1, y1, x2, y2], outline=color, width=4)
                            draw2.rectangle([x1, y1, x2, y2], fill=color_a, outline=(0, 0, 0, 0), width=1)
                        else:
                            draw.rectangle([x1, y1, x2, y2], outline=color, width=2)
                            draw2.rectangle([x1, y1, x2, y2], fill=color_a, outline=(0, 0, 0, 0), width=1)
                        text_x = x1
                        text_y = max(0, y1 - 15)
                            
                        
                        text_bbox = draw.textbbox((0, 0), label_type, font=font)
                        text_width = text_bbox[2] - text_bbox[0]
                        text_height = text_bbox[3] - text_bbox[1]
                        draw.rectangle([text_x, text_y, text_x + text_width, text_y + text_height], 
                                    fill=(255, 255, 255, 30))
                        
                        draw.text((text_x, text_y), label_type, font=font, fill=color)
                    except:
                        pass
        except:
            continue
    img_draw.paste(overlay, (0, 0), overlay)
    return img_draw


def process_image_with_refs(image, ref_texts, output_path):

    result_image = draw_bounding_boxes(image, ref_texts, output_path)
    
    return result_image


def find_closest_aspect_ratio(aspect_ratio, target_ratios, width, height, image_size):
    best_ratio_diff = float('inf')
    best_ratio = (1, 1)
    area = width * height
    for ratio in target_ratios:
        target_aspect_ratio = ratio[0] / ratio[1]
        ratio_diff = abs(aspect_ratio - target_aspect_ratio)
        if ratio_diff < best_ratio_diff:
            best_ratio_diff = ratio_diff
            best_ratio = ratio
        elif ratio_diff == best_ratio_diff:
            if area > 0.5 * image_size * image_size * ratio[0] * ratio[1]:
                best_ratio = ratio

    # print(f'width: {width}, height: {height}, best_ratio: {best_ratio}')
    return best_ratio


def dynamic_preprocess(image, min_num=2, max_num=9, image_size=640, use_thumbnail=False):
    orig_width, orig_height = image.size
    aspect_ratio = orig_width / orig_height

    # calculate the existing image aspect ratio
    target_ratios = set((i, j) for n in range(min_num, max_num + 1) for i in range(1, n + 1) for j in range(1, n + 1) if i * j <= max_num and i * j >= min_num)
    
    target_ratios = sorted(target_ratios, key=lambda x: x[0] * x[1])
    # print(f"target_ratios: {target_ratios}")

    target_aspect_ratio = find_closest_aspect_ratio(
        aspect_ratio, 
        target_ratios, 
        orig_width,
        orig_height, 
        image_size
    )
    # print(f"target_aspect_ratio: {target_aspect_ratio}")

    # calculate the target width and height
    target_width = image_size * target_aspect_ratio[0]
    target_height = image_size * target_aspect_ratio[1]
    blocks = target_aspect_ratio[0] * target_aspect_ratio[1]

    # resize the image
    resized_img = image.resize((target_width, target_height))
    processed_images = []
    for i in range(blocks):
        box = (
            (i % (target_width // image_size)) * image_size,
            (i // (target_width // image_size)) * image_size,
            ((i % (target_width // image_size)) + 1) * image_size,
            ((i // (target_width // image_size)) + 1) * image_size
        )
        # split the image
        split_img = resized_img.crop(box)
        processed_images.append(split_img)

    assert len(processed_images) == blocks
    # print(f"Number of processed images: {len(processed_images)}, Blocks: {blocks}")

    if use_thumbnail and len(processed_images) != 1:
        thumbnail_img = image.resize((image_size, image_size))
        processed_images.append(thumbnail_img)
    return processed_images, target_aspect_ratio


def normalize_transform(mean, std):
    if mean is None and std is None:
        transform = None
    elif mean is None and std is not None:
        mean = [0.] * len(std)
        transform = transforms.Normalize(mean=mean, std=std)
    elif mean is not None and std is None:
        std = [1.] * len(mean)
        transform = transforms.Normalize(mean=mean, std=std)
    else:
        transform = transforms.Normalize(mean=mean, std=std)

    return transform

def format_messages(
        tokenizer: T,
        conversations: List[Dict[str, str]],
        system_prompt: str = "",
):
    if system_prompt is not None and system_prompt != "":
        sys_prompt = {
            "role": "system",
            "content": system_prompt,
        }
        conversations = [sys_prompt] + conversations

    sft_prompt = tokenizer.apply_chat_template(
        conversations,
    )

    return sft_prompt


def text_encode(tokenizer, text: str, bos: bool = True, eos: bool = False):
    """
    Encode text with optional BOS/EOS tokens.
    
    Note: Qwen2VL tokenizer has bos_token_id=None, so we skip BOS for Qwen.
    The chat template handles special tokens automatically.
    """
    t = tokenizer.encode(text, add_special_tokens=False)
    bos_id = tokenizer.bos_token_id
    eos_id = tokenizer.eos_token_id

    # Only add BOS if tokenizer has one AND bos=True
    if bos and bos_id is not None:
        t = [bos_id] + t
    
    # Only add EOS if tokenizer has one AND eos=True
    if eos and eos_id is not None:
        t = t + [eos_id]

    return t

def load_pil_images(conversations: List[Dict[str, str]]) -> List[Image.Image]:
    pil_images = []

    for message in conversations:
        pil_image = None
    
        if message["role"].lower() == "user":
            if isinstance(message["content"], List):
                for d in message["content"]:
                    if d.get("type", "") == "image":
                        # Support both "image" (Qwen format) and "data" keys
                        image_path = d.get("image") or d.get("data", "")
                        pil_image = load_image(image_path)

            elif isinstance(message["content"], Dict):
                if message["content"].get("type", "") == "image":
                    # Support both "image" (Qwen format) and "data" keys
                    image_path = message["content"].get("image") or message["content"].get("data", "")
                    pil_image = load_image(image_path)

            if pil_image is not None:
                pil_images.append(pil_image)

    return pil_images


class BaseTransform(ABC):

    def set_rng(self, *args, **kwargs):
        pass

    def __call__(self, *args, **kwargs) -> torch.Tensor:
        pass

    @property
    def default_shape(self):
        raise NotImplementedError


class BasicImageTransform(BaseTransform):
    def __init__(
        self, 
        mean: Optional[Tuple[float, float, float]] = (0.5, 0.5, 0.5),
        std: Optional[Tuple[float, float, float]] = (0.5, 0.5, 0.5),
        normalize: bool = True
    ):
        self.mean = mean
        self.std = std
    
        transform_pipelines = [
            transforms.ToTensor()
        ]

        normalize = normalize_transform(mean, std) if normalize else nn.Identity()
        if normalize is not None:
            transform_pipelines.append(normalize)

        self.transform = transforms.Compose(transform_pipelines)
    
    def __call__(self, x):
        x = self.transform(x)
        return x

class NoEOSTextStreamer(TextStreamer):
    
    def on_finalized_text(self, text: str, stream_end: bool = False):
        eos_text = self.tokenizer.decode([self.tokenizer.eos_token_id], skip_special_tokens=False)
        text = text.replace(eos_text, "\n")
        print(text, flush=True, end="")




# @title Create datacollator

import torch
import math
from dataclasses import dataclass
from typing import Dict, List, Any, Tuple
from PIL import Image, ImageOps
from torch.nn.utils.rnn import pad_sequence
import io

# Use local functions (Qwen-compatible) instead of DeepSeek's versions
# from deepseek_ocr.modeling_deepseekocr import (
#     format_messages,
#     text_encode,
#     BasicImageTransform,
#     dynamic_preprocess,
# )


@dataclass
class DeepQwenDataCollator:
    """
    Data collator for DeepQwen model using Qwen2VL tokenizer.
    
    This collator processes images using DeepSeek OCR's dynamic cropping algorithm
    while maintaining compatibility with Qwen2VL's tokenization format.
    
    Key token mappings (Qwen2VL):
        - image_token: <|image_pad|> (id=151655)
        - vision_start: <|vision_start|> (id=151652)
        - vision_end: <|vision_end|> (id=151653)
        - eos_token: <|im_end|> (id=151645)
        - NO bos_token (bos_token_id is None)
    
    Args:
        tokenizer: Qwen2VL Tokenizer
        model: Model
        image_size: Size for image patches (default: 640)
        base_size: Size for global view (default: 1024)
        crop_mode: Whether to use dynamic cropping for large images
        train_on_responses_only: If True, only train on assistant responses (mask user prompts)
    """
    tokenizer: T
    model: Any
    image_size: int = 640
    base_size: int = 1024
    crop_mode: bool = True
    train_on_responses_only: bool = True

    def __init__(
        self,
        tokenizer,
        model,
        image_size: int = 640,
        base_size: int = 1024,
        crop_mode: bool = True,
        train_on_responses_only: bool = True,
        max_length: int = None,
    ):
        self.tokenizer = tokenizer
        self.model = model
        self.image_size = image_size
        self.base_size = base_size
        self.crop_mode = crop_mode
        self.dtype = model.dtype  # Get dtype from model
        self.train_on_responses_only = train_on_responses_only
        self.max_length = max_length  # None means no truncation
        
        # Qwen2VL specific token IDs
        # <|image_pad|> = 151655
        self.image_token_id = getattr(tokenizer, 'image_token_id', None)
        if self.image_token_id is None:
            # Fallback: try to get from added_tokens or use default Qwen2VL value
            self.image_token_id = 151655  # Qwen2VL's <|image_pad|>
        
        self.image_token = tokenizer.decode([self.image_token_id], skip_special_tokens=False)
        
        # Vision wrapper tokens for Qwen2VL format
        self.vision_start_token_id = getattr(tokenizer, 'vision_start_token_id', 151652)
        self.vision_end_token_id = getattr(tokenizer, 'vision_end_token_id', 151653)

        self.image_transform = BasicImageTransform(
            mean=(0.5, 0.5, 0.5),
            std=(0.5, 0.5, 0.5),
            normalize=True
        )
        self.patch_size = 16
        self.downsample_ratio = 4

        # Qwen2VL has NO bos_token (bos_token_id is None)
        # The chat template handles conversation formatting
        self.bos_id = tokenizer.bos_token_id  # Will be None for Qwen2VL
        self.eos_id = tokenizer.eos_token_id  # 151645 for Qwen2VL
        self.pad_token_id = tokenizer.pad_token_id  # 151643 for Qwen2VL

    def deserialize_image(self, image_data) -> Image.Image:
        """Convert image data (bytes dict, PIL Image, or file path) to PIL Image in RGB mode"""
        if isinstance(image_data, Image.Image):
            return image_data.convert("RGB")
        elif isinstance(image_data, str):
            # File path - load lazily
            image = load_image(image_data)
            if image is None:
                raise ValueError(f"Failed to load image from path: {image_data}")
            return image.convert("RGB")
        elif isinstance(image_data, dict) and 'bytes' in image_data:
            image_bytes = image_data['bytes']
            image = Image.open(io.BytesIO(image_bytes))
            return image.convert("RGB")
        else:
            raise ValueError(f"Unsupported image format: {type(image_data)}")

    def calculate_image_token_count(self, image: Image.Image, crop_ratio: Tuple[int, int]) -> int:
        """Calculate the number of tokens this image will generate"""
        num_queries = math.ceil((self.image_size // self.patch_size) / self.downsample_ratio)
        num_queries_base = math.ceil((self.base_size // self.patch_size) / self.downsample_ratio)

        width_crop_num, height_crop_num = crop_ratio

        if self.crop_mode:
            img_tokens = num_queries_base * num_queries_base + 1
            if width_crop_num > 1 or height_crop_num > 1:
                img_tokens += (num_queries * width_crop_num + 1) * (num_queries * height_crop_num)
        else:
            img_tokens = num_queries * num_queries + 1

        return img_tokens

    def process_image(self, image: Image.Image) -> Tuple[List, List, List, List, Tuple[int, int]]:
        """
        Process a single image based on crop_mode and size thresholds

        Returns:
            Tuple of (images_list, images_crop_list, images_spatial_crop, tokenized_image, crop_ratio)
        """
        images_list = []
        images_crop_list = []
        images_spatial_crop = []

        if self.crop_mode:
            # Determine crop ratio based on image size
            if image.size[0] <= 640 and image.size[1] <= 640:
                crop_ratio = (1, 1)
                images_crop_raw = []
            else:
                images_crop_raw, crop_ratio = dynamic_preprocess(
                    image, min_num=2, max_num=9,
                    image_size=self.image_size, use_thumbnail=False
                )

            # Process global view with padding
            global_view = ImageOps.pad(
                image, (self.base_size, self.base_size),
                color=tuple(int(x * 255) for x in self.image_transform.mean)
            )
            images_list.append(self.image_transform(global_view).to(self.dtype))

            width_crop_num, height_crop_num = crop_ratio
            images_spatial_crop.append([width_crop_num, height_crop_num])

            # Process local views (crops) if applicable
            if width_crop_num > 1 or height_crop_num > 1:
                for crop_img in images_crop_raw:
                    images_crop_list.append(
                        self.image_transform(crop_img).to(self.dtype)
                    )

            # Calculate image tokens
            num_queries = math.ceil((self.image_size // self.patch_size) / self.downsample_ratio)
            num_queries_base = math.ceil((self.base_size // self.patch_size) / self.downsample_ratio)

            tokenized_image = ([self.image_token_id] * num_queries_base + [self.image_token_id]) * num_queries_base
            tokenized_image += [self.image_token_id]

            if width_crop_num > 1 or height_crop_num > 1:
                tokenized_image += ([self.image_token_id] * (num_queries * width_crop_num) + [self.image_token_id]) * (
                    num_queries * height_crop_num)

        else:  # crop_mode = False
            crop_ratio = (1, 1)
            images_spatial_crop.append([1, 1])

            # For smaller base sizes, resize; for larger, pad
            if self.base_size <= 640:
                resized_image = image.resize((self.base_size, self.base_size), Image.LANCZOS)
                images_list.append(self.image_transform(resized_image).to(self.dtype))
            else:
                global_view = ImageOps.pad(
                    image, (self.base_size, self.base_size),
                    color=tuple(int(x * 255) for x in self.image_transform.mean)
                )
                images_list.append(self.image_transform(global_view).to(self.dtype))

            num_queries = math.ceil((self.base_size // self.patch_size) / self.downsample_ratio)
            tokenized_image = ([self.image_token_id] * num_queries + [self.image_token_id]) * num_queries
            tokenized_image += [self.image_token_id]

        return images_list, images_crop_list, images_spatial_crop, tokenized_image, crop_ratio

    def process_single_sample(self, messages: List[Dict]) -> Dict[str, Any]:
            """
            Process a single conversation into model inputs.
            
            Expected message format (Qwen2.5-VL native style):
            [
                {
                    "role": "user",
                    "content": [
                        {"type": "image", "image": <PIL.Image or path or bytes>},
                        {"type": "text", "text": "Describe this image."}
                    ]
                },
                {
                    "role": "assistant", 
                    "content": [{"type": "text", "text": "This is a description..."}]
                }
            ]
            
            Also supports string content for backward compatibility.
            """

            # --- 1. Setup ---
            tokenized_str = []
            images_seq_mask = []
            images_list, images_crop_list, images_spatial_crop = [], [], []

            prompt_token_count = -1  # Index to start training
            assistant_started = False

            # Qwen2VL has NO bos_token, so we don't add one

            for message in messages:
                role = message["role"].lower()  # Normalize role to lowercase
                content = message["content"]

                # Check if this is the assistant's turn
                if role == "assistant":
                    if not assistant_started:
                        # This is the split point. All tokens added *so far*
                        # are part of the prompt.
                        prompt_token_count = len(tokenized_str)
                        assistant_started = True

                # Process content based on format
                if isinstance(content, list):
                    # Qwen2.5-VL native format: content is a list of typed items
                    content_parts = []
                    
                    for item in content:
                        item_type = item.get("type", "")
                        
                        if item_type == "image":
                            # Get image data from various possible keys
                            image_data = item.get("image") or item.get("data")
                            if image_data is not None:
                                pil_image = self.deserialize_image(image_data)
                                
                                # Process the image through DeepSeek's encoder
                                img_list, crop_list, spatial_crop, tok_img, _ = self.process_image(pil_image)
                                
                                images_list.extend(img_list)
                                images_crop_list.extend(crop_list)
                                images_spatial_crop.extend(spatial_crop)
                                
                                # Add image placeholder tokens
                                tokenized_str.extend(tok_img)
                                images_seq_mask.extend([True] * len(tok_img))
                                
                        elif item_type == "text":
                            text = item.get("text", "")
                            
                            # For assistant, append EOS at the end of all text
                            if role == "assistant" and item == content[-1]:
                                if self.tokenizer.eos_token:
                                    text = f"{text.strip()}{self.tokenizer.eos_token}"
                            
                            # Tokenize the text
                            tokenized_text = text_encode(self.tokenizer, text, bos=False, eos=False)
                            tokenized_str.extend(tokenized_text)
                            images_seq_mask.extend([False] * len(tokenized_text))
                            
                else:
                    # Legacy format: content is a string (backward compatibility)
                    text_content = content
                    
                    # For assistant, append EOS token
                    if role == "assistant" and self.tokenizer.eos_token:
                        text_content = f"{text_content.strip()}{self.tokenizer.eos_token}"
                    
                    # Tokenize the text
                    tokenized_text = text_encode(self.tokenizer, text_content, bos=False, eos=False)
                    tokenized_str.extend(tokenized_text)
                    images_seq_mask.extend([False] * len(tokenized_text))

            # --- 2. Validation and Final Prep ---
            # If we never found an assistant message, we're in a weird state
            # (e.g., user-only prompt). We mask everything.
            if not assistant_started:
                print("Warning: No assistant message found in sample. Masking all tokens.")
                prompt_token_count = len(tokenized_str)

            # # DEBUG: Print after processing
            # print(f"[DEBUG] tokenized_str length: {len(tokenized_str)}")
            # print(f"[DEBUG] images_seq_mask length: {len(images_seq_mask)}, True count: {sum(images_seq_mask)}")
            # print(f"[DEBUG] images_list length: {len(images_list)}")
            # print(f"[DEBUG] images_crop_list length: {len(images_crop_list)}")
            # print(f"[DEBUG] prompt_token_count: {prompt_token_count}")

            # Prepare image tensors
            images_ori = torch.stack(images_list, dim=0)
            images_spatial_crop_tensor = torch.tensor(images_spatial_crop, dtype=torch.long)

            if images_crop_list:
                images_crop = torch.stack(images_crop_list, dim=0)
            else:
                images_crop = torch.zeros((1, 3, self.base_size, self.base_size), dtype=self.dtype)

            return {
                "input_ids": torch.tensor(tokenized_str, dtype=torch.long),
                "images_seq_mask": torch.tensor(images_seq_mask, dtype=torch.bool),
                "images_ori": images_ori,
                "images_crop": images_crop,
                "images_spatial_crop": images_spatial_crop_tensor,
                "prompt_token_count": prompt_token_count, # This is now accurate
            }

    def __call__(self, features: List[Dict[str, Any]]) -> Dict[str, torch.Tensor]:
        """
        Collate batch of samples.
        
        Expected feature format:
        {
            "prompt": str,  # The user's question/instruction
            "response": str,  # The assistant's response
            "image": PIL.Image or bytes dict  # The image
        }
        
        This will be converted to Qwen2.5-VL native conversation format:
        [
            {
                "role": "user", 
                "content": [
                    {"type": "image", "image": <PIL.Image>},
                    {"type": "text", "text": "<prompt>"}
                ]
            },
            {
                "role": "assistant", 
                "content": [{"type": "text", "text": "<response>"}]
            }
        ]
        """
        batch_data = []

        # Process each sample
        for feature in features:
            try:
                # Get image from either 'image' or 'image_path' key (lazy loading support)
                image_data = feature.get('image') or feature.get('image_path')
                if image_data is None:
                    raise ValueError("Sample missing both 'image' and 'image_path' keys")

                # Use Qwen2.5-VL native message format
                # content is a list of typed items: {"type": "image", ...} or {"type": "text", ...}
                messages = [
                    {
                        "role": "user",
                        "content": [
                            {"type": "image", "image": image_data},
                            {"type": "text", "text": feature['prompt']}
                        ]
                    },
                    {
                        "role": "assistant",
                        "content": [
                            {"type": "text", "text": feature["response"]}
                        ]
                    }
                ]
                
                processed = self.process_single_sample(messages)
                batch_data.append(processed)
            except Exception as e:
                print(f"Error processing sample: {e}")
                continue

        if not batch_data:
            raise ValueError("No valid samples in batch")

        # Extract lists
        input_ids_list = [item['input_ids'] for item in batch_data]
        images_seq_mask_list = [item['images_seq_mask'] for item in batch_data]
        prompt_token_counts = [item['prompt_token_count'] for item in batch_data]

        # Pad sequences using Qwen2VL's pad_token_id (151643 = <|endoftext|>)
        input_ids = pad_sequence(input_ids_list, batch_first=True, padding_value=self.pad_token_id)
        images_seq_mask = pad_sequence(images_seq_mask_list, batch_first=True, padding_value=False)

        # Truncate to max_length if specified (prevents OOM on long sequences)
        if self.max_length is not None and input_ids.shape[1] > self.max_length:
            input_ids = input_ids[:, :self.max_length]
            images_seq_mask = images_seq_mask[:, :self.max_length]
            # Adjust prompt_token_counts if they exceed max_length
            prompt_token_counts = [min(p, self.max_length) for p in prompt_token_counts]

        # Create labels
        labels = input_ids.clone()

        # Mask padding tokens
        labels[labels == self.pad_token_id] = -100

        # Mask image tokens (model shouldn't predict these)
        labels[images_seq_mask] = -100

        # Mask user prompt tokens when train_on_responses_only=True (only train on assistant responses)
        if self.train_on_responses_only:
            for idx, prompt_count in enumerate(prompt_token_counts):
                if prompt_count > 0:
                    labels[idx, :prompt_count] = -100

        # Create attention mask
        attention_mask = (input_ids != self.pad_token_id).long()

        images_batch = []
        for item in batch_data:
            images_batch.append((item['images_crop'], item['images_ori']))

        images_spatial_crop = torch.cat([item['images_spatial_crop'] for item in batch_data], dim=0)

        return {
            "input_ids": input_ids,
            "attention_mask": attention_mask,
            "labels": labels,
            "images": images_batch,
            "images_seq_mask": images_seq_mask,
            "images_spatial_crop": images_spatial_crop,
        }