File size: 27,688 Bytes
10b127e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import pandas as pd 
import ast
import numpy as np

arabic_letters = {
    'A': 'أ',
    'B': 'ب',
    'C': 'ج',
    'D': 'د',
    'E': 'ه',
    'F': 'و',
    'G': 'ز',
    'H': 'ح',
    'I': 'ط',
    'J': 'ي',
    'K': 'ك',
    'L': 'ل',
    'M': 'م',
    'N': 'ن',
    'O': 'ع',
    'P': 'ف',
    'Q': 'ص',
    'R': 'ر',
    'S': 'س',
    'هـ': 'ه',
    'ا': 'أ'
}

def translate_numbers(text: str) -> str:
    english_to_arabic = {
        '0': '٠', '1': '١', '2': '٢', '3': '٣', '4': '٤',
        '5': '٥', '6': '٦', '7': '٧', '8': '٨', '9': '٩'
    }
    
    translation_table = str.maketrans(english_to_arabic)
    return text.translate(translation_table)


def mcq_eval(pred: str, gt: str):
    pred = pred.strip()
    gt = gt.strip()
    if len(pred) > 2 and pred[0] == '(' and pred[2] == ')':
        pred = pred[1]
    if len(gt) > 2 and gt[0] == '(' and gt[2] == ')':
        gt = gt[1]
    if len(pred) == 0: return False
    pred = pred[0]
    gt = gt[0]
    pred = arabic_letters.get(pred, pred)
    gt = arabic_letters.get(gt, gt)
    return pred == gt

def create_options_prompt(row_data, option_candidate):
    available_keys = set(row_data.keys()) & set(option_candidate)
    options = {cand: row_data[cand] for cand in available_keys if row_data[cand]}
    sorted_options = dict(sorted(options.items()))
    options_prompt = f"هناك عدة خيارات:\n"
    for key, item in sorted_options.items():
        if pd.notna(item) and item != "nan":
            arabic_key = arabic_letters[key]
            options_prompt += f"{arabic_key}. {item}\n"
    return options_prompt.rstrip("\n")

def mmbench_doc_to_text(doc):
    option_candidate = ["A", "B", "C", "D", "E"]
    options = create_options_prompt(doc, option_candidate)
    question = f"{doc['hint']} {doc['question']} {options}" if pd.notna(doc["hint"]) and doc["hint"] != "nan" else f"{doc['question']} {options}"
    return f"{question}\nأجب بحرف الخيار من الاختيارات المعطاة مباشرة."

def mmbench_eval(pred, gt):
    return mcq_eval(pred, gt)

def mme_doc_to_text(doc):
    question = doc["question"].strip()
    return question

def mme_eval(pred: str, gt: str):
    pred = pred.strip()
    if pred == "صح":
        pred = 'نعم'
    return gt in pred
    
def default_eval(pred: str, gt: str):
    pred = pred.strip()
    gt = gt.strip()
    return pred == gt


def iconqa_doc_to_text(doc):
    return doc['question']

def iconqa_eval(pred, gt):
    return mcq_eval(pred, gt)

def mmmu_parse_options(options):
    option_letters = [arabic_letters[chr(ord("A") + i)] for i in range(len(options))]
    choices_str = "\n".join([f"{option_letter}. {option}" for option_letter, option in zip(option_letters, options)])
    return choices_str




MMMU_MULTI_CHOICE_PROMPT = "أجب بحرف الخيار من الاختيارات المعطاة مباشرة."
MMMU_OPEN_ENDED_PROMPT = "أجب عن السؤال باستخدام كلمة أو عبارة واحدة."
CONTEXT_KEYWORD = "السياق"

def mmmu_doc_to_text(doc):
    question = doc["question"]
    if doc["question_type"] == "multiple-choice":
        # Weirdly, data["options"] is a string in MMMU Huggingface dataset
        parsed_options = mmmu_parse_options(ast.literal_eval(doc["options"].replace("،", ",")))
        question = f"{question}\n{parsed_options}\n\n{MMMU_MULTI_CHOICE_PROMPT}"
    else:
        question = f"{question}\n\n{MMMU_OPEN_ENDED_PROMPT}"
    return question

def mmmu_eval(pred, gt):
    return mcq_eval(pred, gt)

def gqa_doc_to_text(doc):
    question = doc["question"]
    post_prompt = "\nأجب عن السؤال باستخدام كلمة أو عبارة واحدة."
    return f"{question}{post_prompt}"

def gqa_eval(pred, gt):
    return gt in pred

def realworldqa_doc_to_text(doc):
    question = doc["question"].strip()
    pre_prompt = "المستخدم\nالسؤال: "
    return f"{pre_prompt}{question}"

def realworldqa_eval(pred, gt):
    return mcq_eval(pred, gt)

def vqav2_doc_to_text(doc):
    post_prompt = "\nأجب على السؤال باستخدام كلمة أو عبارة واحدة."
    return f"{doc['question']}{post_prompt}"

def vizwiz_doc_to_text(doc):
    post_prompt = "\nعندما تكون المعلومات المقدمة غير كافية، أجب بـ 'لا يمكن الإجابة'.\nأجب عن السؤال باستخدام كلمة واحدة أو عبارة قصيرة."
    text = f"{doc['question'].capitalize()}{post_prompt}"
    return text

def vizwiz_eval(pred: str, gt: str):
    try:
        _ = ast.literal_eval(gt)
        gt = gt.replace(" ", ", ")
        gt = ast.literal_eval(gt)
    except:
        gt = gt.strip()
    pred = pred.strip()
    if pred == gt:
        return True
    for x in gt:
        if x in pred:
            return True
    return False

def pope_doc_to_text(doc):
    question = doc["question"].strip()
    return f"{question}\nأجب عن السؤال باستخدام كلمة واحدة أو عبارة قصيرة."

def pope_eval(pred, gt):
    pred = pred.strip()
    gt = gt.strip()
    return gt in pred

def countbench_doc_to_text(_):
    return "كم عدد الأشياء الموجودة في الصورة؟\nأجب برقم فقط."

def countbench_eval(pred, gt):
    pred = pred.strip()
    gt = gt.strip()
    return translate_numbers(pred) == translate_numbers(gt)

def diagramsMMMU_eval(pred, gt):
    pred = pred.strip()
    gt = gt.strip()
    if len(pred) == 0: return False
    if len(gt) == 1:
        return pred[0] == gt
    pred = translate_numbers(pred)
    gt = translate_numbers(gt)
    return gt in pred

def diagramsMMMU_doc_to_text(doc):
    return mmmu_doc_to_text(doc)

def medicalMMMU_eval(pred, gt):
    return mcq_eval(pred, gt)

def medicalMMMU_doc_to_text(doc):
    return mmmu_doc_to_text(doc)

def medicalMMMUPro_eval(pred, gt):
    return mcq_eval(pred, gt)

def medicalMMMUPro_parse_options(options):
    option_letters = [arabic_letters[chr(ord("A") + i)] for i in range(len(options))]
    choices_str = "\n".join([f"{option_letter}. {option}" for option_letter, option in zip(option_letters, options)])
    return choices_str


def medicalMMMUPro_doc_to_text(doc):
    post_prompt="أجب بحرف الخيار من الخيارات المعطاة مباشرة."
    question = doc["question"]
    # Weirdly, data["options"] is a string in MMMU Huggingface dataset
    parsed_options = medicalMMMUPro_parse_options(ast.literal_eval(doc["options"].replace("،", ",")))
    question = f"{question}\n{parsed_options}\n\n{post_prompt}"
    return question


def mmt_doc_to_text(doc):
    question_text = "سؤال: <image>\n" + doc["question"].strip()

    options = []
    for option in ["A", "B", "C", "D", "E", "F", "G", "H", "I"]:
        option_text = doc.get(option)
        if option_text and option_text.strip():
            options.append(f"{arabic_letters[option]}: {option_text.strip()}")

    options_text = "\n".join(options) if options else ""

    formatted_question = f"{question_text}\n{options_text}"
    post_prompt = "\nأجب عن السؤال باستخدام حرف واحد من الخيارات المعطاة."
    formatted_question = f"{formatted_question}{post_prompt}"

    return formatted_question

def medicalmmt_doc_to_text(doc):
    return mmt_doc_to_text(doc)

def medicalmmt_eval(pred, gt):
    return mcq_eval(pred, gt)

def seed_doc_to_text(doc):
    question = doc["question"]
    question += "\n" + f"أ. {doc['choice_a']}\n"
    question += f"ب. {doc['choice_b']}\n"
    question += f"ج. {doc['choice_c']}\n"
    question += f"د. {doc['choice_d']}"
    return f"{question}\nأجب بحرف الخيار من الاختيارات المعطاة مباشرة."

def seed_eval(pred, gt):
    return mcq_eval(pred, gt)

def hallucinationmmt_doc_to_text(doc):
    return mmt_doc_to_text(doc)

def hallucinationmmt_eval(pred, gt):
    return mcq_eval(pred, gt)

def vqammt_doc_to_text(doc):
    return mmt_doc_to_text(doc)

def vqammt_eval(pred, gt):
    return mcq_eval(pred, gt)



def mutliimagemmt_doc_to_text(doc):
    return mmt_doc_to_text(doc)

def mutliimagemmt_eval(pred, gt):
    return mcq_eval(pred, gt)


def our_options_to_str(options):
    option_prompt_str = ""
    for i, option in enumerate(options):
        option_choice = chr(ord("A") + i)
        option_choice = arabic_letters[option_choice]
        option_prompt_str += f"{option_choice}. {option}\n"

    option_prompt_str = option_prompt_str.rstrip("\n")
    return option_prompt_str

def our_doc_to_text(doc):
    question_text = "سؤال:\n" + doc["question"].strip()
    options = our_options_to_str(doc["options"])
    options_text = options
    formatted_question = f"{question_text}\n{options_text}"
    post_prompt = "\nأجب عن السؤال باستخدام حرف واحد من الخيارات المعطاة."
    formatted_question = f"{formatted_question}{post_prompt}"
    return formatted_question


def isidocvqa_doc_to_text(doc):
    return our_doc_to_text(doc)

def isidocvqa_eval(pred, gt):
    return mcq_eval(pred, gt) 

def patddocvqa_doc_to_text(doc):
    return our_doc_to_text(doc)

def patddocvqa_eval(pred, gt):
    return mcq_eval(pred, gt)

def celebvqa_doc_to_text(doc):
    return our_doc_to_text(doc)

def celebvqa_eval(pred, gt):
    return mcq_eval(pred, gt)

def countriesvqa_doc_to_text(doc):
    return our_doc_to_text(doc)

def countriesvqa_eval(pred, gt):
    return mcq_eval(pred, gt)

def foodvqa_doc_to_text(doc):
    return our_doc_to_text(doc)

def foodvqa_eval(pred, gt):
    return mcq_eval(pred, gt)

def objectcoco_doc_to_text(doc):
    return doc['question']

def objectcoco_eval(pred, gt):
    return mcq_eval(pred, gt)

def blink_doc_to_text(doc):
    return doc['question']

def blink_eval(pred, gt):
    return mcq_eval(pred, gt)

def examsv_doc_to_text(doc):
    return doc['question']

def examsv_eval(pred, gt):
    return mcq_eval(pred, gt)

def chartqa_doc_to_text(doc):
    return doc['question']

def chartqa_eval(pred, gt):
    return mcq_eval(pred, gt)

def mtvqa_doc_to_text(doc):
    return doc['question']

def create_one_query(problem):
    demo_prompt = ""

    question = problem["question"]
    unit = problem["unit"]
    choices = problem["choices"]
    precision = problem["precision"]
    question_type = problem["question_type"]
    answer_type = problem["answer_type"]

    if question_type == "multi_choice":
        assert answer_type == "text"
        hint_text = f"تلميح: يرجى الإجابة على السؤال وتقديم حرف الخيار الصحيح، مثل أ أو ب أو ج أو د، في النهاية."
    else:
        assert answer_type in ["integer", "float", "list"]
        if answer_type == "integer":
            hint_text = f"تلميح: يرجى الإجابة على السؤال الذي يتطلب إجابة بعدد صحيح وتقديم القيمة النهائية، مثل 1 أو 2 أو 3، في النهاية."

        elif answer_type == "float" and precision == 1:
            hint_text = f"تلميح: يرجى الإجابة على السؤال الذي يتطلب رقمًا عشريًا بمنزلة عشرية واحدة وتقديم القيمة النهائية، مثل 1.2 أو 1.3 أو 1.4، في النهاية."

        elif answer_type == "float" and precision == 2:
            hint_text = f"تلميح: يرجى الإجابة على السؤال الذي يتطلب رقمًا عشريًا بمنزلتين عشريتين وتقديم القيمة النهائية، مثل 1.23 أو 1.34 أو 1.45، في النهاية."

        elif answer_type == "list":
            hint_text = f"تلميح: يرجى الإجابة على السؤال الذي يتطلب قائمة بايثون كإجابة وتقديم القائمة النهائية، مثل [1, 2, 3] أو [1.2, 1.3, 1.4]، في النهاية."

    hint_text = translate_numbers(hint_text)    

    question_text = f"سؤال: {question}"
    if unit:
        question_text += f" (الوحدة: {unit})"

    if choices and choices != 'None':
        texts = ["الاختيارات:"]
        choices = ast.literal_eval(choices.replace("' '", '", "'))
        for i, choice in enumerate(choices):
            texts.append(f"({arabic_letters[chr(ord('A')+i)]}) {choice}")
        choices_text = "\n".join(texts)
    else:
        choices_text = ""


    prompt = "الحل: "

    elements = [question_text, choices_text, hint_text, prompt]
    test_query = "\n".join([e for e in elements if e != ""])

    query = demo_prompt + "\n\n" + test_query
    query = query.strip()
    return query

def mathvista_doc_to_text(doc):
    problem = {
        "question_type": doc["question_type"],
        "answer_type": doc["answer_type"],
        "question": doc["question"],
        "unit": doc["unit"] if "unit" in doc else "",
        "choices": doc["choices"],
        "answer": doc["answer"] if "answer" in doc else None,
        "precision": doc["precision"] if "precision" in doc else 0,
    }
    query_prompt = create_one_query(problem)
    return query_prompt


def infographicsvqa_doc_to_text(doc):
    return doc['question']

def infographicsvqa_eval(pred, gt):
    return mcq_eval(pred, gt)

def agrovqa_doc_to_text(doc):
    return doc['question']

def agrovqa_eval(pred, gt):
    return mcq_eval(pred, gt)

def diagramsvqa_doc_to_text(doc):
    return doc['question']

def diagramsvqa_eval(pred, gt):
    return mcq_eval(pred, gt)

def tablesvqa_doc_to_text(doc):
    return doc['question']

def tablesvqa_eval(pred, gt):
    return mcq_eval(pred, gt)

def scienceqa_doc_to_text(doc):
    context, question, choices = doc["hint"], doc["question"], doc["choices"]
    choices = ast.literal_eval(choices)
    len_choices = len(choices)
    options = [arabic_letters[chr(ord("A") + i)] for i in range(len_choices)]
    choices_str = "\n".join([f"{option}. {choice}" for option, choice in zip(options, choices)])
    if context:
        context = f"السياق: {context}\n"

    post_prompt = "\n.أجب بحرف الخيار من الخيارات المعطاة مباشرة"
    return f"{context}{question}\n{choices_str}{post_prompt}"

def scienceqa_eval(pred, gt):
    gt = arabic_letters[chr(ord('A') + int(gt))]
    return mcq_eval(pred, gt)

def ocrisi_doc_to_text(doc):
    return doc['question']

def cer(pred, gt):
    d = np.zeros((len(gt) + 1, len(pred) + 1))
    for i in range(len(gt) + 1):
        d[i, 0] = i
    for j in range(len(pred) + 1):
        d[0, j] = j
    
    for i in range(1, len(gt) + 1):
        for j in range(1, len(pred) + 1):
            if gt[i-1] == pred[j-1]:
                d[i, j] = d[i-1, j-1]
            else:
                substitution = d[i-1, j-1] + 1
                insertion = d[i, j-1] + 1
                deletion = d[i-1, j] + 1
                d[i, j] = min(substitution, insertion, deletion)
    
    error = d[len(gt), len(pred)]
    total_chars = len(gt)
    cer = error / total_chars
    
    return cer

from fast_edit_distance import edit_distance
def fastcer(pred, gt):
    ed = edit_distance(pred, gt, max_ed=1000)
    if len(pred) > len(gt):
        return ed / len(pred)
    return ed /len(gt)


def ocrisi_eval(pred: str, gt: str):
    pred = pred.strip()
    gt = gt.strip()
    return fastcer(pred, gt) <= 0.3

def evarest_doc_to_text(doc):
    return doc['question']

def evarest_eval(pred: str, gt: str):
    pred = pred.strip()
    gt = gt.strip()
    return fastcer(pred, gt) <= 0.5

def historicalbooks_doc_to_text(doc):
    return doc['question']

def historicalbooks_eval(pred: str, gt: str):
    pred = pred.strip()
    gt = gt.strip()
    return fastcer(pred, gt) <= 0.2

def khatt_doc_to_text(doc):
    return doc['question']

def khatt_eval(pred: str, gt: str):
    pred = pred.strip()
    gt = gt.strip()
    return fastcer(pred, gt) <= 0.3

def patsocr_doc_to_text(doc):
    return doc['question']

def patsocr_eval(pred: str, gt: str):
    pred = pred.strip()
    gt = gt.strip()
    return fastcer(pred, gt) <= 0.5

def arabicocr_doc_to_text(doc):
    return doc['question']

def arabicocr_eval(pred: str, gt: str):
    pred = pred.strip()
    gt = gt.strip()
    return fastcer(pred, gt) <= 0.4


def culturevideovqa_doc_to_text(doc):
    return doc['question']

def culturevideovqa_eval(pred, gt):
    return mcq_eval(pred, gt)


def videomme_doc_to_text(doc):
    prompt = """الترجمات النصية لهذا الفيديو مدرجة أدناه:
{subtitles}
اختر أفضل إجابة للسؤال التالي متعدد الخيارات بناءً على الفيديو. أجب فقط بالحرف (أ، ب، ج، أو د) للخيار الصحيح.
{question}
{options}
أفضل إجابة هي:"""
    subtitles = doc["subtitles"]
    question = doc["question"]
    options = doc["options"]
    options_str = "\n".join([f"{option}" for i, option in enumerate(options)])
    return prompt.format(subtitles=subtitles, question=question, options=options_str)

def videomme_eval(pred, gt):
    return mcq_eval(pred, gt)
    
def geochat_doc_to_text(doc):
    pre_prompt = "أجب على السؤال التالي بكلمة أو جملة.\n"
    return pre_prompt + doc['question']

def muribench_doc_to_text(doc):
    return doc['question']

def muribench_eval(pred, gt):
    return mcq_eval(pred, gt)

def mmlu_doc_to_text(doc):
    question = doc['Question']
    options = []
    for i in range(1, 6):
        if doc[f'Option {i}'] is not None:
            options.append(doc[f'Option {i}'])
    parsed_options = mmmu_parse_options(options)
    context = doc['Context']
    if context: 
        return f"{question}\n{CONTEXT_KEYWORD}: {context}\n{parsed_options}\n{MMMU_MULTI_CHOICE_PROMPT}"
    return f"{question}\n{parsed_options}\n{MMMU_MULTI_CHOICE_PROMPT}"

def mmlu_eval(pred, gt):
    return mcq_eval(pred, gt)

def henna_doc_to_text(doc):
    return doc['question']


mtvqa_user_prompt = """Here are three examples to guide your evaluation:
Example 1:
Question: "ما هي اللغة المستخدمة في النص؟"
Predicted Answer: "العربية"
Ground Truth: "اللغة العربية"
Score: 1

Example 2:
Question: "ما هو موضوع النص؟"
Predicted Answer: "إثنان"
Ground Truth: "الحب و الكراهية"
Score: 0

Example 3:
Question: "ما هو عدد صفحات الكتاب؟"
Predicted Answer: "الصورة لا تحتوي على عدد صفحات الكتاب."
Ground Truth: "غير معروف"
Score: 1

Now, for each new pair of sentences, analyze their semantic similarity and provide a score of 1 for similar meanings or 0 for different meanings. Always consider the context and potential variations in expressing the same concept.
Question: "{question}"
Predicted Answer: "{pred}"
Ground Truth: "{gt}"
Score: """

geochat_user_prompt = """Here are three examples to guide your evaluation:
Example 1:
Question: "أجب على السؤال التالي بكلمة أو جملة.\n<image>\n كم عدد قوارب الصيد الموجودة في الصورة؟"
Predicted Answer: "قارب صيد واحد"
Ground Truth: "يوجد قارب صيد واحد في الصورة."
Score: 1

Example 2:
Question: "أجب على السؤال التالي بكلمة أو جملة.\n<image>\nكم عدد السيارات الصغيرة الموجودة في الصورة؟"
Predicted Answer: "عدد السيارات الصغيرة هو 10."
Ground Truth: "هناك سيارة صغيرة واحدة في الصورة."
Score: 0

Example 3:
Question: "أجب على السؤال التالي بكلمة أو جملة.\n<image>\nهل يمكنك إخباري بعدد الأشجار الموجودة في أعلى الصورة؟"
Predicted Answer: "توجد عدة أشجار في أعلى الصورة، ومن الصعب تحديد العدد الدقيق بسبب تداخلها."
Ground Truth: "يوجد بعض الأشجار في الأعلى"
Score: 1

Now, for each new pair of sentences, analyze their semantic similarity and provide a score of 1 for similar meanings or 0 for different meanings. Always consider the context and potential variations in expressing the same concept.
Question: "{question}"
Predicted Answer: "{pred}"
Ground Truth: "{gt}"
Score: """

mathvista_user_prompt = """Here are three examples to guide your evaluation:
Example 1:
Question: "سؤال: أي كائن سيتأثر أكثر إذا تم القضاء على الطحالب؟\nالاختيارات:\n(أ) البلطي\n(ب) برغوث الماء الشائع\n(ج) خنفساء الغوص العظيمة\n(د) الشرغوف\nتلميح: يرجى الإجابة على السؤال وتقديم حرف الخيار الصحيح، مثل أ أو ب أو ج أو د، في النهاية.\nالحل: "
Predicted Answer: "ب) برغوث الماء الشائع\nالحل: ب"
Ground Truth: "برغوث الماء الشائع"
Score: 1

Example 2:
Question: "سؤال: تتقاطع منصفات الزاويتين الداخليتين للمثلث △ABC في النقطة O، إذا كانت ∠A = 110°، فكم تساوي ∠BOC؟\nالاختيارات:\n(أ) ١٣٥°\n(ب) ١٤٠°\n(ج) ١٤٥°\n(د) ١٥٠°\nتلميح: يرجى الإجابة على السؤال وتقديم حرف الخيار الصحيح، مثل أ أو ب أو ج أو د، في النهاية.\nالحل: "
Predicted Answer: "لحساب ∠BOC، نستخدم حقيقة أن مجموع الزوايا في المثلث هو 180°. إذا كانت زاوية A = 110°، فإن مجموع الزاويتين B + C هو 70° (لأن 180° - 110° = 70°).<br>\nعندما تتقاطع منصفات الزاويتين B وC عند النقطة O، يمكن حساب ∠BOC باستخدام العلاقة الخاصة بزوايا المنصفات:\n2∠BOC = 180° + ∠A.<br>\nلذا:<br>\n2∠BOC = 180° + 110° = 290°<br>\n∠BOC = 290° / 2 = 145°<br>\nوبذلك، الخيار الصحيح هو (ج) ١٤٥°."
Ground Truth: "ب"
Score: 0

Example 3:
Question: "سؤال: مشتقة f(x) عند x=2 هي ____ عند x=5\nالاختيارات:\n(أ) أكبر من\n(ب) تساوي\n(ج) أصغر من\nتلميح: يرجى الإجابة على السؤال وتقديم حرف الخيار الصحيح، مثل أ أو ب أو ج أو د، في النهاية.\nالحل: "
Predicted Answer: "الوح بالاعتماد على توضيح الصورة، مشتقة الدالة f(x) = |2x - 3| + 1 عند نقاط مختلفة تعتمد على كيفية تغير ميل المنحنى. عند x=2 (يمين رأس الدالة)، الميل هو +2. عند x=5، الميل أيضًا +2 لأن هذا الجزء من الدالة خطي بزاوية صعودية مستمرة.\n\nلذا، مشتقة f(x) عند x=2 تساوي مشتقتها عند x=5.\n\nالإجابة الصحيحة هي (ب) تساوي."
Ground Truth: "تساوي"
Score: 1

Now, for each new pair of sentences, analyze their semantic similarity and provide a score of 1 for similar meanings or 0 for different meanings. Always consider the context and potential variations in expressing the same concept.
Question: "{question}"
Predicted Answer: "{pred}"
Ground Truth: "{gt}"
Score: """

vqav2_user_prompt = """Here are three examples to guide your evaluation:
Example 1:
Question: "ما لون الحوض؟\nأجب على السؤال باستخدام كلمة أو عبارة واحدة."
Predicted Answer: "أبيض"
Ground Truth: "أبيض"
Score: 1

Example 2:
Question: "كم عدد الأشخاص في الصورة؟\nأجب على السؤال باستخدام كلمة أو عبارة واحدة."
Predicted Answer: "اثنان"
Ground Truth: "٣"
Score: 0

Example 3:
Question: "ما هو شكل الطائرة الورقية؟\nأجب على السؤال باستخدام كلمة أو عبارة واحدة."
Predicted Answer: "سداسي"
Ground Truth: "سداسي الأضلاع"
Score: 1

Now, for each new pair of sentences, analyze their semantic similarity and provide a score of 1 for similar meanings or 0 for different meanings. Always consider the context and potential variations in expressing the same concept.
Question: "{question}"
Predicted Answer: "{pred}"
Ground Truth: "{gt}"
Score: """



henna_user_prompt = """Here are three examples to guide your evaluation:
Example 1:
Question: "هل تظهر علامات الحضارة البشرية مثل البناء أو الطرق في الصورة؟"
Predicted Answer: "لا"
Ground Truth: "لا، لا تظهر في الصورة علامات الحضارة البشرية مثل البناء أو الطرق، وتبدو المنطقة طبيعية وغير معبأة بالتطورات البشرية."
Score: 1

Example 2:
Question: "ما العناصر المعمارية المرئية على باب اليمن التي تدل على قدمه؟"
Predicted Answer: "أ"
Ground Truth: "العناصر المعمارية المرئية على باب اليمن والتي تدل على قدمه تشمل الأبراج الدفاعية المستديرة، الزخارف والنقوش على الجدران، والأقواس الضخمة للبوابة التي تعود إلى العمارة الإسلامية التقليدية."
Score: 0

Example 3:
Question: "ما هي المظاهر الثقافية أو الاجتماعية التي يمكن استنتاجها من الملابس التقليدية للأشخاص في الصورة؟"
Predicted Answer: "استناداً إلى الصورة، يمكن استنتاج عدة مظاهر ثقافية أو اجتماعية من الملابس التقليدية للأشخاص:\n\n1. الزي التقليدي: يرتدي العديد من الأشخاص في الصورة ملابس تقليدية، مما يشير إلى أن هذا قد يكون مهرجاناً أو احتفالاً ثقافياً أو ممارسة اجتماعية حيث يرتدي الناس ملابس تقليدية.\n\n2. التنوع في الملابس: تشير مجموعة الألوان والأنماط المختلفة للملابس إلى وجود تنوع ثقافي أو اجتماعي داخل المجتمع. قد يمثل هذا تنوعاً في الأصول أو الأعراق أو الممارسات الثقافية المختلفة.\n\n3. الأزياء الرسمية: تشير وجود أشخاص يرتدون ملابس رسمية، بما في ذلك الأوشحة، إلى أن هذا قد يكون حدثاً اجتماعياً أو ثقافياً رسمياً أو مهنياً.\n\n4. التجمعات الجماعية: يشير وجود العديد من الأشخاص الذين يرتدون ملابس تقليدية إلى أن"
Ground Truth: "الملابس التقليدية التي يرتديها الأشخاص تشير إلى التمسك بالتراث والثقافة اليمنية، والتي تعبر عن هوية المجتمع وتنوعه الثقافي."
Score: 1

Now, for each new pair of sentences, analyze their semantic similarity and provide a score of 1 for similar meanings or 0 for different meanings. Always consider the context and potential variations in expressing the same concept.
Question: "{question}"
Predicted Answer: "{pred}"
Ground Truth: "{gt}"
Score: """

fuzz_eval_system_prompt = "You are an expert in natural language understanding and semantic similarity. Your task is to evaluate the semantic similarity between two given sentences: a predicted answer and a ground truth answer. You should output a score of 1 if the sentences are semantically similar, and 0 if they are not."