ahmedheakl commited on
Commit
10b127e
·
verified ·
1 Parent(s): cb2eba1

Create utils.py

Browse files
Files changed (1) hide show
  1. utils.py +711 -0
utils.py ADDED
@@ -0,0 +1,711 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ import pandas as pd
2
+ import ast
3
+ import numpy as np
4
+
5
+ arabic_letters = {
6
+ 'A': 'أ',
7
+ 'B': 'ب',
8
+ 'C': 'ج',
9
+ 'D': 'د',
10
+ 'E': 'ه',
11
+ 'F': 'و',
12
+ 'G': 'ز',
13
+ 'H': 'ح',
14
+ 'I': 'ط',
15
+ 'J': 'ي',
16
+ 'K': 'ك',
17
+ 'L': 'ل',
18
+ 'M': 'م',
19
+ 'N': 'ن',
20
+ 'O': 'ع',
21
+ 'P': 'ف',
22
+ 'Q': 'ص',
23
+ 'R': 'ر',
24
+ 'S': 'س',
25
+ 'هـ': 'ه',
26
+ 'ا': 'أ'
27
+ }
28
+
29
+ def translate_numbers(text: str) -> str:
30
+ english_to_arabic = {
31
+ '0': '٠', '1': '١', '2': '٢', '3': '٣', '4': '٤',
32
+ '5': '٥', '6': '٦', '7': '٧', '8': '٨', '9': '٩'
33
+ }
34
+
35
+ translation_table = str.maketrans(english_to_arabic)
36
+ return text.translate(translation_table)
37
+
38
+
39
+ def mcq_eval(pred: str, gt: str):
40
+ pred = pred.strip()
41
+ gt = gt.strip()
42
+ if len(pred) > 2 and pred[0] == '(' and pred[2] == ')':
43
+ pred = pred[1]
44
+ if len(gt) > 2 and gt[0] == '(' and gt[2] == ')':
45
+ gt = gt[1]
46
+ if len(pred) == 0: return False
47
+ pred = pred[0]
48
+ gt = gt[0]
49
+ pred = arabic_letters.get(pred, pred)
50
+ gt = arabic_letters.get(gt, gt)
51
+ return pred == gt
52
+
53
+ def create_options_prompt(row_data, option_candidate):
54
+ available_keys = set(row_data.keys()) & set(option_candidate)
55
+ options = {cand: row_data[cand] for cand in available_keys if row_data[cand]}
56
+ sorted_options = dict(sorted(options.items()))
57
+ options_prompt = f"هناك عدة خيارات:\n"
58
+ for key, item in sorted_options.items():
59
+ if pd.notna(item) and item != "nan":
60
+ arabic_key = arabic_letters[key]
61
+ options_prompt += f"{arabic_key}. {item}\n"
62
+ return options_prompt.rstrip("\n")
63
+
64
+ def mmbench_doc_to_text(doc):
65
+ option_candidate = ["A", "B", "C", "D", "E"]
66
+ options = create_options_prompt(doc, option_candidate)
67
+ question = f"{doc['hint']} {doc['question']} {options}" if pd.notna(doc["hint"]) and doc["hint"] != "nan" else f"{doc['question']} {options}"
68
+ return f"{question}\nأجب بحرف الخيار من الاختيارات المعطاة مباشرة."
69
+
70
+ def mmbench_eval(pred, gt):
71
+ return mcq_eval(pred, gt)
72
+
73
+ def mme_doc_to_text(doc):
74
+ question = doc["question"].strip()
75
+ return question
76
+
77
+ def mme_eval(pred: str, gt: str):
78
+ pred = pred.strip()
79
+ if pred == "صح":
80
+ pred = 'نعم'
81
+ return gt in pred
82
+
83
+ def default_eval(pred: str, gt: str):
84
+ pred = pred.strip()
85
+ gt = gt.strip()
86
+ return pred == gt
87
+
88
+
89
+ def iconqa_doc_to_text(doc):
90
+ return doc['question']
91
+
92
+ def iconqa_eval(pred, gt):
93
+ return mcq_eval(pred, gt)
94
+
95
+ def mmmu_parse_options(options):
96
+ option_letters = [arabic_letters[chr(ord("A") + i)] for i in range(len(options))]
97
+ choices_str = "\n".join([f"{option_letter}. {option}" for option_letter, option in zip(option_letters, options)])
98
+ return choices_str
99
+
100
+
101
+
102
+
103
+ MMMU_MULTI_CHOICE_PROMPT = "أجب بحرف الخيار من الاختيارات المعطاة مباشرة."
104
+ MMMU_OPEN_ENDED_PROMPT = "أجب عن السؤال باستخدام كلمة أو عبارة واحدة."
105
+ CONTEXT_KEYWORD = "السياق"
106
+
107
+ def mmmu_doc_to_text(doc):
108
+ question = doc["question"]
109
+ if doc["question_type"] == "multiple-choice":
110
+ # Weirdly, data["options"] is a string in MMMU Huggingface dataset
111
+ parsed_options = mmmu_parse_options(ast.literal_eval(doc["options"].replace("،", ",")))
112
+ question = f"{question}\n{parsed_options}\n\n{MMMU_MULTI_CHOICE_PROMPT}"
113
+ else:
114
+ question = f"{question}\n\n{MMMU_OPEN_ENDED_PROMPT}"
115
+ return question
116
+
117
+ def mmmu_eval(pred, gt):
118
+ return mcq_eval(pred, gt)
119
+
120
+ def gqa_doc_to_text(doc):
121
+ question = doc["question"]
122
+ post_prompt = "\nأجب عن السؤال باستخدام كلمة أو عبارة واحدة."
123
+ return f"{question}{post_prompt}"
124
+
125
+ def gqa_eval(pred, gt):
126
+ return gt in pred
127
+
128
+ def realworldqa_doc_to_text(doc):
129
+ question = doc["question"].strip()
130
+ pre_prompt = "المستخدم\nالسؤال: "
131
+ return f"{pre_prompt}{question}"
132
+
133
+ def realworldqa_eval(pred, gt):
134
+ return mcq_eval(pred, gt)
135
+
136
+ def vqav2_doc_to_text(doc):
137
+ post_prompt = "\nأجب على السؤال باستخدام كلمة أو عبارة واحدة."
138
+ return f"{doc['question']}{post_prompt}"
139
+
140
+ def vizwiz_doc_to_text(doc):
141
+ post_prompt = "\nعندما تكون المعلومات المقدمة غير كافية، أجب بـ 'لا يمكن الإجابة'.\nأجب عن السؤال باستخدام كلمة واحدة أو عبارة قصيرة."
142
+ text = f"{doc['question'].capitalize()}{post_prompt}"
143
+ return text
144
+
145
+ def vizwiz_eval(pred: str, gt: str):
146
+ try:
147
+ _ = ast.literal_eval(gt)
148
+ gt = gt.replace(" ", ", ")
149
+ gt = ast.literal_eval(gt)
150
+ except:
151
+ gt = gt.strip()
152
+ pred = pred.strip()
153
+ if pred == gt:
154
+ return True
155
+ for x in gt:
156
+ if x in pred:
157
+ return True
158
+ return False
159
+
160
+ def pope_doc_to_text(doc):
161
+ question = doc["question"].strip()
162
+ return f"{question}\nأجب عن السؤال باستخدام كلمة واحدة أو عبارة قصيرة."
163
+
164
+ def pope_eval(pred, gt):
165
+ pred = pred.strip()
166
+ gt = gt.strip()
167
+ return gt in pred
168
+
169
+ def countbench_doc_to_text(_):
170
+ return "كم عدد الأشياء الموجودة في الصورة؟\nأجب برقم فقط."
171
+
172
+ def countbench_eval(pred, gt):
173
+ pred = pred.strip()
174
+ gt = gt.strip()
175
+ return translate_numbers(pred) == translate_numbers(gt)
176
+
177
+ def diagramsMMMU_eval(pred, gt):
178
+ pred = pred.strip()
179
+ gt = gt.strip()
180
+ if len(pred) == 0: return False
181
+ if len(gt) == 1:
182
+ return pred[0] == gt
183
+ pred = translate_numbers(pred)
184
+ gt = translate_numbers(gt)
185
+ return gt in pred
186
+
187
+ def diagramsMMMU_doc_to_text(doc):
188
+ return mmmu_doc_to_text(doc)
189
+
190
+ def medicalMMMU_eval(pred, gt):
191
+ return mcq_eval(pred, gt)
192
+
193
+ def medicalMMMU_doc_to_text(doc):
194
+ return mmmu_doc_to_text(doc)
195
+
196
+ def medicalMMMUPro_eval(pred, gt):
197
+ return mcq_eval(pred, gt)
198
+
199
+ def medicalMMMUPro_parse_options(options):
200
+ option_letters = [arabic_letters[chr(ord("A") + i)] for i in range(len(options))]
201
+ choices_str = "\n".join([f"{option_letter}. {option}" for option_letter, option in zip(option_letters, options)])
202
+ return choices_str
203
+
204
+
205
+ def medicalMMMUPro_doc_to_text(doc):
206
+ post_prompt="أجب بحرف الخيار من الخيارات المعطاة مباشرة."
207
+ question = doc["question"]
208
+ # Weirdly, data["options"] is a string in MMMU Huggingface dataset
209
+ parsed_options = medicalMMMUPro_parse_options(ast.literal_eval(doc["options"].replace("،", ",")))
210
+ question = f"{question}\n{parsed_options}\n\n{post_prompt}"
211
+ return question
212
+
213
+
214
+ def mmt_doc_to_text(doc):
215
+ question_text = "سؤال: <image>\n" + doc["question"].strip()
216
+
217
+ options = []
218
+ for option in ["A", "B", "C", "D", "E", "F", "G", "H", "I"]:
219
+ option_text = doc.get(option)
220
+ if option_text and option_text.strip():
221
+ options.append(f"{arabic_letters[option]}: {option_text.strip()}")
222
+
223
+ options_text = "\n".join(options) if options else ""
224
+
225
+ formatted_question = f"{question_text}\n{options_text}"
226
+ post_prompt = "\nأجب عن السؤال باستخدام حرف واحد من الخيارات المعطاة."
227
+ formatted_question = f"{formatted_question}{post_prompt}"
228
+
229
+ return formatted_question
230
+
231
+ def medicalmmt_doc_to_text(doc):
232
+ return mmt_doc_to_text(doc)
233
+
234
+ def medicalmmt_eval(pred, gt):
235
+ return mcq_eval(pred, gt)
236
+
237
+ def seed_doc_to_text(doc):
238
+ question = doc["question"]
239
+ question += "\n" + f"أ. {doc['choice_a']}\n"
240
+ question += f"ب. {doc['choice_b']}\n"
241
+ question += f"ج. {doc['choice_c']}\n"
242
+ question += f"د. {doc['choice_d']}"
243
+ return f"{question}\nأجب بحرف الخيار من الاختيارات المعطاة مباشرة."
244
+
245
+ def seed_eval(pred, gt):
246
+ return mcq_eval(pred, gt)
247
+
248
+ def hallucinationmmt_doc_to_text(doc):
249
+ return mmt_doc_to_text(doc)
250
+
251
+ def hallucinationmmt_eval(pred, gt):
252
+ return mcq_eval(pred, gt)
253
+
254
+ def vqammt_doc_to_text(doc):
255
+ return mmt_doc_to_text(doc)
256
+
257
+ def vqammt_eval(pred, gt):
258
+ return mcq_eval(pred, gt)
259
+
260
+
261
+
262
+ def mutliimagemmt_doc_to_text(doc):
263
+ return mmt_doc_to_text(doc)
264
+
265
+ def mutliimagemmt_eval(pred, gt):
266
+ return mcq_eval(pred, gt)
267
+
268
+
269
+ def our_options_to_str(options):
270
+ option_prompt_str = ""
271
+ for i, option in enumerate(options):
272
+ option_choice = chr(ord("A") + i)
273
+ option_choice = arabic_letters[option_choice]
274
+ option_prompt_str += f"{option_choice}. {option}\n"
275
+
276
+ option_prompt_str = option_prompt_str.rstrip("\n")
277
+ return option_prompt_str
278
+
279
+ def our_doc_to_text(doc):
280
+ question_text = "سؤال:\n" + doc["question"].strip()
281
+ options = our_options_to_str(doc["options"])
282
+ options_text = options
283
+ formatted_question = f"{question_text}\n{options_text}"
284
+ post_prompt = "\nأجب عن السؤال باستخدام حرف واحد من الخيارات المعطاة."
285
+ formatted_question = f"{formatted_question}{post_prompt}"
286
+ return formatted_question
287
+
288
+
289
+ def isidocvqa_doc_to_text(doc):
290
+ return our_doc_to_text(doc)
291
+
292
+ def isidocvqa_eval(pred, gt):
293
+ return mcq_eval(pred, gt)
294
+
295
+ def patddocvqa_doc_to_text(doc):
296
+ return our_doc_to_text(doc)
297
+
298
+ def patddocvqa_eval(pred, gt):
299
+ return mcq_eval(pred, gt)
300
+
301
+ def celebvqa_doc_to_text(doc):
302
+ return our_doc_to_text(doc)
303
+
304
+ def celebvqa_eval(pred, gt):
305
+ return mcq_eval(pred, gt)
306
+
307
+ def countriesvqa_doc_to_text(doc):
308
+ return our_doc_to_text(doc)
309
+
310
+ def countriesvqa_eval(pred, gt):
311
+ return mcq_eval(pred, gt)
312
+
313
+ def foodvqa_doc_to_text(doc):
314
+ return our_doc_to_text(doc)
315
+
316
+ def foodvqa_eval(pred, gt):
317
+ return mcq_eval(pred, gt)
318
+
319
+ def objectcoco_doc_to_text(doc):
320
+ return doc['question']
321
+
322
+ def objectcoco_eval(pred, gt):
323
+ return mcq_eval(pred, gt)
324
+
325
+ def blink_doc_to_text(doc):
326
+ return doc['question']
327
+
328
+ def blink_eval(pred, gt):
329
+ return mcq_eval(pred, gt)
330
+
331
+ def examsv_doc_to_text(doc):
332
+ return doc['question']
333
+
334
+ def examsv_eval(pred, gt):
335
+ return mcq_eval(pred, gt)
336
+
337
+ def chartqa_doc_to_text(doc):
338
+ return doc['question']
339
+
340
+ def chartqa_eval(pred, gt):
341
+ return mcq_eval(pred, gt)
342
+
343
+ def mtvqa_doc_to_text(doc):
344
+ return doc['question']
345
+
346
+ def create_one_query(problem):
347
+ demo_prompt = ""
348
+
349
+ question = problem["question"]
350
+ unit = problem["unit"]
351
+ choices = problem["choices"]
352
+ precision = problem["precision"]
353
+ question_type = problem["question_type"]
354
+ answer_type = problem["answer_type"]
355
+
356
+ if question_type == "multi_choice":
357
+ assert answer_type == "text"
358
+ hint_text = f"تلميح: يرجى الإجابة على السؤال وتقديم حرف الخيار الصحيح، مثل أ أو ب أو ج أو د، في النهاية."
359
+ else:
360
+ assert answer_type in ["integer", "float", "list"]
361
+ if answer_type == "integer":
362
+ hint_text = f"تلميح: يرجى الإجابة على السؤال الذي يتطلب إجابة بعدد صحيح وتقديم القيمة النهائية، مثل 1 أو 2 أو 3، في النهاية."
363
+
364
+ elif answer_type == "float" and precision == 1:
365
+ 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."
366
+
367
+ elif answer_type == "float" and precision == 2:
368
+ 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."
369
+
370
+ elif answer_type == "list":
371
+ 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."
372
+
373
+ hint_text = translate_numbers(hint_text)
374
+
375
+ question_text = f"سؤال: {question}"
376
+ if unit:
377
+ question_text += f" (الوحدة: {unit})"
378
+
379
+ if choices and choices != 'None':
380
+ texts = ["الاختيارات:"]
381
+ choices = ast.literal_eval(choices.replace("' '", '", "'))
382
+ for i, choice in enumerate(choices):
383
+ texts.append(f"({arabic_letters[chr(ord('A')+i)]}) {choice}")
384
+ choices_text = "\n".join(texts)
385
+ else:
386
+ choices_text = ""
387
+
388
+
389
+ prompt = "الحل: "
390
+
391
+ elements = [question_text, choices_text, hint_text, prompt]
392
+ test_query = "\n".join([e for e in elements if e != ""])
393
+
394
+ query = demo_prompt + "\n\n" + test_query
395
+ query = query.strip()
396
+ return query
397
+
398
+ def mathvista_doc_to_text(doc):
399
+ problem = {
400
+ "question_type": doc["question_type"],
401
+ "answer_type": doc["answer_type"],
402
+ "question": doc["question"],
403
+ "unit": doc["unit"] if "unit" in doc else "",
404
+ "choices": doc["choices"],
405
+ "answer": doc["answer"] if "answer" in doc else None,
406
+ "precision": doc["precision"] if "precision" in doc else 0,
407
+ }
408
+ query_prompt = create_one_query(problem)
409
+ return query_prompt
410
+
411
+
412
+ def infographicsvqa_doc_to_text(doc):
413
+ return doc['question']
414
+
415
+ def infographicsvqa_eval(pred, gt):
416
+ return mcq_eval(pred, gt)
417
+
418
+ def agrovqa_doc_to_text(doc):
419
+ return doc['question']
420
+
421
+ def agrovqa_eval(pred, gt):
422
+ return mcq_eval(pred, gt)
423
+
424
+ def diagramsvqa_doc_to_text(doc):
425
+ return doc['question']
426
+
427
+ def diagramsvqa_eval(pred, gt):
428
+ return mcq_eval(pred, gt)
429
+
430
+ def tablesvqa_doc_to_text(doc):
431
+ return doc['question']
432
+
433
+ def tablesvqa_eval(pred, gt):
434
+ return mcq_eval(pred, gt)
435
+
436
+ def scienceqa_doc_to_text(doc):
437
+ context, question, choices = doc["hint"], doc["question"], doc["choices"]
438
+ choices = ast.literal_eval(choices)
439
+ len_choices = len(choices)
440
+ options = [arabic_letters[chr(ord("A") + i)] for i in range(len_choices)]
441
+ choices_str = "\n".join([f"{option}. {choice}" for option, choice in zip(options, choices)])
442
+ if context:
443
+ context = f"السياق: {context}\n"
444
+
445
+ post_prompt = "\n.أجب بحرف الخيار من الخيارات المعطاة مباشرة"
446
+ return f"{context}{question}\n{choices_str}{post_prompt}"
447
+
448
+ def scienceqa_eval(pred, gt):
449
+ gt = arabic_letters[chr(ord('A') + int(gt))]
450
+ return mcq_eval(pred, gt)
451
+
452
+ def ocrisi_doc_to_text(doc):
453
+ return doc['question']
454
+
455
+ def cer(pred, gt):
456
+ d = np.zeros((len(gt) + 1, len(pred) + 1))
457
+ for i in range(len(gt) + 1):
458
+ d[i, 0] = i
459
+ for j in range(len(pred) + 1):
460
+ d[0, j] = j
461
+
462
+ for i in range(1, len(gt) + 1):
463
+ for j in range(1, len(pred) + 1):
464
+ if gt[i-1] == pred[j-1]:
465
+ d[i, j] = d[i-1, j-1]
466
+ else:
467
+ substitution = d[i-1, j-1] + 1
468
+ insertion = d[i, j-1] + 1
469
+ deletion = d[i-1, j] + 1
470
+ d[i, j] = min(substitution, insertion, deletion)
471
+
472
+ error = d[len(gt), len(pred)]
473
+ total_chars = len(gt)
474
+ cer = error / total_chars
475
+
476
+ return cer
477
+
478
+ from fast_edit_distance import edit_distance
479
+ def fastcer(pred, gt):
480
+ ed = edit_distance(pred, gt, max_ed=1000)
481
+ if len(pred) > len(gt):
482
+ return ed / len(pred)
483
+ return ed /len(gt)
484
+
485
+
486
+ def ocrisi_eval(pred: str, gt: str):
487
+ pred = pred.strip()
488
+ gt = gt.strip()
489
+ return fastcer(pred, gt) <= 0.3
490
+
491
+ def evarest_doc_to_text(doc):
492
+ return doc['question']
493
+
494
+ def evarest_eval(pred: str, gt: str):
495
+ pred = pred.strip()
496
+ gt = gt.strip()
497
+ return fastcer(pred, gt) <= 0.5
498
+
499
+ def historicalbooks_doc_to_text(doc):
500
+ return doc['question']
501
+
502
+ def historicalbooks_eval(pred: str, gt: str):
503
+ pred = pred.strip()
504
+ gt = gt.strip()
505
+ return fastcer(pred, gt) <= 0.2
506
+
507
+ def khatt_doc_to_text(doc):
508
+ return doc['question']
509
+
510
+ def khatt_eval(pred: str, gt: str):
511
+ pred = pred.strip()
512
+ gt = gt.strip()
513
+ return fastcer(pred, gt) <= 0.3
514
+
515
+ def patsocr_doc_to_text(doc):
516
+ return doc['question']
517
+
518
+ def patsocr_eval(pred: str, gt: str):
519
+ pred = pred.strip()
520
+ gt = gt.strip()
521
+ return fastcer(pred, gt) <= 0.5
522
+
523
+ def arabicocr_doc_to_text(doc):
524
+ return doc['question']
525
+
526
+ def arabicocr_eval(pred: str, gt: str):
527
+ pred = pred.strip()
528
+ gt = gt.strip()
529
+ return fastcer(pred, gt) <= 0.4
530
+
531
+
532
+ def culturevideovqa_doc_to_text(doc):
533
+ return doc['question']
534
+
535
+ def culturevideovqa_eval(pred, gt):
536
+ return mcq_eval(pred, gt)
537
+
538
+
539
+ def videomme_doc_to_text(doc):
540
+ prompt = """الترجمات النصية لهذا الفيديو مدرجة أدناه:
541
+ {subtitles}
542
+ اختر أفضل إجابة للسؤال التالي متعدد الخيارات بناءً على الفيديو. أجب فقط بالحرف (أ، ب، ج، أو د) للخيار الصحيح.
543
+ {question}
544
+ {options}
545
+ أفضل إجابة هي:"""
546
+ subtitles = doc["subtitles"]
547
+ question = doc["question"]
548
+ options = doc["options"]
549
+ options_str = "\n".join([f"{option}" for i, option in enumerate(options)])
550
+ return prompt.format(subtitles=subtitles, question=question, options=options_str)
551
+
552
+ def videomme_eval(pred, gt):
553
+ return mcq_eval(pred, gt)
554
+
555
+ def geochat_doc_to_text(doc):
556
+ pre_prompt = "أجب على السؤال التالي بكلمة أو جملة.\n"
557
+ return pre_prompt + doc['question']
558
+
559
+ def muribench_doc_to_text(doc):
560
+ return doc['question']
561
+
562
+ def muribench_eval(pred, gt):
563
+ return mcq_eval(pred, gt)
564
+
565
+ def mmlu_doc_to_text(doc):
566
+ question = doc['Question']
567
+ options = []
568
+ for i in range(1, 6):
569
+ if doc[f'Option {i}'] is not None:
570
+ options.append(doc[f'Option {i}'])
571
+ parsed_options = mmmu_parse_options(options)
572
+ context = doc['Context']
573
+ if context:
574
+ return f"{question}\n{CONTEXT_KEYWORD}: {context}\n{parsed_options}\n{MMMU_MULTI_CHOICE_PROMPT}"
575
+ return f"{question}\n{parsed_options}\n{MMMU_MULTI_CHOICE_PROMPT}"
576
+
577
+ def mmlu_eval(pred, gt):
578
+ return mcq_eval(pred, gt)
579
+
580
+ def henna_doc_to_text(doc):
581
+ return doc['question']
582
+
583
+
584
+ mtvqa_user_prompt = """Here are three examples to guide your evaluation:
585
+ Example 1:
586
+ Question: "ما هي اللغة المستخدمة في النص؟"
587
+ Predicted Answer: "العربية"
588
+ Ground Truth: "اللغة العربية"
589
+ Score: 1
590
+
591
+ Example 2:
592
+ Question: "ما هو موضوع النص؟"
593
+ Predicted Answer: "إثنان"
594
+ Ground Truth: "الحب و الكراهية"
595
+ Score: 0
596
+
597
+ Example 3:
598
+ Question: "ما هو عدد صفحات الكتاب؟"
599
+ Predicted Answer: "الصورة لا تحتوي على عدد صفحات الكتاب."
600
+ Ground Truth: "غير معروف"
601
+ Score: 1
602
+
603
+ 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.
604
+ Question: "{question}"
605
+ Predicted Answer: "{pred}"
606
+ Ground Truth: "{gt}"
607
+ Score: """
608
+
609
+ geochat_user_prompt = """Here are three examples to guide your evaluation:
610
+ Example 1:
611
+ Question: "أجب على السؤال التالي بكلمة أو جملة.\n<image>\n كم عدد قوارب الصيد الموجودة في الصورة؟"
612
+ Predicted Answer: "قارب صيد واحد"
613
+ Ground Truth: "يوجد قارب صيد واحد في الصورة."
614
+ Score: 1
615
+
616
+ Example 2:
617
+ Question: "أجب على السؤال التالي بكلمة أو جملة.\n<image>\nكم عدد السيارات الصغيرة الموجودة في الصورة؟"
618
+ Predicted Answer: "عدد السيارات الصغيرة هو 10."
619
+ Ground Truth: "هناك سيارة صغيرة واحدة في الصورة."
620
+ Score: 0
621
+
622
+ Example 3:
623
+ Question: "أجب على السؤال التالي بكلمة أو جملة.\n<image>\nهل يمكنك إخباري بعدد الأشجار الموجودة في أعلى الصورة؟"
624
+ Predicted Answer: "توجد عدة أشجار في أع��ى الصورة، ومن الصعب تحديد العدد الدقيق بسبب تداخلها."
625
+ Ground Truth: "يوجد بعض الأشجار في الأعلى"
626
+ Score: 1
627
+
628
+ 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.
629
+ Question: "{question}"
630
+ Predicted Answer: "{pred}"
631
+ Ground Truth: "{gt}"
632
+ Score: """
633
+
634
+ mathvista_user_prompt = """Here are three examples to guide your evaluation:
635
+ Example 1:
636
+ 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: "
637
+ Predicted Answer: "ب) برغوث الماء الشائع\nالحل: ب"
638
+ Ground Truth: "برغوث الماء الشائع"
639
+ Score: 1
640
+
641
+ Example 2:
642
+ 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: "
643
+ 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°."
644
+ Ground Truth: "ب"
645
+ Score: 0
646
+
647
+ Example 3:
648
+ 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: "
649
+ 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."
650
+ Ground Truth: "تساوي"
651
+ Score: 1
652
+
653
+ 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.
654
+ Question: "{question}"
655
+ Predicted Answer: "{pred}"
656
+ Ground Truth: "{gt}"
657
+ Score: """
658
+
659
+ vqav2_user_prompt = """Here are three examples to guide your evaluation:
660
+ Example 1:
661
+ Question: "ما لون الحوض؟\nأجب على السؤال باستخدام كلمة أو عبارة واحدة."
662
+ Predicted Answer: "أبيض"
663
+ Ground Truth: "أبيض"
664
+ Score: 1
665
+
666
+ Example 2:
667
+ Question: "كم عدد الأشخاص في الصورة؟\nأجب على السؤال باستخدام كلمة أو عبارة واحدة."
668
+ Predicted Answer: "اثنان"
669
+ Ground Truth: "٣"
670
+ Score: 0
671
+
672
+ Example 3:
673
+ Question: "ما هو شكل الطائرة الورقية؟\nأجب على السؤال باستخدام كلمة أو عبارة واحدة."
674
+ Predicted Answer: "سداسي"
675
+ Ground Truth: "سداسي الأضلاع"
676
+ Score: 1
677
+
678
+ 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.
679
+ Question: "{question}"
680
+ Predicted Answer: "{pred}"
681
+ Ground Truth: "{gt}"
682
+ Score: """
683
+
684
+
685
+
686
+ henna_user_prompt = """Here are three examples to guide your evaluation:
687
+ Example 1:
688
+ Question: "هل تظهر علامات الحضارة البشرية مثل البناء أو الطرق في الصورة؟"
689
+ Predicted Answer: "لا"
690
+ Ground Truth: "لا، لا تظهر في الصورة علامات الحضارة البشرية مثل البناء أو الطرق، وتبدو المنطقة طبيعية وغير معبأة بالتطورات البشرية."
691
+ Score: 1
692
+
693
+ Example 2:
694
+ Question: "ما العناصر المعمارية المرئية على ��اب اليمن التي تدل على قدمه؟"
695
+ Predicted Answer: "أ"
696
+ 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."
697
+ Score: 0
698
+
699
+ Example 3:
700
+ Question: "ما هي المظاهر الثقافية أو الاجتماعية التي يمكن استنتاجها من الملابس التقليدية للأشخاص في الصورة؟"
701
+ 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"
702
+ Ground Truth: "الملابس التقليدية التي يرتديها الأشخاص تشير إلى التمسك بالتراث والثقافة اليمنية، والتي تعبر عن هوية المجتمع وتنوعه الثقافي."
703
+ Score: 1
704
+
705
+ 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.
706
+ Question: "{question}"
707
+ Predicted Answer: "{pred}"
708
+ Ground Truth: "{gt}"
709
+ Score: """
710
+
711
+ 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."