LTPO/data.py at main · Scout-UCAS/LTPO · GitHub

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
"""
Data API
"""
import json

from datasets import Dataset, load_dataset, load_from_disk, concatenate_datasets
from prompts import gsm8k_prompt, asdiv_aug_prompt, math_500_prompt, aime_prompt, strategy_qa_prompt, du_prompt

def get_dataset(data_name_or_path, tokenizer, prompt_idx):
    """
    Args:
        data_name_or_path: dataset name or path
        tokenizer: tokenizer
        prompt_idx: which query prompt to use
    Returns:
        dataset: dataset
    """

    ### Load dataset ###
    if "gsm8k" in data_name_or_path.lower():
        try:
            dataset = load_from_disk(data_name_or_path)['test']
        except:
            dataset = load_dataset("openai/gsm8k", "socratic")["test"]
        question_col = "question"
        answer_col = "answer"

    elif "asdiv-aug" in data_name_or_path.lower():
        try:
            dataset = load_from_disk(data_name_or_path)['test']
        except:
            dataset = load_dataset("xuyige/ASDiv-Aug")["test"]
        question_col = "question"
        answer_col = "answer"

    elif "math-500" in data_name_or_path.lower():
        try:
            dataset = load_from_disk(data_name_or_path)['test']
        except:
            dataset = load_dataset("HuggingFaceH4/MATH-500")["test"]
        question_col = "problem"
        answer_col = "answer"

    elif "aime_2024" in data_name_or_path.lower():
        try:
            dataset = load_from_disk(data_name_or_path)
        except:
            dataset = load_dataset("Maxwell-Jia/AIME_2024")['train']
        question_col = "Problem"
        answer_col = "Answer"

    elif "aime2025" in data_name_or_path.lower():
        try:
            dataset = load_from_disk(data_name_or_path)
        except:
            dataset = concatenate_datasets([
                load_dataset("opencompass/AIME2025", "AIME2025-I")['test'],
                load_dataset("opencompass/AIME2025", "AIME2025-II")['test'],
            ])
        question_col = "question"
        answer_col = "answer"

    elif "strategyqa" in data_name_or_path.lower():
        return get_strategyqa(tokenizer, prompt_idx)

    elif "date_understanding" in data_name_or_path.lower():
        dataset = load_dataset("maveriq/bigbenchhard", "date_understanding")['train']
        question_col = "input"
        answer_col = "target"

    else:
        raise ValueError(f"Unsupported dataset: {data_name_or_path}")

    # preprocess dataset
    def preprocess_function(examples):
        '''
        Preprocess dataset

        Args:
            examples: dataset examples

        Returns:
            formatted: formatted dataset
        '''
        prompt = []
        formatted = []
        answers = examples[answer_col]
        questions = examples[question_col]
        for q in questions:
            if "gsm8k" in data_name_or_path.lower():
                messages = gsm8k_prompt(q, prompt_idx)
            elif "asdiv-aug" in data_name_or_path.lower():
                messages = asdiv_aug_prompt(q, prompt_idx)
            elif "math-500" in data_name_or_path.lower():
                messages = math_500_prompt(q, prompt_idx)
            elif "aime" in data_name_or_path.lower():
                messages = aime_prompt(q, prompt_idx)
            elif "date_understanding" in data_name_or_path.lower():
                messages = du_prompt(q, prompt_idx)
            else:
                raise ValueError(f"Unsupported dataset: {data_name_or_path}")

            prompt.append(messages)
            formatted.append(tokenizer.apply_chat_template(
                messages, tokenize=False, add_generation_prompt=True
            ))
        if "aime" in data_name_or_path.lower() and "2025" in data_name_or_path.lower():
            answers = [ans.replace('^\circ', '') for ans in answers]
        if "date_understanding" in data_name_or_path.lower():
            answers = [ans[1] for ans in answers]
        return {
            "prompt": prompt,
            "formatted": formatted,
            "question": questions,
            "answer": answers,
        }

    dataset = dataset.map(preprocess_function, batched=True, load_from_cache_file=False)
    return dataset

def get_strategyqa(tokenizer, prompt_idx):
    prompt = []
    formatted = []
    answers = []
    questions = []
    with open('strategyqa_train.json', 'r') as f:
        data = json.load(f)
    for ins in data:
        q, a = ins['question'], ins['answer']
        msg = strategy_qa_prompt(q, prompt_idx)
        questions.append(q)
        answers.append(a)
        prompt.append(msg)
        formatted.append(tokenizer.apply_chat_template(
            msg, tokenize=False, add_generation_prompt=True
        ))
    return Dataset.from_dict({
        "prompt": prompt,
        "formatted": formatted,
        "question": questions,
        "answer": answers,
    })