Files
MatBench/layer1/ALL/QASC-process.py
2025-05-28 10:55:34 +08:00

45 lines
1.5 KiB
Python

import json
# 读取 JSON 文件
def transform_choices(old_choices):
# 提取所有 "text" 值
text_list = [choice["text"] for choice in old_choices]
# 提取所有 "label" 值
label_list = [choice["label"] for choice in old_choices]
return {
"text": text_list,
"label": label_list
}
def transform_json(file_path):
with open(file_path, 'r', encoding='utf-8') as f:
data = json.load(f)
new_json=[]
for item in data:
question = item["question"]["stem"]
choices = item["question"]["choices"]
answer_index = item["answerKey"]
new_choices =transform_choices(choices)
# 构造新的 JSON 数据
transformed_data = {
"question": question,
"choices":new_choices,
"answer": f"[ANSWER]{answer_index}[/ANSWER]",
"prompt":"You MUST include the letter(s) of the correct answer (separated by comma if there are many) within the following tags: [ANSWER] and [/ANSWER]. No explanations and other information. Only return the '[ANSWER]<answer>[/ANSWER]'. We require this because we use automatic parsing."
}
new_json.append(transformed_data)
return new_json
input_path = '/home/ubuntu/50T/fsy/benchmark-dataset-third/ALL/QASC-dev-mat.json'
output_path = '/home/ubuntu/50T/fsy/benchmark-dataset-third/ALL/10-QASC.json'
transformed_data = transform_json(input_path)
with open(output_path, 'w', encoding='utf-8') as f:
json.dump(transformed_data, f, ensure_ascii= False, indent=2)