src/output_extract.py

import json
import os
from datasets import load_dataset
from math_verify import parse
import tqdm
import argparse
def extra_revise(response:str):
    # 提取response中的Critique和Revised Answer
    if "Revised Answer" in response:
        revised_answer = response.split("Revised Answer")[1].strip()
    else:
        revised_answer = response
    return revised_answer
    
def content_parse(data):
    return data["response"]["body"]["choices"][0]["message"]["content"]
def extract_from_dir(dir_path, dataset="mathhard", extra_re=False):
    if dataset == "mathhard":
        ds = load_dataset("lighteval/MATH-Hard", "default")["test"]
    elif dataset == "mathhard_train":
        ds = load_dataset("lighteval/MATH-Hard", "default")["train"]
    elif dataset == "aime24":
        ds = load_dataset("math-ai/aime24")["test"]
    else:
        raise ValueError(f"Dataset {dataset} is not supported yet.")
    idx_map = {}    
    for file in os.listdir(dir_path):
        if file.startswith("output_"):
            with open(os.path.join(dir_path, file), "r") as f:
                for idx,line in enumerate(f):
                    data = json.loads(line)
                    problem = ds[idx]["problem"]
                    solution = ds[idx]["solution"]
                    solution_ = "$" + solution + "$"
                    answer = str(parse(solution_))
                    if idx not in idx_map:
                        idx_map[idx] = {
                            "problem": problem,
                            "solution": solution,
                            "answer": answer,
                            "outputs": []
                        }
                    if extra_re:
                        idx_map[idx]["outputs"].append(extra_revise(content_parse(data)))
                    else:
                        idx_map[idx]["outputs"].append(content_parse(data))
    # 写入合并后的 jsonl 文件
    
    output_file = os.path.join(dir_path, "parallel_merged_output.jsonl")
    with open(output_file, "w") as fout:
        for item in idx_map.values():
            fout.write(json.dumps(item, ensure_ascii=False) + "\n")


if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument("--input_dir", type=str, required=True)
    parser.add_argument("--extra_re",action="store_true")
    parser.add_argument("--dataset", type=str, required=True)
    args = parser.parse_args()
    extract_from_dir(args.input_dir, args.dataset, args.extra_re)