File size: 4,966 Bytes
d4a7e84 | 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 | import os
import json
from math_verify import parse, verify
from typing import Dict, Any
from collections import Counter
import argparse
def simple_verify(gold_answer: str, output_text: str) -> Dict[str, Any]:
"""
使用math_verify进行简单验证
与原有infer.py保持一致
"""
try:
# gold_answer = "$" + gold_answer + "$"
gold = parse(gold_answer)
answer = parse(output_text)
verify_result = verify(gold, answer)
extracted_answer = str(answer) if answer is not None else None
except Exception as e:
verify_result = False
extracted_answer = None
print(f"解析失败: {e}")
return {
'extracted_predicted': extracted_answer,
'is_correct': verify_result
}
def majority_verify(input_path, output_path,all_samples=False,clip_sample_num=4):
total_len = 0
majority_correct_len = 0
correct_count_dict = {}
correct_indices = []
with open(input_path, 'r', encoding='utf-8') as f:
for idx, line in enumerate(f):
data = json.loads(line)
solution = data['solution']
if all_samples:
outputs_list = data['outputs']
else:
if len(data['outputs']) < clip_sample_num:
outputs_list = data['outputs']
else:
outputs_list = data['outputs'][:clip_sample_num]
answer_counts = {}
answer_verify_counts = {}
verify_infos = []
for output in outputs_list:
verify_info = simple_verify(solution, output)
extracted_answer = verify_info['extracted_predicted']
normalized_answer = str(extracted_answer).strip() if extracted_answer is not None else None
verify_infos.append((normalized_answer, verify_info['is_correct']))
if normalized_answer is None:
continue
answer_counts[normalized_answer] = answer_counts.get(normalized_answer, 0) + 1
if normalized_answer not in answer_verify_counts:
answer_verify_counts[normalized_answer] = {'correct': 0, 'total': 0}
answer_verify_counts[normalized_answer]['total'] += 1
if verify_info['is_correct']:
answer_verify_counts[normalized_answer]['correct'] += 1
if not answer_counts:
correct_count_dict[idx] = 0
total_len += 1
continue
# 打印每个问题的答案出现次数
print(f"问题{idx} 答案出现次数：")
for ans, cnt in answer_counts.items():
print(f"（{ans}）：{cnt}次")
# 找到出现次数最多的答案
final_answer = max(answer_counts.items(), key=lambda x: x[1])[0]
verify_info_majority = answer_verify_counts.get(final_answer, {'correct': 0, 'total': 0})
verify_result = verify_info_majority['correct'] > 0
if verify_result:
majority_correct_len += 1
correct_indices.append(idx)
# 统计所有 sample 里正确的个数
correct_count = sum(1 for _, is_correct in verify_infos if is_correct)
correct_count_dict[idx] = correct_count
total_len += 1
acc = majority_correct_len / total_len if total_len > 0 else 0.0
print(f"total_len: {total_len}, majority_correct_len: {majority_correct_len}")
print(f"majority accuracy: {acc}")
result = {
'acc': acc,
'correct_count_dict': correct_count_dict,
'correct_indices': correct_indices
}
with open(output_path, 'w', encoding='utf-8') as fout:
json.dump(result, fout, ensure_ascii=False, indent=2)
dist = Counter(correct_count_dict.values())
print("每个问题 sample 正确数量的分布:")
for k in sorted(dist.keys(), reverse=True):
print(f"{k} 个 sample 全对的问题数: {dist[k]}")
# input_path = "/home/tianqiu/tts_schedule/batch_infer/results/mathhard/Qwen_Qwen2.5-7B-Instruct/Sequence_new_prompt/batch_data/batch_1/parallel_merged_output.jsonl"
# output_path = "/home/tianqiu/tts_schedule/batch_infer/results/mathhard/Qwen_Qwen2.5-7B-Instruct/Sequence_new_prompt/batch_data/batch_1/acc.jsonl"
# majority_verify(input_path, output_path)
if __name__ == "__main__":
parser = argparse.ArgumentParser()
parser.add_argument("--input_path", type=str, required=True)
parser.add_argument("--output_path", type=str, required=True)
parser.add_argument("--all_samples", action="store_true")
parser.add_argument("--clip_sample_num", type=int, required=True)
args = parser.parse_args()
input_path = args.input_path
output_path = args.output_path
all_samples = args.all_samples
clip_sample_num = args.clip_sample_num
majority_verify(input_path, output_path, False, clip_sample_num) |