(output, mode='qa', extract_answer=False)
| 16 | |
| 17 | |
| 18 | def extract_answer_fn(output, mode='qa', extract_answer=False): |
| 19 | if extract_answer == False and mode not in ['infogen', 'summary', 'research']: |
| 20 | if mode == 'qa': |
| 21 | return output.strip() |
| 22 | pred_answer_lines = output.replace("\n\n", "\n").strip().split('\n') |
| 23 | pred_answer = '\n'.join(pred_answer_lines[-3:]) |
| 24 | return pred_answer |
| 25 | extracted_text = '' |
| 26 | if mode == 'codegen': |
| 27 | pattern = r'```python\s*(.*?)\s*```' # Extract the code between ```python and ``` |
| 28 | matches = re.findall(pattern, output, re.DOTALL | re.IGNORECASE) |
| 29 | if matches: |
| 30 | extracted_text = matches[-1].strip() # Take the last match |
| 31 | elif mode in ['infogen', 'summary', 'research']: |
| 32 | pattern_info = "**Final Information" |
| 33 | if "</think>\n" in output: |
| 34 | extracted_text = output.split("</think>\n")[-1].split("<|begin_click_link|>")[0].replace(pattern_info, "").strip(':**').strip('\n').strip("```").strip() # 提取</think>后面的内容 |
| 35 | if mode == 'infogen': |
| 36 | extracted_text = '\n'.join(extracted_text.replace("\n\n", "\n").split('\n')[:5]) # 只保留前5行 |
| 37 | elif pattern_info in output: |
| 38 | extracted_text = output.split(pattern_info)[-1].split("<|begin_click_link|>")[0].strip('\n').strip(':**').strip("```").strip() # 提取**Final Information**后面的内容 |
| 39 | if mode == 'infogen': |
| 40 | extracted_text = '\n'.join(extracted_text.replace("\n\n", "\n").split('\n')[:5]) # 只保留前5行 |
| 41 | else: |
| 42 | # extracted_text = "No helpful information found." |
| 43 | extracted_text = '\n'.join(output.strip().replace("</think>\n", "").replace("\n\n", "\n").split('\n')[-5:]) # 若没提取到，只保留最后5行 |
| 44 | if mode == 'research': |
| 45 | extracted_text = extracted_text[:6000] |
| 46 | else: |
| 47 | extracted_text = extracted_text[:2500] |
| 48 | elif mode in ['math', 'choose', 'qa']: |
| 49 | pattern = r'\\boxed\{(.*)\}' |
| 50 | matches = re.findall(pattern, output) |
| 51 | if matches: |
| 52 | extracted_text = matches[-1] # Take the last match |
| 53 | else: |
| 54 | pattern = 'ANSWER:' |
| 55 | if pattern in output: |
| 56 | extracted_text = output.split(pattern)[-1].strip('**').strip() |
| 57 | if mode in ['choose']: |
| 58 | inner_pattern = r'\\text\{(.*)\}' |
| 59 | inner_matches = re.findall(inner_pattern, extracted_text) |
| 60 | if inner_matches: |
| 61 | extracted_text = inner_matches[-1] # Take the last match |
| 62 | extracted_text = extracted_text.strip("()") |
| 63 | return extracted_text |
| 64 | |
| 65 | |
| 66 | async def llm_evaluate_equivalence_single( |
no outgoing calls
no test coverage detected