|
|
|
|
|
import re
|
|
|
|
|
|
def clean_worker_model_answer(raw_answer_text: str, prompt_version: str) -> tuple[str, bool]:
|
|
|
"""
|
|
|
Cleans the raw answer text from the worker LLM.
|
|
|
Returns:
|
|
|
- cleaned_answer (str)
|
|
|
- is_correctly_formatted_output (bool): True if CoT format was followed (if CoT was used), True for other prompt types by default.
|
|
|
"""
|
|
|
answer = raw_answer_text.strip()
|
|
|
is_correctly_formatted_output = True
|
|
|
|
|
|
if prompt_version == "COT":
|
|
|
match = re.search(r"Final Answer:\s*(.*)", answer, re.IGNORECASE | re.DOTALL)
|
|
|
if match:
|
|
|
answer = match.group(1).strip()
|
|
|
else:
|
|
|
is_correctly_formatted_output = False
|
|
|
|
|
|
print(f"\nINFO (Worker Output): COT prompt used, but 'Final Answer:' marker not found. Cleaning applied to full raw output. Raw Preview: \"{raw_answer_text[:100]}...\"")
|
|
|
|
|
|
prefixes_to_remove = [
|
|
|
"Answer is:", "Answer:", "The answer is:", "The final answer is ", "Expert Answer:",
|
|
|
"答案是：", "答案：", "答案是", "答案", "好的，答案是：", "好的，答案是", "了解，答案是：", "了解，答案是"
|
|
|
]
|
|
|
|
|
|
temp_answer = answer
|
|
|
for prefix in prefixes_to_remove:
|
|
|
if temp_answer.lower().startswith(prefix.lower()):
|
|
|
temp_answer = temp_answer[len(prefix):].strip()
|
|
|
break
|
|
|
answer = temp_answer
|
|
|
|
|
|
if answer.startswith("- "): answer = answer[2:].strip()
|
|
|
if answer.startswith("* "): answer = answer[2:].strip()
|
|
|
|
|
|
if len(answer) > 1 and ((answer.startswith('"') and answer.endswith('"')) or \
|
|
|
(answer.startswith("'") and answer.endswith("'"))):
|
|
|
answer = answer[1:-1]
|
|
|
|
|
|
if len(answer) > 1 and (answer.endswith(".") or answer.endswith("。")):
|
|
|
answer = answer[:-1].strip()
|
|
|
|
|
|
if answer.startswith("> "): answer = answer[2:].strip()
|
|
|
if answer.startswith(">"): answer = answer[1:].strip()
|
|
|
|
|
|
return answer.strip(), is_correctly_formatted_output |
