Simple parser to extract the \boxed{answer} parts from LLM completions

extract_boxed.py

from typing import Optional

def extract_boxed_solution(text: str) -> Optional[str]:
    """
    Extracts the content of the last `\boxed{}` in a given LaTeX-style text.

    Args:
        text (str): The input string containing LaTeX-style content.

    Returns:
        Optional[str]: The extracted content inside the last `\boxed{}` if found 
        and properly matched, otherwise `None`.

    Example:
        >>> extract_boxed_solution("The result is \\boxed{42}.")
        '42'
        >>> extract_boxed_solution("Unmatched \\boxed{42")
        None
    """
    try:
        start_index = text.rindex("\\boxed{")
        content_start = start_index + 7
        bracket_count = 1
        current_pos = content_start

        while bracket_count > 0 and current_pos < len(text):
            if text[current_pos] == "{":
                bracket_count += 1
            elif text[current_pos] == "}":
                bracket_count -= 1
            current_pos += 1

        if bracket_count == 0:
            content = text[content_start : current_pos - 1].strip()
            return content
        else:
            print("Error: Unmatched brackets in the text")
            return None

    except ValueError:
        print("No boxed solution found in the text")
        return None
    except Exception as e:
        print(f"Error processing text: {str(e)}")
        return None