Function parse_evaluation_response

eval/src/eval_common.py:228–258

Parse XML-formatted model response to extract analysis and score. Returns: (analysis, score) tuple on success, None on failure.

(text, testpoint)

def parse_evaluation_response(text, testpoint):
    """Parse XML-formatted model response to extract analysis and score.

    Returns:
        (analysis, score) tuple on success, None on failure.
    """
    if text is None:
        return None

    analysis_match = re.search(r"<analysis>(.*?)</analysis>", text, re.DOTALL)
    score_match = re.search(r"<score>(.*?)</score>", text, re.DOTALL)

    if analysis_match is None or score_match is None:
        print("Warning: <analysis> or <score> tags not found in response")
        return None

    try:
        analysis = ast.literal_eval(analysis_match.group(1).strip())
        score = ast.literal_eval(score_match.group(1).strip())
    except (ValueError, SyntaxError) as e:
        print(f"Warning: failed to parse response: {e}")
        return None

    if len(testpoint) != len(analysis) or len(testpoint) != len(score):
        print(
            f"Warning: length mismatch - testpoint({len(testpoint)}) "
            f"vs analysis({len(analysis)}) vs score({len(score)})"
        )
        return None

    return analysis, score


def make_result(index, testpoint, prompt, img_path, output_text, result_json=None):