Function _normalize_ground_truth

src/benchmarks/aime/evaluate.py:64–73

Normalize AIME ground truth to a plain integer string. AIME answers are always integers 0-999, but some datasets include formatting like '336^\\circ'. Extract the integer.

(ground_truth: str)

def _normalize_ground_truth(ground_truth: str) -> str:
    """Normalize AIME ground truth to a plain integer string.

    AIME answers are always integers 0-999, but some datasets include
    formatting like '336^\\circ'. Extract the integer.
    """
    import re

    m = re.search(r"\d+", ground_truth)
    return m.group(0) if m else ground_truth


def verify_answer(model_response: str, ground_truth: str) -> bool: