From 6b720462c8b22a02411427493fdf7c222537f6f4 Mon Sep 17 00:00:00 2001 From: Yizheng Huang Date: Tue, 15 Sep 2026 17:47:10 -0700 Subject: [PATCH] evaluator: preserve signed GSM8K ground-truth answers --- proeval/evaluator/predictor.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/proeval/evaluator/predictor.py b/proeval/evaluator/predictor.py index 87302f9..30d1b7d 100644 --- a/proeval/evaluator/predictor.py +++ b/proeval/evaluator/predictor.py @@ -197,10 +197,10 @@ def _cmp(pred, gt): return 0.0 if str(pred).strip() == str(gt).strip() else 1.0 def _gt(ans): - m = re.search(r"####\s*([0-9,]+(?:\.[0-9]+)?)", str(ans)) + m = re.search(r"####\s*([+-]?[0-9,]+(?:\.[0-9]+)?)", str(ans)) if m: return m.group(1).replace(",", "") - m = re.search(r"([0-9,]+(?:\.[0-9]+)?)", str(ans)) + m = re.search(r"([+-]?[0-9,]+(?:\.[0-9]+)?)", str(ans)) return m.group(1).replace(",", "") if m else str(ans).strip() return DatasetConfig(