Upload 143 files

hugging/td_fuse/grpo_reason.py

@@ -252,15 +252,27 @@ def verify_answer(problem: Dict, response: str) -> float:
         says_one_heavier = any(w in answer_text for w in ["is heavier", "weighs more", "is more"])
         return 1.0 if (has_same and not says_one_heavier) else 0.0
 
-    # For yes/no answers: check that the answer agrees, not just mentions the word
+    # For yes/no answers: check overall sentiment of the answer
     if expected in ("yes", "no"):
-        # Look at the first sentence of the answer for the actual conclusion
-        first_sentence = answer_text.split(".")[0] if "." in answer_text else answer_text[:100]
         if expected == "yes":
-            return 1.0 if ("yes" in first_sentence or "do " in first_sentence or "does " in first_sentence or "can " in first_sentence) and "no" not in first_sentence[:20] else 0.0
+            # Positive indicators anywhere in answer
+            yes_words = ["yes", "do breathe", "does breathe", "does need", "do need",
+                         "correct", "true", "indeed", "absolutely", "certainly",
+                         "they do", "it does", "they can", "it can"]
+            # Negative indicators that would flip it
+            no_words = ["no,", "no.", "not ", "cannot", "can't", "doesn't", "don't",
+                        "is not", "are not", "do not"]
+            has_yes = any(w in answer_text for w in yes_words)
+            has_no = any(w in answer_text[:50] for w in no_words)  # Only check start for negation
+            return 1.0 if has_yes and not has_no else 0.0
         else:
-            has_no = any(w in first_sentence for w in ["no", "not", "cannot", "can't", "doesn't", "neither"])
-            return 1.0 if has_no else 0.0
+            # For "no" answers
+            no_words = ["no", "not", "cannot", "can't", "doesn't", "don't",
+                        "neither", "isn't", "aren't", "we can't", "not necessarily",
+                        "not definitely", "doesn't mean"]
+            yes_override = answer_text.strip().startswith("yes")
+            has_no = any(w in answer_text for w in no_words)
+            return 1.0 if has_no and not yes_override else 0.0
 
     # For "once" (subtract 5 from 25 trick):
     if expected == "once":