Upload consensus_fusion.py with huggingface_hub

consensus_fusion.py

@@ -1,14 +1,26 @@
 from typing import List, Dict, Any
 import re
 
-def _normalize_answer(ans: str) -> str:
-    """Normalize an answer string for comparison (remove spaces, lowercase, strip LaTeX wrappers)."""
+try:
+    from math_verify import parse, verify
+    MATH_VERIFY_AVAILABLE = True
+except ImportError:
+    MATH_VERIFY_AVAILABLE = False
+
+def _normalize_answer(ans: str) -> Any:
+    """Uses math_verify to parse the answer for robust comparison."""
+    if MATH_VERIFY_AVAILABLE:
+        try:
+            return parse(str(ans))
+        except:
+            return str(ans)
+    
+    # Legacy fallback
     s = str(ans).strip()
     s = re.sub(r'\$', '', s)
     s = re.sub(r'\\(?:approx|approx|cdot|,|;|\s)', ' ', s)
     s = s.replace("\\", "").replace("{", "").replace("}", "")
     s = s.replace(" ", "").lower()
-    # Normalize floats: "3.0" == "3"
     try:
         f = float(s)
         s = str(int(f)) if f == int(f) else str(round(f, 6))
@@ -23,10 +35,18 @@ def normalize_answers(answers: List[str]) -> Dict[str, List[int]]:
         clean = _normalize_answer(ans)
         matched = False
         for key in list(normalized_groups.keys()):
-            if _normalize_answer(key) == clean:
-                normalized_groups[key].append(idx)
-                matched = True
-                break
+            key_clean = _normalize_answer(key)
+            
+            if MATH_VERIFY_AVAILABLE:
+                if verify(clean, key_clean):
+                    normalized_groups[key].append(idx)
+                    matched = True
+                    break
+            else:
+                if key_clean == clean:
+                    normalized_groups[key].append(idx)
+                    matched = True
+                    break
         if not matched:
             normalized_groups[ans] = [idx]
     return normalized_groups