Fix winograd task comparison

src/custom_tasks/winograd_task.py

@@ -14,7 +14,7 @@ def winograd_eval_fn(golds: list[str], predictions: list[str], formatted_doc: Do
     if len(predictions)  > 1:
         raise ValueError("Predictions should have one item")
     # do some santizations, since some models produce more info
-    pred = re.sub('<[^>]+>', '', predictions[0]) # remove xml tags
+    pred = re.sub('<[^>]+>', '', predictions[0]).strip() # remove xml tags
     return 1 if pred == golds[0] else 0
     
 winograd_acc_metric = CorpusLevelMetric(