udpate

src/leaderboard/read_evals.py

@@ -114,7 +114,7 @@ class EvalResult:
     def to_dict(self, task_class):
         """Converts the Eval Result to a dict compatible with our dataframe display"""
         # ignore uncertainty for overall calculation
-        scores = [v for k, v in self.results.values() if v is not None and k != 'uncertainty']
+        scores = [v for k, v in zip(self.results.keys(), self.results.values()) if v is not None and k != 'uncertainty']
         average = sum(scores) / len(scores)
         # average = sum([v for v in self.results.values() if v is not None]) / len(task_class)
         data_dict = {