Correct comp id computation

utils.py

@@ -337,6 +337,9 @@ def get_benchmarks() -> Tuple[pd.DataFrame, Dict[str, Dict[str, float]]]:
     for run in runs:
         uid = run.config.get("uid", None)
         model = run.config.get("model", None)
+        # Any run without a competition_id was for competition 2.
+        comp_name = run.config.get("competition_id", "B7_MULTI_CHOICE")
+        comp_id = COMP_NAME_TO_ID.get(comp_name, 2)
         if not uid or not model:
             continue
         samples = list(
@@ -360,9 +363,7 @@ def get_benchmarks() -> Tuple[pd.DataFrame, Dict[str, Dict[str, float]]]:
         if not has_all_keys:
             continue
         
-        # Any run without a competition ID was for competition 2.
-        comp_name = sample.get("competition_id", "B7_MULTI_CHOICE")
-        comp_ids.append(COMP_NAME_TO_ID.get(comp_name, 2))
+        comp_ids.append(comp_id)
         timestamps.append(datetime.datetime.fromtimestamp(sample["_timestamp"]))
         mmlu.append(sample["mmlu.acc,none"])
         mmlu_pro.append(sample["mmlu_pro"])