update

src/leaderboard/read_evals.py

@@ -88,18 +88,22 @@ class EvalResult:
             # We average all scores of a given metric (mostly for mmlu)
 
             def post_process_results(results: dict) -> dict:
+                # {'nq_open': {'em': 0.018005540166204988, 'em_stderr': 0.0022134216580395583}}
                 res_copy = results.copy()
 
-                for k, v in res_copy.items():
-                    if "exact_match" in k:
-                        results[k.replace("exact_match", "em")] = v
+                for task_name in res_copy.keys():
+                    entry_copy = results[task_name].copy()
 
-                res_copy = results.copy()
+                    for k, v in entry_copy.items():
+                        if "exact_match" in k:
+                            results[task_name][k.replace("exact_match", "em")] = v
+
+                    entry_copy = results[task_name].copy()
 
-                for k, v in res_copy.items():
-                    if "," in k:
-                        tokens = k.split(",")
-                        results[tokens[0]] = v
+                    for k, v in entry_copy.items():
+                        if "," in k:
+                            tokens = k.split(",")
+                            results[task_name][tokens[0]] = v
 
                 return results