Fix push details to hub (#98)

---------

Co-authored-by: Clémentine Fourrier <[email protected]>
Co-authored-by: [email protected] <[email protected]>

Author: NathanHB

src/lighteval/logging/evaluation_tracker.py

@@ -462,8 +462,7 @@ def recreate_metadata_card(self, repo_id: str, model_name: str = None) -> None:
         last_results_file_path = hf_hub_url(repo_id=repo_id, filename=last_results_file, repo_type="dataset")
         f = load_dataset("json", data_files=last_results_file_path, split="train")
         results_dict = f["results"][0]
-        value = results_dict.pop("all")
-        new_dictionary = {"all": value}
+        new_dictionary = {"all": results_dict}
         new_dictionary.update(results_dict)
         results_string = json.dumps(new_dictionary, indent=4)
 

src/lighteval/logging/info_loggers.py

@@ -490,12 +490,21 @@ def aggregate(self, task_dict: dict[str, LightevalTask], bootstrap_iters: int =
                         hlog_warn(f"{task_name}, {metric_name} got an OVERFLOW ERROR when computing stderr.")
 
         # We group subtasks which belong to the same parent task, like MMLU, to compute an average on them
+        # and compute an average of all metrics
         grouped_tasks = collections.defaultdict(list)
-        for k in self.metric_aggregated.keys():
+        suite_average = {}
+        suite_nb = {}
+
+        # Build aggregation
+        for k, metrics in self.metric_aggregated.items():
             if "|" in k:
                 suite, task, fewshot = k.split("|")
                 grouped_tasks[f"{suite}|{task.split(':')[0]}:_average|{fewshot}"].append(k)
+            for metric, value in metrics.items():
+                suite_average[metric] = suite_average.get(metric, 0) + value
+                suite_nb[metric] = suite_nb.get(metric, 0) + 1
 
+        # Compute average for sub groups
         for average_task, list_of_subtasks in grouped_tasks.items():
             if len(list_of_subtasks) > 1:
                 metrics = list(self.metric_aggregated[list_of_subtasks[0]].keys())
@@ -504,6 +513,12 @@ def aggregate(self, task_dict: dict[str, LightevalTask], bootstrap_iters: int =
                     for metric in metrics
                 }
 
+        # Compute average for all
+        for metric, value in suite_average.items():
+            suite_average[metric] = value / suite_nb[metric]
+
+        self.metric_aggregated["all"] = suite_average
+
 
 class VersionsLogger:
     """Logger of the tasks versions.