Refactor mmbench_aggregate_dev_results_eval function to generate and …

…save detailed evaluation results

lmms_eval/tasks/mmbench/cn_utils.py

@@ -99,8 +99,16 @@ def mmbench_process_results(doc, results):
 
 def mmbench_aggregate_dev_results_eval(results, args):
     print(f"============= MMBench-CN(Dev) Detailed Results =============")
-    accuracy = mmbench_evaluator.eval_result(results, eval_method="openai")
-    return accuracy * 100
+    overall_acc, category_acc, l2_category_acc = mmbench_evaluator.eval_result(results, eval_method="openai")
+    file = generate_submission_file("mmbench_cn_dev_results.json", args)
+    details_info = {
+        "overall_acc": overall_acc,
+        "category_acc": category_acc,
+        "l2_category_acc": l2_category_acc,
+    }
+    with open(file, "w") as f:
+        json.dump(details_info, f)
+    return overall_acc * 100
 
 
 def mmbench_aggregate_dev_results(results, args):