diff --git a/ais_bench/benchmark/configs/summarizers/groups/mmlu_pro.py b/ais_bench/benchmark/configs/summarizers/groups/mmlu_pro.py
index ba2d9fc7..b7832e89 100644
--- a/ais_bench/benchmark/configs/summarizers/groups/mmlu_pro.py
+++ b/ais_bench/benchmark/configs/summarizers/groups/mmlu_pro.py
@@ -3,3 +3,22 @@
 mmlu_pro_summary_groups = [
     {'name': 'mmlu_pro', 'subsets': ['mmlu_pro_' + c.replace(' ', '_') for c in categories]},
 ]
+
+_mmlu_pro_all = ['mmlu_pro_' + c.replace(' ', '_') for c in categories]
+_mmlu_pro_weights = {
+    'mmlu_pro_math': 1351,
+    'mmlu_pro_physics': 1299,
+    'mmlu_pro_chemistry': 1132,
+    'mmlu_pro_law': 1101,
+    'mmlu_pro_engineering': 969,
+    'mmlu_pro_other': 924,
+    'mmlu_pro_economics': 844,
+    'mmlu_pro_health': 818,
+    'mmlu_pro_psychology': 798,
+    'mmlu_pro_business': 789,
+    'mmlu_pro_biology': 717,
+    'mmlu_pro_philosophy': 499,
+    'mmlu_pro_computer_science': 410,
+    'mmlu_pro_history': 381,
+}
+mmlu_pro_summary_groups.append({'name': 'mmlu_pro-weighted', 'subsets': _mmlu_pro_all, 'weights': _mmlu_pro_weights})
diff --git a/ais_bench/benchmark/openicl/icl_inferencer/icl_bfcl_v3_inferencer.py b/ais_bench/benchmark/openicl/icl_inferencer/icl_bfcl_v3_inferencer.py
index 19f822d6..b7dee555 100644
--- a/ais_bench/benchmark/openicl/icl_inferencer/icl_bfcl_v3_inferencer.py
+++ b/ais_bench/benchmark/openicl/icl_inferencer/icl_bfcl_v3_inferencer.py
@@ -7,6 +7,8 @@
 from aiohttp import ClientSession
 from multiprocessing import BoundedSemaphore
 
+from bfcl_eval.utils import make_json_serializable
+
 from ais_bench.benchmark.openicl.icl_retriever import BaseRetriever
 from ais_bench.benchmark.registry import MODELS
 from ais_bench.benchmark.utils.prompt import PromptList
@@ -600,6 +602,7 @@ async def _inference_multi_turn(self, data: dict, finial_output: FunctionCallOut
         await self.status_counter.case_finish()
         if all_model_response:
             finial_output.tool_calls = all_model_response
+        finial_output.inference_log = make_json_serializable(finial_output.inference_log)
         await self.output_handler.report_cache_info(
             index, prompt_list, finial_output, data_abbr
         )