update

Author: Yunnglin

evalscope/app/utils/visualization.py

@@ -38,14 +38,18 @@ def plot_single_report_sunburst(report_list: List[Report]):
         path = [ReportKey.dataset_name] + categories + [ReportKey.subset_name]
     logger.debug(f'df: \n{df}')
     df[categories] = df[categories].fillna('default')  # NOTE: fillna for empty categories
+    df = df[df[ReportKey.num] > 0]  # NOTE: filter out zero-num rows to avoid ZeroDivisionError in plotly
+    if df.empty:
+        return None
 
     plot = px.sunburst(
         df,
         path=path,
         values=ReportKey.num,
         color=ReportKey.score,
         color_continuous_scale='RdYlGn',  # see https://plotly.com/python/builtin-colorscales/
-        color_continuous_midpoint=np.average(df[ReportKey.score], weights=df[ReportKey.num]),
+        color_continuous_midpoint=np.average(df[ReportKey.score], weights=df[ReportKey.num])
+        if df[ReportKey.num].sum() > 0 else df[ReportKey.score].mean(),
         template=PLOTLY_THEME,
         maxdepth=4
     )

tests/benchmark/test_eval.py

@@ -387,7 +387,7 @@ def test_bfcl_v3(self):
                 'underscore_to_dot': True
             }
         }
-        self._run_dataset_test('bfcl_v3', dataset_args=dataset_args, model='qwen-plus', limit=10)
+        self._run_dataset_test('bfcl_v3', dataset_args=dataset_args, model='qwen-plus', limit=1)
 
     def test_bfcl_v4(self):
         """Test BFCL v4 dataset."""

tests/cli/test_all.py

@@ -51,7 +51,7 @@
     'frames',
     'docmath',
     'needle_haystack',
-    'bfcl_v3',
+    # 'bfcl_v3',
     'hle',
     'tau_bench',
 ]