osworld baselines

src/streamlit_app.py

@@ -206,6 +206,14 @@ def fetch_leaderboard_data():
         # Create DataFrame
         df = pd.DataFrame(results)
         
+        # Adjust evaluated results for osworld-g (do not touch baselines)
+        if not df.empty and 'dataset' in df.columns and 'overall_accuracy' in df.columns:
+            osworld_mask = df['dataset'] == 'osworld-g'
+            if osworld_mask.any():
+                df.loc[osworld_mask, 'overall_accuracy'] = (
+                    df.loc[osworld_mask, 'overall_accuracy'] * 0.90425531914
+                )
+        
         # Process checkpoints: for each base model, find the best checkpoint
         if not df.empty:
             # Group by dataset and base_model