Upload folder using huggingface_hub

baseline/run_baseline.py

@@ -20,7 +20,7 @@ from code_debug_env.models import Action
 try:
     from openai import AsyncOpenAI
 except ImportError:
-    print("Please install openai: pip install openai")
+    print("Please install openai: pip install openai", file=sys.stderr)
     sys.exit(1)
 
 BASE_URL = os.getenv("OPENENV_URL", "http://127.0.0.1:8000")
@@ -102,10 +102,10 @@ async def main(output_format: str = "table"):
     if output_format == "json":
         print(json.dumps({"baseline_results": results, "agent": "openai_api"}))
     else:
-        print("\n=== Baseline Results ===")
+        print("\n=== Baseline Results ===", file=sys.stderr)
         for r in results:
-            print(f"  {r['task_id']:15s}  score={r['best_score']:.3f}  steps={r['steps']}")
-        print(f"\n  avg score: {sum(r['best_score'] for r in results) / len(results):.3f}")
+            print(f"  {r['task_id']:15s}  score={r['best_score']:.3f}  steps={r['steps']}", file=sys.stderr)
+        print(f"\n  avg score: {sum(r['best_score'] for r in results) / len(results):.3f}", file=sys.stderr)
 
 if __name__ == "__main__":
     output = "json" if "json" in sys.argv else "table"

server/app.py

@@ -124,13 +124,21 @@ def run_baseline() -> dict:
     Run the baseline agent on all tasks and return scores.
     This endpoint triggers the baseline inference script.
     """
-    import subprocess, sys, json
+    import subprocess, sys, json, re
     try:
         result = subprocess.run(
             [sys.executable, "baseline/run_baseline.py", "--output", "json"],
             capture_output=True, text=True, timeout=120,
         )
-        return json.loads(result.stdout)
+        
+        # Robustly find JSON in potentially noisy stdout
+        stdout = result.stdout.strip()
+        match = re.search(r'(\{.*\})', stdout, re.DOTALL)
+        if match:
+            return json.loads(match.group(1))
+        
+        raise ValueError(f"No JSON found in baseline output. Stdout: {stdout[:200]}")
+        
     except Exception as e:
         raise HTTPException(status_code=500, detail=str(e))