Add configurable conversation timeout to all benchmarks

benchmarks/gaia/run_infer.py

@@ -303,7 +303,8 @@ def evaluate_instance(
             conversation.send_message(msg)
         else:
             conversation.send_message(instruction)
-        conversation.run()
+        run_timeout = int(os.getenv("CONVERSATION_TIMEOUT", "3600"))
+        conversation.run(timeout=run_timeout)
 
         # Extract answer from conversation history
         model_answer_raw = self._extract_answer_from_history(conversation.state.events)

benchmarks/multiswebench/run_infer.py

@@ -340,7 +340,8 @@ def _log_event(ev):  # keep it simple
             workspace_path=workspace.working_dir,
         )
         conversation.send_message(instruction)
-        conversation.run()
+        run_timeout = int(os.getenv("CONVERSATION_TIMEOUT", "3600"))
+        conversation.run(timeout=run_timeout)
 
         # git add
         workspace.execute_command(f"cd {repo_path} ; git add -A")

benchmarks/openagentsafety/run_infer.py

@@ -434,10 +434,11 @@ def event_callback(event) -> None:
         conversation.send_message(instruction)
 
         # Run conversation with error handling
+        run_timeout = int(os.getenv("CONVERSATION_TIMEOUT", "3600"))
         try:
             with warnings.catch_warnings():
                 warnings.filterwarnings("ignore", category=UserWarning)
-                conversation.run()
+                conversation.run(timeout=run_timeout)
             logger.info(f"Conversation completed for {instance.id}")
         except ValidationError as e:
             logger.warning(f"Validation error from custom events (continuing): {e}")

benchmarks/swebench/run_infer.py

@@ -247,7 +247,8 @@ def _log_event(ev):  # keep it simple
             workspace_path=workspace.working_dir,
         )
         conversation.send_message(instruction)
-        conversation.run()
+        run_timeout = int(os.getenv("CONVERSATION_TIMEOUT", "3600"))
+        conversation.run(timeout=run_timeout)
 
         # git add
         workspace.execute_command(f"cd {repo_path} ; git add -A")

benchmarks/swebenchmultimodal/run_infer.py

@@ -291,7 +291,8 @@ def _log_event(ev):  # keep it simple
         else:
             logger.info("No image_assets found, sending text-only instruction")
             conversation.send_message(instruction)
-        conversation.run()
+        run_timeout = int(os.getenv("CONVERSATION_TIMEOUT", "3600"))
+        conversation.run(timeout=run_timeout)
 
         # git add
         workspace.execute_command(f"cd {repo_path} ; git add -A")

benchmarks/swtbench/run_infer.py

@@ -274,7 +274,8 @@ def _log_event(ev):  # keep it simple
             workspace_path=workspace.working_dir,
         )
         conversation.send_message(instruction)
-        conversation.run()
+        run_timeout = int(os.getenv("CONVERSATION_TIMEOUT", "3600"))
+        conversation.run(timeout=run_timeout)
 
         # git add
         workspace.execute_command(f"cd {repo_path} ; git add -A")