delete exit message

Author: binary-husky

ajet/backbone/trainer_verl.py

@@ -596,8 +596,9 @@ def fit(self):  # noqa: C901
                         context_tracker_arr: List[BaseContextTracker] = self.parallel_env.rollout(
                             tasks, mode="sample", epoch=f"train.{epoch}"
                         )
-                        from ajet import bp
-                        bp("BATCH")
+
+                        # from ajet import bp; bp("BATCH")
+
                         logger.info("end fit rollout")
                         gen_batch_output = self.parallel_env.to_dataproto(context_tracker_arr)
                         logger.info("end dataproto convertion")

ajet/task_rollout/single_worker.py

@@ -128,7 +128,7 @@ def rollout_env_worker(
                     workflow_task=workflow_task,
                 )
             except SwarmReceiveAbortException as exc:  # noqa: BLE001
-                print('SwarmReceiveAbortException caught in rollout_env_worker')
+                # print('SwarmReceiveAbortException caught in rollout_env_worker')
                 return None # type: ignore
             except TestSuccessException as e:
                 logger.success(

ajet/task_runner/swarm_runner.py

@@ -75,7 +75,7 @@ def register_episode_and_wait_output(
                     message = zmq_socket.recv_string()
                 except zmq.Again as e:
                     if should_exit_hard():
-                        logger.warning(f'{episode_uuid} Exiting workflow due to should_exit_hard signal.')
+                        # logger.warning(f'{episode_uuid} Exiting workflow due to should_exit_hard signal.')
                         context_tracker.reset()
                         raise SwarmReceiveAbortException(f"Episode {episode_uuid} aborted due to system exit.")
                     else:
@@ -127,7 +127,7 @@ def execute(self, workflow_task: WorkflowTask) -> BaseContextTracker:
         should_exit_hard = hooks['should_interrupt_hard_fn']
 
         if should_exit_soft() or should_exit_hard():
-            print(f'Exiting workflow worker due to interrupt signal for episode {workflow_task.episode_uuid}.')
+            # print(f'Exiting workflow worker due to interrupt signal for episode {workflow_task.episode_uuid}.')
             raise SwarmReceiveAbortException(f"Episode {workflow_task.episode_uuid} aborted due to interrupt signal.")
 
         # context tracker will trace and gather everything we need for training