diff --git a/slime/backends/megatron_utils/actor.py b/slime/backends/megatron_utils/actor.py index adf9ad3a33..16f1748cc5 100644 --- a/slime/backends/megatron_utils/actor.py +++ b/slime/backends/megatron_utils/actor.py @@ -515,7 +515,7 @@ def save_model(self, rollout_id: int, force_sync: bool = False) -> None: # torch dist may trigger nccl communication during saving. if self.args.offload_train: - reload_process_groups() + self.wake_up() if self.args.async_save: from megatron.training.async_utils import maybe_finalize_async_save @@ -533,7 +533,7 @@ def save_model(self, rollout_id: int, force_sync: bool = False) -> None: save_hf_model(self.args, rollout_id, self.model) if self.args.offload_train: - destroy_process_groups() + self.sleep() @timer def update_weights(self) -> None: