diff --git a/verl/workers/fsdp_workers.py b/verl/workers/fsdp_workers.py index 1d3ce76e..8f9fa6be 100644 --- a/verl/workers/fsdp_workers.py +++ b/verl/workers/fsdp_workers.py @@ -816,7 +816,7 @@ def save_checkpoint(self, local_path, hdfs_path=None): if hdfs_path is not None: print(f'Uploading critic checkpoint to {hdfs_path}') hdfs_io.makedirs(hdfs_path, exist_ok=True) - hdfs_io.copy(src=local_path, dst=hdfs_path) + hdfs_io.copy(src=local_path, dst=hdfs_path, dirs_exist_ok=True) torch.distributed.barrier() if self._is_offload_param: