import os import datetime import shutil from torch.distributed import barrier, init_process_group, destroy_process_group init_process_group(backend="nccl", timeout=datetime.timedelta(seconds=12 * 60 * 60)) from_fp = "/mnt/round-surf/checkpoints/2024-04-15_22-54-10/last_ckpt.pt" to_fp = "/tmp/13b.pt" # Check if the destination exists if os.path.exists(to_fp): print("deleting destination") # Delete if it exists os.remove(to_fp) barrier() print("copy", from_fp, "to", to_fp) # Make the copy shutil.copyfile(from_fp, to_fp) barrier() print("done!") destroy_process_group()