fix: fixed local device name in multinode example.

Author: fjosw

distributed/ddp-tutorial-series/multinode.py

@@ -37,7 +37,7 @@ def __init__(
         self.model = DDP(self.model, device_ids=[self.local_rank])
 
     def _load_snapshot(self, snapshot_path):
-        loc = f"cuda:{self.gpu_id}"
+        loc = f"cuda:{self.local_rank}"
         snapshot = torch.load(snapshot_path, map_location=loc)
         self.model.load_state_dict(snapshot["MODEL_STATE"])
         self.epochs_run = snapshot["EPOCHS_RUN"]