diff --git a/dags/test-train-pytorch.py b/dags/test-train-pytorch.py index 3b3f003..9bde1c5 100644 --- a/dags/test-train-pytorch.py +++ b/dags/test-train-pytorch.py @@ -31,6 +31,7 @@ def run_rank_0(**context): "MASTER_ADDR": fqdn, "MASTER_PORT": MASTER_PORT, "WORLD_SIZE": "2", + "NCCL_IB_GID_INDEX": "0", "RANK": "0", "NCCL_SOCKET_IFNAME": "eth0", # bootstrap/rendezvous через eth "NCCL_IB_DISABLE": "0", # трафик через IB