how can I run the test across multi nodes？

I run the test_internode.py and test_low_latency on H20 (node_num=2, rank_num=16)

What is the correct environment variable setting or what is the full script？

Only I set NCCL_IB_GID_INDEX=3, the test_internode.py can run normally,

but, even I set NCCL_IB_GID_INDEX=3, the test_low_latency.py runs abnormally.

Thank you！