这个错误可能是由于NCCL_P2P_LEVEL设置不正确导致的。你可以尝试将NCCL_P2P_LEVEL设置为0，然后重新运行...
I also tried to set the NCCL_P2P_LEVEL in ~/.nccl.conf, but I get the same result. How should I proceed to set NCCL_P2P_VALUE=NVL ? I have NVLink on my machine (nvidia-smi -m topo shows it) with 8 GPUs. Thanks a lot. KimchiMember sjeaugey commented Apr 29, 2020 Hi, The...