解决方案
1.https://developer.nvidia.com/nccl/nccl-legacy-downloads
找到cuda 对应的NCCL版本,需要账号登录,我的是cuda11.2 +ubuntu20.04
<Download NCCL 2.8.3, for CUDA 11.2, January 06,2021>
<Local installer for Ubuntu 20.04> 下载.deb
sudo dpkg -i nccl-repo-对应版本.deb
sudo apt update
sudo apt install libnccl2=2.8.3-1+cuda11.2 libnccl-dev=2.8.3-1+cuda11.2(换成自己的版本)
参考链接
https://blog.youkuaiyun.com/weixin_45653647/article/details/116306628
https://blog.youkuaiyun.com/fyfugoyfa/article/details/124203296