1. 确认CUDNN+CUDA+pytorch运行正确
importtorchtorch.__version__'1.6.0+cu101'torch.version.cuda'10.1'# fengkai @ archer in ~ [15:53:15] $nvcc-Vnvcc: NVIDIA (R) CudacompilerdriverCopyright (c) 2005-2019NVIDIACorporationBuiltonSun_Jul_28_19:07:16_PDT_2019Cudacompilationtools, release10.1, V10.1.243
2. 上述没有问题后,运行
rm-rf ~/.nv