Closed johnnynunez closed 9 months ago
@johnnynunez cuDNN should be automatically detected by PyTorch and USE_CUDNN=1
isn't explicitly needed.
If you're on JetPack 4, make sure the default docker runtime is nvidia and DOCKER_BUILDKIT=0
thanks @dusty-nv
Should we use export USE_CUDNN=1 to build Pytorch? @dusty-nv