AlexCzar opened this issue 3 years ago · 0 comments
Now that there is official support for ROCm in PyTorch, would it be possible to use that instead of CUDA?