PyTorch is only part of it. There is still a huge amount of CUDA that isn’t just wrapped by PyTorch and isn’t easily portable.
... but not in deep learning or am I missing something important here?
... but not in deep learning or am I missing something important here?