Since it is misisng in the docu (https://pytorch.org/docs/stable/backends.html) I’m wondering what is the default value of torch.backends.cudnn.enabled or does it depend on special things like PyTorch version, hardware etc?
If cudnn is available (which is the case for the binaries shipping with CUDA+cudnn), it’s enabled by default and you can check it via print(torch.backends.cudnn.enabled).
1 Like