A call to torch.cuda.is_available makes an unrelated multi-processing computation crash?

5

Looks like set_start_method did not work for me but mp = mp.get_context('spawn') did. Hope that provides some help.