We read every piece of feedback, and take your input very seriously.
To see all available qualifiers, see our documentation.
torch.cuda.amp.autocast(args...)
torch.autocast("cuda", args...)