if all_args.cuda_deterministic: torch.backends.cudnn.benchmark = False torch.backends.cudnn.deterministic = True注意需要设置 cuda_deterministic == True
cudnn.benchmark == False