Open sfsdfd62 opened 2 years ago
Try torch.backends.cudnn.enabled = False, it resolves my problem, but the execution becomes very slow.
Try torch.backends.cudnn.enabled = False, it resolves my problem, but the execution becomes very slow.
I understood this situation. This is because I'm using GTX 3090, which only supports CUDA version>11.0 and pytorch >some certain release. But it seems that Pytorch used in this work is way too out-dated, so we may need to upgrade it.
My version information is as below: cuDNN:8.5.0 cuda:11.7 pytorch=1.4.0 python=3.6.6 Ubuntu 20.04.4