I am getting the following error when trying to use the efficientNet architectures in torchvision models.
It’s in 0.11.0 release
Thankyou. The issue was with the torchvision version.