transfer recipes for resnet strikes back?

[Oklahomawhore]

hi, thanks for the amazing effort of this library. I want to reproduce the transfrer learning results form RSB, you did mention that transfer learning recipes are similar to A3 #901 . But if there is a full recipe it would be nice.

The results that I would like to reproduce ( i.e. CIFAR-10 and CIFAR-100)

I downloaded the weights from here but the inital accuracy when weights are loaded has a huge defecit on CIFAR100 dataset 76.28 vs 86.9% in the paper.

thank you very much!