apply scale to 2d rel pos bias in levit

only apply scaling after applying 2d rel pos bias

implement SOTA new self-supervised learning technique from facebook f…

…or vision transformers, Dino

implement SOTA new self-supervised learning technique from facebook f…

…or vision transformers, Dino

be able to accept non-square patches, thanks to @FilipAndersson245

fix small bug

release twins svt

for PiT, project to increased dimensions on first grouped conv for de…

…pthwise-conv

add zeroing of weight parameters of batchnorm in levit just before re…

…sidual connection, noticed by @EelcoHoogendoorn