0.5.1

warmup to full decoder discriminator loss over 30k steps

0.5.0

add diff augment with translation and cutout

0.4.3

turn off path length reg by default and update batch size

0.4.2

revert to norm queries for linear attention

0.4.1

bump to allow for webp images

0.4.0

update image linear attention to better variant, and also change adam…

0.3.1

add samples, double network capacity, update readme

0.3.0

add initial conv after 4x4

0.2.3

more aggressive gradient penalty at the first 5000 steps

0.2.2

bump for release