0.22.3

add diffaugment script from zhaos paper, allow user to specify which …

0.22.2

no sync all DDPs that were usedeven if detached, to play it safe

0.22.1

properly manage gradient accumulations for distributed case

0.22.0

add distributed data parallel and offer multi-GPU training on one mac…

0.21.2

use entry points instead of scripts in setup.py, potentially fixing w…

0.21.1

allow for assigning a specific gpu id for training, readying for mult…

0.21.0

add relativistic discriminator loss

0.20.2

bump

0.20.1

make random crop in augmentation a bit less severe

0.20.0

make sure there is an initial conv in generator after 4x4