0.9.0

remove outdated apex in favor of native pytorch AMP

0.8.1

get rid of numpy

0.8.0

use full attention at the center of the unet

0.7.1

greater kernel size in convnext blocks

0.7.0

replace resnets with convnext blocks

0.6.9

fix layernorm

0.6.8

remove stray print

0.6.7

use pre-layernorm with linear attention, and also allow for turning o…

0.6.6

move all constants into configurable class init parameters

0.6.5

fix data not being normalized to range of -1 to 1