0.4.0

remove group norms https://arxiv.org/abs/2312.02696

0.3.1

able to customize attention heads and dim head differently across stages

0.3.0

switch 3x3 convs within convnext to be semi depthwise

0.2.1

make it work with rectangular images

0.2.0

update convnext with global response normalization from updated paper

0.1.0

switch back to regular attention

0.0.22

make sure nested unet can accommodate feature dimensions other than p…

…ower of 2

0.0.21

upgrade to best type of downsample

0.0.20

change the nested RSU from unet squared to using conv3d, in preparati…

…on to make unet squared support video

0.0.19

add ability to finely specify number of self attention blocks per stage