0.4.33

make some progress on sparse causal 2dna attention, for audio

0.4.32

add token shift along time axis for audio tokens for NUWA audio + video

0.4.31

fix some bugs

0.4.30

VAE passed into NUWA does not need the discriminator nor VGG, make a …

…copy and remove them

0.4.29

move vqgan vae to separate file

0.4.26

working NUWA trainer class for moving mnist test

0.4.25

prepare for NUWA to accept video token ids, regardless of whether VAE…

… is supplied on init

0.4.23

make sure discriminator gets separate optimizer and add gradient penalty

0.4.21

some more fixes for discriminator

0.4.20

fix missing activation in discriminator