1.1.5

switch to stable softmax when sampling

1.1.4

1.1.4

1.1.2

1.1.2

1.1.1

add additional stabilization technique, looking at ruDALL-E results

1.1.0

add sandwich norm, from the coqview paper, for stabilizing training e…

1.0.8

no rotary embedding for transformers within CLIP model

1.0.7

enforce latest einops

1.0.6

default shifting tokens to false

1.0.5

bump

1.0.4

Update discord link (permanent now). (#351)