v0.16.8

cast long as float before deriving sinusoidal pos emb

v0.16.7

remove forcing of softmax in f32, in case it is interfering with deep…

v0.16.6

set ability to do warmup steps for each unet during training

v0.16.5

set ability to do warmup steps for each unet during training

v0.16.3

need to keep track of training steps separately for each unet in deco…

v0.16.2

bring in two tricks from the cogview paper for reducing the chances o…

v0.16.1

bring in two tricks from the cogview paper for reducing the chances o…

v0.16.0

do bias-less layernorm manually

v0.15.4

0.15.4

v0.15.3

add ability to specify full self attention on specific stages in the …