0.3.1

add GLU gating for feedforward

0.3.0

one key/value head feature

0.2.2

mixed precision

0.2.1

bump

0.2.0

compressive transformer now works

0.1.2

add ability to customize weight on reconstruction loss

0.1.1

update readme, and fix bug with variable memory layers

0.1.0

enwik8 training now working with compressive transformers

0.0.8

allow memory layers to be specified as a range

0.0.7

fix all device related errors, make sure cuda works