1.1.4

make sure amax for numerical stability is detached

1.1.3

1.1.3

1.1.2

1.1.2

1.1.1

fix axis for torch max used for numerical stability when deriving key…

… prime

1.1.0

update example

1.0.11

remove pytorch-fast-transformers from dep

1.0.10

add a handy method for ProjectionUpdater

1.0.9

expose projection update logic, for easy use in alphafold2 project

1.0.8

expose cross attention, in ready to use Performer in alphafold2 repo

1.0.7

allow for users to turn off bias for attention out projections