0.0.5

may need for another project

0.0.4

pre-layernorm optional

0.0.3

allow for biasing attention matrix

0.0.2

stable softmax

0.0.1

first commit