0.1.9

address https://github.com/lucidrains/FLASH-pytorch/issues/12

0.1.8

add ability to directly handle t5 relative positional bias within GAU

0.1.7

fix laplace activation function, thanks to @boweny-cerebras

0.1.6

add the laplace attention function, which the authors of Mega propose…

v0.1.5

make reducing group for context in noncausal linear attention a hyper…

v0.1.4

fix an error thanks to @ShomyLiu at https://github.com/lucidrains/FLA…

0.1.2

fix mask for linear attention

0.1.1

add token shift, for dramatic improvements, cite @BlinkDL

0.0.15

0.0.15a

mask bug

0.0.14

fix