0.2.23

dropout still helpful for smaller transformers

0.2.22

nvm confused myself

0.2.21

fix can_eos logic in generate

0.2.20

not needed

0.2.19

sos is autohandled

0.2.18

add a masked mean squeeze excite

0.2.17

add a masked mean squeeze excite

0.2.16

oops, should dropout after activation

0.2.15

all dropouts

0.2.14

all dropouts