0.1.6

0.1.6

0.1.5

isolate fastweight mem logic into module and make exportable, so some…

0.1.1a

fix an issue with rotary embeddings and cached kv

0.1.0

seems to tbe working

0.0.25

allow for infini-transformer learning on detached past memories, hand…

0.0.24

handle edge case where different batch samples can have different num…

0.0.23

training works with any sequence length

0.0.22

simplify

0.0.21

able to generate

0.0.20

start using dropout again bc of https://arxiv.org/abs/2305.16179