0.11.1

Full Changelog: 0.11.0...0.11.1

0.11.0

Full Changelog: 0.10.19...0.11.0

0.10.20

address https://github.com/lucidrains/CoLT5-attention/issues/8

0.10.19

add decorators to function

0.10.18

make sure differentiable topk is not performed in float16

0.10.17

make sure last_da is only used for the first backwards segment, when …

…the number of checkpoint segments is greater than 1

0.10.16

disable autocast for coordinate descent functions

0.10.15

prepare for differentiable topk use for another lib

0.10.14

fix query rotary embeddings in routed autoregressive attention

0.10.12

fix autoregressive routed rotary embeddings