1.41.0

add data dependent alibi from the forgetting transformers paper from …

1.40.11

multiple cls tokens support from @pradeep-pyro

1.40.10

address https://github.com/lucidrains/x-transformers/issues/282

1.40.9

Full Changelog: 1.40.8...1.40.9

1.40.8

make sure value residuals work for cross attention blocks too

1.40.7

seeing immediate improvements when investigating a new iclr 2025 paper

1.40.6

1.40.6

1.40.5

address in-attention conditioning for @JLenzy https://github.com/luci…

1.40.4

add ability to reinject input at every layer

1.40.3

needed for a disney research paper