Skip to content

Comment on New attention mechanisms that outperform standard multi-head attention

Comments

I feel like FlashAttention is the relevant baseline here.

FlashAttention is completely orthogonal to this. This work is about speeding up the computation of Q, K and V vectors while FlashAttention is about speeding up the attention algorithm itself.

You could combine the two.

Then they should have. It isn’t clear to be that this change is measurable when the code is optimized. Even in the presented numbers the difference seems meh at times.

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.