Hacker News new | ask | show | jobs
Jagged Flash Attention Optimization (shaped.ai)
24 points by tullie 461 days ago
2 comments
platers 458 days ago
Flash attention natively supports packing multiple variable length sequences into a single call, what is the advantage of jagged flash attention?
link
bbstats 457 days ago
If only there was a link to a page somewhere that could answer this question for you.
link
CapsAdmin 458 days ago
See also

https://github.com/thu-ml/SageAttention and https://github.com/thu-ml/SpargeAttn

link