Add compress_factor for compressed causal attention#2418

jduprat wants to merge 2 commits into

Dao-AILab:mainDao-AILab/flash-attention:mainfrom

jduprat:compress-factorjduprat/flash-attention:compress-factorCopy head branch name to clipboard