Rewrite prefix key-padding masks as k/v truncation to keep SDPA on the flash path (PyTorch) - #80

Rewrite prefix key-padding masks as k/v truncation to keep SDPA on the flash path (PyTorch)#80

arnaudlvq wants to merge 1 commit into

google-research:maingoogle-research/tabfm:mainfrom

arnaudlvq:prefix-mask-flash-attentionarnaudlvq/tabfm:prefix-mask-flash-attentionCopy head branch name to clipboard