Open Models Almanac
Methodsinference & servinginference kernel

single mention · inference & serving

FA4 blockwise attention

Blockwise attention from FA4 used to handle re-tokenization while maintaining efficiency.

The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See inference kernel for the methods filed beside it.

Also called blockwise attention from FA4.

Filed under inference & serving › inference kernel