Open Models Almanac
Methodsinference & servinginference kernel

single mention · inference & serving

Fused QSA attention and KL-loss kernel

Jointly computes sparse attention outputs and KL loss without materializing intermediate results.

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 Fused sparse-attention and KL-loss kernel, fused QSA kernel.

Filed under inference & serving › inference kernel