Open Models Almanac
Methodsinference & servingdecoding strategy

single mention · inference & serving

Three-layer dense MLA speculative drafter

Uses three dense MLA draft layers in BF16 to propose a three-token block.

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

Also called 3 dense MLA draft layers in BF16.

Filed under inference & serving › decoding strategy