Open Models Almanac
Methodsinference & servingdecoding strategy

single source · inference & serving

Recursive shared MTP-head drafting

Applies an MTP head recurrently or recursively to propose several tokens for verification in one target-model pass.

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.

Filed under inference & serving › decoding strategy