Open Models Almanac
Methodsinference & servingdecoding strategy

single mention · inference & serving

MTP draft-length configuration

Configures the number of MTP draft tokens through serving-engine parameters.

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 MTP draft length.

Filed under inference & serving › decoding strategy