Open Models Almanac
Methodstraining objectivedistillation objective

single mention · training objective

Backbone-to-MTP-head knowledge distillation

A knowledge-distillation loss trains the MTP head using the frozen backbone as teacher.

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

Filed under training objective › distillation objective