Open Models Almanac
Methodstraining objectiveauxiliary loss

single mention · training objective

Mask non-target tokens in rollout training data

Excludes prompt, history, tool-output, and prefilled response-prefix tokens from the training loss.

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

Filed under training objective › auxiliary loss