Open Models Almanac
Methodsoptimizationtraining stability

single mention · optimization

Per-token regularization for off-policy stability

Constrains policy updates to a localized neighborhood to tolerate highly stale data and sustain training stability.

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

Also called per-token regularization for off-policy RL.

Filed under optimization › training stability