Open Models Almanac
Methodspost-trainingrollout & RL infrastructure

single mention · post-training

Duration- and acceptance-adjusted per-source oversampling

Assigns per-source scheduling budgets using an oversampling ratio adjusted for source characteristics.

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

Also called oversampling ratio.

Filed under post-training › rollout & RL infrastructure