Open Models Almanac
Methodspost-trainingrollout & RL infrastructure

single mention · post-training

Bounding off-policy sample ratio

Tunes sample dispatch and waiting conditions to bound how far training samples are off-policy.

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 bound the maximum off-policy ratio.

Filed under post-training › rollout & RL infrastructure