Open Models Almanac
Methodspost-trainingpreference optimization

single mention · post-training

Budget-based verbosity control

An implementation detail that penalizes a candidate in a binary comparison when its output exceeds a multiplier of the cold-start model’s estimated verbosity.

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

Filed under post-training › preference optimization