Open Models Almanac
Methodsoptimizationtraining stability

single mention · optimization

Behavioral regularization

Behavioral regularization is a set of penalty mechanisms added to RL training to improve stability and encourage efficient, reliable behavior.1 It penalizes excessive generated-token growth relative to successful rollouts in a group, and applies segment-level penalties for format violations and tool call errors.2

How this definition was written

Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.

  1. ToimproveRLtrainingstabilityandencourageefficient, reliablebehavior, weintroducetwocom- plementary mechanisms — MiMo V2 6 technical report
  2. A group-relative length penalty curbs excessive token growth, while segment-level behavioral penalties address format violations and tool call errors — MiMo V2 6 technical report

Filed under optimization › training stability