rollouts

Definition

Runs of the current policy through the environment, each producing a complete record of actions and outcomes that becomes training signal. In reasoning-model training the standard recipe is to sample many rollouts for the same problem, score each by whether it reached a verifiable answer, and update toward the ones that succeeded. How many you sample per problem is a central cost knob, since generation dominates the compute bill.

Explainers using this term

• Paper Walkthrough: Annotations as Rollouts — Dropping the Ground Truth Into the Group as a Ninth AnswerAnnotations as Rollouts: Efficient and Scalable Reinforcement Learning for Video MLLMs