method · first seen 7 Nov 2026
policy gradient with multi-pass rollout reuse
Also: off-policy reuse of rollouts
Reusing each batch of sampled rollouts for several gradient passes to reduce generation cost in policy optimization.
Mentions per week
PapersCodeDiscussion
Earliest sources
- BRAT: Bounded Residual-Aligned Training for Stable Multi-Pass LLM Post-TrainingPapers · 7 Nov 2026 · Xunlan Zhou, Yuzhe Zhao, Xuhong Huang, Feng Chen, Xuanlin Chen, Lang Feng, Bo An
Latest sources
- BRAT: Bounded Residual-Aligned Training for Stable Multi-Pass LLM Post-TrainingPapers · 7 Nov 2026 · Xunlan Zhou, Yuzhe Zhao, Xuhong Huang, Feng Chen, Xuanlin Chen, Lang Feng, Bo An