VERPO: Verified Evidence Regularized Policy Optimization

T1Covered by 1 source · first reported by arXiv at 24 Sept, 04:00 UTC

arXiv:2609.06100v4 Announce Type: replace-cross Abstract: Verifiable rewards improve language models through reliable task-level feedback, but methods based on Group Relative Policy Optimization (GRPO) apply a sequence-level advantage uniformly across all tokens. This coarse credit assignment…

science

Sources

• T1VERPO: Verified Evidence Regularized Policy OptimizationarXiv24 Sept, 04:00