Jul 2026
ReCo: Reweighting GRPO Against Distributional Concentration
ReCo, a reweighting method that improves Pass@k for large values of k and is comparable to GRPO for small values of k, and replaces the token-level importance ratio with a variance-based ratio that gives larger update scale to non-saturated decision points.
Junoh Park, Junseong Hwang, Won-Yong Cho et al.
· arXiv.org · 0 citations