MDP-GRPO: Stabilized Group Relative Policy Optimization for Multi-Constraint Instruction Following 文章

ArXiv CS.CL2026-06-05NEWSen作者: Mohammad Mahdi Salmani-Zarchi, Zahra Rahimi, Heshaam Faili, Mohammad Javad Dousti
查看原文 →
关系图谱

MDP-GRPO: Stabilized Group Relative Policy Optimization for Multi-Constraint Instruction Following · 相关技术

相关技术

asymmetric KL regularizationprospect-theoretic shapingdual-anchor advantagesmulti-temperature samplingz-score group normalizationMDP-GRPOGRPO