MDP-GRPO: Stabilized Group Relative Policy Optimization for Multi-Constraint Instruction Following
For practitioners training LLMs with verifiable rewards on multi-constraint tasks, MDP-GRPO provides a stable and effective alternative to standard GRPO.
MDP-GRPO stabilizes group-relative policy optimization for multi-constraint instruction following by addressing pathologies in z-score normalization under discrete, low-dispersion rewards, achieving up to 5.0% improvement in strict constraint satisfaction on Llama-3.2-3B.
Reinforcement learning with verifiable rewards is ideal for multi-constraint instruction following, yet standard group-relative policy optimization (GRPO) becomes unstable under discrete, low-dispersion rewards, where within-group reward distributions are frequently homogeneous. We identify and formalize three pathologies of z-score group normalization in this regime: low-variance amplification, mean-centering blindness, and zero-variance collapse. To address them, we propose MDP-GRPO, which stabilizes learning through (1) multi-temperature sampling to increase reward dispersion, (2) dual-anchor advantages to restore gradients in homogeneous groups and stop mean-centering blindness, (3) prospect-theoretic shaping to bound updates and penalize violations based on Kahneman and Tversky's theory, and (4) asymmetric KL regularization. Evaluated on FollowBench, IFEval, and a curated multi-constraint dataset, MDP-GRPO outperforms standard GRPO, improving strict constraint satisfaction by up to 5.0% on Llama-3.2-3B. Our method also enables stable convergence with small group sizes while preserving general capabilities on MMLU and ARC.