Search papers, labs, and topics across Lattice.
The paper introduces GRPO-VPS, an enhancement to Group Relative Policy Optimization (GRPO) that improves LLM reasoning by incorporating verifiable process supervision. GRPO-VPS segments the LLM's generation into discrete steps and tracks the conditional probability of the correct answer at each step, providing interpretable segment-wise progress measurements. Experiments on mathematical and general-domain benchmarks demonstrate that GRPO-VPS achieves more targeted policy updates, leading to improved accuracy and reduced reasoning length compared to GRPO.
LLMs can reason more effectively by directly tracking their own belief in the correct answer throughout the reasoning process, enabling more targeted policy updates.
Reinforcement Learning with Verifiable Rewards (RLVR) has advanced the reasoning capabilities of Large Language Models (LLMs) by leveraging direct outcome verification instead of learned reward models. Building on this paradigm, Group Relative Policy Optimization (GRPO) eliminates the need for critic models but suffers from indiscriminate credit assignment for intermediate steps, which limits its ability to identify effective reasoning strategies and incurs overthinking. In this work, we introduce a model-free and verifiable process supervision via probing the model's belief in the correct answer throughout its reasoning trajectory. By segmenting the generation into discrete steps and tracking the conditional probability of the correct answer appended at each segment boundary, we efficiently compute interpretable segment-wise progress measurements to refine GRPO's trajectory-level feedback. This approach enables more targeted and sample-efficient policy updates, while avoiding the need for intermediate supervision derived from costly Monte Carlo rollouts or auxiliary models. Experiments on mathematical and general-domain benchmarks show consistent gains over GRPO across diverse models: up to 2.6-point accuracy improvements and 13.7% reasoning-length reductions on math tasks, and up to 2.4 points and 4% on general-domain tasks, demonstrating strong generalization.
Tsinghua AI