Self-Guided Process Reward Optimization with Redefined Step-wise Advantage for Process Reinforcement Learning
This work addresses efficiency and theoretical gaps in PRL for LLMs, offering a practical solution for industrial applications, though it is incremental as it builds on existing PRL methods.
The paper tackles the computational overhead and lack of theoretical framework in Process Reinforcement Learning (PRL) for Large Language Models by proposing SPRO, which uses intrinsic process rewards and a novel step-wise advantage estimation, resulting in 3.4x higher training efficiency and a 17.5% test accuracy improvement compared to baseline methods.
Process Reinforcement Learning~(PRL) has demonstrated considerable potential in enhancing the reasoning capabilities of Large Language Models~(LLMs). However, introducing additional process reward models incurs substantial computational overhead, and there is no unified theoretical framework for process-level advantage estimation. To bridge this gap, we propose \textbf{S}elf-Guided \textbf{P}rocess \textbf{R}eward \textbf{O}ptimization~(\textbf{SPRO}), a novel framework that enables process-aware RL through two key innovations: (1) we first theoretically demonstrate that process rewards can be derived intrinsically from the policy model itself, and (2) we introduce well-defined cumulative process rewards and \textbf{M}asked \textbf{S}tep \textbf{A}dvantage (\textbf{MSA}), which facilitates rigorous step-wise action advantage estimation within shared-prompt sampling groups. Our experimental results demonstrate that SPRO outperforms vaniila GRPO with 3.4x higher training efficiency and a 17.5\% test accuracy improvement. Furthermore, SPRO maintains a stable and elevated policy entropy throughout training while reducing the average response length by approximately $1/3$, evidencing sufficient exploration and prevention of reward hacking. Notably, SPRO incurs no additional computational overhead compared to outcome-supervised RL methods such as GRPO, which benefit industrial implementation.