SHAPE: Stage-aware Hierarchical Advantage via Potential Estimation for LLM Reasoning
This addresses token inefficiency and reasoning accuracy for LLM users, representing an incremental improvement over existing process supervision methods.
The paper tackled the problem of limited reasoning capabilities and token inefficiency in LLMs by proposing SHAPE, a framework that formalizes reasoning as a trajectory and uses hierarchical credit assignment, resulting in an average accuracy gain of 3% with 30% reduced token consumption in math reasoning benchmarks.
Process supervision has emerged as a promising approach for enhancing LLM reasoning, yet existing methods fail to distinguish meaningful progress from mere verbosity, leading to limited reasoning capabilities and unresolved token inefficiency. To address this, we propose Stage-aware Hierarchical Advantage via Potential Estimation (SHAPE), a framework that formalizes reasoning as a trajectory through a state space of empirical solvability. SHAPE introduces a hierarchical credit assignment mechanism: at the segment level, it employs a stage-aware advantage function to prioritize efficient breakthroughs in low-potential states; at the token level, it utilizes entropy-driven redistribution to sharpen execution signals. Extensive experiments in math reasoning across three base models and five benchmarks demonstrate that SHAPE achieves an average accuracy gain of 3% with 30% reduced token consumption.