Search papers, labs, and topics across Lattice.
The paper introduces Preplan-Plan-CoT (PPC), a novel framework for mathematical reasoning in LLMs that explicitly incorporates a "preplan" stage to identify the problem type, applicable tools, and potential pitfalls before planning and execution. PPC uses a three-stage synthesis pipeline with a spoiler-score detector to generate clean preplan supervision and a composite GRPO reward to ensure the plan follows from the preplan. Experiments across multiple LLMs and mathematical reasoning benchmarks demonstrate that PPC significantly outperforms existing methods, achieving state-of-the-art results on most metrics without increasing inference token overhead.
LLMs can get a +3% boost in mathematical reasoning accuracy just by explicitly thinking about *what* kind of problem they're facing before diving into *how* to solve it.
Current plan-based reasoning methods improve large language models (LLMs) by inserting a planning stage before execution, giving rise to the question $\rightarrow$ plan $\rightarrow$ cot paradigm. While effective, a closer examination reveals an inherent paradigm-level gap: both the planning and its execution stages decide how to solve a problem, while the prior question of what to solve; recognizing the problem type, the applicable tools, and the foreseeable pitfalls; remains entirely implicit. To bridge this gap, we propose PPC (Preplan-Plan-CoT), a framework that introduces an explicit problem-understanding stage, the preplan, yielding a new question $\rightarrow$ preplan $\rightarrow$ plan $\rightarrow$ cot paradigm. Realizing this paradigm requires safeguarding the conceptual integrity of preplan at both ends. Specifically, we design a three-stage synthesis pipeline with a spoiler-score detector that filters out leakage and spoiler failures to build clean preplan supervision, and a composite GRPO reward enforces that the generated plan genuinely follows from the preplan. Experiments across four backbones and five mathematical reasoning benchmarks show that PPC achieves the best results on 39 of 40 metrics, improving maj@16 and pass@16 by +2.23 and +3.06 over the strongest baseline without introducing additional inference token overhead.
