📄 中文摘要
Chain-of-Thought推理显著提升了大型语言模型在复杂任务上的表现，但由于生成过程较长，导致推理延迟较高。逐步推测推理旨在降低这一成本，但现有方法在准确性、推理速度和资源效率之间存在长期的权衡。提出的ConfSpec是一种置信度门控的级联验证框架，旨在解决这一权衡。关键在于生成与验证之间的不对称性：生成正确的推理步骤需要较大的模型能力，而逐步验证则是一个受限的判别任务，小型草稿模型在其能力范围内经过良好校准，从而实现高效的推理。
ConfSpec：通过置信度门控验证实现高效的逐步推测推理
📄 English Summary
ConfSpec: Efficient Step-Level Speculative Reasoning via Confidence-Gated Verification
Chain-of-Thought reasoning significantly enhances the performance of large language models on complex tasks but incurs high inference latency due to lengthy generation traces. Step-level speculative reasoning aims to alleviate this cost; however, existing approaches face a long-standing trade-off among accuracy, inference speed, and resource efficiency. The proposed ConfSpec is a confidence-gated cascaded verification framework that addresses this trade-off. The key insight lies in the asymmetry between generation and verification: while generating a correct reasoning step requires substantial model capacity, step-level verification is a constrained discriminative task for which small draft models are well-calibrated within their competence range, enabling efficient reasoning.