用少量数据教会大模型按难易程度调整推理深度，减少冗余计算。
Hint Tuning: Less Data Makes Better Reasoners
- 通过指令模型判断题目难度，自动生成三类训练样本
- 仅用1000条数据即实现平均31.5%的推理token减少
- 适合需要高效推理的部署场景，尤其轻量化模型
大型推理模型虽能通过扩展思维链达到高准确率，但常生成5–8个多余词元，对所有题目采用统一冗长推理。本文提出提示调优（Hint Tuning），一种数据高效的推理深度自适应方法。核心思路是：指令模型可作为理想的难度探测器。通过测试指令模型在不同提示下的求解能力，自动构建三类训练样本：无提示（直接回答）、稀疏提示（最小前缀）和完整提示（完整推理）。该方法将抽象的难度标注转化为指令模型与推理模型之间的可度量一致性检验。仅需1000条自标注样本，提示调优在主流推理模型（Qwen3-Thinking、DeepSeek-R1-Distill）上实现了24%–66%的词元减少（平均31.5%），覆盖4B–32B多种规模，在五个基准测试中保持竞争力。相比需海量蒸馏数据或昂贵强化学习的方法，本方法通过简单对齐指令模型能力实现更高效率。代码与数据已开源。
原文摘要 · Abstract (English)
Large reasoning models achieve high accuracy through extended chain-of-thought but generate 5--8 more tokens than necessary, applying verbose reasoning uniformly regardless of problem difficulty. We propose Hint Tuning, a data-efficient approach that teaches models to calibrate reasoning depth. Our key insight: the corresponding instruct model serves as an ideal difficulty probe. By testing what the instruct model can solve with varying guidance, we automatically construct training data across three states: No-Hint (direct answer), Sparse-Hint (minimal prefix), and Full-Hint (complete reasoning). This converts the abstract challenge of difficulty labeling into a measurable consistency check between the instruct and reasoning models. With only 1K self-annotated samples, Hint Tuning achieves 24--66% token reduction (31.5% average) across mainstream reasoning models (Qwen3-Thinking, DeepSeek-R1-Distill) at multiple scales (4B--32B) while maintaining competitive accuracy on five benchmarks. Unlike methods requiring massive distillation datasets or expensive RL, we achieve superior efficiency through simple alignment with the instruct model's capabilities. Code and data are available at https://github.com/redai-infra/hint-tuning.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。