用控制标签引导小模型理性思考，提升推理效率与准确率
D-COT: Disciplined Chain-of-Thought Learning for Efficient Reasoning in Small Language Models
- 引入控制标签规范思维路径，防止小模型过度思考
- 仅用5000样本训练，准确率提升超9%，计算成本大幅降低
- 推理时无需标签仍保持高性能，适合资源受限场景
大型语言模型（LLMs）的思维链（CoT）蒸馏常导致小型语言模型（SLMs）出现‘过度思考’现象，引发性能下降和令牌消耗过多。本文提出受控思维链（D-CoT）框架，通过在训练中引入控制标签（如<TEMP_LOW>用于事实核查，<TEMP_HIGH>用于多角度探索）作为辅助结构，优化思维链轨迹，抑制推理偏差，同时实现令牌减少与性能提升。我们在Qwen3-8B上验证了该方法：仅需5000个训练样本，即可在GPQA-diamond上提升准确率9.9%，在MMLU-Pro（零样本）上提升9.1%，并显著降低计算开销。此外，模型能内化这种结构化思维，在推理时无需显式控制标签仍保持高表现。
原文摘要 · Abstract (English)
Chain-of-Thought (CoT) distillation from Large Language Models (LLMs) often induces "overthinking" in Small Language Models (SLMs), leading to performance degradation and excessive token consumption. In this study, we propose Disciplined Chain-of-Thought (D-CoT), a novel framework that enforces a structured reasoning process using control tags -- such as <TEMP_LOW> for fact-checking and <TEMP_HIGH> for multi-perspective exploration -- as auxiliary scaffolding during training. By optimizing the CoT trajectory, D-CoT suppresses reasoning drift and simultaneously achieves token reduction and performance improvement. We demonstrate the efficacy of our approach on Qwen3-8B: with only 5,000 training samples, D-CoT significantly boosts accuracy on GPQA-diamond by 9.9% and MMLU-Pro (0-shot) by 9.1%, while drastically reducing computational costs. Furthermore, we confirm that the model internalizes this disciplined thought structure, maintaining high performance even without explicit control tags during inference.
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。