让大模型只在必要时才深度思考，大幅减少推理耗时。
AutoL2S: Auto Long-Short Reasoning for Efficient Large Language Models
- 用轻量开关标记动态决定是否长思考
- 推理长度最多缩短71%且准确率几乎不变
- 适合追求高效推理的部署场景
具备推理能力的大语言模型在复杂任务上表现优异，但经过蒸馏后常出现过度思考问题，对简单输入也生成冗长的思维链（CoT），导致推理成本过高。直接缩短推理长度会降低准确性，因简洁推理可能不足且缺乏显式监督。我们提出Auto Long-Short Reasoning（AutoL2S），一种蒸馏框架，使非推理型大模型仅在必要时进行深入思考。AutoL2S首先学习一个轻量级开关标记，结合经验证的长短思维链实现实例级选择。随后利用该开关标记引发的长短思维链轨迹，在类似GRPO的损失函数下优化推理效率，同时保持准确率。实验表明，AutoL2S可将推理长度最多减少71%，准确率损失极小，显著改善了令牌长度与推理时间的权衡关系。
原文摘要 · Abstract (English)
Reasoning-capable large language models (LLMs) achieve strong performance on complex tasks but often exhibit overthinking after distillation, generating unnecessarily long chain-of-thought (CoT) reasoning even for simple inputs and incurring high inference cost. However, naively shortening reasoning length can degrade reasoning accuracy, as concise reasoning may be insufficient for certain inputs and lacks explicit supervision. We propose Auto Long-Short Reasoning (AutoL2S), a distillation framework that empowers non-reasoning LLMs to think thoroughly but only when necessary. AutoL2S first learns a lightweight switching token with verified long-short CoTs to enable instance-wise long-short reasoning selection. Then it leverages long-short reasoning rollouts induced by a switching token in a GRPO-style loss to improve reasoning efficiency while maintaining accuracy. Experiments demonstrate that AutoL2S effectively reduces reasoning length up to 71% with minimal accuracy loss, yielding markedly better trade-off in token length and inference time while preserving accuracy.
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。