让大模型学会何时该深思，只在必要时调用慢推理，提升效率与准确率。
When to Think Deeply: Inhibitory Deliberation for LLM Reasoning
- 根据快速回答的特征判断是否需要慢推理，动态控制思考时机。
- 仅在8.2%的情况下启用慢推理，准确率从47.90%提升至48.92%。
- 比纯随机和基于置信度的基线更精准识别需深思的问题，适合资源受限场景。
推理型大语言模型可通过慢速推理提升解题能力，但对每个输入都启用慢推理成本过高且不必要。本文提出IDPR框架，实现响应条件下的抑制性深思。该框架先生成简洁直觉答案，再通过抑制控制器判断该响应是否应释放或抑制以启动慢推理。控制器基于快速答案及其侧边证据（如置信度、逻辑差异、可解析性、生成成本）进行决策。模型通过配对的快慢输出训练，并在保留验证集上选择抑制阈值，以优先保证准确率。在5000例数学推理测试集上，IDPR仅在8.20%的样本中调用慢推理，准确率由47.90%提升至48.92%；相同慢调用预算下，随机路由准确率降至46.76%，最强置信度基线为48.22%。IDPR还展现出最高修正精度，表明响应条件抑制能更优识别需慢推理的直觉答案。
原文摘要 · Abstract (English)
Reasoning Large Language Models can improve problem-solving performance through deliberative inference, but invoking slow reasoning for every input is computationally expensive and often unnecessary. We propose IDPR, a framework for response-conditioned inhibitory deliberation. IDPR first generates a concise intuitive answer and then uses an inhibition controller to decide whether that specific response should be released or suppressed in favor of slow reasoning. Unlike input-only routers, the inhibition controller conditions on the fast answer and fast-side evidence, including confidence, logit margin, parseability, and generation cost. We train the controller from paired fast-slow outcomes and select the inhibition threshold on a held-out validation set under an accuracy-first slow-call budget. On a held-out 5,000-example mathematical reasoning test set, IDPR invokes slow reasoning on only 8.20% of examples and improves accuracy from 47.90% to 48.92%. Under the same slow-call budget, random routing decreases accuracy to 46.76%, while the strongest confidence-based baseline reaches 48.22%. IDPR also achieves the highest corrective precision, showing that response-conditioned inhibition better identifies fast answers that benefit from slow reasoning.
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。