通过信心引导压缩推理链，让大模型思考更简洁高效。
ConCISE: Confidence-guided Compression in Step-by-step Efficient Reasoning
- 基于置信度识别冗余反思，设计信心注入与提前终止机制。
- 推理长度减少约50%，同时保持高任务准确率。
- 适合需要高效推理的场景，如实时问答与资源受限部署。
大型推理模型（LRMs）通过思维链（CoT）提示在复杂推理任务中表现优异，但常产生冗长输出，增加计算开销。现有基于微调的压缩方法或为事后剪枝，可能破坏推理连贯性；或依赖采样选择，无法彻底去除冗余内容。本文从置信度视角揭示两类冗余现象：信心不足（模型对正确中间步骤反复反思）与终止延迟（已确认答案后仍持续反思）。据此提出ConCISE框架，融合信心注入以提升推理置信度，以及早期终止策略，在置信度足够时停止推理。大量实验表明，基于ConCISE生成数据微调模型，在SimPO下可将推理长度减少约50%，同时维持高任务准确率，实现压缩与性能的更好平衡。
原文摘要 · Abstract (English)
Large Reasoning Models (LRMs) perform strongly in complex reasoning tasks via Chain-of-Thought (CoT) prompting, but often suffer from verbose outputs, increasing computational overhead. Existing fine-tuning-based compression methods either operate post-hoc pruning, risking disruption to reasoning coherence, or rely on sampling-based selection, which fails to remove redundant content thoroughly. To address these limitations, this work begins by framing two key patterns of redundant reflection in LRMs--Confidence Deficit, wherein the model reflects on correct intermediate steps, and Termination Delay, where reflection continues after a verified, confident answer--through a confidence-guided perspective. Based on this, we introduce ConCISE (Confidence-guided Compression In Step-by-step Efficient Reasoning), a framework designed to generate concise reasoning chains, integrating Confidence Injection to boost reasoning confidence, and Early Stopping to terminate reasoning when confidence is sufficient. Extensive experiments demonstrate that compared to baseline methods, fine-tuning LRMs on ConCISE-generated data yields a better balance between compression and task performance, reducing length by up to approximately 50% under SimPO, while maintaining high task 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。