不训练也不部署多个模型，用开关切换实现快慢思考的高效推理。
Gold-Switch: Training-Free Superposition of Slow- and Fast- Thinking LLMs
- 通过轻量级无训练机制动态开关大模型推理能力
- 在保持推理质量前提下降低计算开销，避免过度思考
- 适合资源受限场景，尤其适合需快速响应的推理应用
大型推理模型（LRMs）在结构化任务中表现优异，但常因过度推理导致性能下降和资源浪费。传统方法是同时部署普通大模型（LLM）和推理模型（LRM），通过预测输入是否需要推理来路由，但多模型部署成本高且不切实际。本文提出一种无需训练的叠加部署策略：在推理时选择性地从LRM中‘遗忘’部分信息，通过开关控制推理行为，降低计算量的同时保留必要推理能力。基于奇异值累积能量分析，识别最优低秩投影，精准调节推理强度，实现高效推理。
原文摘要 · Abstract (English)
Large Reasoning Models (LRMs) excel in structured tasks by emulating deliberate human reasoning but often suffer from overthinking, degrading performance and wasting resources. One possible baseline is to deploy both LLM and LRM, then route input by predicting whether it requires reasoning and may cause overthinking. However, deploying multiple models can be costly or impractical. We propose a superposed deployment strategy with a lightweight, training-free regulation to optimize inference by switching one model on and off. Instead of routing, we selectively unlearn from LRM at inference, scaling down computation while preserving reasoning. By analyzing the cumulative energy of singular values, we identify optimal low-rank projections to adjust reasoning just right.
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。