HybridInfer：热感知强化学习三层LLM推理路由
原标题：HybridInfer: Thermal-Aware Reinforcement-Learning Tier Routing for On-Device, Edge, and Cloud LLM Inference
AI 摘要
研究者提出 HybridInfer，一个面向设备端、边缘和云端三层 LLM 推理的热感知强化学习路由器，使用手机热余量和查询复杂度作为状态，通过离线训练的 Q-learning 策略选择推理层级。在三星 Galaxy S25+ 上对 210 条提示的实测显示，该路由器在最低成本下质量显著优于两个手工调优启发式方法，而始终在设备端推理虽单查询质量有竞争力，但速度慢 3-6 倍且长查询会失败。作者称这是首次在真实硬件上利用设备端热余量来选择不同能力的 LLM 推理层级。
正文节选
HybridInfer: Thermal-Aware Reinforcement-Learning Tier Routing for On-Device, Edge, and Cloud LLM Inference Abstract On-device inference with small language models is attractive on modern smartphones because it keeps user data local, works offline, and incurs no per-query cost, so the on-device tier is the preferred place to answer a query when it is adequate. On-device inference is thermally constrained, however, and I find the constraint is sharper than a slowdown: on a flagship Snapdragon dev