[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"news-7352d2c8-4322-4343-8236-9773eb99b82f":3},{"id":4,"title":5,"summary":6,"original_url":7,"source_id":8,"tags":9,"published_at":20,"created_at":21,"modified_at":22,"is_published":23,"publish_type":24,"image_url":13,"view_count":25},"7352d2c8-4322-4343-8236-9773eb99b82f","本地LLM Agent能效困境：AgentStop实现15-20%能耗优化","在消费级设备上运行本地LLM Agent正面临严峻的能效挑战。来自Brave实验和伦敦帝国学院的研究团队发表的AgentStop论文指出，基于LLM的自主Agent在执行多步复杂任务（如编码、网页问答）时，会产生显著的GPU功耗上升、温度升高和电池消耗。\n\n与单次推理相比，Agentic工作流因迭代推理、工具调用和失败重试导致Token消耗激增，往往在任务未完成时就浪费大量计算资源。AgentStop通过低成本的执行信号（如Token级对数概率）预测任务成功概率，提前终止成功率低的轨迹。实验表明，该方法可在网页问答和编码基准测试中减少15-20%的能耗浪费，同时将任务性能下降控制在5%以内。\n\n这项研究为实现可持续、隐私保护的端侧LLM Agent提供了切实可行的优化路径。随着端侧AI落地加速，能效优化将成为消费级设备上部署Agent能力的关键瓶颈之一。","https:\u002F\u002Farxiv.org\u002Fabs\u002F2605.15206","7437aeb9-930c-4866-a2e9-48003c1a792b",[10,14,17],{"id":11,"name":12,"slug":12,"description":13,"color":13},"0ef8513a-0a26-42f0-b6f9-5b6dadded45c","efficiency",null,{"id":15,"name":16,"slug":16,"description":13,"color":13},"0a93ec8e-ea39-4693-81de-563ca8c173f7","inference",{"id":18,"name":19,"slug":19,"description":13,"color":13},"01598627-1ea6-4b27-a5d8-874971571a71","llm","2026-05-18T22:00:00Z","2026-05-18T22:07:26.288992Z","2026-05-18T22:07:26.289005Z",true,"agent",7]