Liquid AI 发布 LFM2.5-2.6B:高效本地智能体模型
原标题:Deploy local agents everywhere with LFM2.5-2.6B 4 August 4, 2026
AI 摘要
Liquid AI 发布了 LFM2.5-2.6B,一个 2.6B 参数的小型语言模型,专为本地和边缘设备上的智能体部署而设计。该模型通过四阶段后训练(包括 SFT、教师专业化、多领域策略蒸馏和智能体强化学习)在工具使用、指令遵循和多步智能体任务上表现出色,可与 4 倍大的模型竞争。LFM2.5-2.6B 支持高效的 CPU 和 GPU 推理,在 Apple M5 Max 上达到 220 tok/s,在 AMD Ryzen CPU 上达到 113 tok/s,并已与 llama.cpp、vLLM 等推理框架集成。该模型已在 Hugging Face 上发布,并提供浏览器演示。
正文节选
- Best-in-class agent: Competitive with models 4x larger on tool use, instruction following, and multi-step agentic tasks. - Agentic reinforcement learning: Trained inside the most popular agentic harnesses to improve compatibility. - Efficient inference: 220 tok/s on an Apple M5 Max and 113 tok/s on an AMD Ryzen CPU, in under 2.5 GB of memory. LFM2.5-2.6B is pre-trained on ~34T tokens, with a mid-training phase that extends the context window to 128K. Post-training then turns the base model int