Liquid AI 发布 LFM2.5-24B-A2B,已可在 Ollama 本地运行
ollama run lfm2.5:24b https://t.co/MpidEmAorV
Liquid AI 出了个 24B 总参数、每 token 只激活 2B 的本地推理模型,能在笔记本上跑智能体任务,还拿 Qwen3.6-35B-A3B 对标,Ollama 一条命令就能用。
Liquid AI 发布 LFM2.5 家族中最大的 LFM2.5-24B-A2B,总参数 24B,每 token 仅激活 2B。上下文扩展到 131K tokens,词表 128K,预训练数据 33T tokens。它是一个推理模型,面向笔记本和 PC 本地运行多步骤智能体任务,官方称在智能体与工具使用任务上可与参数量多约 50% 的 Qwen3.6-35B-A3B 竞争。基座与后训练权重已在 Hugging Face 开放,可通过 ollama run lfm2.5:24b 直接运行。
ollama run lfm2.5:24b https://t.co/MpidEmAorV
ollama run lfm2.5:24b ollama.com/library/lfm2.5… Liquid AI @liquidai Today we release 𝙻𝙵𝙼𝟸.𝟻-𝟸𝟺𝙱-𝙰𝟸𝙱, the largest 𝙻𝙵𝙼𝟸.𝟻 family member. It is an efficient local model for laptops and PCs to work through multi-step tasks inside popular agent harnesses. > 24B total parameters, 2B active per token > Context expanded to 131K tokens > A 128K-token vocabulary > 33T pretraining tokens > It is a reasoning model > Very easy to fine-tune and customize > Performs competitively with Qwen3.6-35B-A3B, a model with ~50% more total and active parameters, on agentic and tool-use tasks. > Open base and post-trained weights on @huggingface 🤗 1/ 🔗 View Quoted Tweet 💬 5 🔄 3 ❤️ 13 👀 2057 📊 6 ⚡