Rohan Paul· @rohanpaul_ai · X·· 2 小时前AI 评分61
AI 导读
Liquid AI 发布开源权重模型家族 LFM3,其中 1.7B 文本模型超越前代 2.6B 版本。该模型采用 24 层三混合架构,包含 13 层短卷积、5 层 256-token 滑动窗口注意力以及 6 层全局注意力。在 Liquid 运行的 AutomationBench 智能体基准测试中,1.7B 模型取得 41.29 分。
正文 · 原文
Liquid AI just launched LFM3, an open-weight model family whose 1.7B text model beats its 2.6B predecessor.
The gain comes from a tri-hybrid design: of 24 layers, 13 are short convolutions, 5 use 256-token sliding-window attention and 6 attend globally.
That efficiency carries into agent benchmarks run by Liquid, where the 1.7B model scores 41.29 on AutomationBench while Qwen3.5-9B manages 15.10.
来源:Rohan Paul · x.com