OpenBMB · @OpenBMB · X·2026-09-08 21:47·51分钟前
AI 导读

面壁智能 OpenBMB 称赞 @aijoey 将 MiniCPM5-2B 作为本地 worker,在 DGX Spark 上构建证据地图应用 Footnote。该 2B 模型完成从 21 个来源提取 31 条主张并连接推理的全部工作,且支持增量更新。面壁智能称这正是其期望的端侧智能体负载。

OpenBMB@OpenBMB
33AI 编辑部评分,满分 100
2026-09-08 21:47· 51分钟前
AI 导读

面壁智能 OpenBMB 称赞 @aijoey 将 MiniCPM5-2B 作为本地 worker,在 DGX Spark 上构建证据地图应用 Footnote。该 2B 模型完成从 21 个来源提取 31 条主张并连接推理的全部工作,且支持增量更新。面壁智能称这正是其期望的端侧智能体负载。

Great real-world example. @aijoey is using MiniCPM5-2B as local workers to extract quoted findings and build a living evidence map (Footnote app) — all from a 2B model running on DGX Spark. 21 sources → 31 claims → connected reasoning, with the ability to incrementally update the map. This is exactly the kind of practical, on-device agentic workload we hoped MiniCPM5-2B could support. Impressive work!

JoeyBeen putting MiniCPM5-2B to work on my DGX Spark. Built an app called Footnote that turns a research question into a living map of evidence. The setup: • Astra ...

来源:OpenBMB· x.com