community Reddit LocalLLaMA 2026-10-14T15:53 正文 thin 0 words

A 35B multimodal model running on a phone with ~3GB memory? MiniCPM-V 4.7 demo is pretty wild

英文标题:A 35B multimodal model running on a phone with ~3GB memory? MiniCPM-V 4.7 demo is pretty wild

Just saw this demo from OpenBMB for MiniCPM-V 4.7, and the on-device capability is pretty impressive. A 35B-A3B MoE vision-language model running locally on mobile devices is already interesting, but the use cases are what caught my attention: 📷 Reading real-world signs 📄 Understanding comple
打开原文相关 5 · 机会 0

中文

Just saw this demo from OpenBMB for MiniCPM-V 4.7, and the on-device capability is pretty impressive. A 35B-A3B MoE vision-language model running locally on mobile devices is already interesting, but the use cases are what caught my attention: 📷 Reading real-world signs 📄 Understanding comple

中文全文翻译正在补齐,可先查看下方英文原文。

English / 原文

Just saw this demo from OpenBMB for MiniCPM-V 4.7, and the on-device capability is pretty impressive. A 35B-A3B MoE vision-language model running locally on mobile devices is already interesting, but the use cases are what caught my attention: 📷 Reading real-world signs 📄 Understanding comple