返回
原文
返回
AI 评分 33/100在 X 查看原推
导出 Markdown

MiniCPM5-1B 多语言评测与端侧部署

OpenBMB · @OpenBMB · X·2026-09-11 22:00·46分钟前
AI 导读

开发者 luispoveda93 对 MiniCPM5-1B 在西班牙语、加泰罗尼亚语、巴斯克语和加利西亚语上做了零样本评测，覆盖推理、NLI、复述识别、QA 和翻译任务。该模型还被转换为约 1.1GB 的 INT8 LiteRT-LM 模型，可在 Android、iOS、桌面和 IoT 设备上离线推理。面壁智能表示 MiniCPM5-2B 已发布，希望更多开发者探索其多语言与端侧应用。

OpenBMB@OpenBMB
33AI 编辑部评分，满分 100

MiniCPM5-1B 多语言评测与端侧部署

2026-09-11 22:00· 46分钟前
AI 导读

开发者 luispoveda93 对 MiniCPM5-1B 在西班牙语、加泰罗尼亚语、巴斯克语和加利西亚语上做了零样本评测，覆盖推理、NLI、复述识别、QA 和翻译任务。该模型还被转换为约 1.1GB 的 INT8 LiteRT-LM 模型，可在 Android、iOS、桌面和 IoT 设备上离线推理。面壁智能表示 MiniCPM5-2B 已发布，希望更多开发者探索其多语言与端侧应用。

🌍 Multilingual evaluation meets on-device deployment with MiniCPM5-1B!

Developer @luispoveda93 explored MiniCPM5-1B across Spanish, Catalan, Basque, and Galician, evaluating its zero-shot capabilities on reasoning, NLI, paraphrase identification, QA, and translation tasks.

The model was also converted into an ~1.1GB INT8 LiteRT-LM model, bringing MiniCPM5-1B to offline inference across Android, iOS, desktop, and IoT devices.

✨ Highlights: 🧠 Zero-shot evaluation across 4 languages ⚡ INT8 quantization for a compact ~1.1GB model 📱 LiteRT-LM deployment for on-device inference 🌍 Exploring multilingual adaptation beyond high-resource languages

Now that MiniCPM5-2B is out, we’d love to see more developers explore what it can do across different languages, devices, and real-world workflows.

Built something with MiniCPM5-2B? Share your case with us! 🚀

🔗 Project: http://huggingface.co/luispoveda93/MiniCPM5-1B 🤗 MiniCPM5-1B: http://huggingface.co/openbmb/MiniCPM5-1B 🤗 MiniCPM5-2B: http://huggingface.co/openbmb/MiniCPM5-2B

来源：OpenBMB· x.com

导出 Markdown