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: https://t.co/2pLjw81lUY
🤗 MiniCPM5-1B: https://t.co/6BAjgn7bkc
🤗 MiniCPM5-2B: https://t.co/FZOMTZhBjq
来源：@OpenBMB · x.com