Customize with Ease. Deploy Across Devices.
Fine-tuning for new tasks can come at the cost of a model's general capabilities. To address this, we introduce Self-Replay Fine-Tuning—a simple, low-cost way to customize models while preserving its general performance.
When it's time to deploy, MiniCPM-V 4.7 scales from mobile devices to PCs and edge systems. Day-0 hardware support covers Apple, Intel, AMD, Arm and Rockchip, with additional platforms supported through FlagOS, including Huawei, Hygon, Moore Threads and Kunlunxin.
Build with familiar tools throughout your workflow:
🔹 Inference: SGLang, vLLM, llama.cpp, mlx-vlm and Ollama
🔹 Fine-tuning: SWIFT and LLaMA-Factory
🔹 Quantization: GGUF, BNB, AWQ, GPTQ and FP8
The full stack is open source:
Give MiniCPM-V 4.7 a try, and let us know how it performs on your hardware!
🔗 GitHub: http://github.com/OpenBMB/MiniCPM-V
🤗 Hugging Face: http://huggingface.co/collections/openbmb/minicpm-v-47
🌐 ModelScope: http://modelscope.cn/collections/OpenBMB/MiniCPM-V-47
🧪 Web Demo: http://huggingface.co/spaces/openbmb/MiniCPM-V-4.7-Demo
📱 Apps: http://github.com/OpenBMB/MiniCPM-V-Apps
Self-Replay Fine-Tuning: http://openbmb-dev.modelbest.co/en/blog/2026-07-10-rjs-replay