MNN-LLMs

LLMs converted for MNN: on-device inference on CPU and GPU.

Note: these run CPU/GPU only (no NPU) and are not as fast as the GGUF Q4_0 builds in KleidiAi_support — pick those when your device supports it.

Use anywhere

Standard MNN format — runs with any MNN-based LLM runtime, on Android and beyond. No lock-in.

Tuned for FancyAI

These models are built for FancyAI (fancyai-os.com), the free and open-source Android app built around on-device inference.