alibaba/MNN
Lightweight inference engine for on-device LLMs and Edge AI
- Chart
- Not on a chart right now
- Stars
- 16.6k
What it is
MNN is a lightweight deep learning framework developed by Alibaba for on-device model inference and training. It also provides MNN-LLM and MNN-Diffusion to run large language models and Stable Diffusion locally.
Who it is for
Developers deploying deep learning models, LLMs, or diffusion models on mobile phones, PCs, and embedded devices.
Why it is trending
It gained 96 stars today following its recent 3.6.2 release and the announcement of MNN-Pocket-0.3B, a 300M-parameter mobile-app-native language model.
Main features
- Model conversion and inference for TensorFlow, Caffe, ONNX, and Torchscripts
- Hardware acceleration across ARM/x64 CPUs, Metal, OpenCL, Vulkan, CUDA, and NPUs
- FP16, INT8, and INT4 quantization with small mobile library footprints
- Built-in tools including MNN-Converter, MNN-Compress, MNN-CV, and MNN-Train
Summarized on November 17, 2026 from the README and repository data. It can be wrong; check the source before relying on it. Report a wrong summary