#3232nd of 56 in ML frameworks and libraries
mnn
On-device inference and training engine for phones, embedded and PC
Documentation ↗
(opens in a new tab)Website ↗
(opens in a new tab)Repository on GitHub ↗
(opens in a new tab)
- Stars
- 16.2k
- License
- Apache-2.0
- Last commit
- Oct 2026
- Last release
- Jul 2026
- Language
- C++
Missed the latest GitHub check; these facts may be older.
Overview
MNN is a C++ deep learning engine from Alibaba that runs and trains models on phones, embedded devices and PCs. It converts TensorFlow, Caffe, ONNX and TorchScript models and runs them on ARM/x86 CPUs and on Metal, OpenCL, Vulkan and CUDA GPUs, with a Hexagon backend added in 3.6.1. MNN-LLM and MNN-Diffusion run LLMs and Stable Diffusion locally, and Android and iOS chat apps are included.
Who it is for: Mobile and embedded engineers deploying models on-device
Strengths
- Small footprint: about 800KB core .so on Android, about 12MB iOS static library with all options
- Backends for ARM/x86 CPU, Metal, OpenCL, Vulkan, CUDA, CoreML, NNAPI and Hexagon
- Converter covers TensorFlow, Caffe, ONNX and TorchScript, with FP16/Int8 quantization
- Includes MNN-LLM, MNN-Diffusion and Android/iOS demo apps
Weaknesses
- Community discussion is on DingTalk groups, predominantly in Chinese
- QNN is rated B or C in the README support table
- No Docker or HTTP server deployment described; it is a library and toolchain
- Native CPU and NNAPI paths are rated B (bugs or not optimized)
What it needs
- GPU optional
- Models: TensorFlow, Caffe, ONNX, TorchScript, Qwen
Also in ML frameworks and libraries
See all 56| Rank | Project | Score |
|---|---|---|
| 1 | transformersPython library defining open models for text, vision, audio and multimodal tasks | 95 out of 100 |
| 2 | pytorchPython tensor library with GPU support and tape-based autograd for deep learning | 94 out of 100 |
| 3 | ultralyticsPython library and CLI for training and running YOLO vision models | 92 out of 100 |
| #4 | supervisionPython toolkit for loading, annotating and counting computer vision detections | 89 out of 100 |
| #5 | pytorch-image-modelsPyTorch library of image models, pretrained weights and reference training scripts | 89 out of 100 |