MegEngine is a fast, scalable, and easy-to-use deep learning framework with three key features:

• Integration of training and reasoning: the same core for training and reasoning, model structure, quantization, pre- and post-processing, dynamic shape and even derivation can be put into the model for reasoning, and the training and reasoning can be easily aligned to the accuracy
• Ultra-low hardware threshold: relying on algorithms to optimize the occupancy of various key resources, DTR reduces the occupancy of training video memory by 3 times with one click, and the pushdown memory allocation algorithm reduces the occupancy of inference memory to a very low level
• High-efficiency inference on all platforms: You can experience high-performance and precision-aligned inference on x86/Arm/CUDA/RoCM platforms, and more advanced usage can optimize performance and save memory

start using

train

reasoning

Installation Notes

Notice: MegEngine now supports installing Python packages on platforms such as Linux-64bit/Windows-64bit/macos-10.14/Android 7+ and above (MacOS/Android only supports cpu), and supports Python3.5 to Python3.8. For Windows 10 users, you can experience it by installing Windows Subsystem for Linux (WSL), and we also support Windows natively. MegEngine also supports inference on many other platforms.

Install via package manager

The command to install via pip is as follows:

python3 -m pip install --upgrade pip
python3 -m pip install megengine -f https://megengine.org.cn/whl/mge.html

Compile and install from source code

How to get involved

• MegEngine manages the open source community according to the Contributor Covenant. Please read the Code of Conduct for more information.
• Each MegEngine contributor needs to sign a Contributor License Agreement (CLA) to clarify the intellectual property license related to the contributed content. Please refer to the agreement for more details.
• We welcome you to help MegEngine become better by:

We believe that we can build an open and friendly open source community environment to benefit mankind with artificial intelligence.

contact us

resource

Open Source License

MegEngine uses the Apache License, Version 2.0.

Quote MegEngine

If MegEngine is used in your research, it is recommended that you cite the copy in BibTeX format as follows.

@Misc{MegEngine,
  institution = {megvii},
  title =  {MegEngine:A fast, scalable and easy-to-use deep learning framework},
  howpublished = {\url{https://github.com/MegEngine/MegEngine}},
  year = {2020}
}

Copyright (c) 2014-2021 Megvii Inc. All rights reserved.