All Tools
M
Dev ToolsFreeOpen Source
MEGENGINE
Train and deploy models on one low-memory DL framework
Apache-2.0
ABOUT
Teams often train in one framework and rewrite for mobile or Arm inference. MegEngine keeps training, quantization, and deployment on one graph, cuts GPU memory with DTR, and runs the same model on x86, CUDA, Arm, and mobile without a separate export stack.
INSTALL
pip install megengineINTEGRATION GUIDE
1. Train vision models with automatic differentiation on CUDA GPUs
2. Quantize a trained graph and deploy it to Arm or mobile devices
3. Cut GPU memory during training with dynamic tensor rematerialization
4. Serve the same MegEngine model on Linux, Windows, iOS, or Android
TAGS
pythondeep-learningtraininginferencecudaquantizationmegvii