nncf

Why we included this project

NNCF is the compression framework from the OpenVINO project, aimed at the familiar production problem of models that are too large or too slow to deploy comfortably. It covers both post-training and training-time compression, including quantization, pruning, sparsity, and mixed-precision training, letting you shrink a network and cut inference latency without redesigning it. Because it works on PyTorch, ONNX, and OpenVINO models directly, it fits into training pipelines you already have instead of forcing you onto a separate stack. The maintained samples and model zoo document real compression results and give reproducible recipes, which is a practical way to start if your goal is faster inference on Intel hardware.