microsoft/nni

Supports multiple search strategies (Bayesian optimization, evolutionary algorithms, reinforcement learning) and NAS paradigms (one-shot methods like DARTS alongside multi-trial approaches), while providing framework-agnostic compression via pruning and quantization. Integrates with PyTorch, TensorFlow, scikit-learn, XGBoost, and other frameworks, with distributed training support across local machines, remote servers, and cloud platforms. Offers a unified Python API that decouples trial code from search algorithms through a managed experiment system, enabling reproducible AutoML workflows across different hardware configurations.

14,342 stars. No commits in the last 6 months.