General-purpose ML libraries

The broadest part of the list: libraries that train and run models without being tied to one kind of data. It mixes two generations. Classical toolkits such as scikit-learn, XGBoost, LightGBM and CatBoost fit models to tables of features and remain the first choice for structured business data. Deep learning frameworks such as TensorFlow, PyTorch, JAX and Keras build neural networks of any shape and run them on GPUs.

Plenty of entries here are wrappers, bindings or research code from the 2010s that no longer receive commits. The activity column separates the projects that still ship releases from those that stopped, and the archived filter hides repositories their owners have frozen. A library listed under several languages, like XGBoost or MXNet, appears once with every language it serves.

For a new project in Python, scikit-learn plus one gradient boosting library covers most tabular problems, and PyTorch is the common default for neural networks. Use the task filter to narrow this page to deep learning, or open a language page to see what exists outside Python.