lightgbm-org/lightgbm

LightGBM is a machine learning framework written in C++ that implements gradient boosting, a technique where many simple decision trees are combined in sequence, each one learning to correct the mistakes of the previous ones, to produce accurate predictions for tasks like classifying data or ranking items. The framework is designed to be faster and use less memory than comparable tools, while maintaining or improving accuracy. It supports training in parallel across multiple CPU cores or machines, and can also use graphics processing units to accelerate computation. It handles large datasets that would be impractical for some other approaches. LightGBM has official interfaces for Python, R, and C, and the community has created additional bindings for other languages. It has been used in many machine learning competition winning solutions. The readme notes the project moved from the Microsoft GitHub organization to its own organization in March 2026 but remains managed by the same team. The project includes extensive documentation covering installation, available parameters, distributed training, and integration with automated hyperparameter tuning tools. It is licensed under the MIT license.