Install lightgbm with Homebrew, MacPorts, Nix, winget

Project links

• Homepage
• Repository
• Upstream docs
• Package manager page

Commands and aliases

• lightgbm

Project history

The GitHub repository was created on 2016-08-05. The README describes LightGBM as a gradient boosting framework designed for faster training, lower memory usage, better accuracy, parallel and distributed learning, GPU learning, and large-scale data.

The official papers connect the implementation to Microsoft Research work on communication-efficient parallel decision trees in 2016 and the 2017 NIPS paper "LightGBM: A Highly Efficient Gradient Boosting Decision Tree." The paper introduced Gradient-based One-Side Sampling and Exclusive Feature Bundling as key techniques for speeding up GBDT training.

The repository moved from Microsoft/LightGBM to lightgbm-org/LightGBM in March 2026. The maintainers documented the move in issue 7187 and stated that the same maintainers, including the creator of LightGBM, continued managing the official source repository.

Adoption history

The README says LightGBM has been widely used in winning machine-learning competition solutions. Its package footprint spans command-line binaries, Python, R, conda, CRAN, NuGet, Winget, Homebrew, and downstream integrations such as Spark-oriented wrappers and inference converters.

LightGBM's adoption followed a practical need in tabular-data workflows: teams wanted XGBoost-class accuracy with faster training iterations and better memory behavior on large datasets.

How it is used

Users train models from the CLI or language bindings for regression, classification, ranking, and large-scale distributed tasks. Common package-manager use cases include installing the CLI for experiments, installing Python or R bindings for notebooks and pipelines, and installing the library as a dependency of higher-level ML systems.

Why package nerds care

LightGBM is package-nerd significant because it is a research system that became packaging infrastructure: native C++, Python wheels, R packages, GPU builds, distributed modes, and many downstream wrappers all have to agree on the same fast tree learner.

It is also a canonical example of ML packaging complexity, where one upstream project must serve CLI users, language-binding users, GPU users, and distro maintainers without losing performance-sensitive native code paths.

Timeline

• 2016-08-05: GitHub repository created.
• 2016: NIPS paper on communication-efficient parallel decision trees published by LightGBM authors.
• 2017-02-27: Official experiment documentation records the first version of comparison and parallel experiments.
• 2017: NIPS paper "LightGBM: A Highly Efficient Gradient Boosting Decision Tree" published.
• 2018: GPU acceleration paper cited by the README.
• 2020-03-08: Official experiment documentation updated against a then-new master branch.
• 2022: Quantized training paper cited by the README.
• 2026-03: Repository moved from Microsoft/LightGBM to lightgbm-org/LightGBM.

Related projects

• Related projects include XGBoost, scikit-learn integrations, SynapseML, FLAML, Optuna, Treelite, SHAP, ML.NET, ONNX conversion tools, and many language bindings listed by the LightGBM README.

Sources

• https://api.github.com/repos/lightgbm-org/LightGBM
• https://formulae.brew.sh/formula/lightgbm
• https://github.com/lightgbm-org/LightGBM
• https://github.com/lightgbm-org/LightGBM/issues/7187
• https://lightgbm.readthedocs.io/en/stable/Features.html
• https://papers.nips.cc/paper/6907-lightgbm-a-highly-efficient-gradient-boosting-decision-tree
• https://raw.githubusercontent.com/lightgbm-org/LightGBM/master/README.md
• https://raw.githubusercontent.com/lightgbm-org/LightGBM/master/docs/Experiments.rst