m2cgen

Why we included this project

When a trained model has to run somewhere a Python interpreter cannot, m2cgen turns a fitted scikit-learn, XGBoost, or LightGBM estimator into plain source code instead of a runtime dependency. The output is a self-contained function in C, Java, JavaScript, Go, Rust, PHP, or one of the other supported languages, with nothing extra to install or ship. That makes it a workable answer for edge devices, embedded systems, and browser-based tooling where pulling in a full inference stack is impractical. It covers the usual classical model families, from linear models and SVMs to trees, random forests, and boosting, and offers both a Python API and a CLI that can consume a pickled model. Teams that want readable, dependency-free code they can audit and compile themselves should weigh it against ONNX rather than assume a runtime-based exporter is the only path.