coremltools

Why we included this project

Anyone shipping ML features in an iOS, macOS, watchOS, or tvOS app will eventually reach the point where a trained model has to become something Core ML can run. coremltools is Apple's own Python tooling for that step: it converts models trained in PyTorch, TensorFlow, scikit-learn, XGBoost, or LibSVM into the .mlpackage format the on-device runtime consumes. On top of conversion it lets you edit and optimize models directly, shrinking them through quantization to fit mobile constraints, and validate predictions on a Mac before the app ships. Because it is the official pipeline, it is also the most direct route to Neural Engine acceleration and the low power draw of running inference on the device rather than over a network. Teams maintaining a fleet of models across Apple hardware will find the editing primitives useful for retargeting or pruning conversions without retraining.