Core ML Tools

Core ML Tools (coremltools) is Apple’s official Python toolkit for converting, editing, validating and optimising machine learning models for Core ML, the format that runs on iPhone, iPad, Mac and other Apple devices. It converts models from frameworks such as PyTorch and TensorFlow into Core ML models ready to ship in apps. The BSD-licensed project has more than 5,400 stars on GitHub and is under active development in 2026.

Overview

Getting a trained model onto Apple hardware means translating it into Core ML and then squeezing it to fit mobile constraints. Coremltools provides the unified conversion API for that job, supporting neural networks from PyTorch, TensorFlow, Keras and classic formats such as scikit-learn and XGBoost through its converter ecosystem. Beyond conversion, it offers tools to inspect and edit model specifications, validate that a converted model produces the same outputs as the original, and compress models with techniques such as weight quantisation and palettisation so they run faster and take less space on device. It is the standard toolchain behind on-device ML in the Apple ecosystem — the path most teams take to put features like image classification, NLP and generative models into production iOS and macOS apps — and it is maintained by Apple alongside the Core ML framework itself, with documentation and examples on the official site.

Key Features

• Converts PyTorch and TensorFlow models to Core ML format
• Unified conversion API with support for many source frameworks
• Model editing and inspection tools for Core ML specifications
• Validation utilities that compare converted outputs against the original model
• Compression with quantisation and palettisation for smaller, faster models
• Official Apple tooling, actively maintained alongside Core ML

💳 Pricing

Free. Core ML Tools is open source under the BSD-3-Clause license and is provided by Apple at no charge.

Who Is It For

iOS and macOS developers and ML engineers who need to deploy trained models on Apple devices with on-device inference, privacy and low latency.