flower: a framework-agnostic system for building and scaling federated AI

What it solves

Flower simplifies the process of building federated AI systems, allowing developers to train machine learning models across multiple decentralized devices or servers without needing to move the raw data to a central location.

How it works

Flower provides a framework that acts as an orchestration layer for federated learning. It is designed to be framework-agnostic, meaning it can integrate with any ML library—such as PyTorch, TensorFlow, JAX, Hugging Face, or scikit-learn—and can be extended to support custom strategies and communication patterns for distributing model training.

Who it’s for

It is designed for AI researchers and engineers who need to build scalable, customizable federated learning systems, as well as developers looking to implement privacy-preserving AI on edge devices (like Android, iOS, or Raspberry Pi).

Highlights

• Framework Agnostic: Works with almost any ML library, including PyTorch, TensorFlow, Hugging Face, and even NumPy.
• Highly Customizable: Allows users to override components to create new state-of-the-art federated systems.
• Broad Device Support: Includes quickstarts for mobile platforms (Android/TFLite, iOS/CoreML) and embedded devices (Raspberry Pi, Nvidia Jetson).
• Research-Ready: Includes "Flower Baselines," a collection of community-contributed reproductions of popular federated learning publications.