flower

Why we included this project

If you need to train a machine learning model on data that lives across phones, hospitals, or edge devices and cannot be moved to a central server, Flower is a solid open-source choice. It coordinates federated learning runs so models train collaboratively across many clients while each device keeps its raw data. The framework is deliberately agnostic about the underlying ML stack, so your existing PyTorch, TensorFlow, or scikit-learn training loops work as-is. Flower grew out of a research project at the University of Oxford, which is why so many components are built to be customized and extended for experiments. For teams just starting with federated setups, the quickstart examples and an active community lower the barrier despite the underlying distributed-systems complexity.