FedLearner is ByteDance’s open-source framework for multi-party collaborative machine learning, with 899 stars on GitHub.
FedLearner is a federated learning framework that enables joint modelling on data distributed between institutions. In many industries — finance, healthcare, advertising — the most valuable training data is split across organisations that cannot simply pool it, for legal, competitive or privacy reasons. FedLearner lets those parties train shared models together while the underlying data stays where it lives, exchanging only the information the federated protocol requires. Developed at ByteDance, where it was built for real production collaboration scenarios, the framework covers the machinery needed to coordinate training across organisational boundaries, manage the participating roles and complete joint model building end to end. Typical deployments pair a coordinating party with data-holding participants, and training proceeds through rounds of model updates rather than data transfers, so each side keeps custody of its own records from start to finish. The framework targets the deep learning workflows used in recommendation and risk modelling — exactly the domains where the training data is most valuable and where cross-institution collaboration pays off most. It is not a beginner tool: a real deployment assumes serious infrastructure on every side and a partner organisation to federate with.
FedLearner is free and open source under the Apache-2.0 license. Organisations deploy it on their own infrastructure; there are no licence fees or hosted tiers.
FedLearner is for ML teams at banks, hospitals, platforms and enterprises that want to collaborate on models with partners without exchanging raw data. It also serves researchers studying federated learning who want a production-shaped framework rather than a toy simulation.