FedScale is a scalable and extensible open-source federated learning (FL) engine. It provides high-level APIs to implement FL algorithms, deploy and evaluate them at scale across diverse hardware and software backends. FedScale also includes the largest FL benchmark that contains FL tasks ranging from image classification and object detection to language modeling and speech recognition. Moreover, it includes datasets to faithfully emulate FL runtime environments where FL solutions will realistically be deployed.
We are actively developing FedScale, and welcome contributions from the community. Join our slack to keep up to date.
Realistic Datasets
FedScale incorporates the most comprehensive FL datasets to date for evaluating different aspects of real FL deployments, including 20+ realistic datasets across ML tasks, deployment scales, and client system traces.
Scalable Platform
FedScale enables FL benchmarking at scale across multiple backends, including mobile backends for real on-device execution, local backend on your laptop, and cluster backends on GPUs/CPUs for time- and cost-efficient FL.
Extensible Design
FedScale provides high-level APIs to implement new FL algorithms and systems techniques with ease. It enables fair comparison against state-of-the-art solutions across all the benchmarking datasets without extra effort.