FedML is a unified, scalable machine-learning library for distributed training, model serving, and federated learning. It is designed to run training and deployment anywhere, from multi-cloud GPU clusters down to edge servers and smartphones.
Originally created for federated learning research, FedML has grown into a full-stack platform. Its open-source library covers large-scale distributed training, an inference engine for model serving, and on-device training. FEDML Launch, a cross-cloud scheduler, pairs AI jobs with economical GPU resources on any cloud or on-premise cluster and provisions them automatically. The library is the foundation of TensorOpera AI, a commercial generative-AI platform that adds managed MLOps, scheduling, and cloud services on top. FedML is released under the Apache-2.0 license, has over 4,000 stars on GitHub, and was actively updated through 2025, with a large research and industry community around federated and distributed learning.
The FedML library is free and open source under the Apache-2.0 license. TensorOpera AI, the commercial platform built on it, offers paid cloud services for managed training and deployment, making the overall offering freemium.
ML researchers and engineering teams working on federated learning, distributed training, or cost-efficient model deployment across clouds and edge devices.
The project began in the research community around federated learning and is widely cited in that literature. Its repository includes examples for cross-device and cross-silo federated training, model compression, and deployment scenarios, and the team maintains documentation plus community channels on Slack and Discord for researchers and enterprise users who are evaluating federated or distributed approaches to training.