FedML

Overview

FedML is an open-source Python library for large-scale distributed training, model serving, and federated learning. It includes FedML Launch, a cross-cloud scheduler that runs AI jobs across GPU clouds or on-premise clusters. The library forms the foundation of the commercial TensorOpera AI platform.

Best for

Best for
ML engineers and researchers who need a unified framework for distributed training, serving, or federated learning across multiple cloud and on-premises environments

Use cases

• Distributing training of large neural networks across multiple GPUs or nodes
• Deploying models with low-latency serving across cloud and edge infrastructure
• Running federated learning experiments with data distributed across silos

Notes

FedML is an open-source Python library for large-scale distributed training, model serving, and federated learning. It includes FedML Launch, a cross-cloud scheduler that runs AI jobs across GPU clouds or on-premise clusters. The library forms the foundation of the commercial TensorOpera AI platform.

4,045 stars on GitHub. Last updated 2025-10-28. Licensed Apache-2.0.

Use cases

• Distributing training of large neural networks across multiple GPUs or nodes
• Deploying models with low-latency serving across cloud and edge infrastructure
• Running federated learning experiments with data distributed across silos

Pros

• Covers a broad range of ML workloads (training, serving, federated learning) in one library
• Cross-cloud scheduler reduces vendor lock-in for infrastructure
• Active open-source community with over 4,000 GitHub stars

Cons

• Steep learning curve due to the complexity of distributed and federated setups
• Documentation and examples may lag behind the rapid pace of development
• Some advanced features require the commercial TensorOpera platform

Indexed from awesome-llmops and enriched against its public facts.