NVIDIA FLARE

NVIDIA FLARE (Federated Learning Application Runtime Environment) is an open-source SDK for federated learning. It helps ML practitioners adapt existing training workflows (PyTorch, TensorFlow, XGBoost, scikit-learn, NeMo) to a federated setting with minimal code changes, and enables platform teams to deploy secure, privacy-preserving multi-party collaboration.

Choose Your Path

New to FLARE (ML Practitioners)

Start here if you want to federate an existing training script.

• Welcome – What FLARE is and what it supports

• Installation – Install FLARE and set up your environment

• Quick Start – Run a Hello World example and convert your ML code

• Client API – Recommended high-level API for federated training

• Job Recipe API – Pre-built recipes for common FL workflows

• Migration Guide – Upgrade between FLARE versions

• Examples & Tutorials – End-to-end examples and tutorials

Deployment & Security (Production Teams)

Start here if you are deploying FLARE in an organization or consortium.

• Deployment Overview – Provisioning, Docker/Kubernetes, cloud deployment, dashboard

• Admin Commands – Operating and managing a running FL system

• System Configuration – Configuration files and settings

• Preflight Check – Pre-launch validation

• Security Overview – Authentication, authorization, privacy, auditing

• Confidential Computing – Hardware-backed TEEs for end-to-end IP protection

Developers (Advanced / Contributors)

Start here if you want to extend FLARE or build custom workflows.

• Developer Guide – Architecture deep-dives, controllers, filters, and extension points

• API Reference – Full Python API documentation

• Contributing – How to contribute to NVIDIA FLARE

Explore by Use Case

• Industry Use Cases – Real-world deployments across healthcare, finance, government, and more

• Large Models & LLM – Federated fine-tuning, memory management, and optimization for large models

• Edge & Mobile – Mobile training (iOS/Android) and hierarchical FL for large-scale deployments