Welcome to NVIDIA FLARE

What is Federated Computing?

Federated computing coordinates computation across independently controlled sites, bringing approved computation to local data instead of collecting the underlying datasets in one place. Sites exchange application-defined results according to the collaboration’s and each site’s policies. Applications include model training and evaluation, analytics, statistics, site-local data processing, and custom multi-site workflows.

This approach can enable collaboration where privacy, regulation, data sovereignty, intellectual property, data ownership, or the cost of moving data makes central collection impractical.

Federated learning is one application of federated computing. It lets multiple sites collaboratively train or evaluate models using their local data and exchange model updates, metrics, or other approved results. Federated learning includes centralized workflows such as federated averaging, decentralized workflows such as swarm learning, and horizontal or vertical collaboration patterns.

What is NVIDIA FLARE?

NVIDIA FLARE™ (NVIDIA Federated Learning Application Runtime Environment) is a domain-agnostic, open-source, and extensible SDK for federated learning and other federated-computing applications. It allows researchers and data scientists to adapt existing ML and DL workflows to a federated paradigm and enables platform developers to build secure, privacy-preserving solutions for distributed multi-party collaboration.

NVIDIA FLARE builds its federated-learning capabilities on a general federated-computing core. The core provides workflow orchestration, task execution, communication, security, and lifecycle services shared by training, evaluation, analytics, statistics, and custom distributed applications.

Because applications integrate through extensible Python APIs, NVIDIA FLARE is not limited to a fixed model family or framework. Maintained integrations and examples include PyTorch, TensorFlow, JAX, scikit-learn, XGBoost, Hugging Face, NeMo, and Flower.

Keep Data at Its Source

NVIDIA FLARE moves approved computation to participating sites. In a properly designed and governed federation, raw datasets remain at their source and only outputs permitted by the collaboration and each site’s policies leave the site. Application owners and site operators define and enforce those policies for their data, regulatory requirements, and threat model.

NVIDIA FLARE supplies controls for identity, authorization, secure communication, auditing, privacy-preserving techniques, and confidential computing. These controls support a collaboration’s security and governance design; using NVIDIA FLARE alone does not determine which data or results an application is allowed to transmit.

From Development to Production

Applications use the same core programming model across three stages:

  • Simulator runs server and client logic on one system for rapid application development and validation.

  • Proof of Concept (POC) simulates a production deployment on one local host with separate processes and locally generated startup kits.

  • Production runs across provisioned sites with production identities, authorization, networking, policies, and operations.

NVIDIA FLARE provides high-level Client and Recipe APIs, the Collaboration API (Technical Preview), and lower-level controllers, executors, events, filters, and communication components. Researchers can implement new algorithms and workflows, data scientists can adapt existing applications, and platform teams can provision and operate multi-site systems.

Continue from Here

Run and understand a complete local federation in Quick Start. The documentation home then maps the maintained paths for Agent Skills, API selection, examples and tutorials, custom workflow research, deployment and security, and project contribution.

For broader product material, visit the NVIDIA FLARE website and developer site. See Industry Use Cases for applications and supporting material, and What’s New for current release information.