FedRAG

FedRAG is an open-source framework for fine-tuning Retrieval-Augmented Generation (RAG) systems across both centralized and federated architectures.

Comprehensive support for state-of-the-art RAG fine-tuning methods that can be federated with ease.

Seamlessly integrates with popular frameworks including HuggingFace, LlamaIndex, and LangChain — use the tools you already know.

Clean, intuitive abstractions that simplify RAG fine-tuning while maintaining full flexibility and control.

# pypi
pip install fed-rag

# conda-forge
conda install -c conda-forge fed-rag

git clone https://github.com/VectorInstitute/fed-rag.git
cd fed-rag

# install using pip
pip install -e .

# or, install using uv, our package manager tool of choice
uv sync --all-extras --group dev --group docs

For more detailed documentation, visit our official documentation site.

• Getting Started
• Quick Starts
• Examples
• API Reference
• Glossary

Check out our examples directory for more detailed usage examples:

• Basic RAG fine-tuning with federated learning
• Implementing RA-DIT with FedRAG
• Custom federated aggregation strategies
• Integration with popular LLM frameworks

We welcome contributions! Please see our Contributing Guide for more details.

If you use FedRAG in your research, please cite our library:

@software{Fajardo_fed-rag_2025,
author = {Fajardo, Andrei and Emerson, David},
doi = {10.5281/zenodo.15092361},
license = {Apache-2.0},
month = mar,
title = {{fed-rag}},
url = {https://github.com/VectorInstitute/fed-rag},
version = {0.0.27},
year = {2025}
}

FedRAG is released under the Apache License 2.0.

FedRAG is developed and maintained by the Vector Institute.