FlashRAG
A Python toolkit for reproducing and developing retrieval-augmented generation research, bundling 36 preprocessed benchmark datasets and 23 published RAG algorithms behind one customizable pipeline framework.
FlashRAG is a Python toolkit for the reproduction and development of retrieval-augmented generation (RAG) research. It includes 36 preprocessed benchmark RAG datasets and 23 reproduced state-of-the-art RAG algorithms, including seven reasoning-based methods that combine reasoning ability with retrieval, all built from a shared set of components such as retrievers, rerankers, generators and compressors that can be flexibly assembled into custom pipelines.
It simplifies RAG workflow preparation with scripts for corpus processing, retrieval index building and pre-retrieval of documents, and its execution is optimized with tools such as vLLM and FastChat for faster LLM inference and Faiss for vector index management. A companion FlashRAG-UI provides a visual interface for configuring and running the bundled RAG baselines and evaluation scripts without writing code.
It is licensed under MIT.