FlashRAG and MiniRAG
FlashRAG is a comprehensive RAG research framework optimized for benchmarking and experimentation, while MiniRAG is a lightweight, simplified RAG implementation designed to work with smaller language models—making them complementary tools for different scales of RAG deployment (research infrastructure vs. resource-constrained production).
About FlashRAG
RUC-NLPIR/FlashRAG
⚡FlashRAG: A Python Toolkit for Efficient RAG Research (WWW2025 Resource)
Provides modular components (retrievers, rerankers, generators, compressors) for building custom RAG pipelines, plus 36 pre-processed benchmark datasets and implementations of 23 SOTA algorithms including 7 reasoning-based methods that integrate language model reasoning with retrieval. Integrates acceleration tools like vLLM and Faiss, supports multimodal RAG with MLLMs (Llava, Qwen) and CLIP-based retrievers, and includes a visual UI for configuration and evaluation without code.
About MiniRAG
HKUDS/MiniRAG
"MiniRAG: Making RAG Simpler with Small and Open-Sourced Language Models"
Constructs a semantic-aware heterogeneous graph combining text chunks and named entities to reduce dependency on complex semantic understanding, then retrieves knowledge via lightweight topology-aware graph traversal rather than dense embeddings. Supports 10+ graph databases (Neo4j, PostgreSQL, TiDB) and achieves comparable performance to LLM-based RAG with 75% less storage while running small models like Phi-3.5-mini and Qwen2.5-3B on-device.
Related comparisons
Scores updated daily from GitHub, PyPI, and npm data. How scores work