RAGU: A Multi-Step GraphRAG Engine
NSU · hf · 2026-07-20
RAGU is an open-source, multi-step GraphRAG engine that splits knowledge graph construction into extraction and integration phases: it first performs two-phase typed extraction, then uses DBSCAN for deduplication, LLMs for summarization, and Leiden community detection to enhance retrieval quality.
The authors note that the LLM responsible for extraction in the pipeline relies more on linguistic capabilities than broad world knowledge, prompting them to train a compact 7B model named Meno-Lite-0.1. Experiments show that for knowledge graph construction, it improves the harmonic mean by 12.5% over Qwen2.5-32B, and performs comparably on English GraphRAG tasks. On GraphRAG-Bench (Medical), RAGU's evidence recall reaches up to 0.84, outperforming HippoRAG2 in comprehensive QA. The project can be installed via pip install graphragu, supports single-GPU execution, and both code and models are openly available.
More from Research
- Nat Lambert shares a reading list on synthetic data and agentic SFT data — natolambert · 2026-07-22
- WeirdChat catalogs strange model behaviors from more than 100 million sampled responses — JacobSteinhardt · 2026-07-22
- New agentic benchmark shows AI managers escalate to coercion and fake success — Jasmine Brazilek · 2026-07-22
- Ai2’s Asta adds one-click handoff and self-checking deep paper search — allen_ai · 2026-07-22
- DepthART pushes monocular depth to tiny models at 1000 FPS on RTX A6000 — kwangmoo_yi · 2026-07-22
- Meta says SAM 3 and DINOv3 cut 3D volume labeling from a month to 15 minutes — AIatMeta · 2026-07-22