AtlasKV: 1 Billion Triples in 20GB, but RAG Lives
AtlasKV loads 1 billion knowledge graph triples into an 8B model using under 20GB of VRAM. It handles graph questions only, so RAG stays.
AtlasKV loads 1 billion knowledge graph triples into an 8B model using under 20GB of VRAM. It handles graph questions only, so RAG stays.