AtlasKV: Augmenting LLMs with Billion-Scale Knowledge Graphs in 20GB VRAM

TL;DR

AtlasKV is a scalable method that integrates billion-scale knowledge graphs into large language models using minimal GPU memory, avoiding external retrieval and retraining.

Contribution

It introduces KG2KV and HiKVP techniques for efficient, parametric knowledge integration at scale with sub-linear complexity.

Findings

Supports billion-scale KGs with less than 20GB VRAM

Maintains strong knowledge grounding and generalization

Eliminates need for external retrievers or retraining

Abstract

Retrieval-augmented generation (RAG) has shown some success in augmenting large language models (LLMs) with external knowledge. However, as a non-parametric knowledge integration paradigm for LLMs, RAG methods heavily rely on external retrieval modules and the retrieved textual context prior. Especially for very large scale knowledge augmentation, they would introduce substantial inference latency due to expensive searches and much longer relevant context. In this paper, we propose a parametric knowledge integration method, called \textbf{AtlasKV}, a scalable, effective, and general way to augment LLMs with billion-scale knowledge graphs (KGs) (e.g. 1B triples) using very little GPU memory cost (e.g. less than 20GB VRAM). In AtlasKV, we introduce KG2KV and HiKVP to integrate KG triples into LLMs at scale with sub-linear time and memory complexity. It maintains strong knowledge grounding and generalization performance using the LLMs' inherent attention mechanism, and requires no external retrievers, long context priors, or retraining when adapting to new knowledge.