KGCache system speeds up LLM knowledge graph reasoning

Researchers have developed KGCache, a novel in-memory caching system designed to improve the efficiency of Large Language Models (LLMs) when reasoning with knowledge graphs. KGCache stores frequently accessed one-hop knowledge graph neighborhoods, reducing redundant queries across different questions. The system is compatible with both iterative traversal and one-shot planning KGQA paradigms. Evaluations on WebQSP and CWQ datasets demonstrated significant entity reuse, with entity caching accelerating KG retrieval by up to 1.91x and semantic-context caching achieving up to 1.06x full-system speedup. AI

IMPACT This system could significantly speed up LLM applications that rely on knowledge graph grounding, making them more efficient and responsive.

RANK_REASON This is a research paper detailing a new system for improving LLM performance with knowledge graphs. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• KGCache
• knowledge graphs
• Knowledge Graph Question Answering
• Large language models
• WebQSP

• paper
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →