From Prefix Cache to Fusion RAG Cache: Accelerating LLM Inference in Retrieval-Augmented Generation

TL;DR

FusionRAG is a novel inference framework that enhances retrieval-augmented generation by embedding cross-chunk context and selectively recomputing key-value caches, significantly improving quality and efficiency.

Contribution

It introduces FusionRAG, which optimizes both preprocessing and reprocessing in RAG, effectively balancing generation quality and computational efficiency.

Findings

Achieves up to 70% higher normalized F1 scores with less than 15% token recomputation.

Reduces Time to First Token (TTFT) by 2.66x to 9.39x compared to full attention.

Significantly improves generation quality over previous state-of-the-art methods.

Abstract

Retrieval-Augmented Generation enhances Large Language Models by integrating external knowledge, which reduces hallucinations but increases prompt length. This increase leads to higher computational costs and longer Time to First Token (TTFT). To mitigate this issue, existing solutions aim to reuse the preprocessed KV cache of each retrieved chunk to accelerate RAG. However, the lack of cross-chunk contextual information leads to a significant drop in generation quality, leaving the potential benefits of KV cache reuse largely unfulfilled. The challenge lies in how to reuse the precomputed KV cache of chunks while preserving generation quality. We propose FusionRAG, a novel inference framework that optimizes both the preprocessing and reprocessing stages of RAG. In the offline preprocessing stage, we embed information from other related text chunks into each chunk, while in the online reprocessing stage, we recompute the KV cache for tokens that the model focuses on. As a result, we achieve a better trade-off between generation quality and efficiency. According to our experiments, FusionRAG significantly improves generation quality at the same recomputation ratio compared to previous state-of-the-art solutions. By recomputing fewer than 15% of the tokens, FusionRAG achieves up to 70% higher normalized F1 scores than baselines and reduces TTFT by 2.66x-9.39x compared to Full Attention.