ProphetKV: User-Query-Driven Selective Recomputation for Efficient KV Cache Reuse in Retrieval-Augmented Generation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wang, Shihao, Chen, Jiahao, Pan, Yanqi, Huang, Hao, Hao, Yichen, Zou, Xiangyu, Xia, Wen, Zhang, Wentao, Qiu, Chongyang, Wang, Pengfei
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911422157422592
author Wang, Shihao
Chen, Jiahao
Pan, Yanqi
Huang, Hao
Hao, Yichen
Zou, Xiangyu
Xia, Wen
Zhang, Wentao
Qiu, Chongyang
Wang, Pengfei
author_facet Wang, Shihao
Chen, Jiahao
Pan, Yanqi
Huang, Hao
Hao, Yichen
Zou, Xiangyu
Xia, Wen
Zhang, Wentao
Qiu, Chongyang
Wang, Pengfei
contents The prefill stage of long-context Retrieval-Augmented Generation (RAG) is severely bottlenecked by computational overhead. To mitigate this, recent methods assemble pre-calculated KV caches of retrieved RAG documents (by a user query) and reprocess selected tokens to recover cross-attention between these pre-calculated KV caches. However, we identify a fundamental "crowding-out effect" in current token selection criteria: globally salient but user-query-irrelevant tokens saturate the limited recomputation budget, displacing the tokens truly essential for answering the user query and degrading inference accuracy. We propose ProphetKV, a user-query-driven KV Cache reuse method for RAG scenarios. ProphetKV dynamically prioritizes tokens based on their semantic relevance to the user query and employs a dual-stage recomputation pipeline to fuse layer-wise attention metrics into a high-utility set. By ensuring the recomputation budget is dedicated to bridging the informational gap between retrieved context and the user query, ProphetKV achieves high-fidelity attention recovery with minimal overhead. Our extensive evaluation results show that ProphetKV retains 96%-101% of full-prefill accuracy with only a 20% recomputation ratio, while achieving accuracy improvements of 8.8%-24.9% on RULER and 18.6%-50.9% on LongBench over the state-of-the-art approaches (e.g., CacheBlend, EPIC, and KVShare).
format Preprint
id arxiv_https___arxiv_org_abs_2602_02579
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle ProphetKV: User-Query-Driven Selective Recomputation for Efficient KV Cache Reuse in Retrieval-Augmented Generation
Wang, Shihao
Chen, Jiahao
Pan, Yanqi
Huang, Hao
Hao, Yichen
Zou, Xiangyu
Xia, Wen
Zhang, Wentao
Qiu, Chongyang
Wang, Pengfei
Operating Systems
Artificial Intelligence
The prefill stage of long-context Retrieval-Augmented Generation (RAG) is severely bottlenecked by computational overhead. To mitigate this, recent methods assemble pre-calculated KV caches of retrieved RAG documents (by a user query) and reprocess selected tokens to recover cross-attention between these pre-calculated KV caches. However, we identify a fundamental "crowding-out effect" in current token selection criteria: globally salient but user-query-irrelevant tokens saturate the limited recomputation budget, displacing the tokens truly essential for answering the user query and degrading inference accuracy. We propose ProphetKV, a user-query-driven KV Cache reuse method for RAG scenarios. ProphetKV dynamically prioritizes tokens based on their semantic relevance to the user query and employs a dual-stage recomputation pipeline to fuse layer-wise attention metrics into a high-utility set. By ensuring the recomputation budget is dedicated to bridging the informational gap between retrieved context and the user query, ProphetKV achieves high-fidelity attention recovery with minimal overhead. Our extensive evaluation results show that ProphetKV retains 96%-101% of full-prefill accuracy with only a 20% recomputation ratio, while achieving accuracy improvements of 8.8%-24.9% on RULER and 18.6%-50.9% on LongBench over the state-of-the-art approaches (e.g., CacheBlend, EPIC, and KVShare).
title ProphetKV: User-Query-Driven Selective Recomputation for Efficient KV Cache Reuse in Retrieval-Augmented Generation
topic Operating Systems
Artificial Intelligence
url https://arxiv.org/abs/2602.02579