AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

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

Shihao Wang, Jiahao Chen, Yanqi Pan, Hao Huang, Yichen Hao, Xiangyu Zou, Wen Xia, Chongyang Qiu, Wentao Zhang, PengFei Wang

ICML 2026 · Architectures and efficiency › Attention and transformers · approach: KV cache eviction and compression · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
select recompute tokens by user-query attention (ProphetKV) vs CacheBlendKV cache reuse · RULER 8K, Llama3.1-8B-Instruct, 20% recomputeaverage accuracy: 77.27 -> 84.71Improved
select recompute tokens by user-query attention vs CacheBlendKV cache reuse · LongBench, Llama3.1-8B-Instruct, 20% recomputeaverage accuracy: 39.82 -> 50.80Improved
select recompute tokens by user-query attention vs CacheBlendKV cache reuse · LongBench, Qwen2.5-14B-Instruct, 20% recomputeaverage accuracy: 35.40 -> 53.43Improved
select recompute tokens by user-query attention vs CacheBlendKV cache reuse · RULER 8K, Qwen3-14B-Thinking, 20% recomputeaverage accuracy: 71.99 -> 89.89Improved
add 20% partial recomputation vs naive reuseProphetKV · LongBench, Qwen2.5-14B-Instructaverage accuracy: 15.86 -> 53.43Improved
fuse query attention across all layers (uniform average) instead of single layer (ablation)ProphetKV · layer-wise optimal token overlapoverlap ratioImproved
use average layer fusion vs fixed-depth selection (ablation)ProphetKV · fixed-depth ablationaccuracyNo clear effect
use query attention proxy vs CacheBlend under oracle selection (ablation)ProphetKV · RULER-MVrecompute ratio to converge: 0.06 -> 0.04Improved
increase recompute ratio (ablation)ProphetKV · RULER-MV, LongBench MusiqueaccuracyImproved
decrease chunk size (ablation)all KV reuse methods · RULER-MVaccuracyHurt
ProphetKV vs full recomputation latencyProphetKV · TTFT 16K, Llama3.1-8B-InstructTTFT (s): 5.23 -> 1.13Improved
ProphetKV vs EPIC latencyProphetKV · TTFT 16K, Llama3.1-8B-InstructTTFT (s): 1.08 vs 1.13No clear effect

Later papers that compared with this method

1 later accepted paper reports ProphetKV as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy