cs.AINov 4, 2026

EchoKV: Self-Corrective KV Cache Eviction via Iterative Query Feedback

Authors: JianKun Zou, Tongtong Yuan

Organizations: Beijing University of Technology

Abstract

KV cache eviction is not merely a token-ranking problem: the workspace retained for generating eviction queries also determines the quality of those queries. Consequently, an early workspace error can distort later query observations and reinforce an incorrect final eviction decision, particularly when evidence is dispersed across a long context. We present EchoKV, which discovers and addresses this failure mode with iterative correction. It initializes a feedback-compatible bootstrap workspace using proposals at future decode positions, then uses early decode-generated queries to recover the workspace before the remaining lookahead steps. It uses the same number of workspace forward passes as LAQ while adding a tensor re-scoring operation for workspace repair. On LongBench (16 tasks, Mistral-7B-Instruct-v0.2 and Qwen2.5-7B-Instruct), EchoKV shows notable improvements over LAQ on several tasks requiring dispersed evidence, including multi-hop cross-document QA and long single-document comprehension. Across budgets and backbones, EchoKV selectively improves evidence-intensive tasks while maintaining comparable performance on the remaining LongBench subsets. These results identify query--workspace coupling as a practical failure mode of query-driven KV eviction.

Figures & tables

Explore similar work

CardsList
  1. Monte Carlo Estimation for KV Cache Eviction

    Oct 6, 2026Ahsan Bilal, Muhammad Ahmed Mohsin, Muhammad Umer +5LLM Inference EfficiencyKV-Cache Eviction

  2. DropKV: Decoupling Residual-Output Perturbation for Near-Optimal KV-Cache Eviction

    Oct 15, 2026Aozhong Zhang, Selcuk Gurses, Yanxia Deng +8KV-Cache EvictionLLM Inference Acceleration

  3. QEvict: Recoverable Quantized KV Eviction for Attention-Drift-Robust Long-Context Decoding

    Aug 5, 2026Ayushman Garg, Akshita Gupta, Shaswata Bhattacharya +3KV CachingLong-Context Language Model Inference