Research

Paper

AI LLM March 24, 2026

EchoKV: Efficient KV Cache Compression via Similarity-Based Reconstruction

Authors

Yixuan Wang, Shiyu Ji, Yijun Liu, Qingfu Zhu, Wanxiang Che

Abstract

The increasing memory demand of the Key-Value (KV) cache poses a significant bottleneck for Large Language Models (LLMs) in long-context applications. Existing low-rank compression methods often rely on irreversible parameter transformations, sacrificing the flexibility to switch back to full-precision inference when memory is abundant. In this paper, we propose EchoKV, a flexible KV cache compression scheme that enables on-demand transitions between standard and compressed inference. Unlike traditional compression-decompression paradigms, EchoKV utilizes a lightweight network to reconstruct the residual KV components from a partial subset, leveraging intrinsic inter-layer and intra-layer similarities among attention heads. We further introduce a two-stage fine-tuning strategy that allows for rapid, low-cost training (e.g., ~1 A100 GPU-hour for a 7B model). Experimental results on LongBench and RULER demonstrate that EchoKV consistently outperforms existing methods across various compression ratios while maintaining high throughput for short-context scenarios.

Metadata

arXiv ID: 2603.22910
Provider: ARXIV
Primary Category: cs.CL
Published: 2026-03-24
Fetched: 2026-03-25 06:02

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.22910v1</id>\n    <title>EchoKV: Efficient KV Cache Compression via Similarity-Based Reconstruction</title>\n    <updated>2026-03-24T07:58:42Z</updated>\n    <link href='https://arxiv.org/abs/2603.22910v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.22910v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>The increasing memory demand of the Key-Value (KV) cache poses a significant bottleneck for Large Language Models (LLMs) in long-context applications. Existing low-rank compression methods often rely on irreversible parameter transformations, sacrificing the flexibility to switch back to full-precision inference when memory is abundant. In this paper, we propose EchoKV, a flexible KV cache compression scheme that enables on-demand transitions between standard and compressed inference. Unlike traditional compression-decompression paradigms, EchoKV utilizes a lightweight network to reconstruct the residual KV components from a partial subset, leveraging intrinsic inter-layer and intra-layer similarities among attention heads. We further introduce a two-stage fine-tuning strategy that allows for rapid, low-cost training (e.g., ~1 A100 GPU-hour for a 7B model). Experimental results on LongBench and RULER demonstrate that EchoKV consistently outperforms existing methods across various compression ratios while maintaining high throughput for short-context scenarios.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.CL'/>\n    <published>2026-03-24T07:58:42Z</published>\n    <arxiv:primary_category term='cs.CL'/>\n    <author>\n      <name>Yixuan Wang</name>\n    </author>\n    <author>\n      <name>Shiyu Ji</name>\n    </author>\n    <author>\n      <name>Yijun Liu</name>\n    </author>\n    <author>\n      <name>Qingfu Zhu</name>\n    </author>\n    <author>\n      <name>Wanxiang Che</name>\n    </author>\n  </entry>"
}