---
id: 20260507-T0-07
title: "eOptShrinkQ：无损KV缓存压缩新方案"
title_en: "eOptShrinkQ: Near-Lossless KV Cache Compression"
url: https://ai.daily.yangsir.net/daily/20260507-T0-07
issue_date: 2026-05-07
publish_date: 2026-05-06T04:00:00.000Z
category: research
source_name: "arXiv cs.LG (ML)"
source_url: https://arxiv.org/abs/2605.02905
---

# eOptShrinkQ：无损KV缓存压缩新方案

研究人员提出eOptShrinkQ方法，通过最优谱去噪和量化技术实现近乎无损的KV缓存压缩。该技术能显著降低大模型的内存使用，同时保持性能不变。这项突破对于提高大模型推理效率具有重要意义。

## English Version

**eOptShrinkQ: Near-Lossless KV Cache Compression**

Researchers introduce eOptShrinkQ, achieving near-lossless KV cache compression through optimal spectral denoising and quantization. This breakthrough significantly reduces memory usage while maintaining performance, crucial for improving LLM inference efficiency.

---

**来源**：[arXiv cs.LG (ML)](https://arxiv.org/abs/2605.02905)

**详情页**：https://ai.daily.yangsir.net/daily/20260507-T0-07

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*