---
id: 20260709-T0-09
title: "KV-Cache压缩技术对比基准：任务质量与系统性能的权衡"
title_en: "Benchmarking KV-Cache Optimizations Shows Tradeoffs Between Quality and System Performance"
url: https://ai.daily.yangsir.net/daily/20260709-T0-09
issue_date: 2026-07-09
publish_date: 2026-07-08T04:00:00.000Z
category: research
source_name: "arXiv cs.CL (NLP)"
source_url: https://arxiv.org/abs/2607.05399
---

# KV-Cache压缩技术对比基准：任务质量与系统性能的权衡

该研究系统性地对比了多种 KV-Cache 压缩技术在不同模型、任务、预算和推理配置下的表现。研究发现，现有压缩方法在任务质量（如长上下文理解准确率）和系统性能（如吞吐量和延迟）之间存在明显权衡，且之前的研究因评估设置不同而难以直接比较。该工作为开发者选择适合自身场景的 KV-Cache 优化策略提供了统一基准。

## English Version

**Benchmarking KV-Cache Optimizations Shows Tradeoffs Between Quality and System Performance**

This work benchmarks KV-Cache compression techniques across models, tasks, budgets, and serving setups. It reveals clear tradeoffs between task quality (e.g., long-context accuracy) and system performance (throughput, latency), which prior studies couldn't compare due to differing evaluation settings. The benchmark provides a unified reference for choosing the right compression strategy.

---

**来源**：[arXiv cs.CL (NLP)](https://arxiv.org/abs/2607.05399)

**详情页**：https://ai.daily.yangsir.net/daily/20260709-T0-09

---

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