对比两种压缩方法，发现结构化数据中性能更优
Ablation, Statistical Inference, and Validation for KV-Cache Compression
- 用统计验证分离编码器差异与实现波动
- 特征基方法在重尾数据中失效，但结构化数据表现优异
- 语义有效维度随校准预算变化，非真实秩
本研究系统比较了Turbo-Quant与SpectralQuant两种KV缓存压缩方法，通过统计验证方法分离编码器的系统性差异与实现中的随机波动。关键发现表明，基于特征基的方法在重尾数据上因协方差不稳定而失效，但在结构化数据中表现卓越；其有效语义维度（$d_{eff}$）随校准预算调整，而非反映真实数据秩。
原文摘要 · Abstract (English)
This study systematically compares Turbo-Quant and SpectralQuant KV-cache compression, evaluating non-dominated schemes, including WHT rotation with Beta Lloyd-Max and QJL, through a statistical validation methodology that separates systematic codec differences from implementation variance. Key findings reveal that while eigenbasis-based methods fail on heavy-tailed data due to covariance instability, they excel in structured regimes, with the effective semantic dimension ($d_{eff}$) adapting to calibration budgets rather than true data rank. (this is an abstract of the abstract thank you )
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。