English

Ablation, Statistical Inference, and Validation for KV-Cache Compression

Machine Learning 2026-06-14 v1 Artificial Intelligence Information Theory

Abstract

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 (deffd_{eff}) adapting to calibration budgets rather than true data rank. (this is an abstract of the abstract thank you )

Cite

@article{arxiv.2607.09683,
  title  = {Ablation, Statistical Inference, and Validation for KV-Cache Compression},
  author = {Paolo D'Alberto and Ashish Siarasao and Elliott Delaye and Rajeev Patwari},
  journal= {arXiv preprint arXiv:2607.09683},
  year   = {2026}
}

Comments

15 pages, 8 figures, minimum number of citations