Saved in:
Bibliographic Details
Main Author: Patel, Dipkumar
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2603.27467
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917365833269248
author Patel, Dipkumar
author_facet Patel, Dipkumar
contents We compress KV cache entries by quantizing angles in the Fast Walsh-Hadamard domain, where a random diagonal rotation makes consecutive element pairs approximately uniformly distributed on the unit circle. We extend this angular quantizer with per-layer early-boost, which independently configures K and V codebook sizes at each layer, allocating higher precision to a model-specific subset of critical layers. Across seven models (1B to 7B parameters), per-layer early-boost achieves lossless compression on four models and near-lossless quality on six of seven, at 3.28 to 3.67 angle bits per element. Asymmetric norm quantization (8-bit for keys, 4-bit log-space for values) yields 6.56 total bits per element on Mistral-7B with perplexity degradation of +0.0014 and no calibration data. A layer-group sensitivity analysis reveals model-specific bottleneck patterns, including K-dominated versus V-dominated layers and negative-transfer layers where increased precision degrades quality.
format Preprint
id arxiv_https___arxiv_org_abs_2603_27467
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle TurboAngle: Near-Lossless KV Cache Compression via Uniform Angle Quantization
Patel, Dipkumar
Machine Learning
Artificial Intelligence
We compress KV cache entries by quantizing angles in the Fast Walsh-Hadamard domain, where a random diagonal rotation makes consecutive element pairs approximately uniformly distributed on the unit circle. We extend this angular quantizer with per-layer early-boost, which independently configures K and V codebook sizes at each layer, allocating higher precision to a model-specific subset of critical layers. Across seven models (1B to 7B parameters), per-layer early-boost achieves lossless compression on four models and near-lossless quality on six of seven, at 3.28 to 3.67 angle bits per element. Asymmetric norm quantization (8-bit for keys, 4-bit log-space for values) yields 6.56 total bits per element on Mistral-7B with perplexity degradation of +0.0014 and no calibration data. A layer-group sensitivity analysis reveals model-specific bottleneck patterns, including K-dominated versus V-dominated layers and negative-transfer layers where increased precision degrades quality.
title TurboAngle: Near-Lossless KV Cache Compression via Uniform Angle Quantization
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2603.27467