AI Paper Atlas

Research map / Architectures and efficiency / Quantization

STAR-KV: Low-Rank KV Cache Compression via Soft Thresholding for Adaptive Rank Control

Priyansh Bhatnagar, Ashkan Moradifirouzabadi, Se-Hyun Yang, SeungJae Lee, Jungwook Choi, Mingu Kang

ICML 2026 · Architectures and efficiency › Quantization · approach: KV cache quantization · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use learned adaptive rank selection instead of uniform static rank (ablation)STAR-KV at 75% compression · Commonsense zero-shot (6 tasks avg)accuracy: 38.88 -> 60.49Improved
add block-wise Hadamard to mixed-precision quantization (ablation)STAR-KV 75% + 3.2-bit quant · Commonsense zero-shot (5 tasks avg)accuracy: 57.27 -> 59.18Improved
use block-wise Hadamard with mixed precision instead of global Hadamard at uniform 3-bit (ablation)STAR-KV 75% + quant · Commonsense zero-shot (5 tasks avg)accuracy: 58.23 -> 59.18Improved
hybrid decomposition: head-wise keys, joint values instead of joint for both (ablation)STAR-KV · Reconstruction latency vs accuracylatencyImproved
apply soft-threshold STAR-KV at 60% instead of Palu at 50% compressionLLaMA-3.1-8B-Instruct · LongBench (7 tasks avg)score: 32.21 -> 39.58Improved
STAR-KV 60% instead of Palu 60% on RULER 16KLongChat-v1.5-32K · RULER 16K avgaccuracy: 62.94 -> 84.16Improved
compression-loss weight gamma 1.0 / 0.1 / 0.01 (ablation)STAR-KV 60% · Commonsense zero-shot (6 tasks avg)accuracy: 61.28 / 61.53 / 61.33No clear effect
change calibration dataset from FineWeb-Edu to C4 or RedPajama (ablation)STAR-KV 60% on LongChat-v1.5-32K · Commonsense zero-shot (6 tasks avg)accuracy: 61.53 -> 61.39 / 61.18No clear effect
change calibration dataset from FineWeb-Edu to C4 or RedPajama (ablation)STAR-KV 60% on LLaMA-3.1-8B-Inst · Commonsense zero-shot (6 tasks avg)accuracy: 66.43 -> 66.08 / 65.75No clear effect
outlier:inlier split 20:80 vs 10:90 (ablation)STAR-KV 75% + quant · Commonsense zero-shot (5 tasks avg)accuracy: 58.80 -> 59.18Improved
outlier:inlier split 40:60 vs 20:80 (ablation)STAR-KV 75% + quant · Commonsense zero-shot (5 tasks avg)accuracy: 59.18 -> 59.52; compression 20.00 -> 18.82Mixed
combine 60% low-rank STAR-KV with 60% H2O token pruningLongChat-v1.5-32K · Commonsense zero-shot (6 tasks avg)accuracy: 59.95 -> 59.99Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy