WUSH-KV: KV Cache Quantization with Data-Adaptive Transforms

Jiale Chen; Vage Egiazarian; Eldar Kurtić; Torsten Hoefler; Dan Alistarh

Home/Archive/arXiv

arXiv · 2609.38121

WUSH-KV: KV Cache Quantization with Data-Adaptive Transforms

Jiale Chen·Vage Egiazarian·Eldar Kurtić·Torsten Hoefler·Dan Alistarh

View original source ↗Read source document ↗Export metadata ↓

Abstract

KV cache memory and bandwidth costs grow with context length and batch size, which limits efficient long-context inference. To address this bottleneck, we introduce WUSH-KV for low-bit KV-cache quantization. It adapts WUSH, which constructs a data-aware transform from the second-order statistics of both factors in a matrix product to reduce quantization error. WUSH-KV uses calibration data to construct separate key and value transforms, with the value transform folded into the model weights and the key transform applied after RoPE. The transforms can be paired with clipped quantizers. For one such quantizer, QuEST INT, we show that, under mild assumptions, the WUSH transform is near-optimal. With this quantizer, WUSH-KV reduces layerwise reconstruction error and achieves the lowest end-to-end perplexity among other tested transforms. For end-to-end evaluation, we integrate WUSH-KV into SGLang using OSCAR-style percentile-clipped affine quantization. At 2-bit, WUSH-KV performs comparably to or outperforms the OSCAR transform across all evaluated models and downstream tasks.

Explore related subjects

cs.LG

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Jiale Chen, Vage Egiazarian, Eldar Kurtić, Torsten Hoefler, Dan Alistarh. 2026-09-29. WUSH-KV: KV Cache Quantization with Data-Adaptive Transforms. https://arxiv.org/abs/2609.38121

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

Efficient Policy Evaluation with Offline Data Informed Behavior Policy Design

Online Monte Carlo evaluation is a fundamental tool for assessing policy performance in reinforcement learning and sequential decision-making problems arising in operations research. However, achieving accurate estimates often requires extensive online interaction with the environment, which can be costly or impractical in many real-world settings. In this paper, we develop a framework that improves the sample efficiency of online Monte Carlo estimators while preserving unbiasedness. We first derive a closed-form optimal behavior policy that minimizes estimator variance under unbiasedness constraints. We then propose practical algorithms for learning the proposed behavior policy from previously collected offline data, enabling improved online evaluation without requiring estimation of the environment transition model. We provide theoretical analysis that quantifies the resulting variance reduction and analyzes the impact of approximation errors. Empirical studies across diverse environments demonstrate substantial improvements in online sample efficiency compared with standard on-policy Monte Carlo evaluation and existing baseline methods. Our results provide a unified framework for optimal behavior policy design in off-policy evaluation, with applications to reinforcement learning and operations research.

cs.LG↗

arXiv2026

MONOVAB : An Annotated Corpus for Bangla Multi-label Emotion Detection

In recent years, Sentiment Analysis (SA) and Emotion Recognition (ER) have been increasingly popular in the Bangla language, which is the seventh most spoken language throughout the entire world. However, the language is structurally complicated, which makes this field arduous to extract emotions in an accurate manner. Several distinct approaches such as the extraction of positive and negative sentiments as well as multiclass emotions, have been implemented in this field of study. Nevertheless, the extraction of multiple sentiments is an almost untouched area in this language. Which involves identifying several feelings based on a single piece of text. Therefore, this study demonstrates a thorough method for constructing an annotated corpus based on scrapped data from Facebook to bridge the gaps in this subject area to overcome the challenges. To make this annotation more fruitful, the context-based approach has been used. Bidirectional Encoder Representations from Transformers (BERT), a well-known methodology of transformers, have been shown the best results of all methods implemented. Finally, a web application has been developed to demonstrate the performance of the pre-trained top-performer model (BERT) for multi-label ER in Bangla.

cs.LG↗

arXiv2026

Online Regularized Statistical Learning in Reproducing Kernel Hilbert Space With Non-Stationary Data

We study recursive regularized learning algorithms in the reproducing kernel Hilbert space (RKHS) with non-stationary online data streams. We introduce the concept of a random Tikhonov regularization path and decompose the tracking error of the algorithm's output for the regularization path into random difference equations in RKHS. We show that the tracking error vanishes in mean square and almost surely if the regularization path is slowly time-varying. Then, leveraging the monotonicity of inverse operators and the spectral decomposition of compact operators, and introducing the RKHS persistence of excitation condition, we develop a dominated convergence method to prove the mean square and almost sure consistency between the regularization path and the unknown function to be learned. Especially, for independent and non-identically distributed data streams, the mean square and almost sure consistency between the algorithm's output and the unknown function is achieved if the input data's marginal probability measures are slowly time-varying and the average measure over each fixed-length time period is uniformly above a strictly positive finite Borel measure.

cs.LG↗