AI Paper Atlas

Research map / Architectures and efficiency / Quantization

VAR-Q: Tuning-free KV Cache Quantization for Visual Autoregressive Image and Video Generation

Boxun Xu, Jiaji Lu, Yuxuan Yin, Zihu Wang, Ziyue (Alvin) Liu, Yu Wang, Zirui Liu, Peng Li

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

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes VAR-Q, a tuning-free KV cache quantization method for visual autoregressive image and video generation, motivated by KV caches whose statistics differ from language models.

Problem

KV memory in visual autoregressive generation (next-scale prediction for images/video and next-frame prediction for video) exceeds model weight memory at scale, limiting deployment because of the large number of visual tokens.

Approach

Analyzes KV caches of block-wise visual autoregressive decoding, finds their statistics diverge from language-model KV caches, and uses this analysis to design a quantization scheme that requires no tuning.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
group KV quantization per (channel, block) instead of per-channel only (ablation)VAR-Q · Infinity-2B INT4attention output error eps_O (%): 13.13 -> 10.13Improved
add block-axis grouping to channel grouping (ablation)VAR-Q · VAR-d30 INT4FID: 2.83 -> 1.97Improved
add channel-axis grouping (ablation)VAR-Q · VAR-d30 INT4FID: 12.07 -> 1.97Improved
add channel-axis grouping (ablation)VAR-Q · Self-Forcing INT4VBench Total Quality: 67.22 -> 80.99Improved
use block-aware grouping instead of per-token & per-head (ablation)VAR-Q · VAR-d30 INT4FID: 2.21 -> 1.97Improved
align group boundaries with block boundaries (ablation)VAR-Q · Self-Forcing INT4VBench Total Quality: 80.46 -> 80.91Improved
use finer sub-block ratio r=1/4 instead of r=1 (ablation)VAR-Q · Self-Forcing INT4VBench Total Quality: 80.99 -> 81.34Improved
zero top 0.5% outlier channels vs equal-sized inlier set (ablation)LongLive-1.3B KV cache · LongLive-1.3BCLIP score: 0.32 -> 0.28 (outlier); 0.32 (inlier)Hurt
use VAR-Q instead of KIVI at INT3 on InfinityStar 720pKIVI · InfinityStar-8B 720p INT3VBench Total Quality: 75.46 -> 80.59Improved
use VAR-Q instead of FlexGen at INT3 on VAR-d20FlexGen · VAR-d20 INT3FID: 15.33 -> 3.77Improved
quantize at INT4 vs INT3 with Infinity-8BVAR-Q · Infinity-8BGenEval Overall: 83.28 (INT3) -> 74.38 (INT2)Hurt
VAR-Q at INT4 on VAR-d20 vs BF16VAR-Q · VAR-d20FID: 4.11 -> 3.44No clear effect

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