VAR-Q: Tuning-free KV Cache Quantization for Visual Autoregressive Image and Video Generation
Computer Science > Computer Vision and Pattern Recognition arXiv:2611.05380 (cs) [Submitted on 12 Jul 2026] Title: VAR-Q: Tuning-free KV Cache Quantization for Visual Autoregressive Image and Video Generation Authors: Boxun Xu, Jiaji Lu, Yuxuan Yin, Zihu Wang, Ziyue Liu, Yu Wang, Zirui Liu, Peng Li.
计算机科学 > 计算机视觉与模式识别 arXiv:2611.05380 (cs) [提交于 2026 年 7 月 12 日] 标题：VAR-Q：用于视觉自回归图像和视频生成的免调优 KV 缓存量化。作者：Boxun Xu, Jiaji Lu, Yuxuan Yin, Zihu Wang, Ziyue Liu, Yu Wang, Zirui Liu, Peng Li。
Abstract: Visual autoregressive generation spans two dominant paradigms: next-scale prediction for images and video, and next-frame prediction for video. Both decode block-by-block rather than token-by-token, yielding KV caches whose statistics diverge from those of language models. At scale, KV memory exceeds the weights and limits large-scale deployment caused by a large amount of visual tokens.
摘要：视觉自回归生成涵盖了两种主流范式：图像和视频的尺度预测（next-scale prediction），以及视频的帧预测（next-frame prediction）。两者均采用逐块（block-by-block）而非逐词（token-by-token）解码，产生的 KV 缓存统计特性与语言模型存在差异。在大规模应用中，由于视觉词元数量庞大，KV 内存占用超过了权重，限制了大规模部署。
Our analysis of these caches reveals three findings. First, a small set of massive-value channels carries the functional load: ablating these outlier channels collapses generation, whereas ablating a matched set of inlier channels leaves quality intact. Second, the outliers organize along two axes, feature and block, with magnitudes persistent within each cell and shifting across cells. Third, this structure arises from two independent mechanisms: projection bias and RoPE-induced scattering.
我们对这些缓存的分析揭示了三个发现。首先，一小部分高值通道承担了主要功能负载：消融这些异常值通道会导致生成崩溃，而消融同等数量的正常值通道则不会影响质量。其次，这些异常值沿特征和块两个轴组织，其量级在每个单元内保持一致，但在不同单元间发生偏移。第三，这种结构源于两个独立的机制：投影偏差和 RoPE 引起的散射。
Guided by these findings, we propose VAR-Q, a tuning-free KV-cache quantization framework that groups along the same two axes. The generation loop provides block boundaries for free, removing the need for outlier detection or calibration. Paired with bit-packing kernels for runtime memory reduction, VAR-Q delivers low-bit quantization across nine models spanning both paradigms, cutting KV memory by up to 78.5% with negligible quality degradation and enabling 6x larger inference batches.
基于这些发现，我们提出了 VAR-Q，这是一个沿相同两个轴进行分组的免调优 KV 缓存量化框架。生成循环免费提供了块边界，无需进行异常值检测或校准。结合用于运行时内存缩减的位打包内核，VAR-Q 在涵盖两种范式的九个模型上实现了低位量化，在质量下降可忽略不计的情况下，将 KV 内存减少了高达 78.5%，并使推理批次规模扩大了 6 倍。