arXiv 2610.153242026-10-14cs.CV

TempoKV：用于自回归视频生成的预测型KV缓存量化

TempoKV: Predictive KV Cache Quantization for Autoregressive Video Generation

Shanghai Jiao Tong University（上海交通大学）；South China University of Technology（华南理工大学）；Alibaba Group（阿里巴巴集团）

Jiahui Hu, Weile Mo, Peiliang Cai, Jiacheng Liu, Jinkui Ren, Xiantao Zhang, Xiaobing Tu, Linfeng Zhang

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

TempoKV是一种无训练的预测型KV缓存量化框架，针对自回归视频生成设计，通过利用视频KV缓存的时空特性实现最高6.71倍压缩，且生成质量优于现有方法。

摘要：

自回归（AR）视频扩散模型可实现长视频的流式生成，但因KV缓存快速增长而存在显著的内存开销问题。现有KV缓存量化方法主要针对大型语言模型（LLM）设计，难以迁移到视频模型，视频模型的激活值具有不同的统计特性和强时空冗余性。我们观察到视频KV缓存随时间具有较强可预测性，且存在明显不对称性：键（Key）可通过其时间历史高度预测，而值（Value）主要依赖当前键，仅需要少量前序值的历史。基于这些观察，我们提出TempoKV，一种用于自回归视频生成的无训练预测型KV缓存量化框架。TempoKV根据键和值的不同依赖关系进行预测，仅以低精度存储残差。两个预测器均为轻量级，通过在一小批BF16样本上进行线性回归完成一次离线校准，在不同提示和生成过程中保持固定。在Self-Forcing、Causal Forcing和LongCat-Video-13B上的实验表明，TempoKV在可比或更高压缩率下，始终比现有KV缓存量化方法实现更高的生成质量，达到最高6.71倍的KV缓存压缩，且具备最佳的质量-效率权衡。我们的代码将在GitHub上发布。

英文摘要：

Autoregressive (AR) video diffusion models enable streaming generation of long videos but suffer from significant memory overhead due to the rapidly growing KV cache. Existing KV cache quantization methods are primarily designed for LLMs and transfer poorly to video models, where activations exhibit different statistics and strong spatiotemporal redundancy. We observe that the video KV cache is largely predictable over time, with a clear asymmetry. Keys are highly predictable from their temporal history, while values depend primarily on the current key and need a short history of preceding values. Based on these observations, we propose \textbf{\method}, a training-free predictive KV cache quantization framework for AR video generation. \method predicts keys and values according to their distinct dependencies and stores only the residuals at low precision. Both predictors are lightweight, calibrated once offline by linear regression on a small set of BF16 samples, and remain fixed across prompts and generation runs. Experiments on Self-Forcing, Causal Forcing, and LongCat-Video-13B show that \method consistently achieves higher generation quality than existing KV cache quantization methods at comparable or higher compression ratios, reaching up to $6.71\times$ KV cache compression with the best overall quality--efficiency tradeoff. Our code will be released on GitHub.

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。