MuKV：面向长流式视频问答的多粒度 KV 缓存压缩
MuKV: Multi-Grained KV Cache Compression for Long Streaming Video Question-Answering
对做长视频多模态理解的人，这篇给出了一个可直接复用的 KV 缓存压缩思路：在 patch、帧、片段三个粒度上保留视觉词元，并用自注意力与频率双信号筛选词元，再配合半分层检索取回相关缓存，从而在有限显存下支撑长流式视频问答。
原文依据 MuKV extracts visual representations at patch-, frame-, and segment-levels.
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论长流式视频问答仍然具有挑战性，原因在于视觉词元不断增长，而大语言模型（LLM）的推理长度有限。KV 缓存通过 LLM 预填充存储历史词元的键值（KV），使流式问答更加高效。然而，现有方法对每一帧或每两帧都进行缓存，造成冗余的内存占用，并丢失了帧内的细粒度空间细节或跨帧的时序上下文。本文提出 MuKV，该方法以多粒度 KV 缓存压缩模块和半分层检索方法为特色，同时提升长流式视频问答的效率与准确率。对于离线 KV 缓存，MuKV 在块级、帧级和片段级提取视觉表征。多级粒度既保留了局部线索，也保留了全局时序上下文，同时通过由自注意力和频率引导的双信号词元压缩机制保持效率。对于在线问答，MuKV 设计了半分层检索方法，为答案生成检索相关的 KV 缓存。在长流式视频问答基准上的实验表明，MuKV 显著提升了答案准确率，且不牺牲内存与在线问答效率。此外，仅我们的压缩机制本身就在答案准确率、内存和问答效率上相较基线带来了一致的收益，展现出极高的有效性。
英文原文摘要 1394 字符
Long streaming video QA remains challenging due to growing visual tokens and limited reasoning length of large language models (LLMs). KV-caching stores the Key-Value (KV) of the historical tokens via LLM prefill and enables more efficient streaming QA. However, existing methods cache every one or two frames, causing redundant memory usage and losing fine-grained spatial details within frame or temporal contexts across frames. This paper proposes MuKV, a method that features a multi-grained KV cache compression module and a semi-hierarchical retrieval approach to improve both efficiency and accuracy for long streaming VideoQA. For the offline KV cache, MuKV extracts visual representations at patch-, frame-, and segment-levels. The multiple levels of granularity preserve both local cues and global temporal context, while maintaining efficiency with a dual signal token compression mechanism guided by self-attention and frequency. For online QA, MuKV designs a semi-hierarchical retrieval method to retrieve relevant KV caches for answer generation. Experiments on long-streaming VideoQA benchmarks show that MuKV significantly improves answer accuracy, without sacrificing memory and online QA efficiency. Moreover, our compression mechanism alone brings consistent benefits across answer accuracy, memory, and QA efficiency over baselines, showcasing highly effective contribution.