最后但同样重要：用于多模态KV缓存压缩的边界注意力校准
Last But Not Least: Boundary Attention CalibratiON for Multimodal KV Cache Compression
AI总结：
针对多模态大语言模型长视觉上下文中KV缓存压缩导致关键证据丢失的问题，提出BACON方法，通过校准观察窗口注意力与最后查询注意力，并利用层内一致性和层间持久性抑制噪声，在激进压缩下平均提升7.5%性能。
摘要：
多模态大语言模型（MLLMs）实现了强大的视觉-语言推理，但长视觉上下文会扩大KV缓存并增加解码延迟。现有的压缩方法依赖观察窗口注意力进行稳定的token重要性估计，然而这种聚合可能稀释稀疏的视觉证据，并在激进压缩下丢弃答案关键token。因此，我们识别出最后查询注意力作为恢复此类证据的补充来源，但其与答案无关的信号可能误导保留。我们提出BACON，一种即插即用方法，通过最后查询证据校准观察窗口注意力，并通过层内一致性和层间持久性抑制孤立噪声。在多种基准、模型、预算和压缩方法下，BACON在最激进的预算下平均提升多模态KV压缩7.5%，最高提升达30.9%。
英文摘要：
Multimodal Large Language Models (MLLMs) achieve strong vision-language reasoning but incur large KV caches and high decoding latency with long visual contexts. Existing compression methods rely on observation window attention for stable token importance estimation, yet this aggregation can dilute sparse critical evidence and discard answer-relevant tokens under aggressive compression. We identify last query attention as a complementary signal for recovering such evidence, though its irrelevant signals may introduce additional noise. We propose BACON, a plug-and-play method that calibrates observation window attention with last query evidence while suppressing noise through intra-layer coherence and inter-layer persistence. Across diverse benchmarks, models, budgets, and compression methods, BACON improves multimodal KV-cache compression by 7.5% on average under the most aggressive budget, with gains up to 30.9%.
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。
