FAEDKV：用于无偏 KV 缓存压缩的无限窗口傅里叶变换
FAEDKV: Infinite-Window Fourier Transform for Unbiased KV Cache Compression
AI总结：
针对长上下文 LLM 的 KV 缓存内存与计算瓶颈，提出无需训练的 FAEDKV，用 IWDFT 在频域实现无偏、逐层压缩，并在 LongBench 和 Needle-In-A-Haystack 上验证优势。
摘要：
大型语言模型（LLM）在长上下文任务中的有效性，往往受到键值（KV）缓存巨大内存占用和计算需求的阻碍。当前压缩策略（包括 token 驱逐和学习型投影）经常导致有偏表示——要么过度强调近期或高注意力 token，要么反复降低来自较早上下文的信息——并且可能需要代价高昂的模型再训练。我们提出 FAEDKV（Frequency-Adaptive Infinite-Window for KV cache，用于 KV 缓存的频率自适应无限窗口），这是一种新颖的、无需训练的 KV 缓存压缩框架，可确保无偏的信息保留。FAEDKV 通过使用所提出的无限窗口傅里叶变换（IWDFT）将 KV 缓存变换到频域来运行。该方法允许所有 token 对压缩表示作出均衡贡献，从而有效保留早期和近期上下文信息。一项初步频率消融研究识别出用于逐层定向压缩的关键频谱成分。在 LongBench 基准上的实验表明，FAEDKV 相比现有方法具有高达 22% 的优势。此外，在 Needle-In-A-Haystack 任务中，与基于压缩的方法相比，我们的方法表现出优越的、与位置无关的检索准确率。
英文摘要：
The efficacy of Large Language Models (LLMs) in long-context tasks is often hampered by the substantial memory footprint and computational demands of the Key-Value (KV) cache. Current compression strategies, including token eviction and learned projections, frequently lead to biased representations -- either by overemphasizing recent/high-attention tokens or by repeatedly degrading information from earlier context -- and may require costly model retraining. We present FAEDKV (Frequency-Adaptive Infinite-Window for KV cache), a novel, training-free KV cache compression framework that ensures unbiased information retention. FAEDKV operates by transforming the KV cache into the frequency domain using a proposed Infinite-Window Fourier Transform (IWDFT). This approach allows for the equalized contribution of all tokens to the compressed representation, effectively preserving both early and recent contextual information. A preliminary frequency ablation study identifies critical spectral components for layer-wise, targeted compression. Experiments on LongBench benchmark demonstrate FAEDKV's superiority over existing methods by up to 22\%. In addition, our method shows superior, position-agnostic retrieval accuracy on the Needle-In-A-Haystack task compared to compression based approaches.
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。
