arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2410.15252cs.CLcs.AI

无损KV缓存压缩至2%

Lossless KV Cache Compression to 2%

Tencent Hunyuan(腾讯混元);Tokyo Institute of Technology(东京工业大学)

Zhen Yang, J. N. Han, Kan Wu, Ruobing Xie, An Wang, Xingwu Sun, Zhanhui Kang

AI总结:

本文提出跨层潜在注意力(CLLA)架构,通过注意力头/维度缩减、层共享和量化技术将KV缓存压缩至原始大小2%以下,并在大多数任务上实现无损性能。

摘要:

大型语言模型已在众多领域革新了数据处理方式,其处理长上下文推理的能力获得了显著认可。为加速推理,维护键值(KV)缓存内存至关重要。然而,对KV缓存内存日益增长的需求给高效实现带来了重大障碍。本工作提出了一种新颖的架构——跨层潜在注意力(Cross-Layer Latent Attention,CLLA),旨在将KV缓存压缩至其原始大小的2%以下,同时保持可比的性能水平。CLLA将KV缓存压缩的多个方面整合到一个统一的框架中,包括注意力头/维度缩减、层共享和量化技术。我们的大量实验表明,CLLA在大多数任务上实现了无损性能,同时仅使用极少的KV缓存,标志着实用KV缓存压缩方面的一项重大进展。

英文摘要:

Large language models have revolutionized data processing in numerous domains, with their ability to handle extended context reasoning receiving notable recognition. To speed up inference, maintaining a key-value (KV) cache memory is essential. Nonetheless, the growing demands for KV cache memory create significant hurdles for efficient implementation. This work introduces a novel architecture, Cross-Layer Latent Attention (CLLA), aimed at compressing the KV cache to less than 2% of its original size while maintaining comparable performance levels. CLLA integrates multiple aspects of KV cache compression, including attention head/dimension reduction, layer sharing, and quantization techniques, into a cohesive framework. Our extensive experiments demonstrate that CLLA achieves lossless performance on most tasks while utilizing minimal KV cache, marking a significant advancement in practical KV cache compression.

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。

↑