PolarQuant: Leveraging Polar Transformation for Efficient Key Cache Quantization and Decoding Acceleration

doi:arXiv:2502.00527

	PolarQuant: Leveraging Polar Transformation for Efficient Key Cache Quantization and Decoding Acceleration
	Wu, Songhao 1; Lv, Ang 1; Feng, Xiao2 ; Zhang, Yufei 3; Zhang, Xun 3; Yin, Guojun 3; Lin, Wei 3; Yan, Rui 1
	2025-02-01
状态	已发表
摘要	The KV cache in large language models is a dominant factor in memory usage, limiting their broader applicability. Quantizing the cache to lower bit widths is an effective way to reduce computational costs; however, previous methods struggle with quantizing key vectors due to outliers, resulting in excessive overhead. We propose a novel quantization approach called PolarQuant, which efficiently addresses the outlier challenge. We observe that outliers typically appear in only one of two dimensions, which are rotated together by a specific angle when rotary position embeddings are applied. When represented as two-dimensional vectors, these dimensions exhibit well-structured patterns, with radii and angles smoothly distributed in polar coordinates. This alleviates the challenge of outliers on per-channel quantization, making them well-suited for quantization. Thus, PolarQuant divides key vectors into groups of two-dimensional sub-vectors, encoding them as the corresponding quantized radius and the polar angle, rather than quantizing original key vectors directly. PolarQuant achieves the superior efficiency in KV cache quantization and accelerates the decoding process by turning the query-key inner product into a table lookup, all while maintaining the downstream performance of full-precision models.
语种	英语
DOI	arXiv:2502.00527
相关网址	查看原文
出处	Arxiv
收录类别	PPRN.PPRN
WOS记录号	PPRN:121095670
WOS类目	Computer Science, Artificial Intelligence ; Computer Science, Interdisciplinary Applications
文献类型	预印本
条目标识符	https://kms.shanghaitech.edu.cn/handle/2MSLDSTB/507029
专题	信息科学与技术学院_硕士生
通讯作者	Yin, Guojun; Yan, Rui
作者单位	1.Renmin Univ China, Beijing, Peoples R China 2.ShanghaiTech Univ, Shanghai, Peoples R China 3.Meituan, Beijing, Peoples R China
推荐引用方式 GB/T 7714	Wu, Songhao,Lv, Ang,Feng, Xiao,et al. PolarQuant: Leveraging Polar Transformation for Efficient Key Cache Quantization and Decoding Acceleration. 2025.