Rotary Position Embedding (RoPE) enables each attention head to capture multi-frequency information along the sequence dimension and is widely applied in foundation models. However, the nonlinearity introduced by RoPE complicates optimization of the key state in the Key-Value (KV) cache for RoPE-based attention. Existing KV cache compression methods typically store key state before rotation and apply the transformation during decoding, introducing additional computational overhead. This paper introduces EliteKV, a flexible modification framework for RoPE-based models supporting variable KV cache compression ratios. EliteKV first identifies the intrinsic frequency preference of each head using RoPElite, selectively restoring linearity to certain dimensions of key within attention computation. Building on this, joint low-rank compression of key and value enables partial cache sharing. Experimental results show that with minimal uptraining on only 0.6% of the original training data, RoPE-based models achieve a 75% reduction in KV cache size while preserving performance within a negligible margin. Furthermore, EliteKV consistently performs well across models of different scales within the same family.
@article{arxiv.2503.01586,
title = {EliteKV: Scalable KV Cache Compression via RoPE Frequency Selection and Joint Low-Rank Projection},
author = {Yuhao Zhou and Sirui Song and Boyang Liu and Zhiheng Xi and Senjie Jin and Xiaoran Fan and Zhihao Zhang and Wei Li and Xuanjing Huang},
journal= {arXiv preprint arXiv:2503.01586},
year = {2025}
}