CAKE: Cascading and Adaptive KV Cache Eviction with Layer Preferences

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Qin, Ziran, Cao, Yuchen, Lin, Mingbao, Hu, Wen, Fan, Shixuan, Cheng, Ke, Lin, Weiyao, Li, Jianguo
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866918262323806208
author Qin, Ziran
Cao, Yuchen
Lin, Mingbao
Hu, Wen
Fan, Shixuan
Cheng, Ke
Lin, Weiyao
Li, Jianguo
author_facet Qin, Ziran
Cao, Yuchen
Lin, Mingbao
Hu, Wen
Fan, Shixuan
Cheng, Ke
Lin, Weiyao
Li, Jianguo
contents Large language models (LLMs) excel at processing long sequences, boosting demand for key-value (KV) caching. While recent efforts to evict KV cache have alleviated the inference burden, they often fail to allocate resources rationally across layers with different attention patterns. In this paper, we introduce Cascading and Adaptive KV cache Eviction (CAKE), a novel approach that frames KV cache eviction as a "cake-slicing problem." CAKE assesses layer-specific preferences by considering attention dynamics in both spatial and temporal dimensions, allocates rational cache size for layers accordingly, and manages memory constraints in a cascading manner. This approach enables a global view of cache allocation, adaptively distributing resources across diverse attention mechanisms while maintaining memory budgets. CAKE also employs a new eviction indicator that considers the shifting importance of tokens over time, addressing limitations in existing methods that overlook temporal dynamics. Comprehensive experiments on LongBench and NeedleBench show that CAKE maintains model performance with only 3.2% of the KV cache and consistently outperforms current baselines across various models and memory constraints, particularly in low-memory settings. Additionally, CAKE achieves over 10x speedup in decoding latency compared to full cache when processing contexts of 128K tokens with FlashAttention-2. Our code is available at https://github.com/antgroup/cakekv.
format Preprint
id arxiv_https___arxiv_org_abs_2503_12491
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle CAKE: Cascading and Adaptive KV Cache Eviction with Layer Preferences
Qin, Ziran
Cao, Yuchen
Lin, Mingbao
Hu, Wen
Fan, Shixuan
Cheng, Ke
Lin, Weiyao
Li, Jianguo
Computation and Language
Large language models (LLMs) excel at processing long sequences, boosting demand for key-value (KV) caching. While recent efforts to evict KV cache have alleviated the inference burden, they often fail to allocate resources rationally across layers with different attention patterns. In this paper, we introduce Cascading and Adaptive KV cache Eviction (CAKE), a novel approach that frames KV cache eviction as a "cake-slicing problem." CAKE assesses layer-specific preferences by considering attention dynamics in both spatial and temporal dimensions, allocates rational cache size for layers accordingly, and manages memory constraints in a cascading manner. This approach enables a global view of cache allocation, adaptively distributing resources across diverse attention mechanisms while maintaining memory budgets. CAKE also employs a new eviction indicator that considers the shifting importance of tokens over time, addressing limitations in existing methods that overlook temporal dynamics. Comprehensive experiments on LongBench and NeedleBench show that CAKE maintains model performance with only 3.2% of the KV cache and consistently outperforms current baselines across various models and memory constraints, particularly in low-memory settings. Additionally, CAKE achieves over 10x speedup in decoding latency compared to full cache when processing contexts of 128K tokens with FlashAttention-2. Our code is available at https://github.com/antgroup/cakekv.
title CAKE: Cascading and Adaptive KV Cache Eviction with Layer Preferences
topic Computation and Language
url https://arxiv.org/abs/2503.12491