PagedEviction: Structured Block-wise KV Cache Pruning for Efficient Large Language Model Inference

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Chitty-Venkata, Krishna Teja, Ye, Jie, Sun, Xian-He, Kougkas, Anthony, Emani, Murali, Vishwanath, Venkatram, Nicolae, Bogdan
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866915479322361856
author Chitty-Venkata, Krishna Teja
Ye, Jie
Sun, Xian-He
Kougkas, Anthony
Emani, Murali
Vishwanath, Venkatram
Nicolae, Bogdan
author_facet Chitty-Venkata, Krishna Teja
Ye, Jie
Sun, Xian-He
Kougkas, Anthony
Emani, Murali
Vishwanath, Venkatram
Nicolae, Bogdan
contents KV caching significantly improves the efficiency of Large Language Model (LLM) inference by storing attention states from previously processed tokens, enabling faster generation of subsequent tokens. However, as sequence length increases, the KV cache quickly becomes a major memory bottleneck. To address this, we propose PagedEviction, a novel fine-grained, structured KV cache pruning strategy that enhances the memory efficiency of vLLM's PagedAttention. Unlike existing approaches that rely on attention-based token importance or evict tokens across different vLLM pages, PagedEviction introduces an efficient block-wise eviction algorithm tailored for paged memory layouts. Our method integrates seamlessly with PagedAttention without requiring any modifications to its CUDA attention kernels. We evaluate PagedEviction across Llama-3.1-8B-Instruct, Llama-3.2-1B-Instruct, and Llama-3.2-3B-Instruct models on the LongBench benchmark suite, demonstrating improved memory usage with better accuracy than baselines on long context tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2509_04377
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle PagedEviction: Structured Block-wise KV Cache Pruning for Efficient Large Language Model Inference
Chitty-Venkata, Krishna Teja
Ye, Jie
Sun, Xian-He
Kougkas, Anthony
Emani, Murali
Vishwanath, Venkatram
Nicolae, Bogdan
Machine Learning
KV caching significantly improves the efficiency of Large Language Model (LLM) inference by storing attention states from previously processed tokens, enabling faster generation of subsequent tokens. However, as sequence length increases, the KV cache quickly becomes a major memory bottleneck. To address this, we propose PagedEviction, a novel fine-grained, structured KV cache pruning strategy that enhances the memory efficiency of vLLM's PagedAttention. Unlike existing approaches that rely on attention-based token importance or evict tokens across different vLLM pages, PagedEviction introduces an efficient block-wise eviction algorithm tailored for paged memory layouts. Our method integrates seamlessly with PagedAttention without requiring any modifications to its CUDA attention kernels. We evaluate PagedEviction across Llama-3.1-8B-Instruct, Llama-3.2-1B-Instruct, and Llama-3.2-3B-Instruct models on the LongBench benchmark suite, demonstrating improved memory usage with better accuracy than baselines on long context tasks.
title PagedEviction: Structured Block-wise KV Cache Pruning for Efficient Large Language Model Inference
topic Machine Learning
url https://arxiv.org/abs/2509.04377