AdaptCache: KV Cache Native Storage Hierarchy for Low-Delay and High-Quality Language Model Serving

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Feng, Shaoting, Li, Hanchen, Du, Kuntai, Gu, Zhuohan, Liu, Yuhan, Yao, Jiayi, Ray, Siddhant, Shen, Samuel, Cheng, Yihua, Ananthanarayanan, Ganesh, Jiang, Junchen
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917205984149504
author Feng, Shaoting
Li, Hanchen
Du, Kuntai
Gu, Zhuohan
Liu, Yuhan
Yao, Jiayi
Ray, Siddhant
Shen, Samuel
Cheng, Yihua
Ananthanarayanan, Ganesh
Jiang, Junchen
author_facet Feng, Shaoting
Li, Hanchen
Du, Kuntai
Gu, Zhuohan
Liu, Yuhan
Yao, Jiayi
Ray, Siddhant
Shen, Samuel
Cheng, Yihua
Ananthanarayanan, Ganesh
Jiang, Junchen
contents Large language model (LLM) applications often reuse previously processed context, such as chat history and documents, which introduces significant redundant computation. Existing LLM serving systems address such redundant computation by storing the KV caches of processed context and loading the corresponding KV cache when a new request reuses the context. Further, as these LLM applications scale, the total size of KV caches becomes excessively large and requires both DRAM and SSD for full storage. However, prior work that stores KV caches in DRAM and SSD suffers from high loading delays, as most KV cache hits come from SSD, which is slow to load. To increase the KV cache hit rate on DRAM, we identify lossy KV cache compression as a promising approach. We design a lossy compression system that decides the compression algorithm, compression rate and device placement for each KV cache entry to maximise DRAM hits and minimise loading delay without significantly degrading generation quality. Compared to various static compression baselines across three tasks, our system AdaptCache achieves 1.43--2.4 x delay savings at the same quality and 6--55% quality improvements at the same delay.
format Preprint
id arxiv_https___arxiv_org_abs_2509_00105
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle AdaptCache: KV Cache Native Storage Hierarchy for Low-Delay and High-Quality Language Model Serving
Feng, Shaoting
Li, Hanchen
Du, Kuntai
Gu, Zhuohan
Liu, Yuhan
Yao, Jiayi
Ray, Siddhant
Shen, Samuel
Cheng, Yihua
Ananthanarayanan, Ganesh
Jiang, Junchen
Operating Systems
Artificial Intelligence
Machine Learning
Large language model (LLM) applications often reuse previously processed context, such as chat history and documents, which introduces significant redundant computation. Existing LLM serving systems address such redundant computation by storing the KV caches of processed context and loading the corresponding KV cache when a new request reuses the context. Further, as these LLM applications scale, the total size of KV caches becomes excessively large and requires both DRAM and SSD for full storage. However, prior work that stores KV caches in DRAM and SSD suffers from high loading delays, as most KV cache hits come from SSD, which is slow to load. To increase the KV cache hit rate on DRAM, we identify lossy KV cache compression as a promising approach. We design a lossy compression system that decides the compression algorithm, compression rate and device placement for each KV cache entry to maximise DRAM hits and minimise loading delay without significantly degrading generation quality. Compared to various static compression baselines across three tasks, our system AdaptCache achieves 1.43--2.4 x delay savings at the same quality and 6--55% quality improvements at the same delay.
title AdaptCache: KV Cache Native Storage Hierarchy for Low-Delay and High-Quality Language Model Serving
topic Operating Systems
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2509.00105