Cover Image

When Does Value-Aware KV Eviction Help? A Fixed-Contract Diagnostic for Non-Monotone Cache Compression

Fuente: arXiv

Saved in:

Bibliographic Details
Main Authors:	Zhang, Ruijie, Liang, Haozhe, Chang, Da, Hu, Li, Kong, Fanqi, Yin, Huaxiao, Li, Yu
Format:	Preprint
Published:	2026
Subjects:	Machine Learning Artificial Intelligence
Online Access:	Acceder al recurso
Tags:	Add Tag No Tags, Be the first to tag this record!

Similar Items

Beyond Token Eviction: Mixed-Dimension Budget Allocation for Efficient KV Cache Compression
by: Miao, Ruijie, et al.
Published: (2026)

SABlock: Semantic-Aware KV Cache Eviction with Adaptive Compression Block Size
by: Chen, Jinhan, et al.
Published: (2025)

EVICPRESS: Joint KV-Cache Compression and Eviction for Efficient LLM Serving
by: Feng, Shaoting, et al.
Published: (2025)

CONF-KV: Confidence-Aware KV Cache Eviction with Mixed-Precision Storage for Long-Horizon LLM
by: Li, Yubo, et al.
Published: (2026)

A Simple Plug-in for Improving Eviction-Based KV Cache Compression
by: Lin, Yuping, et al.
Published: (2026)

CAKE: Cascading and Adaptive KV Cache Eviction with Layer Preferences
by: Qin, Ziran, et al.
Published: (2025)

MomentKV: Closing the Directional Gap in KV Cache Eviction for Long-Context Inference
by: Li, Yu, et al.
Published: (2026)

RetentiveKV: State-Space Memory for Uncertainty-Aware Multimodal KV Cache Eviction
by: Liu, Sihao, et al.
Published: (2026)

GraphKV: Breaking the Static Selection Paradigm with Graph-Based KV Cache Eviction
by: Li, Xuelin, et al.
Published: (2025)

AudioKV: KV Cache Eviction in Efficient Large Audio Language Models
by: Wang, Yuxuan, et al.
Published: (2026)

G-KV: Decoding-Time KV Cache Eviction with Global Attention
by: Liao, Mengqi, et al.
Published: (2025)

Taming the Fragility of KV Cache Eviction in LLM Inference
by: Feng, Yuan, et al.
Published: (2025)

CriticalKV: Optimizing KV Cache Eviction from an Output Perturbation Perspective
by: Feng, Yuan, et al.
Published: (2025)

ForesightKV: Optimizing KV Cache Eviction for Reasoning Models by Learning Long-Term Contribution
by: Dong, Zican, et al.
Published: (2026)

RDKV: Rate-Distortion Bit Allocation for Joint Eviction and Quantization of the KV Cache
by: Zhang, Junkai, et al.
Published: (2026)

In-context KV-Cache Eviction for LLMs via Attention-Gate
by: Zeng, Zihao, et al.
Published: (2024)

DynamicKV: Task-Aware Adaptive KV Cache Compression for Long Context LLMs
by: Zhou, Xiabin, et al.
Published: (2024)

LookaheadKV: Fast and Accurate KV Cache Eviction by Glimpsing into the Future without Generation
by: Ahn, Jinwoo, et al.
Published: (2026)

MadaKV: Adaptive Modality-Perception KV Cache Eviction for Efficient Multimodal Long-Context Inference
by: Li, Kunxi, et al.
Published: (2025)

MPCache: MPC-Friendly KV Cache Eviction for Efficient Private LLM Inference
by: Zeng, Wenxuan, et al.
Published: (2025)

ReST-KV: Robust KV Cache Eviction with Layer-wise Output Reconstruction and Spatial-Temporal Smoothing
by: An, Yongqi, et al.
Published: (2026)

Reformulating KV Cache Eviction Problem for Long-Context LLM Inference
by: Mai, Tho, et al.
Published: (2026)

LAVa: Layer-wise KV Cache Eviction with Dynamic Budget Allocation
by: Shen, Yiqun, et al.
Published: (2025)

Hierarchical Adaptive Eviction for KV Cache Management in Multimodal Language Models
by: Ma, Xindian, et al.
Published: (2026)

Ada-KV: Optimizing KV Cache Eviction by Adaptive Budget Allocation for Efficient LLM Inference
by: Feng, Yuan, et al.
Published: (2024)

R-KV: Redundancy-aware KV Cache Compression for Reasoning Models
by: Cai, Zefan, et al.
Published: (2025)

LLMs Know What to Drop: Self-Attention Guided KV Cache Eviction for Efficient Long-Context Inference
by: Wang, Guangtao, et al.
Published: (2025)

Lookahead Q-Cache: Achieving More Consistent KV Cache Eviction via Pseudo Query
by: Wang, Yixuan, et al.
Published: (2025)

Judge Q: Trainable Queries for Optimized Information Retention in KV Cache Eviction
by: Liu, Yijun, et al.
Published: (2025)

Rethinking KV Cache Eviction via a Unified Information-Theoretic Objective
by: Yang, Jiaming, et al.
Published: (2026)

IndexMem: Learned KV-Cache Eviction with Latent Memory for Long-Context LLM Inference
by: Yang, Xintong, et al.
Published: (2026)

OjaKV: Context-Aware Online Low-Rank KV Cache Compression
by: Zhu, Yuxuan, et al.
Published: (2025)

AhaKV: Adaptive Holistic Attention-Driven KV Cache Eviction for Efficient Inference of Large Language Models
by: Gu, Yifeng, et al.
Published: (2025)

Key, Value, Compress: A Systematic Exploration of KV Cache Compression Techniques
by: Javidnia, Neusha, et al.
Published: (2025)

Not All Tokens Are Worth Caching: Learning Semantic-Aware Eviction for LLM Prefix Caches
by: Fang, Shaoke, et al.
Published: (2026)

NACL: A General and Effective KV Cache Eviction Framework for LLMs at Inference Time
by: Chen, Yilong, et al.
Published: (2024)

VEDA: Efficient LLM Generation Through Voting-based KV Cache Eviction and Dataflow-flexible Accelerator
by: Wang, Zhican, et al.
Published: (2025)

CaliDrop: KV Cache Compression with Calibration
by: Su, Yi, et al.
Published: (2025)

How Much Cache Does Reasoning Need? Depth-Cache Tradeoffs in KV-Compressed Transformers
by: Wang, Xiao
Published: (2026)

ReCalKV: Low-Rank KV Cache Compression via Head Reordering and Offline Calibration
by: Yan, Xianglong, et al.
Published: (2025)