Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Make Each Token Count: Towards Improving Long-Context Performance with KV Cache Eviction

About

The key-value (KV) cache is a major bottleneck in long-context inference, where memory and computation grow with sequence length. Existing KV eviction methods reduce this cost but typically degrade performance relative to full-cache inference. Our key insight is that full-cache attention is not always optimal: in long contexts, irrelevant tokens can dilute attention away from useful evidence, so selective, learnable eviction can improve generation rather than merely approximate the full cache. We introduce a global retention-based KV eviction method that learns each token's future utility under a unified memory budget. Lightweight retention gates assign utility scores to cached KV entries, and a shared final scoring projection calibrates these scores across all layers and heads. This enables a single global eviction policy in which tokens from different layers, heads, and modalities compete directly for cache capacity. We further provide theoretical analysis showing that preferentially retaining useful tokens reduces attention dilution, and we justify geometric retention as a query-agnostic proxy for future utility. Across diverse long-context language and vision-language reasoning, and multi-turn dialogue benchmarks, our method substantially reduces KV memory while matching or surpassing full-cache inference. These results suggest that learned, globally calibrated KV eviction is not only a compression technique, but also a mechanism for improving long-context reasoning.

Ngoc Bui, Hieu Trung Nguyen, Arman Cohan, Rex Ying• 2026

Related benchmarks

TaskDatasetResultRank
Long-context ReasoningLongBench v2
Average Score31.44
88
Video ReasoningVideo-MME--
55
Image ReasoningMMStar
Accuracy71.85
27
Image ReasoningMathVision Mini
Accuracy52.63
27
Image ReasoningMMMU Pro Vision
Accuracy43.87
27
Video ReasoningVideoMMMU adaptation
Accuracy39
27
Video ReasoningVideoMMMU comprehension
Accuracy61.33
27
Video ReasoningVideoMathQA MCQ
Accuracy36.43
27
Multi-turn visual dialogueMMDU 45K
Accuracy3.79
18
Showing 9 of 9 rows

Other info

GitHub

Follow for update