RetentiveKV: State-Space Memory for Uncertainty-Aware Multimodal KV Cache Eviction
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866915981860798464 |
|---|---|
| author | Liu, Sihao Xiong, YuFan Jiang, Zhonghua Wang, Zhaode Zhang, chengfei lv Shengyu |
| author_facet | Liu, Sihao Xiong, YuFan Jiang, Zhonghua Wang, Zhaode Zhang, chengfei lv Shengyu |
| contents | Multimodal Large Language Models face severe challenges in computational efficiency and memory consumption due to the substantial expansion of the visual KV cache when processing long visual contexts. Existing KV cache compression methods typically rely on the "persistence of importance" hypothesis to prune tokens. However, this approach proves fragile in multimodal settings due to two key issues: 1) Visual tokens display "deferred importance," initially exhibiting low salience but becoming pivotal during later decoding, which can lead to premature eviction. 2) Discrete pruning disrupts the inherent spatial continuity of visual cues. To address these challenges, we propose RetentiveKV, an entropy-driven KV cache optimization method that reformulates KV eviction from "discrete context truncation" to "continuous memory evolution" based on State Space Models. Our method leverages information entropy to quantify the information potential of low-attention tokens and integrates tokens scheduled for eviction into a continuous state space through entropy-guided state transitions, enabling their dynamic reactivation when semantic relevance arises during subsequent decoding. Extensive experiments on multimodal benchmarks demonstrate that RetentiveKV achieves 5.0 times KV cache compression and 1.5 times decoding acceleration. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_04075 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | RetentiveKV: State-Space Memory for Uncertainty-Aware Multimodal KV Cache Eviction Liu, Sihao Xiong, YuFan Jiang, Zhonghua Wang, Zhaode Zhang, chengfei lv Shengyu Machine Learning Artificial Intelligence Computation and Language Multimodal Large Language Models face severe challenges in computational efficiency and memory consumption due to the substantial expansion of the visual KV cache when processing long visual contexts. Existing KV cache compression methods typically rely on the "persistence of importance" hypothesis to prune tokens. However, this approach proves fragile in multimodal settings due to two key issues: 1) Visual tokens display "deferred importance," initially exhibiting low salience but becoming pivotal during later decoding, which can lead to premature eviction. 2) Discrete pruning disrupts the inherent spatial continuity of visual cues. To address these challenges, we propose RetentiveKV, an entropy-driven KV cache optimization method that reformulates KV eviction from "discrete context truncation" to "continuous memory evolution" based on State Space Models. Our method leverages information entropy to quantify the information potential of low-attention tokens and integrates tokens scheduled for eviction into a continuous state space through entropy-guided state transitions, enabling their dynamic reactivation when semantic relevance arises during subsequent decoding. Extensive experiments on multimodal benchmarks demonstrate that RetentiveKV achieves 5.0 times KV cache compression and 1.5 times decoding acceleration. |
| title | RetentiveKV: State-Space Memory for Uncertainty-Aware Multimodal KV Cache Eviction |
| topic | Machine Learning Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2605.04075 |