ECOKV: Geometry-Aware KV Cache Eviction via Complementary Diversity Metrics

• URL: http://arxiv.org/abs/2609.06663v1
• Date: Sun, 06 Sep 2026 15:13:48 GMT
• Title: ECOKV: Geometry-Aware KV Cache Eviction via Complementary Diversity Metrics
• Abstract summary: multimodal Large Language Models (MLLMs) excel in diverse tasks, but their scalability remains limited by the memory and computational overhead of KV cache storage.<n>Recent KV cache eviction approaches incorporate a cosine similarity-based diversity metric with importance metrics to selectively retain key-value pairs.<n>We propose a geometry-aware composite metric that jointly leverages Euclidean distance and cosine similarity to capture token diversity from complementary perspectives.
• Score: 15.835203260282746
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Although multimodal Large Language Models (MLLMs) excel in diverse tasks, their scalability remains limited by the memory and computational overhead of KV cache storage. Recent KV cache eviction approaches incorporate a cosine similarity-based diversity metric with importance metrics to selectively retain critical key-value pairs. However, cosine similarity involves normalization that discards magnitude information, and it often yields uniformly high similarity values across layers due to the anisotropy property of hidden representations. In our study ECOKV, we rigorously deconstruct the capabilities of existing diversity metrics. Moving beyond simple measurement, we propose a geometry-aware composite metric that jointly leverages Euclidean distance and cosine similarity to capture token diversity from complementary perspectives. Furthermore, we use these two metrics to estimate the redundancy level of each attention head, allowing adaptive weighting between diversity and importance scores during token selection. Finally, we demonstrate that the observation window commonly employed to preserve recent tokens can be substantially reduced, thereby allocating more cache capacity to informative tokens and yielding consistent improvements. Extensive experiments demonstrate that ECOKV achieves state-of-the-art performance under various compression ratios and can be seamlessly integrated with existing KV cache eviction methods. We further analyze the relationship between importance and diversity, and examine redundancy patterns across layers and attention heads.

Related papers

• CateKV: On Sequential Consistency for Long-Context LLM Inference Acceleration [67.17709877327232]
  Cate KV is a hybrid KV cache method that retains only critical token information for consistent heads.<n>Cate KV reduces memory usage by up to $2.72times$ and accelerates decoding by $2.18times$ in single-sample inputs.
  arXiv  Detail & Related papers  (2026-08-31T06:02:37Z)
• Centering before Pruning: Lightweight Geometry Correction for Diversity-Based Visual Token Pruning in LVLMs [7.371189496638082]
  We find that similarities between raw visual tokens are strongly concentrated in the positive range, limiting their ability to distinguish non-redundant tokens.<n>We propose the textbfCentered Geometry textbfPruner (Cen-Prune), which measures subset diversity using centered cosine similarity.<n>Cen-Prune provides robust improvements in overall performance across existing diversity-based pruners.
  arXiv  Detail & Related papers  (2026-08-31T05:13:29Z)
• DivRL: Disentangled Self-Similarity Rewards for Diverse Subject-Driven Generation [52.23323703015865]
  We propose a post-training framework called DivRL that jointly optimize identity consistency and structural diversity simultaneously.<n>We introduce a Negative Self-Similarity Measure (nSSM) to quantify structural diversity, and Visual Semantic Matching (VSM) to evaluate identity consistency.
  arXiv  Detail & Related papers  (2026-06-22T21:22:42Z)
• Multi-label Classification with Panoptic Context Aggregation Networks [61.82285737410154]
  This paper introduces the Deep Panoptic Context Aggregation Network (PanCAN), a novel approach that hierarchically integrates multi-order geometric contexts.<n>PanCAN learns multi-order neighborhood relationships at each scale by combining random walks with an attention mechanism.<n>Experiments on NUS-WIDE, PASCAL VOC,2007, and MS-COCO benchmarks demonstrate that PanCAN consistently achieves competitive results.
  arXiv  Detail & Related papers  (2025-12-29T14:16:21Z)
• CommonKV: Compressing KV Cache with Cross-layer Parameter Sharing [54.34080239841088]
  CommonKV is a training-free method for cross-layer KV cache compression through adjacent parameters sharing.<n>We show that the proposed method consistently outperforms existing low-rank and cross-layer approaches at various compression ratios.
  arXiv  Detail & Related papers  (2025-08-22T06:55:45Z)
• Homogeneous Keys, Heterogeneous Values: Exploiting Local KV Cache Asymmetry for Long-Context LLMs [28.554617623874176]
  We show a fundamental yet previously overlooked asymmetry in KV caches.<n>Key-value asymmetry reveals a critical limitation in existing compression methods.<n>We propose a training-free compression framework (AsymKV) that combines homogeneity-based key merging with a mathematically proven lossless value compression.
  arXiv  Detail & Related papers  (2025-06-04T16:10:44Z)
• More Tokens, Lower Precision: Towards the Optimal Token-Precision Trade-off in KV Cache Compression [71.42818367729573]
  In large language models (LLMs), the memory usage of KV cache has become a critical bottleneck during inference.<n>The mainstream KV compression methods, including KV pruning and KV quantization, primarily focus on either token or precision dimension separately.<n>In this paper, we comprehensively investigate the token-precision trade-off in KV cache compression.
  arXiv  Detail & Related papers  (2024-12-17T09:20:31Z)
• IME: Integrating Multi-curvature Shared and Specific Embedding for Temporal Knowledge Graph Completion [97.58125811599383]
  Temporal Knowledge Graphs (TKGs) incorporate a temporal dimension, allowing for a precise capture of the evolution of knowledge. We propose a novel Multi-curvature shared and specific Embedding (IME) model for TKGC tasks. IME incorporates two key properties, namely space-shared property and space-specific property.
  arXiv  Detail & Related papers  (2024-03-28T23:31:25Z)
• Regularizing Variational Autoencoder with Diversity and Uncertainty Awareness [61.827054365139645]
  Variational Autoencoder (VAE) approximates the posterior of latent variables based on amortized variational inference. We propose an alternative model, DU-VAE, for learning a more Diverse and less Uncertain latent space.
  arXiv  Detail & Related papers  (2021-10-24T07:58:13Z)
• Beyond Single Instance Multi-view Unsupervised Representation Learning [21.449132256091662]
  We impose more accurate instance discrimination capability by measuring the joint similarity between two randomly sampled instances. We believe that learning joint similarity helps to improve the performance when encoded features are distributed more evenly in the latent space.
  arXiv  Detail & Related papers  (2020-11-26T15:43:27Z)
• Learning Flat Latent Manifolds with VAEs [16.725880610265378]
  We propose an extension to the framework of variational auto-encoders, where the Euclidean metric is a proxy for the similarity between data points. We replace the compact prior typically used in variational auto-encoders with a recently presented, more expressive hierarchical one. We evaluate our method on a range of data-sets, including a video-tracking benchmark.
  arXiv  Detail & Related papers  (2020-02-12T09:54:52Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.