VisCache: Visual KV Cache Pruning for Efficient Vision Large Language Model Inference

arXiv:2608.24063 · cs.CV, cs.AI · Submitted 2026-08-25 · Read on arXiv

cs.CV, cs.AI

Submitted: 2026-08-25

Updated: 2026-08-25

Code: https://github.com/Wlklk/VisCache

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• KV cache — 85× in this paper · explained in 3 episodes
• Accuracy — 37× in this paper · explained in 1 episode
• Benchmark — 34× in this paper · explained in 2 episodes
• Pruning — 32× in this paper · explained in 4 episodes
• Redundancy — 22× in this paper · explained in 2 episodes
• Retention Ratio — 21× in this paper
• Quantization — 13× in this paper · explained in 12 episodes
• Transformer — 9× in this paper · explained in 2 episodes
• Ablation — 8× in this paper · explained in 1 episode
• Generalization — 4× in this paper · explained in 13 episodes
• Video Summarization — 4× in this paper
• Ablation study — 3× in this paper · explained in 1 episode
• Instruction Tuning — 3× in this paper · explained in 2 episodes
• Maximal Marginal Relevance — 3× in this paper
• Robustness — 3× in this paper · explained in 5 episodes
• Time To First Token — 3× in this paper
• Ablation Studies — 2× in this paper · explained in 1 episode
• Adaptation — 2× in this paper · explained in 1 episode
• Aggregation — 2× in this paper · explained in 1 episode
• Alignment — 2× in this paper · explained in 1 episode
• Arithmetic Progression — 2× in this paper
• Computational overhead — 2× in this paper · explained in 2 episodes
• Consistency — 2× in this paper · explained in 3 episodes
• Core — 2× in this paper · explained in 2 episodes
• Foundation Model — 2× in this paper · explained in 3 episodes
• Foundation Models — 2× in this paper · explained in 7 episodes
• Geometric Progression — 2× in this paper
• Multimodal Fusion — 2× in this paper · explained in 3 episodes
• Multimodal Reasoning — 2× in this paper · explained in 5 episodes
• Scout-based Temporal Redundancy Filtering — 2× in this paper
• Temporal Reasoning — 2× in this paper · explained in 2 episodes
• Action Sequence — 1× in this paper
• Contextual Understanding — 1× in this paper · explained in 2 episodes
• Contrastive Learning — 1× in this paper · explained in 5 episodes
• Cosine Similarity — 1× in this paper · explained in 3 episodes
• Edge Devices — 1× in this paper · explained in 2 episodes
• Embeddings — 1× in this paper · explained in 2 episodes
• Fine-Tuning — 1× in this paper · explained in 5 episodes
• Grounding — 1× in this paper · explained in 3 episodes
• Hallucination — 1× in this paper · explained in 4 episodes
• In-Context Learning — 1× in this paper · explained in 4 episodes
• Large Language Model (LLM) — 1× in this paper · explained in 3 episodes
• Modality Gap — 1× in this paper · explained in 4 episodes
• Model-Agnostic — 1× in this paper · explained in 2 episodes
• Multi-Agent Framework — 1× in this paper · explained in 3 episodes
• Optimal Transport — 1× in this paper · explained in 5 episodes
• Parabolic Formulation — 1× in this paper
• Scaling Laws — 1× in this paper · explained in 4 episodes
• Semantic Alignment — 1× in this paper · explained in 3 episodes
• Sparsity — 1× in this paper · explained in 4 episodes
• Training-Free Framework — 1× in this paper · explained in 2 episodes
• Transformer Models — 1× in this paper · explained in 2 episodes
• Vision-language models — 1× in this paper · explained in 2 episodes
• Visual Question Answering — 1× in this paper

Sources

• Qwen Technical Report
• Qwen3-VL Technical Report
• Qwen2.5-VL Technical Report
• PyramidKV: Dynamic KV Cache Compression based on Pyramidal Information Funneling
• Palu: Compressing KV-Cache with Low-Rank Projection
• What do Vision Transformers Learn? A Visual Exploration
• ReFoCUS: Reinforcement-guided Frame Optimization for Contextual Understanding
• LLaVA-OneVision: Easy Visual Task Transfer
• LLaVA-NeXT-Interleave: Tackling Multi-image, Video, and 3D in Large Multimodal Models
• LaCo: Efficient Layer-wise Compression of Visual Tokens for Multimodal Large Language Models
• ZSMerge: Zero-Shot KV Cache Compression for Memory-Efficient Long-Context LLMs
• KIVI: A Tuning-Free Asymmetric 2bit Quantization for KV Cache
• HoliTom: Holistic Token Merging for Fast Video Large Language Models
• LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understanding
• ShadowKV: KV Cache in Shadows for High-Throughput Long-Context LLM Inference
• FlatQuant: Flatness Matters for LLM Quantization
• TokenCarve: Information-Preserving Visual Token Compression in Multimodal Large Language Models
• MEDA: Dynamic KV Cache Allocation for Efficient Multimodal Long-Context Inference
• PrefixKV: Adaptive Prefix KV Cache is What Vision Instruction-Following Models Need for Efficient Generation
• Tarsier: Recipes for Training and Evaluating Large Video Description Models

Related papers

• Loss Knows Best: Detecting Annotation Errors in Videos via Loss Trajectories
• AnchorWeave: World-Consistent Video Generation with Retrieved Local Spatial Memories
• Benchmarking the Robustness of Foundation Models for Mammography under Domain Shift
• MambaX-Net: Dual-Input Mamba-Enhanced Cross-Attention Network for Longitudinal MRI Segmentation
• TeleOCR: Navigating Document Parsing Across Digital and Camera-Captured Documents
• A Survey on Efficient Vision-Language-Action Models