STaR-KV: Spatio-Temporal Adaptive Re-weighting for KV Cache Compression in GUI Vision-Language Models
2026-06-01 • Computer Vision and Pattern Recognition
Computer Vision and Pattern RecognitionArtificial Intelligence
AI summaryⓘ
The authors address a problem with memory use in vision-language GUI agents, where the memory needed grows too large as the system processes more interactions. They found that previous ways to compress this memory made wrong assumptions about how the attention works in the model. To fix this, they created STaR-KV, a new method that adjusts how important information is weighted both in space and over time without needing extra training. This approach reduces memory use by about 40% while keeping accuracy high and adding almost no extra computation. Their method was tested on multiple benchmarks and outperformed existing compression techniques.
Vision-Language ModelsGraphical User Interface (GUI)Key-Value (KV) CacheMemory CompressionAttention MechanismSpatial Mutual InformationEntropy TemperatureTemporal StabilityGPU MemoryKV Cache Compression
Authors
Yuhang Han, Wenzheng Yang, Yujie Chen, Xiangqi Jin, Yaojie Zhang, Siteng Huang, Linfeng Zhang
Abstract
Vision-language-model-based graphical user interface (GUI) agents have shown broad automation capabilities, yet deployment is bottlenecked by a key-value (KV) cache that grows linearly with interaction steps. For instance, UI-TARS-1.5-7B consumes 76 GB of GPU memory on merely five screenshots, approaching the capacity of mainstream 80 GB accelerators. Existing KV compression methods share two structural assumptions: aggregating visual-token importance into a single shared saliency map, and applying a fixed top-B cutoff to the fused score distribution. Pilot measurements refute both: spatial specialization lives at the attention-subspace level and migrates across layers, while the score distribution drifts in shape along a trajectory. We propose STaR-KV (Spatio-Temporal Adaptive Re-weighting), a training-free KV cache compression framework that calibrates token importance along three axes: (i) subspace-aware scoring driven by online spatial mutual information; (ii) a temporal stability discount that suppresses redundant cache entries from persistently attended subspaces; and (iii) an entropy-derived temperature that adaptively reshapes the score distribution. Across four GUI benchmarks, STaR-KV achieves the strongest average accuracy among state-of-the-art KV compression methods (e.g., GUIKV, SnapKV) at matched budgets, with no compression-stage FLOPs overhead (-0.07%) and cutting peak GPU memory by nearly 40% at a 20% KV-cache budget. Code is available at https://github.com/kawhiiiileo/STaR-KV.