STAR: Stage-Wise Attention-Guided Token Reduction for Efficient Large Vision-Language Models Inference

VLM
Re-assign community