HiRED: Attention-Guided Token Dropping for Efficient Inference of High-Resolution Vision-Language Models | Read Paper on Bytez