Multimodal large language models (MLLMs) process hundreds or thousands of visual tokens per image, incurring prohibitive inference costs. While existing visi…
机构:新加坡国立大学
来源:arXiv 2609.10346 | AI4Papers 论文推荐平台