Fetching the paper…

Treat Visual Tokens as Text? But Your MLLM Only Needs Fewer Efforts to See · Around