I didn't see a link to the code, but there's a GitHub for it. https://github.com/Ucas-HaoranWei/GOT-OCR2.0

(Edit: I found it, it was at the top of the paper, the Arxiv html view formatted it weird on mobile)

I ran the demo on their own paper and it went off the rails on page 7. I’m not sure if it is related to the nearby LaTeX or if it is some token limit.