Demo UI vs Inference in code

#24
by WoutDeRijck - opened

When I take an image (scan) of a document and just simply prompt to put everything to markdown, the difference is remarkable between the demo UI and inference in code.

So I am wondering, the files that I find on this hf repo, are these files that are used by the demo? Or what would be giving this difference in performance?

When I clone the hf repo and take do AutoModel.from_pretrained to the local model, I get the same results as the web demo. Just for my information, where exactly is the difference?

Sign up or log in to comment