I tried parsing a few PDFs with small version but it gave MISSING_PAGE on some of the pages, while the base model was able to convert the pages perfectly.
So adding this support for model selection at cli level to help both myself and others. :)
Sounds good.
I'm a bit worried about backward compatibility though. Maybe instead of redownloading the small model, we can check if the old format is present and then move it to the correct spot?
This should close: https://github.com/facebookresearch/nougat/issues/59
I tried parsing a few PDFs with small version but it gave
MISSING_PAGE
on some of the pages, while the base model was able to convert the pages perfectly.So adding this support for model selection at cli level to help both myself and others. :)