Such nice work! 1.3B with top accuracy and speed. That's insane. I'll throw my table-heavy pdfs into it to see how it works. But from the benchmark, I saw that all document parsing models didn't perform really well in terms of old scans. It seems like this is a far-from-solved problem.
Such nice work! 1.3B with top accuracy and speed. That's insane. I'll throw my table-heavy pdfs into it to see how it works. But from the benchmark, I saw that all document parsing models didn't perform really well in terms of old scans. It seems like this is a far-from-solved problem.
Author of OXR here. Happy to answer any quetion here!