We have set up an instance of eScriptorium that's available online, and we've found a collection of images that might be relevant to our question. Now we're training an OCR model on those images in eScriptorium.

The site is live at https://artdata.pythonanywhere.com/ !