Changing the text generation method.

#4
by Prabhav55221 - opened

Hi Team,

I have been using this implementation of DONUT for a use case regarding invoices. First of all, thank you for this codebase - It is really helpful.

However, I was wondering if there a provision to change the text generator model being used by this pipeline. Is there a particular reason that options like GPT2 were not used to generate the text from image?

Owner

Hi,

This was merely a finetune. I think your question is better directed to the authors of the original Donut paper.

to-be changed discussion status to closed

Sign up or log in to comment