[LayoutLMv3] Add TensorFlow implementation

[ChrisFugl]

Adds a TensorFlow implementation of LayoutLMv3.

We have TensorFlow weights available that we have tested are able to produce the same results as the PyTorch models at microsoft/layoutlmv3-base and microsoft/layoutlmv3-large. Can you help us upload the TF weights to those locations?

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you read the contributor guideline,
  Pull Request section?
• Was this discussed/approved via a Github issue or the forum? Please add a link
  to it if that's the case.
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

@LysandreJik @NielsRogge

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[ChrisFugl]

@ChrisFugl the underlying issue is fixed -- can you rebase the PR please? 🙏

We should be able to merge the PR after that 🎉

@gante I followed your instructions on how to rebase, but I am afraid that it messed things up a bit. There are now 45 files changed in this PR - most of which this PR haven't even touched upon.

I resolved all conflicts while rebasing by consistently accepting the incoming changes. The tests still pass.

I think the problems with rebasing must be because I merged upstream/main into this PR prior to the first rebase.

Sorry about the rebasing issues. I am not sure what the right course of action is now?

[gante]

@ChrisFugl can you try the solutions from this StackOverflow page?

The commits will be squashed at merge time, so everything will be fine as long as the diff we see here is correct :)

[ChrisFugl]

@gante Ah that was simple :) Thanks! Should be all good now.

[gante]

@ChrisFugl Fantastic, I'll merge the PR now! 🧡

Thank you so much for participating through the whole process, even with all the hiccups. LayoutLMv3 is a very useful model, I'm confident TF users will deeply appreciate your contribution 🤗

There are two post-merge action points. As always, a helping hand is appreciated (but we will do them otherwise):

• Find and fix the tokenizer problem that is preventing the conversion of microsoft/layoutlmv3-base-chinese (and convert the model)
• [EDIT: we are going to write about it today, ~6pm CEST] Announce the TF model on social media (twitter and LinkedIn)! Unannounced changes are changes that most users will miss. [EDIT2: shared here -- https://twitter.com/joao_gante/status/1564650841688006656 ; https://www.linkedin.com/posts/gante_tensorflow-ai-microsoft-activity-6970415100406951936-dywm?utm_source=share&utm_medium=member_desktop]

[ChrisFugl]

Thank you @gante - and for your help with getting the PR merged in 😁 Super nice to get it merged in.

I can give a shot at the tokenizer problem 👌

Happy to help with sharing as well. Already shared the LinkedIn post. We will probably also do some more spreading of the news in my team :)