Add the Bamba Model

[fabianlim]

What does this PR do?

This PR merges the BambaModel, which is a hybrid mamba2 architecture with SwiGLU. The checkpoints are jointly trained by IBM, Princeton, and UIUC.

The implementation is based off ai21labs/Jamba-v0.1 and the mamba2 implementation ported over to HF for the codestral model.

cc: @ani300, @raghukiran1224

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?

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

[Rocketknight1]

Hi @fabianlim, do you have a paper reference for this model or any details on the trained checkpoints?

[fabianlim]

@Rocketknight1 thanks for reaching out. Yes my colleagues are preparing a paper and a GitHub repo with the (training) code. And checkpoints will be 1.8T, 2T, 2.2T, and an sft model. We will update the PR accordingly.

cc: @raghukiran1224

[raghukiran1224]

The data used is all open, we plan to share any and all details on what the community would want! Open source is the name of the game 😄

[Rocketknight1]

Cool! @molbap will be the point of contact at Hugging Face for this PR, so feel free to ping me or him if you have any questions as you're working on it.