Prepare for release.

[Narsil]

What does this PR do?

Fixes # (issue)

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.

[vrdn-23]

Thanks for the awesome work @Narsil !
Quick question: Would it be possible for us to update to the latest candle version for this release? Are there any blockers that might be preventing us from using the latest release of candle?

[Narsil]

Well it will likely take time, given we're 3 minors behind, and we have some modifications on top of core, but we'll try to update it too. Depending on difficulty it may land in this release or later.

Any specific reason you want to update candle version ?

[vrdn-23]

I was hoping that with the latest candle ecosystem now supporting the Deberta-V2 model (huggingface/candle#2743), this would enable TEI to be able to support the model sooner?
There are more than a few issues that crop up every now and then regarding support for the Deberta-V2 model, so I assumed maybe this was a blocker for moving ahead with that.

Let me know if you have any insight on this or if I'm missing something obvious here.

[Narsil]

I was hoping that with the latest candle ecosystem now supporting the Deberta-V2 model (huggingface/candle#2743), this would enable TEI to be able to support the model sooner?

Not at all, TEI uses it's own modeling because having actual fast layers is important (where raw candle focuses on simplicity). It's easier to port from candle than transformers, but it's still a port regardless, so the actual candle version doesn't matter.

I'm quite confused though deberta models seem to be fill masks, any link to deberta being finetuned for embeddings (to have a usable reference frame) ?

[vrdn-23]

I'm quite confused though deberta models seem to be fill masks, any link to deberta being finetuned for embeddings (to have a usable reference frame) ?

Apologies for not making it clearer @Narsil. The requests for the deberta-v2 model is so that we can access the classification/re-ranking variants of it.
(rel #199, #281, #354, #221).