improve(llama): Faster apply_rotary_pos_emb

[fpgaminer]

What does this PR do?

Faster implementation for apply_rotary_pos_emb in modeling_llama.py.

Please see issue #22683 for code that verifies the correctness of the change.

NOTE: Not marking as fixing the above issue, as speed is still not as good as before.

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.

@gante

[HuggingFaceDocBuilderDev]

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

[gante]

Thanks for improving the performance @fpgaminer 🙏

@amyeroberts for context, this comment shows that a) it gets exactly the same numerical output b) it is faster than the previous version

[amyeroberts]

🔥 🔥 🔥 - thanks for updating and for the time to validate and benchmark 🙏

[neggert]

Should a similar patch be applied to GPT-NeoX?

[amyeroberts]

@neggert I believe it can be added to GPT-NeoX too - very happy to review a PR if you'd like to add!