Add support for Llama3.3

[jorgeantonio21]

https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct

[zackangelo]

There were no architectural changes afaik, it should already work with the existing example, just change the repo URL.

[theHausdorffMetric]

I added the respective repo url to the llama example and created a pull request:
#2677

[LaurentMazare]

Were you able to try it out? I don't think it would fit in the memory of a single gpu so this would be better suited for llama_multiprocess that can use multiple gpus. Would be great if someone can give it a spin there and check that it works.

[zackangelo]

Ah, you're right @LaurentMazare.

I don't believe the multiprocess example has been updated to create the new rope tensors that were introduced in Llama 3.1. But it should work otherwise.