Arthur

Thanks for educating me on inheritance, I understand your use-case and how this can be confusing. The problem is that in order to keep the information of the content of...

Not really in favor of this, if the problem is with the way accelerate handles a for loop, should be solved in accelerate. cc @sgugger

Hey! Could you make sure the CI tests are green? Can review then!

Cool! Reviewing now

👍🏻 on @younesbelkada 's answer, on almost all of our attention modules, the attention should be passed and there are not reason to give them a default value because this...

Hey! Thanks will review now

Yes! Quick fix, use the slow tokenizer. Otherwise I'll open a PR to add template processing! Thanks for reporting!

This should have been fixed by #22959

When you say the same, what do you mean?

Few things here. Llama has no official model so make sure the one you are using is up to date and has the same eos token id for the model.config...