Fix cache_position initialisation for generation with use_cache=False

[nurlanov-zh]

What does this PR do?

Fixes #30482

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?

@ArthurZucker @younesbelkada @gante

[ArthurZucker]

Thanks! Let's maybe avoid relying on the model's cache position initialization !

[gante]

LGTM, thank you for the fix 👍

@ArthurZucker tbh, when we have use_cache=False, we shouldn't even be creating or using cache_positions at all (as the name of the variable indicates, its purpose is cache-related). ATM, the only dependency is the causal mask creation function, which uses cache_positions as a proxy for the sequence length (i.e. we can easily rearrange the code to avoid using it when there is no cache :) )

[ArthurZucker]

Yes @gante, agreed would split both paths a bit better

[ArthurZucker]

Can you just rebase on main? 🤗

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

[nurlanov-zh]

Can you just rebase on main? 🤗

@ArthurZucker , done ✔️

[ArthurZucker]

Thanks 🤗