[LoRA] use removeprefix to preserve sanity.

[sayakpaul]

What does this PR do?

We have multiple recurrences of using the following patterns in our LoRA codebase:

• k.replace(f"{prefix}.", "")
• k[len(f"{prefix}.") :]

It's just mind-twisting sometimes and I think removeprefix() just makes it easier.

I have run the integration tests and ensured they are passing.

[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.

[BenjaminBossan]

LGTM, thanks.

Just for other reviewers: The change is not equivalent, as this will strictly check for prefixes, i.e. the new code corresponds to k[len(f"{prefix}.") :] if k.startswith(prefix) else k. Also, removeprefix was introduced in Python 3.9, but 3.8 is EOL so this should be fine.

[sayakpaul]

The change is not equivalent, as this will strictly check for prefixes, i.e. the new code corresponds to k[len(f"{prefix}.") :] if k.startswith(prefix) else k.

@BenjaminBossan also to be clear that in our codebase prefix is quite exclusively used to handle what you meant. So, IMO, removeprefix() is a better alternative to tackle that. Am I mistaken to think that?

[BenjaminBossan]

So, IMO, removeprefix() is a better alternative to tackle that. Am I mistaken to think that?

No, I agree, I just wanted to highlight the difference in case there is an example where the "prefix" is actually allowed to be in the middle of the string.

[sayakpaul]

For the purposes of this PR and for the use of prefix in the LoRA codebase, I think we're safe there.

[a-r-r-o-w]

Just some nits. Personal preference is to use f-strings but either is okay

[sayakpaul]

Fix for the failing tests #11504.