[docs] Update index and quicktour

[stevhliu]

Starts implementing some of the things we discussed in our docs plan:

• clarify the difference between using AutoPeftModel (inference) and PeftConfig + get_peft_model (train) and splits the quicktour more generally along these lines
• add API reference for AutoPeftModel classes
• cleans up the index with a Space for discovering what models/methods are supported given a task (let me know if you think this is easier and cleaner)
• create separate subsection for LoRA guides

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint.

[BenjaminBossan]

Thanks, Steven, this looks like it will help give a better structure to the PEFT docs and make the intro easier to understand for new users.

I left a couple of comments, please take a look.

[BenjaminBossan]

I have no idea about iframes, but can the height be dynamic, so that we don't have to scroll for longer lists?

[stevhliu]

Tried a couple things but I couldn't get it to work. I just set a larger height for now

[BenjaminBossan]

Nice space. Is it somehow updated automatically or would it need to be updated each time we add a new model?

Also, I'd reword the description a little bit:

Discover what models and PEFT methods are supported for a given task

This could give the impression that other models don't work at all. However, almost all models work, if they're not supported, it's only a matter of setting the config correctly. I think this is important to highlight, as some other similar libraries are actually only working with specific model architectures.

[stevhliu]

Currently, the underlying dataset would need to be updated each time a new model is added or if a new method is supported. Let me know if you have any suggestions for automating it though, that'd be nice!

[BenjaminBossan]

Technically, it would be possible. For example, we could set up a cron job or GitHub CI to parse the README (assuming that should be the single source of truth) with pandoc, extract the table, and create an update the on the dataset. But that seems like overkill for now. Probably we, the maintainers, just need to remember to update the dataset each time a new model is supported.

[BenjaminBossan]

Thanks for improving the docs, LGTM. I think it would be good to have another approval by Sourab or Younes before merging.

[younesbelkada]

Impressive work @stevhliu ! Thanks a lot for this great work