Update docs around audio and vision

[stevhliu]

As part of the strategy to center the docs around the different modalities, this PR updates the quickstart to include audio and vision examples. This improves the developer experience by making audio and vision content more discoverable, enabling users working in these modalities to also quickly get started without digging too deeply into the docs.

Other changes include:

• Moved the installation guide to the Get Started section because it should be part of a user's onboarding to the library before exploring tutorials or how-to's.
• Updated the native TF code at creating a tf.data.Dataset because it was throwing an error. The to_tensor() bit was redundant and removing it fixed the error (please double-check me here!).
• Added some UI components to the quickstart so it's easier for users to navigate directly to the relevant section with context about what to expect.
• Reverted to the code tabs for content that don't have any framework-specific text. I think this saves space compared to the code blocks. We'll still use the code blocks if the torch text is different from the tf text.

Let me know what you think, especially if we should include some code samples for training a model in the audio/vision sections. I left this out since we already showed it in the NLP section. I want to keep the focus on using Datasets to load and process a dataset, and not so much the training part. Maybe we can add links to the Transformers docs instead?

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[mariosasko]

Love the changes!

[mariosasko]

Let me know what you think, especially if we should include some code samples for training a model in the audio/vision sections. I left this out since we already showed it in the NLP section. I want to keep the focus on using Datasets to load and process a dataset, and not so much the training part. Maybe we can add links to the Transformers docs instead?

We plan to address this with end-to-end examples (for each modality) more focused on preprocessing than the ones in the Transformers docs.

[lhoestq]

Awesome thanks !

Let me know what you think, especially if we should include some code samples for training a model in the audio/vision sections. I left this out since we already showed it in the NLP section.

I'd add the conversion to pytorch DataLoader and TF Dataset as well for audio and vision, if it's not too much information. But I think the training loop itself needs to be part of an end-to-end example as @mariosasko suggested.

As soon as there is a pytorch DataLoader or a TF Dataset, it means we're ready for training, I don't think it's necessary to show the complexity of one particular task/model and how it's trained in the quickstart.

Maybe the quickstart can redirect to its corresponding end-to-end example for those who would like to see a complete example with more context ? Something like "Want to see more in a concrete example ? See how a dataset can be prepared for speech recognition with transformers, etc."

[lhoestq]

Love the links you provided ! Later I think we can provide end-to-end examples in the datasets doc itself using a super simpler training loop instead of redirecting to the transformers doc that uses the Trainer

My last comments:

[lhoestq]

Color gradients look all good now, thanks !

CI failures are unrelated to this PR btw - you can ignore them.

Feel free to merge if it's all good for you @stevhliu