docs: Musicgen melody model card

[AshAnand34]

What does this PR do?

This pull request updates the documentation for the MusicGen Melody model in docs/source/en/model_doc/musicgen_melody.md. The changes aim to simplify and enhance the clarity of the documentation by restructuring the content, adding examples, and improving formatting.

Documentation Improvements

Overview and Structure

• Reorganized the content to provide a concise overview of the MusicGen Melody model, highlighting its key features and differences from the original MusicGen.
• Simplified the explanation of the model's architecture, breaking it down into three main components: text encoder, MusicGen Melody decoder, and audio decoder.

Examples and Usage

• Replaced lengthy code snippets with streamlined examples for generating music using text and audio prompts, including text-only and unconditional generation scenarios.
• Added detailed examples for using tools like Demucs for melody isolation and quantization techniques for memory optimization.

Formatting and Accessibility

• Introduced a sidebar navigation and collapsible sections for better readability and user experience.
• Updated links to external resources and added tooltips for key concepts like quantization and guidance scale.

#36979

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).

Who can review?

Anyone in the community is free to review the PR once the tests have passed. Feel free to tag
members/contributors who may be interested in your PR.

@stevhliu

[stevhliu]

Good start on this big doc!

[stevhliu]

Suggested change

	[MusicGen Melody](https://huggingface.co/papers/2306.05284) is a single-stage, auto-regressive Transformer model designed for high-quality music generation, conditioned on both text and audio prompts. Unlike its predecessor, MusicGen Melody uses the audio prompt as a direct melodic guide, allowing for more precise control over the generated music.
	[MusicGen Melody](https://huggingface.co/papers/2306.05284) builds on top of the [MusicGen](./musicgen) model by adding a melody-guided generation approach to enable more controllable audio generation. The model is conditioned on both input text and chromagram. A chromagram better captures the harmonic and melodic features of music.
	Unlike MusicGen, MusicGen Melody uses the audio prompt as a conditional signal for the generated audio sample and the conditional text and audio signals are concatenated to the decoder's hidden states.

[stevhliu]

Suggested change

	You can find all the original [MusicGen Melody](https://huggingface.co/models?sort=downloads&search=facebook%2Fmusicgen) checkpoints on the Hugging Face Hub.
	You can find all the original MusicGen Melody checkpoints under the [AI at Meta](https://huggingface.co/facebook/models?search=musicgen-melody) organization.

[stevhliu]

Suggested change

	> [!TIP]
	> Click on the MusicGen Melody models in the right sidebar for more examples of how to apply the model to various music generation tasks.
	> [!TIP]
	> This model was contributed by [ylacombe](https://huggingface.co/ylacombe).
	>
	> Click on the MusicGen Melody models in the right sidebar for more examples of how to apply the model to various music generation tasks.

[stevhliu]

Suggested change

	The example below demonstrates how to generate music conditioned on an audio melody and a text description using the [`AutoModel`] class.
	The example below demonstrates how to generate music with [`Pipeline`] or the [`AutoModel`] class.

[stevhliu]

import torch
from transformers import pipeline

pipeline = pipeline("text-to-audio", model="facebook/musicgen-melody", device=0, torch_dtype="auto")
pipeline("80s pop track with bassy drums and synth")

[stevhliu]

Suggested change

	**Unconditional Generation**
	- The example below demonstrates unconditional generation.

[stevhliu]

Suggested change

	**Generation Configuration**
	You can inspect and update the model's generation configuration.
	- The generation config stores the default parameters that control the generation process such as sampling, guidance scale, and number of generated tokens.
	Any arguments passed to the [`~GenerationMixin.generate`] method supersedes the parameters in the generation config.

[stevhliu]

Suggested change

	### Other Information
	- **Checkpoint Conversion**: Convert original checkpoints using the script at `src/transformers/models/musicgen_melody/convert_musicgen_melody_transformers.py`.
	- **`head_mask`**: The `head_mask` argument is only effective with `attn_implementation="eager"`.
	- **Sampling**: For best results, use sampling (`do_sample=True`).
	- The `head_mask` argument is only effective with `attn_implementation="eager"`.
	- For best results, set `do_sample=True`.

[stevhliu]

Remove this section and replace with the below. Remember each code snippet should be indented under its list item

Suggested change

	## Model Structure
	- [`MusicgenMelodyForCausalLM`] can be used as a standalone decoder model. Load it by specifying the correct config or accessing it through the `.decoder` attribute of [`MusicgenMelodyForConditionalGeneration`].
	[`MusicgenMelodyForConditionalGeneration`] can be used as a composite model that includes the text and audio encoder.

[stevhliu]

Add a few more notes:

- Ensure you're using a 32kHz checkpoint of the Encodec model because MusicGen was trained on it.