MusicGen (#1020)

* Add MusicGen model

* add benchmarks

* change to from_pretrained

* symlinks

* add readme and requirements

* fix readme

* readme

musicgen/README.md

@@ -0,0 +1,31 @@
+# MusicGen
+
+An example of Meta's MusicGen model in MLX.[^1] MusicGen is used to generate
+music from text descriptions.
+
+### Setup
+
+Install the requirements:
+
+```
+pip install -r requirements.txt
+```
+
+### Example
+
+An example using the model:
+
+```python
+import mlx.core as mx
+from music_gen import MusicGen
+from utils import save_audio
+
+model = MusicGen.from_pretrained("facebook/musicgen-medium")
+
+audio = model.generate("happy rock")
+
+save_audio("out.wav", audio, model.sampling_rate)
+```
+
+[^1]: Refer to the [arXiv paper](https://arxiv.org/abs/2306.05284) and
+  [code](https://github.com/facebookresearch/audiocraft/blob/main/docs/MUSICGEN.md) for more details.