Instructions to run on Mac

[fire17]

Hi there , really excited about open source music gen :)
Thanks for the awesome work!

Was wondering how to run this on a Mac (currently got the m3 max)
Couldn't find instructions in the readme

Thanks a lot and all the best

Ps - also can't wait to see music gen speed getting faster that RT - ie 30 seconds of audio in less than 30 seconds.. or just a pure Omni RealTime music model, to generate music continuesly, like a player, or dj
I'm betting on a year or less haha, that would be epic

[a43992899]

Related issue: #2

I need to export the model to llama.cpp, implement the audio tokenizer in cpp.

[thinkyhead]

It appears the infer.py script will need to be extended to use MPS where it currently only uses CUDA. Not all CUDA features are available on MPS, so there may be some blocks there. Parts of TorchAudio are also dependent on CUDA for speed, so some parts of the generation may just have to be run on CPU. If it can be made to work by simple substitution of "mps" for "cuda" in the script, that would be great! But, it won't be that easy, and we need some smarties to come along to help.