Feature requests

[bennmann]

Please implement all the features outlined in the official Musicians Guide or provide a roadmap:

https://github.com/ace-step/ACE-Step-1.5/blob/main/docs/en/ace_step_musicians_guide.md

Also, if you can make multi-GPU work for the DiT, your repo will cater to many more users as bigger ACE-step models are released. Could get you a lot more stars.

[ServeurpersoCom]

Hi
Check out #6, you can help me test it :)
More atomic PRs would be appreciated, there's quite a lot added at once and I'm working solo.
For multi-GPU, I don't have a setup to test it so I can't implement it for now. Besides, when bigger models arrive we'll have time to revisit.
Since ACE-Step's goal is to democratize audio models and we're already bringing quantization, I doubt multi-GPU is urgent right now. Trying to stay KISS so I can keep everything under control.

[ServeurpersoCom]

Roadmap added

[ServeurpersoCom]

All modes are supported The engine is finished; all that remains is the user-friendly part.

[ServeurpersoCom]

The "lego" and "complete" modes are not very efficient due to the original DiT.