mac Apple Silicon MPS device support

[jcy0720]

What would you like to report?

Dear fairchem team,

I, as a Apple Silicon (M2) mac user, cannot help but use the option device="cpu", as the cuda support for mac has came to an end.

However, Apple provides another way to utilize GPUs, called MPS.

Fortunately, torch seems to support the mps device as its backend device by device = torch.device("mps"), and the availability check commands such as torch.backends.mps.is_built() and torch.backends.mps.is_available().

Also, there already has been an attempt to raise a related issue before:
#815
and a closed pull request:
#1542

Do I have any immediate workaround for using mps device? If not, could you consider adding the mps device setting for the model inference?

Thank you for this great project and the free-distribution of the model.

[rayg1234]

Hi, sorry about the inconvenience, we haven't been able to prioritize mac support due to resource constraints on our side. If you can make an argument for your use case or if many others in the community really want this feature, that would help us consider supporting it in the future. Or alternatively feel free to contribute to the codebase if you can make it work!