Update subcore.py

[harmeetsingh-work]

Making it compatible with apple silicon.

[ltdrdata]

The patch affects not only the functionality used in subpack but also impacts all instances of torch.load across ComfyUI.

It seems appropriate to conditionally call with map_location only when it is silicon.

https://github.com/ltdrdata/ComfyUI-Impact-Subpack/blob/74db20c95eca152a6d686c914edc0ef4e4762cb8/modules/subcore.py#L139C5-L139C18

[harmeetsingh-work]

@ltdrdata the code change:

device = torch.device("mps" if torch.backends.mps.is_available() else "cuda" if torch.cuda.is_available() else "cpu")
orig_torch_load = torch.load(map_location=device)

It loads 'mps' if mps(apple silicon) is available else 'cuda' if cuda is available else it will use 'cpu'.

[ltdrdata]

It loads 'mps' if mps(apple silicon) is available else 'cuda' if cuda is available else it will use 'cpu'.

If it is not MPS, no device should be applied. It should be determined by the callsite.
Therefore, it should not be applied to orig_torch_load.