Add support for Apple Silicon (torch.mps)

[kiyokiku]

This PR adds support for Apple Silicon backend (torch.mps).

• Added support for torch.mps, enabling execution on Apple Silicon devices.
• Updated the device detection order from (cuda -> cpu) to (cuda -> mps -> cpu).

This PR is expected to resolve #187.

[kiyokiku]

@microsoft-github-policy-service agree

[zhenhuaplan]

Has anyone successfully used the GPU of the Apple M1 self-developed chip in the MacBook? I still can't use GPU acceleration properly here

[kiyokiku]

@zhenhuaplan

Could you run the following Python script and let me know the output?

import torch

print("torch.backends.mps:", torch.backends.mps.is_available())
print("torch.mps:", torch.mps.is_available())

On my environment, the output is as follows:

>>> print("torch.backends.mps:", torch.backends.mps.is_available())
torch.backends.mps: True
>>> print("torch.mps:", torch.mps.is_available())
torch.mps: True

[zhenhuaplan]

@zhenhuaplan

Could you run the following Python script and let me know the output?

import torch

print("torch.backends.mps:", torch.backends.mps.is_available())
print("torch.mps:", torch.mps.is_available())

On my environment, the output is as follows:

>>> print("torch.backends.mps:", torch.backends.mps.is_available())
torch.backends.mps: True
>>> print("torch.mps:", torch.mps.is_available())
torch.mps: True

@kiyokiku I am now using torch==2.4.0 instead of torch==2.6.0, and now I can use mps, torch.backends.mps.is_available() results in True, torch.mps.is_available() results in an error, This is because torch has no mps attribute

All in all, I can now use the GPU for acceleration, thanks

[kiyokiku]

For compatibility, I have replaced torch.mps with torch.backends.mps.