Request: Add Metal GPU Apple Silicon Support

[Runtrons]

It would be fantastic to see support for Metal GPU on Apple Silicon in Wan2.1. This enhancement could significantly improve performance for users on macOS devices. This would involve:

• Replacing CUDA device initialization with MPS (e.g., torch.device("mps" if torch.backends.mps.is_available() else "cpu")).
• Adjusting precision from float64 to float32 in areas like sinusoidal embeddings and rotary encodings.
• Handling random generation by creating tensors on CPU before moving them to MPS.
• Replacing CUDA-specific calls (e.g., torch.cuda.empty_cache(), torch.cuda.amp.autocast) with conditional alternatives.
• Adding a compatibility layer during model loading for device differences.

I’m willing to help contribute to this effort and collaborate with anyone interested in making this happen.

[bakhti-ai]

Hi there, I forked and adjusted the repo to make it work on my macbook m1 pro. After adjustments it worked on macos. I opened a PR for this original repo as well. You can check it out here #69

[Runtrons]

You rock! Really appreciate your contribution!

[FurkanGozukara]

Hi there, I forked and adjusted the repo to make it work on my macbook m1 pro. After adjustments it worked on macos. I opened a PR for this original repo as well. You can check it out here #69

i cant imagine speed :D what are the step speeds for 1.3b 480p and 14b 720p?

[bakhti-ai]

Hi there, I forked and adjusted the repo to make it work on my macbook m1 pro. After adjustments it worked on macos. I opened a PR for this original repo as well. You can check it out here #69

i cant imagine speed :D what are the step speeds for 1.3b 480p and 14b 720p?

I haven't tested 14b 720p locally and not going to :D
I tried 1.3b, 480p, frame_num 8 and I think it took around 10mins to generate the video so 1min per step I guess

[bakhti-ai]

Hi there, I forked and adjusted the repo to make it work on my macbook m1 pro. After adjustments it worked on macos. I opened a PR for this original repo as well. You can check it out here #69

i cant imagine speed :D what are the step speeds for 1.3b 480p and 14b 720p?

I haven't tested 14b 720p locally and not going to :D I tried 1.3b, 480p, frame_num 8 and I think it took around 10mins to generate the video so 1min per step I guess

Then I increased frame_num to 24 to generate a little longer video, then waited around 1,5 hours but it was only 20% I just stopped it there