BUG: Parameter --execution_provider=['gpu'] in module FaceSwapper is not in ['cpu']

[alexjolig]

Describe the bug
Everytime I try to choose a different execution provider, I face this error:
LoadingException: Parameter --execution_provider=['gpu'] in module FaceSwapper is not in ['cpu']

To Reproduce
Steps to reproduce the behavior:

python sin.py --source="cool_photo.jpg" --target="not_a_porn.mp4" --frame-processor FaceSwapper FaceEnhancer --output="result.mp4" --many-faces --execution-provider=cuda

If I choose cuda or gpu I'll face this error

Here's the full stack trace:

Traceback (most recent call last):
  File "projects/sinner/sin.py", line 55, in <module>
    Sin().run()
  File "projects/sinner/sin.py", line 46, in run
    BatchProcessingCore(parameters=self.parameters).run()
  File "projects/sinner/sinner/BatchProcessingCore.py", line 110, in run
    current_processor = BaseFrameProcessor.create(processor_name, self.parameters)
  File "projects/sinner/sinner/processors/frame/BaseFrameProcessor.py", line 26, in create
    return handler_class(**params)
  File "projects/sinner/sinner/processors/frame/FaceSwapper.py", line 97, in __init__
    super().__init__(parameters)
  File "projects/sinner/sinner/processors/frame/BaseFrameProcessor.py", line 49, in __init__
    super().__init__(self.parameters)
  File "projects/sinner/sinner/validators/AttributeLoader.py", line 60, in __init__
    raise LoadingException(self.errors)
sinner.validators.LoaderException.LoadingException: Parameter --execution_provider=['gpu'] in module FaceSwapper is not in ['cpu']

• Laptop Info:
  MacBook Pro 2023 M2. 10-core GPU, 8-core CPU

Sanity Check

• [* ] I have the latest code from the github repository
• [* ] I have followed the installation guide

[pozitronik]

It is not an error. This message means that your current hardware/software configuration supports only the 'cpu' execution provider for onnx runtime.

[alexjolig]

@pozitronik So is there something I can do or it's just what it is?

[pozitronik]

I'll try to explain it as much, as I understand it myself.
There's ONNX Runtime, a framework that can run AI models on different types of providers; you can read there about.
You've installed the ONNX Runtime package that supports only the CpuProvider on your hardware (which utilizes the CPU, obviously). I suppose you have to find and install the onnxruntime-silicon package, which may support some kind of hardware acceleration.
It seems that onnxruntime-silicon builds can be found there.
I would appreciate it if you could try to check on this; I can't do it myself, as there's no Mac nearby.

[alexjolig]

@pozitronik Thanks for the explanation. I'd definitely try it and let you know how it goes.

[alexjolig]

Maybe not the right thread, but just wanted to mention that since installing pyvirtualcam is not supported for Mac M1/M2, the work around is to install it from source:

pip install git+https://github.com/letmaik/pyvirtualcam

One other thing that can help is that you replace it in requirements by this:

git+https://github.com/letmaik/pyvirtualcam; platform_system == "Darwin" and platform.machine == "arm64" pyvirtualcam; platform_system != "Darwin" or platform.machine != "arm64"

I tested those solutions and they worked for me.

Maybe you can add Apple silicon hints to your README file.

[pozitronik]

Maybe you can add Apple silicon hints to your README file.

Thanks, i definitely will do that.