set-soft-AudioSeparation/doc/Internals.md

Internals

Call sequence

• The node creates a demixer class with: get_demixer(model_data, device, models_dir)
• This function creates a Demixer object for the correct demix type
• The DemixerGeneric.init() calls load_model
• load_model:
  • Calls get_model to get a proper model object without weights
  • Calls a loader for the container (ONNX, safetensors)
    • load_safetensors loads the weights