A bit late but it seems like you’re hitting a common snag with CoreML conversions. Some PyTorch operations, like pythonop, aren’t supported. A good next step is to trace the model’s forward pass and identify where things might be breaking. From there, tools like TorchScript can help make the conversion smoother.

I ran into similar challenges while converting the Wav2Lip model to CoreML and posted a guide with tips and steps that might help:

https://github.com/Ialzouby/Wav2Lip-PyTorch-To-CoreML/

Hope it helps, and good luck!