I convert a PyTorch transformer BERT model with coremltool into a
.mlpackage model and deploy it to my iOS device. This works well so far.
When I try to optimize the PyTorch model with
BetterTransformer.transform before, the conversion process to the
.mlpackage model fails with the following error message:
PyTorch convert function for op '_nested_tensor_from_mask' not implemented.
Has anyone had any experience with this so far?