I'm working with a style transfer model trained with pytorch in google colaboratory and then converted to an ML package. When I bring it into xcode and try to preview the asset I see the following error.
There was a problem decoding this Core ML document
missingMetadataField(named: "inputSchema")
I've been able to train and convert models as .mlmodel files, I'm only seeing this issue with .mlpackage files.
I'm using xcode 13 beta, which as far as I know is the only version of xcode that can handle ML packages/programs at the moment, and I'm using the coremltools beta to handle the conversion. Prior to the conversion, or if I convert to an ML Model instead it seems to work just fine.
Is this a problem with how the model is being structured, or converted? Is this a problem with how I've set up my xcode environment/swift project? Is there some way to update the metadata associated with ML packages to make sure the missing input schema is included?
This was recently acknowledged as an Xcode 13 bug here: https://github.com/apple/coremltools/issues/1301
