For many mlmodel files you can count the number of layers using coremltools or something like Netron.
But a model built with MLImageClassifierBuilder is really a pipeline consisting of two other models: 1. Vision FeaturePrint, 2. a logistic regression. The Vision FeaturePrint model is the actual neural network but this is kept secret by Apple. It is built into the OS and you cannot look inside it.
Ok, thank you for your answer.
Another question: is there a way to know the time complexity of a model built with MLImageClassifierBuilder?
Thank you again,