this is indeed the case. i will submit a feedback as suggested. Still, is there any resources to understand what is supported by what hardware of the MIL and how CoreML runtime decides how to run model ?