1、The original pytorch model is converted to coreml .mlpackage, using ALL backend you can see the operator running on ANE, but I intercepted part of the model and then use ALL backend, the operator can only run on cpu+gpu, I just intercepted part of the original model. 2、And when using coremltools conversion set minimum_deployment_target=ct.target.iOS15, using ALL the backend actually runs on ANE, but when minimum_deployment_target=ct.target.iOS16, using ALL the backend actually runs on cpu+gpu
