Support the possibility to separate the model implementation and the input/output conversion

• Download the correct explanation model when an explanation invocation is made