That was exactly my question. Namely, djmodel is trained, but is not used for predictions. So, the question arises: “if we don’t use this model to make predictions, then what did we train it for?”
And the opposite question for the inference_model. Namely, how can we use a model for predictions without having trained it first?