Inference uses the same configs as the training and instantiates the model load function from the configs.