Since the inference model and the djmodel are two independent models, how can weights learned by djmodel be transferred to the inference model to make a prediction?
They just use the LSTM_cell that was trained using djmodel as one of the processing steps in the music generation model.