Keras custom loss function - how to access actual truth values and predictions

Question

I am working with time series forecasting with Keras LSTM. I take the last n_input_steps occurrences of the series and try to predict one step forward. For example, if my ti