I try to build a model that looks like a encoder-decoder model, but the difference is that I use the same LSTM model in both for loops.
To merge hidden states from t-1 an