- 2018-04-25 2 layer 128 LSTM, 1 dense, 8k steps, 100-batch, ler: 0.03644, loss: 5.188
- 2018-04-25 1 layer 128 LSTM, 1 dense, 15k steps, 100-batch, ler: 0.02336, loss: 4.798
- 2018-04-28 1 layer 128 LSTM, 1 dense could be trained to decode 6 real-world examples without error