* Implement support for the LSTM model to be able to handle, and train on word / token embeddings rather than just on-hot encodings.