Seq2Seq Model used (Keras) to predict text from a document preserving context and long term dependencies