Modifying Memories In Transformer Models For Nlp