So I have a small doubt regarding saving embeddings. I would like to train a masked language model on some text corpus and then save the final trained embeddings weights to load into a NMT decoder. Is this possible? Otherwise how can I go about saving embeddings from a trained model? Thank you in advance!