What we care about are the weights that are obtained after training the model. These weights are extracted and they act as word embeddings. This is part 1/2 for training word embeddings.