GPU-accelerated generation of text embeddings used for question-answering retrieval.