chunked/nltk_chunking/model_doc_rag/chunk_12.txt

For language generation
tasks, we find that RAG models generate more specific, diverse and factual language than a state-of-the-art
parametric-only seq2seq baseline.