chunked/nltk_chunking/model_doc_openai-gpt/chunk_4.txt

We demonstrate that large gains on these tasks can be realized by generative pretraining of a
language model on a diverse corpus of unlabeled text, followed by discriminative fine-tuning on each specific task.