chunked/nltk_chunking/model_doc_t5/chunk_4.txt

Our systematic study compares pretraining objectives, architectures, unlabeled datasets, transfer
approaches, and other factors on dozens of language understanding tasks.