chunked/nltk_chunking/model_doc_megatron_gpt2/chunk_10.txt

Our BERT model achieves SOTA results on the RACE dataset (90.9% compared to SOTA accuracy
of 89.4%).