NVIDIA

BERT for TensorFlow

BERT is a method of pre-training language representations which obtains state-of-the-art results on a wide array of NLP tasks.