BERT and other Transformer encoder architectures have been very successful in natural language processing (NLP) for computing vector-space representations of text, both in advancing the state of the art in academic benchmarks as well as in large-scale applications like Google Search. BERT has been available for TensorFlow since it was created, but originally relied on non-TensorFlow Python code to transform raw text into model inputs.
Making BERT Easier with Preprocessing Models From TensorFlow Hub Fine tune BERT for Sentiment analysis using TensorFlow Hub blog.tensorflow.org |