Nvidia Inference Engine Keeps BERT Latency Within A Millisecond

It’s a shame when your data scientists dial in the accuracy on a deep learning model to a very high degree, only to be forced to gut the model for