optimized BERT transformer inference on NVIDIA GPU. https://arxiv.org/abs/2210.03052 - View it on GitHub