Good work by well-known reputable authors. The gains in training efficiency and compute cost versus widely used text-encoding models like RoBERTa and XLNet are

Good work by well-known reputable authors. The gains in training efficiency and compute cost versus widely used text-encoding models like RoBERTa and XLNet are significant. Thank you for sharing this on HN!