by Google Research · Natural Language Processing
Efficient pre-training method that trains models as discriminators to identify replaced tokens instead of predicting masked ones, achieving faster convergence than BERT.
No ratings yet — the breakdown appears once this model has reviews.