Fetching the paper…

TaCL: Improving BERT Pre-training with Token-aware Contrastive Learning · Around