prajjwal1/bert-medium-mnli

These BERT variants were introduced in the paper Well-Read Students Learn Better: On the Importance of Pre-training Compact Models . These models are trained on MNLI.MNLI: 75.86%. MNLI-mm: 77.03%. These models were trained for 4 epochs. These are pre-trained for four epochs and are trained for the first time . The models are based on the Tensorflow checkpoint found in the official Google BERT repository . They are trained using Pytorch pre-training models from the official BERT version of Google's BERT . The BERT variant is based on a model that was trained on the MNLI checkpoint .,