The RSE-BERT-large-Transfer is trained with 2 relations including:
1) entailment
2) paraphrase


The BERT-large-uncased model is used as initialization. 

It can be used ideally for Transfer datasets - (Downstream Tasks).