Train with MLM loss augmentation

Is it possible to train an NER model with MLM loss too? Does the trainer api support two datacollators on two datasets?