Developing Trustworthy Language Models With Consistent Predictions