Text this: Applying SoftTriple Loss for Supervised Language Model Fine Tuning