Add interpretability?

[sam-writer]

Not sure what the scope of FitBERT should be, but it might be cool to add AllenNLPs new interpretability tooling:

https://demo.allennlp.org/masked-lm

screenshot:

The only example code I could find was in the tests