Have you implemented any of this in practice? Eg are you benchmarking models?

I've done some for classification, ranking, and other sorts of non-generative tasks.