[Feature] model evaluation + metrics

[Gregory-Pereira]

I would like to have some way to compare efficiency of models, especially because we are using this repo to showcase our granite-7b-lab image. This unfortunately will have to be implemented in a recipe-by-recipe way because the model evaluation metrics very per application / problem type.
/cc @MichaelClifford