FARM is doing test set evaluation when certain conditions are met. The results are only logged but it is impossible / hard to get them with downstream Python code. I think the test set evaluation results should be stored in a variable so they can be read.

What do you think? As always I would crate a PR.