When using FARMs Evaluator for QA evaluation we only have EM and f1 across all input examples.

The official SQuAD eval script has both, an overall score and at the same time scores for text answers vs no_answer.
We should also have both inside the Evaluator Class.