Finetuning mt5 for question answering using run_qa_seq2seq

I want to finetuning mt5-small for question answering in tamil using a translated SQuaD dataset with the run_qa_seq2seq script. How do i display the evaluation metrics like f1 and exact match?