CodaLab evaluation

[jlmmTese]

Hello,

I'm evaluating my model on the CODALAB platform but my score is very low, despite the good performance on the dev split. I would like to ask a few simple questions to increase my performance, if possible.

1. Does the CodaLab evaluation take into account the "Na" - no relation instances?
2. I have in total 392,158 instances. How many of these instances have the no relation label? My model predicted there are only 5000 instances with no prediction. Is this the expected number or should it be higher?
3. What is the expected encoding in the .json submission file ? Is it UTF-8?

Thank you very much.