Evaluation script

[moha23]

Hi, will it be possible to release an evaluation script to reproduce the numbers reported in the paper? I loaded the weights into my model implementation and the resulting metrics for the test set are very different compared to the reported numbers. May be there are differences in some parameter / implementation. An evaluation script would be very helpful.