Our models trained with the InSeNT approach. These are the checkpoints that we used to run the evaluations reported in our paper.