When I use Oim Loss to train the model, In the fisrt several iterations, loss performs normally and reduce with the epoch increasing. But the loss suddenly became to 'NAN' and in the next iterations it was always 'NAN'. I check the data input and there is not 'nan' input. So, is this problem related to your Oim Loss method? Does anybody else meet this problem? I have run other two loss methods and they performed normally.
When I use Oim Loss to train the model, In the fisrt several iterations, loss performs normally and reduce with the epoch increasing. But the loss suddenly became to 'NAN' and in the next iterations it was always 'NAN'. I check the data input and there is not 'nan' input. So, is this problem related to your Oim Loss method? Does anybody else meet this problem? I have run other two loss methods and they performed normally.