when training, here happens 'Gradient overflow. Skipping step, loss scaler 0 reducing loss scale to' #22

Activity

1. mongeoroo commented on Sep 12, 2023

   

   mongeoroo

   on Sep 12, 2023

   Last edited by mongeoroo

   More actions

   I also experienced the same warning. But even with the warning, the network was trained well and the test accuracy is the same as the figure reported in the paper.