Closed this issue 2 years ago · 1 comments
Comparison between original (orange) / PTL training (blue)
*difference in train_loss is due to the change from epoch-based to batch-based stats