You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I am trying to reproduce results (for a publication) for a simple dense NN on a CPU. Training the model in PL 1.5.7 through 1.5.10 produce the same results, however 1.6.0 produces different results.
Could this be due to the "current epoch/global step boundary" change in 1.6.0?
Val loss on the first epoch is different and the step is off by 1. Any ideas on how to alter training to reproduce across versions?
2022-05-24 14:19:42,345 - pytorch_lightning.utilities.rank_zero - INFO - Epoch 0, global step 16: 'val_loss' reached 0.69475 (best 0.69475), saving model to
2022-05-24 14:21:36,574 - pytorch_lightning.utilities.distributed - INFO - Epoch 0, global step 15: val_loss reached 0.68147 (best 0.68147), saving model to
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
-
I am trying to reproduce results (for a publication) for a simple dense NN on a CPU. Training the model in PL 1.5.7 through 1.5.10 produce the same results, however 1.6.0 produces different results.
Could this be due to the "current epoch/global step boundary" change in 1.6.0?
Val loss on the first epoch is different and the step is off by 1. Any ideas on how to alter training to reproduce across versions?
Here is my trainer config below.
Beta Was this translation helpful? Give feedback.
All reactions