Learning rate scheduler's epoch off by one when resuming from checkpoint #1772
Labels
bug
Something isn't working
duplicate
This issue or pull request already exists
help wanted
Open to be worked on
🐛 Bug
Currently lr_scheduler's state is updated after the checkpoint callback, so what is being saved here is last epoch's state.
Note: I think this has the same fix as #1464, but I'm posting it here because (1) I got rekt by this again, (2) in case it's not the same bug, and (3) #1464 is not fixed.
To Reproduce
Steps to reproduce the behavior:
Install using
pip install git+https://github.com/PytorchLightning/pytorch-lightning.git@master --upgrade
Let the model train until convergence. And then reload a saved model and see how it continues:
The
PrintingCallback
would print:and so on.
Expected behavior
The
PrintingCallback
should print:Environment
This is ran on Google colab.
https://colab.research.google.com/drive/1pkCSMaApyjH40jwrdl4aQLVYjnGP3JzD?usp=sharing
Additional context
Related to #1463 and #1464.
The text was updated successfully, but these errors were encountered: