Stochastic Weight Averaging for steps instead of epochs #9861
Unanswered
MaveriQ
asked this question in
code help: NLP / ASR / TTS
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
Hi
I am looking for guidance on how to adapt the SWA callback for a step based averaging and scheduler update. In its present form, both actions take place in the function
on_train_epoch_start
. Some of this will have to be moved into 'training_step' function.I will appreciate any pointers on how to go about that. Thanks
Beta Was this translation helpful? Give feedback.
All reactions