Fine tune arguments to learn new knowledge without forgetting previous #996
Unanswered
SutirthaChakraborty
asked this question in
Q&A
Replies: 1 comment 1 reply
-
|
Continual Learning might help you |
Beta Was this translation helpful? Give feedback.
1 reply
-
|
how do we do that? |
Beta Was this translation helpful? Give feedback.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
What should be the fine-tuning argument to make sure we are just adding knowledge to the existing pretrained model, without affecting their past training and learn few of my new words with that ?
When I am retraining, it's forgetting the pretrained vocab. I uses small learning rate and wd.
Do anyone has the perfect argument values to do that ?
Beta Was this translation helpful? Give feedback.
All reactions