-
-
Notifications
You must be signed in to change notification settings - Fork 849
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Scheduler implementation of Continual Pre-Training of Large Language Models: How to (re)warm your model? #1273
Scheduler implementation of Continual Pre-Training of Large Language Models: How to (re)warm your model? #1273
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks! Lgtm
README.md
Outdated
@@ -797,6 +797,7 @@ early_stopping_patience: 3 | |||
lr_scheduler: # 'one_cycle' | 'log_sweep' | empty for cosine | |||
lr_scheduler_kwargs: | |||
cosine_min_lr_ratio: # decay lr to some percentage of the peak lr, e.g. cosine_min_lr_ratio=0.1 for 10% of peak lr | |||
cosine_constant_lr_ratio: # freeze lr at some percentage of the step, e.g. cosine_constant_lr_ratio=0.8 means start cosine_min_lr at 80% of training step |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Would be nice to have a link to the arxiv paper referenced here too
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
i put paper link to it.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
There is a test failure w
E TypeError: AxolotlTrainingArguments.init() got an unexpected keyword argument 'cosine_constant_lr_ratio'
thank you i add in arguments |
Scheduler implementation of Continual Pre-Training of Large Language Models: How to (re)warm your model? (https://arxiv.org/pdf/2308.04014.pdf)
Description
almost identical to consine min lr but it adds constant ratio which freezes lr at some percent of training step.
Motivation and Context
This scheduler has been proven for continual pretraining.
How has this been tested?
Made test code
Screenshots (if appropriate)
Types of changes
Social Handles (Optional)