Skip to content
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

[Feature] Implement REX lr scheduler #222

Merged
merged 7 commits into from
Mar 2, 2024
Merged

Conversation

kozistr
Copy link
Owner

@kozistr kozistr commented Mar 2, 2024

Problem (Why?)

related to #217

Solution (What/How?)

  • implement REX lr scheduler

Other changes (bug fixes, small refactors)

N/A

Notes

N/A

@kozistr kozistr added the feature request Request features label Mar 2, 2024
@kozistr kozistr self-assigned this Mar 2, 2024
Copy link

codecov bot commented Mar 2, 2024

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 100.00%. Comparing base (fd717fc) to head (a3b1bfb).

Additional details and impacted files
@@            Coverage Diff            @@
##              main      #222   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           81        82    +1     
  Lines         5912      5946   +34     
=========================================
+ Hits          5912      5946   +34     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

@kozistr kozistr merged commit da65344 into main Mar 2, 2024
4 checks passed
@kozistr kozistr deleted the feature/rex-lr-scheduler branch March 2, 2024 09:36
@clementpoiret
Copy link

Hi!
For the sake of curiosity, why does the impl differs from the original paper?
Shouldn't it be self.max_lr * ((1.0 - progress) / (.5 + (1 - progress) / 2.0))?

@kozistr
Copy link
Owner Author

kozistr commented Jul 7, 2024

Hi! For the sake of curiosity, why does the impl differs from the original paper? Shouldn't it be self.max_lr * ((1.0 - progress) / (.5 + (1 - progress) / 2.0))?

sorry for the late response, i missed it

as you mentioned, the formula self.max_lr * ((1.0 - progress) / (.5 + (1 - progress) / 2.0)) is right! actually, it is equal to my implementation (min_lr + (max_lr - min_lr) * ((1.0 - progress) / (1.0 - progress / 2.0))) when min_lr is 0.

image

the reason why i added min_lr is for the flexibility and further usage of the REX scheduler : )

hope it helps you!

@clementpoiret
Copy link

Ah great! I understand, thanks for your reply :)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

2 participants