Skip to content

lr scheduler - update global states into optimizer #29

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

Closed
lessw2020 opened this issue Feb 1, 2024 · 2 comments
Closed

lr scheduler - update global states into optimizer #29

lessw2020 opened this issue Feb 1, 2024 · 2 comments
Labels
better_engineering Repo code quality improvements

Comments

@lessw2020
Copy link
Contributor

lr scheduler currently maintains two global states to implement the full lr warmup and decay.
We want to remove these:
"nit: we can make these two arguments still as function arguments below, but we use:

functools.partial to pack these two arguments when adding to the optimizer.

global config works for now. This can be done as a follow up"

@awgu
Copy link
Collaborator

awgu commented Apr 26, 2024

cc: @tianyu-l is this issue done?

@tianyu-l
Copy link
Contributor

tianyu-l commented May 3, 2024

Seems not. @awgu

https://github.com/pytorch/torchtitan/blob/main/torchtitan/lr_scheduling.py#L10

cc: @tianyu-l is this issue done?

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

No branches or pull requests

4 participants