Skip to content

[training] use the lr when using 8bit adam.#9796

Merged
sayakpaul merged 3 commits intomainfrom
use-lr-optim-class
Oct 31, 2024
Merged

[training] use the lr when using 8bit adam.#9796
sayakpaul merged 3 commits intomainfrom
use-lr-optim-class

Conversation

@sayakpaul
Copy link
Member

@sayakpaul sayakpaul commented Oct 29, 2024

What does this PR do?

Oops.

@a-r-r-o-w FYI.

@sayakpaul sayakpaul requested a review from linoytsaban October 29, 2024 12:13
@HuggingFaceDocBuilderDev
Copy link

HuggingFaceDocBuilderDev commented Oct 29, 2024

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

linoytsaban
linoytsaban approved these changes Oct 29, 2024
View reviewed changes
Copy link
Collaborator

@linoytsaban linoytsaban left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Nice catch!

@sayakpaul
Copy link
Member Author

sayakpaul commented Oct 30, 2024

@linoytsaban sorry I think I went too soon with the PR. I realized that we ALWAYS pack "lr" in params_to_optimize irrespective of the optimizer class we're using.

So, I think you were right not to pass the lr. However, when using "prodigy" we were still passing lr. So, to remove the inconsistency, I omitted those instances in this PR. Could you please give this another look?

Cc: @a-r-r-o-w for the changed Cog scripts.

a-r-r-o-w
a-r-r-o-w approved these changes Oct 30, 2024
View reviewed changes
Copy link
Contributor

@a-r-r-o-w a-r-r-o-w left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Looks good! IIUC, since LR is already packed with parameters, no need to pass it to optimizer_class when creating it - this inconsistency was there in some of our scripts?

@sayakpaul
Copy link
Member Author

sayakpaul commented Oct 30, 2024

Correct. And this PR resolves that.

@sayakpaul sayakpaul merged commit 8ce37ab into main Oct 31, 2024
@sayakpaul sayakpaul deleted the use-lr-optim-class branch October 31, 2024 10:21
a-r-r-o-w pushed a commit that referenced this pull request Nov 1, 2024
@sayakpaul @linoytsaban @a-r-r-o-w
[training] use the lr when using 8bit adam. (#9796)
5d5af74 
* use the lr when using 8bit adam.

* remove lr as we pack it in params_to_optimize.

---------

Co-authored-by: Linoy Tsaban <57615435+linoytsaban@users.noreply.github.com>
sayakpaul added a commit that referenced this pull request Dec 23, 2024
@sayakpaul @linoytsaban
[training] use the lr when using 8bit adam. (#9796)
2094e7a 
* use the lr when using 8bit adam.

* remove lr as we pack it in params_to_optimize.

---------

Co-authored-by: Linoy Tsaban <57615435+linoytsaban@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@linoytsaban linoytsaban linoytsaban approved these changes

@a-r-r-o-w a-r-r-o-w Awaiting requested review from a-r-r-o-w

Assignees

No one assigned

Labels

None yet

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

4 participants

@sayakpaul @HuggingFaceDocBuilderDev @linoytsaban @a-r-r-o-w