Add foreach to optimizers #10
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
facebook-github-bot
added
the
CLA Signed
|
Nice! I will take a look on Monday. |
|
Have you tried running https://github.com/facebookresearch/schedule_free/blob/main/schedulefree/test_schedulefree.py |
|
I implemented it in all of the main optimizers, both closure and non-closure versions. I skipped some of the renaming in the foreach versions for the sake of simplicity on the in-place operations. The code passes the test suite just fine in its current state. |
|
Ok, we will look this over during the next few days, sorry for the delay things are hectic at the moment. |
|
This looks to be working correctly, thanks! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
I rewrote adamw_schedulefree to support using
torch._foreachoperations as an option. According to my profiler, this made the optimizer step go from taking ~85ms to ~65ms (the model used was the SD1.5 Unet). In theory, this should also increase peak memory usage during the optimizer step (according to Pytorch's documentation on the built-in optimizers) -- I have left it as enabled by default regardless since this is the default behavior for built-in optimizers.I think it would obviously be best to add this to all included optimizers, but before I do that I would like to hear some feedback on code style to make sure everything I add is up to standards.