adding optimizer overlap for FSDP #203
Closed
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
chauhang
reviewed
Sep 16, 2023
chauhang
reviewed
Sep 16, 2023
src/llama_recipes/finetuning.py
Outdated
| print(f"setting up optimizer overlap") | ||
| optim_kwargs = {"lr": train_config.lr} | ||
| _apply_optimizer_in_backward( | ||
| optimizer_class=optim.AdamW, |
There was a problem hiding this comment.
If the optimizer_in_backward_available flag is set and user has selected AnyPrecisionAdamW, will be good to add for that case as well, unless there is a restriction on which optimizers support this feature.
There was a problem hiding this comment.
Added for Anyprecision as well, as per my test it works for anyprecision as well.
chauhang
reviewed
Sep 16, 2023
chauhang
suggested changes
Sep 16, 2023
There was a problem hiding this comment.
@HamidShojanazeri Thanks for this PR. Left few comments. It will be great to capture a memory profile as well
rohan-varma
reviewed
Sep 18, 2023
rohan-varma
reviewed
Sep 18, 2023
rohan-varma
approved these changes
Nov 1, 2023
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Outdated suggestions cannot be applied.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.
What does this PR do?
This PR adds Optimizer overlap that bring addition memory savings by fusing the gradinet calculation and parameter update steps.
for 7B --> max reserved memory saving is 7%
---> allocated memory 4%
---> active memory 4%
Feature/Issue validation/testing
Please describe the tests that you ran to verify your changes and relevant result summary. Provide instructions so it can be reproduced.
Please also list any relevant details for your test configuration.
Test A without optimizer_overlap
https://gist.github.com/HamidShojanazeri/08ee3d23bdb0fa60466071dee1efda1f
Test B with optimizer_overlap
https://gist.github.com/HamidShojanazeri/3d1147012e9db130dd7cebf75d3caa64
Logs with/out anyprecision
Logs with AdamW
Before submitting
Pull Request section?
to it if that's the case.
Thanks for contributing 🎉!