fix bnb multi gpu training #2714
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
younesbelkada
approved these changes
Apr 26, 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. |
6 tasks
jambayk
added a commit
to microsoft/Olive
that referenced
this pull request
Apr 27, 2024
## Describe your changes There is a bug in accelerate where it assumes that bnb (loaded_in_4bit) models on multiple gpus cannot be trained but this is not the case. On AML computes with multiple GPUs, the model doesn't have any weights on device 0 which is the default device for accelerator and catches the bug https://github.com/huggingface/accelerate/blob/e82de1215ae701b6bf567eb705615c656e7f55c7/src/accelerate/accelerator.py#L1374. This has been fixed on main by huggingface/accelerate#2714 but is not in a release yet. We workaround this bug for current stable releases of accelerate by setting `ACCELERATE_TORCH_DEVICE` to the same device as the first device the model is on. ` ## Checklist before requesting a review - [ ] Add unit tests for this change. - [ ] Make sure all tests can pass. - [ ] Update documents if necessary. - [ ] Lint and apply fixes to your code by running `lintrunner -a` - [ ] Is this a user-facing change? If yes, give a description of this change to be included in the release notes. - [ ] Is this PR including examples changes? If yes, please remember to update [example documentation](https://github.com/microsoft/Olive/blob/main/docs/source/examples.md) in a follow-up PR. ## (Optional) Issue link
This was referenced May 14, 2024
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 fixes a bug that users who want to perform multi-gpu training with bnb gets when the first gpu is not used. In this PR, you see that we did not allow multi-gpu training before and all the code after that assume that we are in a single gpu setup. Since then, we added the possibility to perform multi-gpu training using naive PP. However, the logic after the multi-gpu check stayed the same as you can see on main . A quick fix would be to just add an elif statement.
Fixes #2713 #2429