Repository navigation
FIX CI: Targeting fused parameters with LoRA - #5430
Merged
albertvillanova merged 1 commit intoApr 1, 2026
Merged
albertvillanova merged 1 commit into
albertvillanova merged 1 commit into
Conversation
See huggingface#5429 In this test, LoRA was targeting up_proj but not gate_proj. In Transformers, these two layers are now fused. Therefore, it is no longer possible to target just one of the two. The test is now changed to target both. It would be possible to add a workaround and set zeros for the layer that is not being targeted, but that would be a bigger task. So if we can accept targeting both, I propose to just do that.
|
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. |
kashif
approved these changes
Apr 1, 2026
Member
Author
|
@kashif It looks like the failing smoke test is unrelated and the dev tests are awaiting the PEFT fix linked above. If that's okay, feel free to merge. |
albertvillanova
approved these changes
Apr 1, 2026
albertvillanova
left a comment
Member
There was a problem hiding this comment.
Thanks. I'm merging this.
This file contains hidden or 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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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?
In this test, LoRA was targeting
up_projbut notgate_proj. In Transformers, these two layers are now fused. Therefore, it is no longer possible to target just one of the two. The test is now changed to target both.It would be possible to add a workaround and set zeros for the layer that is not being targeted, but that would be a bigger task. So if we can accept targeting both, I propose to just do that.
Note that during creation of the error message alerting the user to the fact that targeting just one is not possible, there was an error in PEFT. This error is being resolved via huggingface/peft#3127. That PR also fixes the
KeyError: 'qwen2_vl'we see in the dev dependency CI.Fixes subsequent issue found in #5429 (comment):
Before submitting
AI writing disclosure
We welcome the use of AI tools to help with contributions. For transparency and to help us improve our review process, please indicate the level of AI involvement in this PR.
Who can review?
@albertvillanova
Note
Low Risk
Low risk: test-only change that broadens LoRA target modules to align with upstream fused layers, with no runtime/library behavior changes.
Overview
Adjusts the
RewardTrainerMoE PEFT regression test to includegate_projalongsideup_projinLoraConfig(target_modules=...), reflecting that these projections are now fused in recent Transformers and cannot be targeted independently.Written by Cursor Bugbot for commit 2265519. This will update automatically on new commits. Configure here.