Draft: Merge LoRA Adapters with AWQ BaseModels #2418
Draft
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
Whadup
changed the title
BenjaminBossan
requested changes
Mar 11, 2025
There was a problem hiding this comment.
Thanks for adding merging capabilities to AWQ. I only skimmed the PR so far, but could you please:
- Also implement the
unmergemethod? It should be very similar to themergemethod, but remove the delta weight - There should be a unit test to ensure that merging works, e.g. similar to this test (without DoRA).
- Let's run
make styleon your changes.
|
@BenjaminBossan Thanks for looking into it already! Your three points are on my agenda, I will give you a ping when I commit the changes. |
|
Great, thanks a lot. |
|
This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread. |
|
@Whadup Do you still plan on working on this? |
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Outdated suggestions cannot be applied.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.
This PR extends the AwqLoraLinear class to allow merging in of LoRA Adapters.
Instead of re-quantizing the whole model, we use the original quantization scales and zeros.