Add support for adapter loading in mllama #669
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
|
The PR is ready from my side. Loading randomized adapter results in garbage output while the model answers correctly without it. Slowly changing the noise in the randomized adapter also slowly degrades the quality of the model's output again indicating that adapters are being loaded correctly. |
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Outdated suggestions cannot be applied.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.
Supporting adapter loading in VLMs, specifically mllama models. This is more complex than we had initially anticipated as a VLM is under the hood a composite model with multiple trunks which requires separating and loading the lora weights independently for each trunk. Further, currently we don't have lora adapters for certain layers in one of the trunks (cross attention layers) which requires careful handling. Finally, the linear layer classes currently used don't support adapter weights (currently investigating).