[Core] Ensure LoRA linear respect the base_layer's tp_size and tp_rank #25487
Conversation
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
![gemini-code-assist[bot]](https://avatars.githubusercontent.com/in/956858?s=60&v=4){kind=small}
There was a problem hiding this comment.
Code Review
This pull request refactors the handling of tensor parallelism tp_size and tp_rank in LoRA layers to ensure consistency with the base layer, which is a solid improvement for maintainability and correctness. The changes also include some good optimizations to avoid unnecessary collective operations when tp_size is 1. Additionally, this PR fixes a critical bug in lora_weights.py where input_dim and output_dim properties were returning incorrect values. The changes are well-implemented and look good.
jeejeelee
commented
Sep 23, 2025
| @@ -48,11 +48,11 @@ def optimize(self) -> "LoRALayerWeights": | |||
|
|
|||
| @property | |||
| def input_dim(self) -> int: | |||
| return self.lora_a.shape[0] | |||
| return self.lora_a.shape[1] | |||
There was a problem hiding this comment.
BTW, fix the dimension mismatch
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
jeejeelee
force-pushed
the
lora-tp-optimize
branch
from
941d3d2 to
05e8559
Compare
FeiDaLI
pushed a commit
to FeiDaLI/vllm
that referenced
this pull request
Sep 25, 2025
vllm-project#25487) Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
yewentao256
pushed a commit
that referenced
this pull request
Oct 3, 2025
#25487) Signed-off-by: Jee Jee Li <pandaleefree@gmail.com> Signed-off-by: yewentao256 <zhyanwentao@126.com>
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Outdated suggestions cannot be applied.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.
Purpose
Ensure all LoRA linear respect the base_layer's tp_size and tp_rank
Test Plan
Test Result
Essential Elements of an Effective PR Description Checklist
supported_models.mdandexamplesfor a new model.