[Misc] Fix backward compatibility from #23030 #23070
Conversation
|
👋 Hi! Thank you for contributing to the vLLM project. 💬 Join our developer Slack at https://slack.vllm.ai to discuss your PR in #pr-reviews, coordinate on features in #feat- channels, or join special interest groups in #sig- channels. Just a reminder: PRs would not trigger full CI run by default. Instead, it would only run Once the PR is approved and ready to go, your PR reviewer(s) can run CI to test the changes comprehensively before merging. To run CI, PR reviewers can either: Add 🚀 |
![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 aims to fix a backward compatibility issue with multi-modal embeddings. The change modifies a return value in vllm/multimodal/base.py. However, this change introduces a type inconsistency within the from_seq_group function, where one return path now provides a dict while another provides a MultiModalKwargs object. This violates the function's type hint and can lead to runtime errors. This should be addressed to ensure consistent behavior.
Signed-off-by: Roger Wang <hey@rogerw.me>
|
All CI has passed in 5ea6a70. Force merging since all the following PRs only address pre-commit and doc build |
…ct#23070) Signed-off-by: Roger Wang <hey@rogerw.me> Co-authored-by: Roger Wang <hey@rogerw.me>
…ct#23070) Signed-off-by: Roger Wang <hey@rogerw.me> Co-authored-by: Roger Wang <hey@rogerw.me>
…ct#23070) Signed-off-by: Roger Wang <hey@rogerw.me> Co-authored-by: Roger Wang <hey@rogerw.me>
…ct#23070) Signed-off-by: Roger Wang <hey@rogerw.me> Co-authored-by: Roger Wang <hey@rogerw.me> Signed-off-by: Duncan Moss <djm.moss@gmail.com>
…ct#23070) Signed-off-by: Roger Wang <hey@rogerw.me> Co-authored-by: Roger Wang <hey@rogerw.me>
…ct#23070) Signed-off-by: Roger Wang <hey@rogerw.me> Co-authored-by: Roger Wang <hey@rogerw.me> Signed-off-by: Xiao Yu <xiao.yu@amd.com>
…ct#23070) Signed-off-by: Roger Wang <hey@rogerw.me> Co-authored-by: Roger Wang <hey@rogerw.me>
…ct#23070) Signed-off-by: Roger Wang <hey@rogerw.me> Co-authored-by: Roger Wang <hey@rogerw.me>
3 participants
Purpose
#23030 broke a v0 compatibility with mm embeddings as input and therefore the multimodal test. This PR fixes it.
Test Plan
pytest -v -s --ignore models/multimodal/generation/test_whisper.py --ignore models/multimodal/test_tensor_schema.py models/multimodal -m core_modelTest Result
(Optional) Documentation Update
Essential Elements of an Effective PR Description Checklist
supported_models.mdandexamplesfor a new model.