Skip Qwen 1 in CI because remote code is no longer compatible with Transformers #22536
Conversation
…ansformers Signed-off-by: Harry Mellor <[email protected]>
![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 addresses a CI failure by skipping tests for the Qwen/Qwen-7B-Chat model on newer versions of the transformers library. The change correctly adds a max_transformers_version constraint to the model's test registry entry. This is a reasonable and effective solution to prevent test failures due to upstream package incompatibilities with an older model. The implementation is correct and I have no further feedback.
|
👋 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 🚀 |
github-actions
bot
added
the
ready
…ansformers (vllm-project#22536) Signed-off-by: Harry Mellor <[email protected]> Signed-off-by: Jinzhen Lin <[email protected]>
…ansformers (vllm-project#22536) Signed-off-by: Harry Mellor <[email protected]> Signed-off-by: Noam Gat <[email protected]>
…ansformers (vllm-project#22536) Signed-off-by: Harry Mellor <[email protected]> Signed-off-by: Paul Pak <[email protected]>
…ansformers (vllm-project#22536) Signed-off-by: Harry Mellor <[email protected]> Signed-off-by: Diego-Castan <[email protected]>
…ansformers (vllm-project#22536) Signed-off-by: Harry Mellor <[email protected]>
…ansformers (vllm-project#22536) Signed-off-by: Harry Mellor <[email protected]>
…ansformers (vllm-project#22536) Signed-off-by: Harry Mellor <[email protected]> Signed-off-by: Xiao Yu <[email protected]>
…ansformers (vllm-project#22536) Signed-off-by: Harry Mellor <[email protected]>
Tests using this model are failing on main because the remote code used for the HF reference is incompatible with Transformers v4.55.0.
Since it's a very old model we can just skip the test instead of trying to fix https://huggingface.co/Qwen/Qwen-7B-Chat.