[CI/Build] Update causal-conv1d and lm-eval#22141
[CI/Build] Update causal-conv1d and lm-eval#22141DarkLight1337 wants to merge 2 commits intovllm-project:mainfrom
Conversation
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
DarkLight1337
added
the
ready
|
👋 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 🚀 |
mergify
bot
added
documentation
![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 updates two key dependencies, causal-conv1d and lm-eval. The update to causal-conv1d to version 1.5.2 is a positive change that simplifies the build process by allowing it to be managed through pip-compile. However, the update to lm-eval pins the dependency to a specific git commit hash. While this may be necessary to resolve an immediate compatibility issue, it introduces potential maintenance and security risks. I've added comments suggesting the addition of explanatory notes in the code to clarify the reason for this pinning and to facilitate future updates to a stable release.
| && rm -rf vllm \ | ||
| && python3 -m pip install -e tests/vllm_test_utils \ | ||
| && python3 -m pip install lm-eval[api]==0.4.4 \ | ||
| && python3 -m pip install git+https://github.com/EleutherAI/lm-evaluation-harness.git@bc811365ef3917a15da6aac9b7feafb4ba0fb42f#egg=lm-eval[api] \ |
There was a problem hiding this comment.
Pinning a dependency to a specific commit hash can introduce maintenance and security risks. It makes the build less reproducible if the commit history is changed (e.g., force-push) and bypasses the security vetting of official releases. If this specific commit is necessary to unblock a temporary issue, please add a comment explaining the reason for this pin and which issue/PR it relates to. This will help with future maintenance and make it clear when this can be reverted to a stable, versioned release.
| opencv-python-headless >= 4.11.0 # required for video test | ||
| datamodel_code_generator # required for minicpm3 test | ||
| lm-eval[api]==0.4.8 # required for model evaluation test | ||
| lm-eval[api] @ git+https://github.com/EleutherAI/lm-evaluation-harness.git@bc811365ef3917a15da6aac9b7feafb4ba0fb42f # required for model evaluation test |
There was a problem hiding this comment.
Pinning a dependency to a specific commit hash, as done here for lm-eval, can be risky for long-term maintenance and security. It makes it difficult to track updates and can lead to build instability if the commit is ever removed or the repository history is altered. It's generally better to use a versioned release. If this pin is a temporary measure to resolve a specific compatibility issue, it would be very helpful to add a comment here explaining the context, linking to the relevant issue or PR, and outlining the conditions under which this can be updated to a released version.
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
|
Hmm seems |
|
I'll just update |
|
Prefer #22409, closing |
1 participant
Essential Elements of an Effective PR Description Checklist
supported_models.mdandexamplesfor a new model.Purpose
causal-conv1d==1.5.2is now compatible withuv pip compile, so let's update it to simplify the buildlm-evalto work with [Deprecation] Removeprompt_token_idsarg fallback inLLM.generateandLLM.embed#18800Test Plan
Test Result
(Optional) Documentation Update