Use smaller embedding model when not testing model specifically

[hmellor]

Replaces:

• intfloat/e5-mistral-7b-instruct with intfloat/multilingual-e5-small in places where the model itself is not the test subject
• intfloat/multilingual-e5-large with intfloat/multilingual-e5-small for marginal time reduction

This reduction in parameters is 560M & 7B -> 118M.

[github-actions]

👋 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 fastcheck CI which starts running only a small and essential subset of CI tests to quickly catch errors. You can run other CI tests on top of those by going to your fastcheck build on Buildkite UI (linked in the PR checks section) and unblock them. If you do not have permission to unblock, ping simon-mo or khluu to add you in our Buildkite org.

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 ready label to the PR or enable auto-merge.

🚀

[DarkLight1337]

Let's update vllm/test_utils.py accordingly as well.

[DarkLight1337]

cc @khluu can you add this model to S3?

[hmellor]

@CatherineSue I just wanted to check with you that I've not changed anything that will significantly reduce the coverage of Mistral embedding models?

[hmellor]

cc @khluu can you add this model to S3?

We're likely about to switch to EFS so no need to upload to S3