[DO NOT MERGE] VLM offline benchmark with MMMU-Pro vision

[ywang96]

Opened this PR to share the mini offline batch inference benchmark with MMMU-Pro vision.

We may iterate on this and add support for other frameworks for comparison purposes.

[github-actions]

👋 Hi! Thank you for contributing to the vLLM project.
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 do one of these:

• Add ready label to the PR
• Enable auto-merge.

🚀

[comaniac]

Found that asyncio doesn't work because both initialize_llm and sample_hf_requests are blocking. Add to_thread to create async coroutine to really overlap them. Here is the initialization time before and after the commit (with 350 unique images):
Before: 130s
After: 73s