[Bugfix] Fix incorrect llava next feature size calculation

[zyddnys]

The old code produces incorrect feature size and leads to

Attempted to assign XXX = XXX multimodal tokens to YYY placeholders

An image with size 700x840 can trigger this

[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.

🚀

[DarkLight1337]

Can you show the code you used to initialize vLLM and pass the image into the model? I cannot repro this issue. Also, can you run collect_env.py to show the details about your environment?

[zyddnys]

this only happen to llava next models using siglip so400m as text encoder

[DarkLight1337]

Which HF repo is this model from?