[Model] Officially support Emu3 with Transformers backend

[hmellor]

This PR adds an explicit redirect from Emu3ForConditionalGeneration to TransformersForMultimodalLM. This means that:

• The Transformers backend will be used to load this model
• Users won't see warnings about the Transformers backend being fallback behaviour

In the docs I have added a paragraph explaining what it means for a model to be officially supported via the Transformers backend.

Closes #11008

[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.

🚀

[gemini-code-assist]

Code Review

This pull request officially adds support for the Emu3 model by redirecting it to the Transformers backend, and updates the documentation accordingly. The code changes are logical and well-structured. My main feedback is to add a corresponding test case for the newly supported model to ensure its correctness and prevent future regressions, as is standard practice in this repository.

[DarkLight1337]

Thanks, hope that this would encourage more people to consider Transformers backend before opening their own PRs!

[DarkLight1337]

PTAL at multi-modal tests failure

[hmellor]

It's a weight mapping issue. We've had a TODO in the code for a while to simplify the weight mapping so I've made #21382 to do that ahead of this PR.

[mergify]

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @hmellor.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

[hmellor]

Fixing the hf_to_vllm_mapper revealed an issue where the running stats are not being loaded into the nn.BatchNorm3d in Emu3's vision tower. Currently working on a fix.

[hmellor]

The failing mapper test was using model.named_parameters() as reference but the Emu3 checkpoint also contains buffers (stats for batch norm).

I have modified the test to ignore buffers that it may find in checkpoints.