[WIP] Multimodal model support for V1 TPU

[mgoin]

Based on and requires #11936

Currently only focused on usability and correctness, not performance.

This does not deal with pre-compiling the encoder forward pass, so in the event that the model is passed in image/video/audio that is a new shape, it will force compilation during runtime.

Tested Examples

Image:

VLLM_USE_V1=1 python llava_tpu.py
...
Prompt 1: What do you see in this image?
Response:  The image features a tall tower with a spire, surrounded by a beautiful cherry blossom tree. The tree is filled with pink flowers, creating a stunning contrast against the tower. The blossoms are scattered throughout the tree, with some closer to the top and others near the bottom. The scene

Prompt 2: What colors are most prominent in this image?
Response:  The most prominent colors in this image are pink and white, as they are associated with the cherry blossoms and the sky.

Audio:

VLLM_USE_V1=1 python examples/offline_inference/offline_inference_audio_language.py --model-type qwen2_audio
Processed prompts: 100%|████████████████████████████████████| 1/1 [00:33<00:00, 33.90s/it, est. speed input: 12.80 toks/s, output: 1.42 toks/s]
The recited content in the audio is: 'First words I spoke in the original coronavirus a little feat of practical poetry Mary had a little lamb its fleece was white as snow and everywhere that Mary went the lamb was sure to go.'

[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.

🚀

[robertgshaw2-redhat]

cc @bvrockwell - FYI