[Model][Bugfix] Support TP for PixtralHF ViT

[mgoin]

Currently running a PixtralHF model with TP will cause an input shape error while profiling through ViT attention:

RuntimeError: Error in model execution (input dumped to /tmp/err_execute_model_input_20241117-211655.pkl): shape '[1, 20480, 16, 64]' is invalid for input of size 5242880

Verified with MMMU eval on an FP8 model

Server:

vllm serve nm-testing/pixtral-12b-FP8-dynamic --max-num-seqs 30 --max-model-len 30000 --limit-mm-per-prompt image=5 --tensor-parallel-size 4

Client:

python -m eval.run eval_vllm --model_name nm-testing/pixtral-12b-FP8-dynamic --url http://0.0.0.0:8000 --output_dir output/ --eval_name "mmmu"
Waiting for VLLM server to come online at http://0.0.0.0:8000/health ...
Timeout is 120s
Waiting for server (0s) ...
Server is up!
Loading lmms-lab/MMMU [validation]: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 900/900 [00:10<00:00, 89.20it/s]
Querying model: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 900/900 [01:52<00:00,  7.99it/s]
100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 900/900 [00:00<00:00, 13029.97it/s]
================================================================================
Metrics:
{
    "explicit_prompt_relaxed_correctness": 0.51,
    "anywhere_in_answer_relaxed_correctness": 0.51
}
================================================================================

[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.

🚀

[ywang96]

🦭

[DarkLight1337]

There seems to be some issue with V1 eval.