Skip device and quant Pydantic validation to make plugin device work

[Yikun]

What this PR does / why we need it?

Since 4c2b38c , it converted configs to Pydantic dataclasses and enable Pydantic checks. Due to config.Device has strict validation, run the model test on the plugin device (vllm-ascend in my case) will not work and raise the error:

FAILED tests/singlecard/test_offline_inference.py::test_models[5-half-Qwen/Qwen2.5-0.5B-Instruct]
- pydantic_core._pydantic_core.ValidationError: 2 validation errors for DeviceConfig
device.literal['auto','cuda','neuron','cpu','tpu','xpu','hpu']
  Input should be 'auto', 'cuda', 'neuron', 'cpu', 'tpu', 'xpu' or 'hpu' [type=literal_error, input_value='npu', input_type=str]

See also here: https://github.com/vllm-project/vllm-ascend/actions/runs/15303026138/job/43048387981?pr=958

This patch try to resolve it by skip device field validation to make plugin device work.

How was this patch tested?

# Apply patch, do regresion test and CI passed
$ pytest -sv tests/singlecard/test_offline_inference.py::test_model
tests/singlecard/test_offline_inference.py::test_models PASSED

[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.

🚀

[Yikun]

cc @hmellor @simon-mo (who involved in 4c2b38c) Would you mind taking a quick look?

BTW, I considered to monkey patch the device in vllm-ascend but it doesn't works because the config initilizatized before apply patch, so we have to fix this on vLLM.

[hmellor]

This solution is fine as a hotfix. In future it might be nicer to implement our own Pydantic validator that doesn't error if it detects a plugin device

[Yikun]

CI passed, it's ready to go.

[DarkLight1337]

FYI this argument has been deprecated by #18399

[Potabk]

@DarkLight1337 ci passed, hope can merge

[DarkLight1337]

Can you also skip the validation for the quantization method to fix the failing quant test shown in #18852? Then I'll force-merge this