[Feature]: Add vllm gpu backend support

[yuhao-zh]

Is your feature request related to a problem?

The PP provided by Sgalang is not very entriety, but vllm support PP for vast majority of models

Describe the Solution you'd like

add feature vllm backend

Alternatives Considered (Optional)

or monkey patch for sglang

Additional Context (Optional)

No response