Failure to specify the maximum model length when using the vllm engine causes loading failure！

[tensorflowt]

Feature request / 功能建议

When using the vllm engine of xinference, it is expected to support commonly used inference parameters for easy use.
For example, the maximum model length.

Motivation / 动机

When using a model that xinference already supports, but the default model maximum length cannot be directly loaded due to GPU graphics card resource limitations, the model loading fails. At this time, it can be solved by registering the model, but it is more troublesome. I hope to expose the commonly used vllm inference parameters for easy use.

Your contribution / 您的贡献

xinference launch --model-engine vllm \
--model-name deepseek-r1-distill-qwen \
--size-in-billions 32 \
--model-format pytorch \
--quantization none \
--max-model-len 18080 ```

[qinxuye]

--max_model_len 18080

In this way to specify model len for vllm.

But we'd like to unify all this option for various engines.