feat: Support VLM in reference_hf

[gaocegege]

Motivation

Thanks for the project! It’s like a handy little Swiss Army knife—small but efficient.

I’ve been experimenting with QWen2 VL model and ran into some issues with bitsandbytes, so I wanted to double-check if everything works with reference_hf. Turns out, the script didn’t support VLM initially, so I spent some time adding that functionality.

I hope this PR can serve as an entry point for me to get familiar with the contribution flow of SGLang. I am glad to have the chance to play more with SGLang and explore further. (although not much time :-)

Modifications

scripts/playground/renference_hf.py

• Implemented a new function to support VLM.
• Introduced a new command-line argument --model-type=text,vlm to direct execution to the appropriate function based on the model type.

Checklist

• Format your code according to the Contributor Guide.
• Add unit tests as outlined in the Contributor Guide.
• Update documentation as needed, including docstrings or example tutorials.

[gaocegege]

By the way, the link https://github.com/sgl-project/sglang/blob/main/docs/references/contributor_guide.md in the GitHub issue template is 404.