[Usage]: Option to set system_fingerprint?

[vrdn-23]

Your current environment

The output of `python collect_env.py`

How would you like to use vllm

I would like to know if there is an option to set the system_fingerprint returned in Chat Completions API via vLLM. I looked through the code base but couldn't find anything concrete, so thought I should ask here. I realize that OpenAI has announced they're deprecating this particular return value soon, but I think there is an inherent value in having a field which can help distinguish between servers that are being run in-house.

I would be happy to raise a PR if this functionality doesn't exist currently, but I'd be hesitant to do so, if the system_fingerprint field is going to be removed soon according to OpenAI's deprecation timeline.

For more context, we are interested in using that response field to be able to identify which GPUs our models are running on when the request is being served.

cc @njhill @robertgshaw2-redhat @russellb would love to hear your thoughts/concerns!

Before submitting a new issue...

• Make sure you already searched for relevant issues, and asked the chatbot living at the bottom right corner of the documentation page, which can answer lots of frequently asked questions.

[russellb]

We could have an option to allow setting it to whatever value you want. I don't see any harm in that.

[vrdn-23]

Awesome! Thanks for the quick turnaround @russellb. I'll make a draft PR for review sometime this week then!
My only concern was since OpenAI seems to have decided to deprecate the field, whether vLLM will continue to keep the field or phase it out pretty soon (in which case I don't want to waste effort writing it up).

[russellb]

Do you have a reference to the deprecation?

[vrdn-23]

https://developers.openai.com/api/reference/resources/chat