[Bug]: how to get purely deterministic output for gpt-oss-120b?

[tonyaw]

Your current environment

The output of python collect_env.py

Your output of `python collect_env.py` here

🐛 Describe the bug

I'm using vllm v0.10.1 to serve "openai/gpt-oss-120b".
I need to get purely deterministic output. (Same input shall get exactly same output).

I found following ticket:
https://huggingface.co/openai/gpt-oss-20b/discussions/23
But I still can't get deterministic output even with following configuration:

"seed": 42,  # should be a very large number?
"temperature": 0,
"n": 1,
"top_p": 0.01,  # top_p must be in (0, 1]
"extra_body": {
    "top_k": 1,
    "min_p": 1,
    "repeat_penalty": 1,
},

Could you please help to guide me?
Is it possible to get purely deterministic output via vllm v0.10.1 + "openai/gpt-oss-120b"?
If Yes, could you please help to provide the parameter for openai client?

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.

[ZJY0516]

You can take a look at this https://docs.vllm.ai/en/latest/usage/reproducibility.html

[tonyaw]

@ZJY0516 ,
Thanks for your info!
I think the doc you provided is to make vllm software itself be deterministic.
Besides that, how to make output of "openai/gpt-oss-120b" be deterministic via vllm?
Any suggestion here?

[ZJY0516]

As I understand it, the model's output is guaranteed to be deterministic if two conditions are met: 1) vLLM itself is configured for deterministic execution, and 2) a deterministic sampling strategy (e.g., setting temperature=0) is employed.