Feature request: setting seed parameter of OpenAI's chat completions API

[haukelicht]

Thank you for creating and maintaining this awesome project!

OpenAI recently introduced the seed parameter to make their models' text generation and chat completion behavior (more) reproducible (see https://cookbook.openai.com/examples/reproducible_outputs_with_the_seed_parameter).

I think it would be great if you could enable users of your package to control this parameter when using OpenAI models as a backend (i.e., in the files here: https://github.com/iryna-kondr/scikit-llm/tree/main/skllm/models/gpt)

The seed parameter could be hard-coded

scikit-llm/skllm/llm/gpt/clients/openai/completion.py

Line 50 in 0bdea94

temperature=0.0, messages=messages, **model_dict

similar to setting temperature=0.0.

Alternatively, users could pass seed=<SEED> via **kwargs.

[OKUA1]

Hello @haukelicht,

Thank you for your suggestion. Are you aware whether the seed parameter plays any role if the temperature is set to 0? I have to admit that I did not dig deeper into this topic, but always had an impression that the seed is only relevant when the temperature is positive, otherwise the model is already (almost) deterministic.

But again, I might be completely wrong here.