refactor: use config classes instead of input arguments

[inspiralpatterns]

llmflows/llmflows/llms/openai.py

Lines 38 to 42 in 6ce79a8

	model: str = "text-davinci-003",
	temperature: float = 0.7,
	max_tokens: int = 500,
	max_retries: int = 3,
	api_key: Union[str, None] = None,

In such cases, I would appreciate having a dataclass, pydantic.BaseModel, or protocol like OpenAIConfig so that I avoid having tons of input arguments for class initialisation. Furthermore, you can also fully validate the configs and catch input errors early in your code.

[stoyan-stoyanov]

don't you think this way is more explicit and easier to read?

[vickychudinov]

I think it depends on the number and nature of arguments. If we have more than say 5-6 then a dataclass is justified, below that - its easier the way you have it.

[inspiralpatterns]

don't you think this way is more explicit and easier to read?

I agree to some extent, however, if you make it into a dataclass or a pydantic class you will also gain validation out of the box. Moreover, if you decide to use configuration files to set these parameters, you will gain serialisation and deserialisation out of the box.