refactor: consider data types for logically-grouped variables

[inspiralpatterns]

llmflows/llmflows/llms/openai.py

Line 80 in 6ce79a8

return text_result, call_data, model_config

I would always consider whether to make my code more informative and cleaner by e.g. making a specific data type out of a tuple.

Consider e.g.

from dataclasses import dataclass

@dataclass
class OpenAIResult:
    generated_text: str
    call_data: CallData
    model_config: ModelConfig

where CallData and ModelConfig are other data types as well: everything becomes much more self-documented. Furthermore, you don't have to build maps (i.e. dict) on the fly inside the methods.

You could also define a common interface, e.g. BaseLLMResult and have OpenAIResult extend it.

[stoyan-stoyanov]

I am trying to avoid defining these until I start adding LLMs from other providers. Any idea on how we can make this generic enough beforehand?

[inspiralpatterns]

I am trying to avoid defining these until I start adding LLMs from other providers. Any idea on how we can make this generic enough beforehand?

Yes: you can define abtract data types and extend them whenever you get into a different provider e.g., OpenAIResult could just extend BaseLLMResult - if we assume that you will get a result most times. 😀

The beauty of having abstract interfaces is that you do not have to think about all the possible providers at once.

[stoyan-stoyanov]

Reevaluating this since I added support for another LLM and the need for this becomes more obvious. My worry is that dataclasses are not serializable by default