Usage and Meta classes are required in responses but specific to OpenAI-like APIs

[aimeos]

Prism is currently tailored to OpenAI and compatible APIs, esp. the Usage and Meta classes that are required in responses. Other LLM APIs doesn't provide this information or different ones, e.g. the number of used and still available credits.

Therefore, usage and meta information should be treated as optional and provider specific. The only numbers that are available across many providers are "used" and "available" but the values have different meanings across providers.

[sixlive]

I'm open to this. I think I need to think a little more deeply about this

[ChrisB-TL]

All or some of both usage and meta are used in many none OpenAI compatible providers.

Most of the properties are already optional. If we need to make more of them optional, that's fine but I wouldn't support making them provider specific (not least as it's a fairly big API change for no benefit).

[aimeos]

All or some of both usage and meta are used in many none OpenAI compatible providers.

Mainly the current ones, not the ones I'm implementing at the moment.