OpenAI compatibility issues

Playing around this afternoon swapping in gemini.

However, ran into the following issues:

  • pydantic support isn’t equal - integer Literals aren’t supported (only Str)
  • usage=None - can’t see a way to get token usage (ie completion.usage.completion_tokens doesn’t exist)

So this still requires some adaptation for multi-model support/model swapping. But super happy to see it’s coming!