As mentioned in #11 it should be easy to support base models that are deployed on azure by passing the `AzureOpenAI` client instead of the default one.