Need a better way to handle this then just to remove all variables until a common set is found. * OpenAI documentation https://platform.openai.com/docs/api-reference/chat/create e.g. Mistral models on Azure AI don't support `frequency_penalty` or `presence_penalty`. There might be a way to query and endpoint first to determine what variables are supported.