How can developers optimize the performance of azure openai models