can anyone help with how we can prevent application hitting rate limit on groq apart from caching .we want the models to behave in a specific conversational style which requires meta prompts to be fed to the model each call is there a way how can we avoid it without requiring to fine tune the model and self host it