Usually OpenAI response is within 2 seconds, but llama2 takes 20+ seonds for the same.
How are they tackling the performance?