LLMs aren't the fastest thing in the world, how much data can you realistically parse per second?