Could anyone describe what “slow” means in this context? I also have tried to implement gpt4 into my app for production, but I would find that the full time for the stream to end would take ~40 seconds to show a full ~500 token length response. Is this inline with what other people have or unordinary?