It looks like GPT-4-32k is rolling out

This is great information:

If we’re benchmarking tokens per second we might as well do the same for humans :laughing:

So i did some digging and found the study “Oral Reading Fluency of College Graduates: Toward a Deeper Understanding of College Ready Fluency” it suggests that college graduates can read somewhere between 138 and 287 words per minute.

When converted to words per second, this is approximately 2.3 to 4.78 words per second. Assuming that one token is equivalent to 0.75 words, the reading rate in tokens per second would be approximately 1.73 to 3.59 tokens per second.

Link to the full article here.


But the impressive thing is that this isn’t reading rate, it’s writing rate! The reading rate for GPT-4 appears to be nearly instant, I’m guessing 5000 tokens per second or so. So it literally crushes humans.

But in responding, or writing, it beats most humans hands down. I’m guessing my response here took me maybe 1 or 2 tokens per second at best!

1 Like

Indeed, I should probably have made that a bit more clear :laughing:

Yep what I was thinking was:

Do we really need GPT to output faster than humans can read?

1 Like

The problem is one of latency – there may be post-processing (for example, moderation!) that needs to happen before the text can be presented to the user, and that post-processing might not even be able to start until the full generation is complete.

Output rate and latency are two different things

Output rate refers to how quickly something is produced, while latency refers to the time delay or waiting time before something is received.

1 Like

Is anybody currently using GPT-4 actually getting 8K now?

I ask because of this issue: GPT-4 Token Limit Reduced?

Using the paid API, it’s fine:

'object': 'chat.completion', 'created': 1684985539, 'model': 'gpt-4-0314', 'usage': {'prompt_tokens': 4555, 'completion_tokens': 411, 'total_tokens': 4966},


I thought I explained how output rate puts a lower limit on latency in many use cases.
Was that somehow unclear?

Not everyone is using the GPT output to be read by humans
Sometimes GPT output can go into chains where you use the output for another input. (just like autoGPT for example)

Or output code and run it, is another example

You could also make GPT make some decisions and execute commands.

1 Like

Good point @Neoony :heart:

I’m not saying faster isn’t better, it definitely is, I’m just saying that comparing GPT’s output tokens per second against human reading speed may be relevant when determining when to add more server capacity.

Mixing output rate and latency is not helpful in this case, the output rate is relevant when benchmarking the model, the latency is relevant if we’re benchmarking the networking.

1 Like

I want a gpt4 api to complete the testing of physical robots, who can help me? Please email me

In my primary use, humans are in the loop and read/edit the AI response. The AI does 90% of the drudgery work. Here there is no stringent latency requirement, since it could take several minutes for the human to get to the response. So to @N2U’s point, the output rate being faster than a human can read applies here.

However, with AI agents, and machine-to-machine interactions, I can really feel the latency kick in. So far, nothing “mission critical” being done here, but I can see that if you have an app and an impatient user on the other end, this could be disastrous for your app. But I’m curious how many people here are in this situation, and what is your use case? And is it “mission critical”, or more of a hassle.

1 Like

Benchmarking in a vacuum isn’t interesting to any real product. I am measuring end user experience.
User presses button. User has to wait until they see the result. How long does this take?

In your situation with an end user waiting, I would stream the answer back to them. The first token latency is a few seconds, just like it is in the Playground.

Above we were mostly taking last token latency and/or token output rates per second, and not first token streaming latency, which is probably the most interesting performance parameter for your application.

I already said that I must post process the data before I can show it.

(One common case of this is moderation solutions.)

So, no, I can in fact not stream the data, as I already said, and token generation rate directly impacts the latency to result for the user, as I already said, so generation at significantly faster than reading speed is both useful and important, as I already said.

Another case where generation faster than human reading speed is also important, is when generating steps to execute, such as in plugins.

So, I stand by the assessment that GPT-4 generation speed currently is a significant impediment.

I cannot find anything in OpenAI docs to support this. They only mention moderating to " You can use our free moderation endpoint and safety best practices to help you keep your app safe."

So I could interpret this to mean, in your case, stream to your users, then send the full result to moderation (after it’s done), and if it comes back flagged, the put the user on cooldown, or something similar, and see what happened.

I think you are taking something too literally, but feel free to share where you have to moderate everything before it goes out. If this were the case, the streaming feature is in violation of that policy. This is a contradiction.

But if you are referring to some other moderation or post processing thing, then yeah, as a batch, you aren’t getting past last token latency. But no AI model will have great responsive last token latency, at least for the next 6 months, until sub-quadratic models, that work and are good (and running on smoking hot FPGA’s or ASIC’s), come out.

I understand the argument, end user latency is a combination of multiple things, I’m sure people are aware of this. Benchmarking one thing separately still provides us with relevant information, that can be used by other people.

@jwatte I’m sorry if my information about human reading speed it’s relevant to you, it may be for other people, if you’re having issues with latency I’ll advise that you create a separate topic about this, as it will aid other people in helping you. This thread is for the (very-small) early rollout of GPT-4-32K

I’m expecting most people will be using the 32k while streaming to the end user, running an agent on GPT-4-32k is just too expensive :laughing:

1 Like

What a weird thing to say. It’s absolutely critical to benchmark all components of a process in order to understand where bottlenecks are occurring.


My latest use of 32k is to have long conversations with GPT and to analyze legal terms of service agreements. So more of a onesie-twosie thing, not production.

But I have been running 32k in parallel with 8k for model redundancy, randomly for 25% of all traffic. Nothing earth shattering or mission critical, just redundancy to get data out without retries.


Ha ha, what a coincidence, I’ve been using it for legal texts as well, more specifically to add legal explanations to specific context, very useful stuff :laughing:

1 Like