Why do I get incomplete response and output

Asked to generate some sample data and always get about 25~26 records taking a good few seconds which I suppose is due to compute limitations? I’m using the plus version but seems the performance is same with free or plus.

And more importantly why do I only get 25 records when I asked to generate 100 records, seems like its capped to minimise the compute use?

Hi @AIowl,

yes, ChatGPT limits the output to a certain level. A small example: If you would say that you would have 1 million results and there would be no limits, then you could quickly cripple the service.

However, there is a way to get your 100 results, even if it is a bit more complicated:

Input 1 (Prompt & Specification how to awnswer):

“Your Prompt”, Split your answer in four replies, send me first the first answer now when I send “next” send the second one.

Input 2 (let ChatGPT continue)


and so on…

This works reliable for me, hope this helps you as well :slight_smile:

1 Like

One other way I’d advise would be to use the Playground, there you can increase the tokens used for your reply, I think standard config of 256 is used via regular access, this also fits the aprox. 25 Results you are getting

Please note that there might be a different pricing model so I’d advice to work with this if the previous suggested solution does not work for you.

Playground - OpenAI API

Given your example, I would respond with following prompt:

Continue starting with transaction date 2023-03-10:

What’s curious is that your example ended with basically the current date. I wonder if it would produce more results if you said something like Starting with date 2023-01-01 in your initial prompt.

thanks, what linus suggested here works for me, not ideal but it works

1 Like

I am also facing this issue. Whenever am increase the token, its still getting the incomplete the answer


The GPT models are not trained to produce long form output, you can request longer output in the prompt, but the best way is to make several API calls and concatenate the results.

1 Like

The original poster here ran the model out of output tokens. You can see this by the incomplete line.

“Continue generating” would also appear for a token-limited output, so I can conclude this is recalling a chat to make the screenshot, or so early the feature didn’t exist yet.

63 transactions produced from the same prompt asking for 100, with AI making one less column to consume tokens, and the “continue button” appears:

You can see the known max_token value that ChatGPT uses. The table markdown includes lots of extra space tokens not seen in the UI after markdown->HTML conversion, extra spaces that just burn through the token count:

Are you similarly using ChatGPT? The text looks abruptly cut off in the middle?

Copy the text of the reply with the copy button.
Go to https://tiktokenizer.vercel.app/ and clear the box until it reports 0 tokens.
Paste the response into the box.
See if the output isn’t exactly the expected maximum output length of ChatGPT.

API will report the “finish reason” as being “length” if the max_token parameter sent to the API isn’t large enough to hold the full response.