I set the max-token 300 by api request, but a maximum of 20 is used, why and how to fix it?
If you make the same request in the Open AI Playground, then use all 300
Are you using the same model?
GPT-3 is notoriously bad at understanding word counts with a zero-shot prompt.
When you use the api, the max tokens is used for both the prompt and completion. Your prompt may be using most of them and not leaving enough for the completion
I don’t know about the playground. But maybe that is just for the completion and being added to the prompt token count in the background. Just a thought. Might be the case
How long is your prompt?
I use “text-davinci-003” model
I would recommend using one or two examples in your prompt… and using a stop-token.
I thought you said you requested and had a problem with the completion length?
No, as it turned out, the problem is that the request is being made, but an incomplete response is returned, and in the settings you can see that the secret keys were not used, although I have them inserted in the request
But your screenshots upthread show that you did get a result.
Sounds like you need to reach out to support. Good luck!
Already wrote 3 times, even a return message does not come (((