Hello everyone,
I am calling OpenAI API to deal with a text understanding task.
The problem is, the output is very unstable. Sometimes it works very well, and sometimes it just give any random text. I tried to make it very clear in the prompt but so far no success.
Could you provide some hints to make the output be stable and consistent?
Many thanks
The lack of what endpoint, what model, what API parameters, what output features, what system prompt, what presentation of the text or text or document extraction, its format or language, what instruction, or anything to go on would leave the normal person guessing.
I however, am a mind reader:
Alongside parameters like āmodelā or āmessagesā in you chat completions request, use a parameter "top_p": 0.01
Drop that in, and see if you get more consistently useful results between trials.