By giving the same input. How can I make chat-gpt's output more stable or consistency?

Hi, I’m trying to build my own language model. However, because chat-gpt always giving different output due to the same question I provided, it causes my model can’t make the best result steadily and correctly everytime. Is there any solution can make chat-gpt always response the same result? It don’t need to be totally the same. Or there’s other solution can fix my situation?

If you are trying to make your own application, using one of OpenAI’s models, then you will find API parameters that you can adjust to control the “creativity”, the variety of outputs between runs.

temperature = 1 (default)
top_p = 1 (default)

Reduce either or both of these, all the way to 0 to make the AI produce the same thing almost every time.

1 Like

Thanks!! Will it depends on what model you use? Or any model will output almost the same since the temperature is close to 0?

Each AI model will have different training, and thus respond differently to your input.

What reducing the sampling parameters does is weight and confine the output to only the most likely words (tokens).

Low parameter values is what would give you identical output from identical input on the same AI model in successive runs, without creative use of language in the response generation.