Hi there, I am using davinci to create a chatbot, but when I use playgrounds I get very consistent results, but when using the API is completely random. At the beginning of the prompt, I use a small text to define what is the conversation about. while in playgrounds it respects the text, the API does not. Any Idea?
#Different results between the API and Playgrounds
blissful bone
rough flower
Are you saving the conversation history and sending the entire converastion history every time with the API? You need to do this
potent flint
Same problem here. With API I get much different answers. I assume that is problem with history but if I give my php app same curl request it return a much different answers. Do I must have \n before stop points( "Human" and "AI")? I give history like string Human: Who are you? Ai: I am an Ai assistant. Human: Can you explain quantum computers?
blazing sorrel
I'm also using text-davinci-003 with mathematical prompts. It works perfectly in the Playground but the output in my web app using the API is really bad and illogical
And I've seen that a lot of people have this issue but I have not found anyone who has been able to solve this yet
manic totem
I'm also using text-davinci-003 with mathematical prompts. It works perfectly in...
might need to adjust the settings? like temperature, etc?
blazing sorrel
might need to adjust the settings? like temperature, etc?
Unfortunately, my settings are exactly the same on both platforms. It's very strange. Temperature is 0 and max tokens are set to 50