Hi everyone, the standard product description generator uses a davinci model. I’m wondering if it’s possible to achieve the same result using gpt-3.5-turbo? It’s obviously faster and cheaper. Advice and tips would be appreciated!
It’s more expensive and wouldn’t provide much of an advantage capability wise for this use
Yes, it’s definitely possible to that @op
You should read chat completionn docs
Keep in mind:
text-davinci-003
and chatGPT are different models, the key difference being that the text-davinci-003 series models are trained for completions while the gpt-3.5-turbo models are trained for conversation(chat completions), this means that the prompts for the models will be different for the same task. You prompt will have to be in chat form.
@sps thanks for the reply. Correct they are different models and that’s whats making me a bit confused. I want to build a product description generator that collects user input from multiple text fields and puts it into one single “prompt” or “chat” in this case which then should give me ability to use the gpt-3.5-turbo model if possible. I hope it is! What do you think?
It’s pretty easily doable.
Simply set a system message and then the user message based on those text inputs and send the chat completion request.
LMK if run into any roadblocks.
Hi @sps, Honestly I’m a bit confused. These are the parameters in that OpenAI has on their chat completion api reference:
{
“model”: “gpt-3.5-turbo”,
“messages”: [{“role”: “user”, “content”: “Hello!”}]
}
What confuses me is understanding what this means. What does “content” mean is this context?
Content = the content of message object.
This is based on ChatML.
The model generated messages will have "role": "assistant"
Thank you @sps ! so in this case, I wouldn’t use the parameters I mentioned but rather use the role of “assistant”? And the the content would be the “prompt” with the inputs the user creates right?
Generally gpt-3.5-turbo uses conversational prompts with user role to generate chat completions which are received with the role assistant.
However, yes, you can use it that way for completions.