I’d recommend reading our tutorial on prompt design. There are a few things outlined there that you’ll want to implement, including:
Providing instructions (there are currently no instructions in your prompt)
Having a clear separator (### is recommended instead of the # in your prompt)
Only using the needed response length (yours is 150 tokens, while less than half that are needed, which makes extensive repetition possible)
Provide multiple stop sequences when needed (in your example, adding “Pain:” in addition to “###” should do, assuming you preface the completion with “Pain:”)
Use penalty settings to decrease repetition (yours are currently set to 0)
Try other models (e.g. davinci-instruct-beta)
Could you implement these and let me know if you’re still facing any issues?
You can see a field called “Stop Sequences” on the right side of the Playground. Here, you can add any characters that, if generated, will act as a “stop sequence” to stop the API from continuing to generate.
For example, by adding the stop sequence of “Pain:”, the completion can’t go on to generate another “Pain:” statement.