I’m extracting the content string from the API response, and for my implementation, I don’t want any new line characters in the response as it causes a bug in my program. Short of programatically editing the string, is there a way to prevent the API response from including these? I thought my prompt was firm enough, but has about a 50% success rate.
My prompt to handle this “IMPORTANT: The answer MUST always be a single string, with no new lines, NEVER put a blank line or blank lines in the response, NEVER use the special character \n and do not include blank lines.”
An example of what the API will return “engagement, and goal alignment. \n\nWe will maintain effective communication”
I like this logit_bias approach. Well found! Although I wouldn’t bet my horses on it ….
What is the guarantee that the \n token’s ID will remain the same in the future?
IMHO I do believe that removing the \n characters programmatically is more sustainable.
After almost a year of API testing (in JavaScript) I am used to formatting the inferences as I’d like them to be. Consistency is close to 100%.
Tip: learn Regex. This a very powerful tool when it comes to string manipulation. Also ChatGPT can help you craft good Regexes.
Thank you both, I hadn’t seen the logit_bias parameter before, thanks for sharing this. I’ve used it as suggested and so far it’s working for my program. I’ll switch to regex if the problem persists, thanks for the recommendation!