I’ve encountered what seems to be a bug when using the Playground, and the Ada engine.
My silly prompt (“Guy Fieri was the mayor of flavortown.”) is repeatedly hit with “The model predicted a completion that begins with a stop sequence, resulting in no output. Consider adjusting your prompt or stop sequences.” This is despite the fact that I have not specified any stop sequences. And even if I do specify stop sequences (and make them unusual/unlikely to be generated) I still find that I’m getting this error.
Has this happened to anyone else? Any suggestions?
edit: this has now also occurred with the DaVinci engine, and happens even if I attempt to enter the prompt multiple times.
Have you found solution for this?
I’m struggling with this too. Did anyone find a workaround?
There is a default stop sequence i think its <:endoftext:> or somthing similar… It happens when the model thinks it’s done with its completion before it hits your max responce length. It is not a bug and happens for all models.
The problem isn’t so much that it ends the completion, that’s actually a good thing.
The problem is when you attempt to generate the next completion, it simply doesn’t generate anything after that. Not unless you add more manual input to get it going again.
Is there a way to remove the default stop sequence?
A workaround is to just type a space or a newline after (or some other combination of whitespaces).
You’ll get a warning that using whitespace is inefficient due to how tokenisation works, but it should help you bypass the stop sequences issue.
same here i need a solution too please open ai reply fast as you can we struggling
I overrided the previous warning with “Warning: Your text ends in a trailing space, which causes worse performance due to how the API splits text into tokens.” by adding the trailing space, if you need an instant reply then try this.
Try checking if there is a proper space after a full stop(.) and comma(,) in your input sentence. It will get sorted.