Can Instruct models be trained?

Hey, I’m wondering if there’s a way to train Instruct models based on user feedback and specialise them in certain niches? Is there a way to penalize them when a wrong completion is generated?

Going through the documentation I can’t seem to find a way of doing that, but worth asking here.

Thanks for the reply. Well, sometimes the completion is off-topic or the quality is just poor. I was wondering if there is at least a way to penalise the model when the completion is say rejected by the human for different reasons (off-topic, poor quality, etc).


It would be nice if at least the completion rating could be done via the API .

1 Like