The logit_bias parameter is great for black listing tokens. But what if I want to do the opposite? Ban all tokens except for the ones I white list. Is there an elegant way to currently accomplish this?
This would be incredibly useful for classification tasks when we want the output to always be one of several labels.