Force Tokens as opposed to Bad_Word_Ids

I saw this on Stackoverflow, and thought I would take this here. In other words, if I was to put “hey” into this parameter, hey would appear somewhere in the output.

2 Likes

Maybe something like this works although I must say haven’t tried it. Also check this feature request and feel free to react/comment there. If implemented it would allow per token control, e.g. force “hey” by using a high positive score.