Early LLMs used to have this often. I think's that where the "repetition penalty" parameter comes from. I suspect output quality can be improved with better sam

Early LLMs used to have this often. I think's that where the "repetition penalty" parameter comes from. I suspect output quality can be improved with better sampling parameters.