When I used "virgin" LLama in a chat context, the results were very inconsistent and useless. It would output repeating phrases (even with 2.1 repeat penalty) o

When I used "virgin" LLama in a chat context, the results were very inconsistent and useless. It would output repeating phrases (even with 2.1 repeat penalty) or many emojis. Vicuna was better, but still had an "openAI" personality to it.