Nice extension, but can you fix it to follow different prompt formats. For example, it doesn't respect or use <|end_of_text|> tokens for ChatML models and keeps outputting more and more text.