A finetuned LLaMa model on an adversarial objective. Same license as original LLaMa model.