Learning from Response not Preference: A Stackelberg Approach for LLM Detoxification using Non-parallel Data

Re-assign community