Fine-tune Llama 3 using Direct Preference Optimization