Aligning CodeLLMs with Direct Preference Optimization