Instruction-tuning is an effective avenue for making large language models (LLMs) better at
following real users' instructions. However, it's challenging in aligning to human preference …