STAIF: A Stage-wise Optimization for Complex Instruction Following

Jian Hong, Chen Cheng, Quan Liu +2

Following complex instructions with multiple explicit constraints remains a fundamental challenge for large language models (LLMs). Existing alignment methods, such as DPO, optimiz…