Reverse Preference Optimization for Complex Instruction Following
This addresses a critical challenge in making LLMs more reliable for real-world applications with complex instructions, representing a strong specific gain rather than a foundational breakthrough.
The paper tackles the problem of aligning large language models with complex instructions containing multiple constraints by proposing Reverse Preference Optimization (RPO), which dynamically reverses constraints to reduce noise in preference pairs and improve instruction following, resulting in average improvements of 4.6 and 2.5 points over DPO on two benchmarks and surpassing GPT-4o with a 70B model.
Instruction following (IF) is a critical capability for large language models (LLMs). However, handling complex instructions with multiple constraints remains challenging. Previous methods typically select preference pairs based on the number of constraints they satisfy, introducing noise where chosen examples may fail to follow some constraints and rejected examples may excel in certain respects over the chosen ones. To address the challenge of aligning with multiple preferences, we propose a simple yet effective method called Reverse Preference Optimization (RPO). It mitigates noise in preference pairs by dynamically reversing the constraints within the instruction to ensure the chosen response is perfect, alleviating the burden of extensive sampling and filtering to collect perfect responses. Besides, reversal also enlarges the gap between chosen and rejected responses, thereby clarifying the optimization direction and making it more robust to noise. We evaluate RPO on two multi-turn IF benchmarks, Sysbench and Multi-IF, demonstrating average improvements over the DPO baseline of 4.6 and 2.5 points (on Llama-3.1 8B), respectively. Moreover, RPO scales effectively across model sizes (8B to 70B parameters), with the 70B RPO model surpassing GPT-4o.