Self-alignment, whereby models learn to improve themselves without human annotation, is a rapidly growing research area. However, existing techniques often fail to improve complex reasoning tasks due to the difficulty of assigning correct rewards. An orthogonal approach that is known to improve correctness is self-consistency, a method applied at inference time based on multiple sampling in order to find the most consistent answer. In this work, we extend the self-consistency concept to help train models. We thus introduce self-consistency preference optimization (ScPO), which iteratively trains consistent answers to be preferred over inconsistent ones on unsupervised new problems. We show ScPO leads to large improvements over conventional reward model training on reasoning tasks such as GSM8K and MATH, closing the gap with supervised training with gold answers or preferences, and that combining ScPO with standard supervised learning improves results even further. On ZebraLogic, ScPO finetunes Llama-3 8B to be superior to Llama-3 70B, Gemma-2 27B, and Claude-3 Haiku.

翻译：自对齐是一种无需人工标注、模型通过自我学习实现性能提升的快速发展的研究领域。然而，现有技术常因难以分配正确奖励而在复杂推理任务上改进有限。一种已知能提升正确性的正交方法是自一致性——一种在推理时基于多次采样以寻找最一致答案的方法。本研究将自一致性概念扩展至模型训练过程，提出了自一致性偏好优化（ScPO）。该方法通过在无标注的新问题上迭代训练，使模型偏好一致答案而非不一致答案。实验表明，在GSM8K和MATH等推理任务上，ScPO相比传统奖励模型训练实现了显著提升，缩小了与使用黄金答案或偏好进行监督训练的差距；且ScPO与标准监督学习结合能获得进一步改进。在ZebraLogic任务中，经ScPO微调的Llama-3 8B模型性能超越了Llama-3 70B、Gemma-2 27B及Claude-3 Haiku。