paper-with-me

Papers

Self-Consistency Preference Optimization

2024-11-06 · Archiki Prasad, Weizhe Yuan, Richard Yuanzhe Pang, Jing Xu, Maryam Fazel-Zarandi, Mohit Bansal, Sainbayar Sukhbaatar, Jason Weston, Jane Yu

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.

📄 PDF Abstract BibTeX arXiv:2411.04109

Code (0)

등록된 구현이 없습니다.

Tasks

GSM8KMath

Similar Papers 제목 키워드 기반

Self-Consistency of the Internal Reward Models Improves Self-Rewarding Language Models

2025-02-13 · Xin Zhou, Yiwen Guo, Ruotian Ma, Tao Gui 외

Aligning Large Language Models (LLMs) with human preferences is crucial for their deployment in real-world applications. Recent advancements in Self-Rewarding Language Models suggest that an LLM can use its internal rewa…

Atomic Consistency Preference Optimization for Long-Form Question Answering

2025-05-14 · Jingfeng Chen, Raghuveer Thirukovalluru, Junlin Wang, Kaiwei Luo 외

Large Language Models (LLMs) frequently produce factoid hallucinations - plausible yet incorrect answers. A common mitigation strategy is model alignment, which improves factual accuracy by training on curated factual an…

FormLong Form Question AnsweringQuestion Answering

SGDPO: Self-Guided Direct Preference Optimization for Language Model Alignment

2025-05-18 · Wenqiao Zhu, Ji Liu, Lulu Wang, Jun Wu 외

Direct Preference Optimization (DPO) is broadly utilized for aligning Large Language Models (LLMs) with human values because of its flexibility. Despite its effectiveness, it has been observed that the capability of DPO …

Language ModelingLanguage Modelling

VideoGPA: Distilling Geometry Priors for 3D-Consistent Video Generation

2026-01-30 · Hongyang Du, Junjie Ye, Xiaoyan Cong, Runhao Li 외 arxiv

While recent video diffusion models (VDMs) produce visually impressive results, they fundamentally struggle to maintain 3D structural consistency, often resulting in object deformation or spatial drift. We hypothesize th…

Video Generation

M3PO: Multimodal-Model-Guided Preference Optimization for Visual Instruction Following

2025-08-17 · Ruirui Gao, Emily Johnson, Bowen Tan, Yanfei Qian arxiv

Large Vision-Language Models (LVLMs) hold immense potential for complex multimodal instruction following, yet their development is often hindered by the high cost and inconsistency of human annotation required for effect…

visual instruction following