COIG-P: A High-Quality and Large-Scale Chinese Preference Dataset for Alignment with Human Values
M-A-P Team, Siwei Wu, Jincheng Ren, Xinrun Du, Shuyue Guo, Xingwei Qu, Yiming Liang, Jie Liu, Yunwen Li, Tianyu Zheng, Boyu Feng, Huaqing Yuan, Zenith Wang, Jiaheng Liu, Wenhao Huang, Chenglin Cai, Haoran Que, Jian Yang, Yuelin Bai, Zekun Moore Wang, Zhouliang Yu, Qunshu Lin
2025-04-09
Summary
This paper talks about COIG-P, a huge collection of Chinese question-answer pairs that helps AI chatbots learn to give better, safer answers by understanding what humans prefer.
What's the problem?
Existing Chinese training data for AI is too small, limited to specific topics, and needs lots of human checkers, making it hard to teach AI good manners and helpfulness in Chinese.
What's the solution?
COIG-P uses other AI models to automatically create millions of Chinese examples showing good vs bad answers, covering topics like chatting, coding, and stories, plus a special scoring AI to check quality.
Why it matters?
This helps create Chinese AI assistants that are more helpful and polite, like better customer service bots or study helpers that give culturally appropriate answers.
Abstract
Aligning large language models (LLMs) with human preferences has achieved remarkable success. However, existing Chinese preference datasets are limited by small scale, narrow domain coverage, and lack of rigorous data validation. Additionally, the reliance on human annotators for instruction and response labeling significantly constrains the scalability of human preference datasets. To address these challenges, we design an LLM-based Chinese preference dataset annotation pipeline with no human intervention. Specifically, we crawled and carefully filtered 92k high-quality Chinese queries and employed 15 mainstream LLMs to generate and score chosen-rejected response pairs. Based on it, we introduce COIG-P (Chinese Open Instruction Generalist - Preference), a high-quality, large-scale Chinese preference dataset, comprises 1,009k Chinese preference pairs spanning 6 diverse domains: Chat, Code, Math, Logic, Novel, and Role. Building upon COIG-P, to reduce the overhead of using LLMs for scoring, we trained a 8B-sized Chinese Reward Model (CRM) and meticulously constructed a Chinese Reward Benchmark (CRBench). Evaluation results based on AlignBench liu2024alignbenchbenchmarkingchinesealignment show that that COIG-P significantly outperforms other Chinese preference datasets, and it brings significant performance improvements ranging from 2% to 12% for the Qwen2/2.5 and Infinity-Instruct-3M-0625 model series, respectively. The results on CRBench demonstrate that our CRM has a strong and robust scoring ability. We apply it to filter chosen-rejected response pairs in a test split of COIG-P, and our experiments show that it is comparable to GPT-4o in identifying low-quality samples while maintaining efficiency and cost-effectiveness. Our codes and data are released in https://github.com/multimodal-art-projection/COIG-P.