Shanghai, China – In a significant stride for financial technology, Shanghai University of Finance and Economics (SUFE) has unveiled its first R1-class reasoning large language model (LLM) specifically designed for the financial domain. Developed by Professor Liwen Zhang of the School of Statistics and Data Science at SUFE and his Financial AI Large Language Model Lab (SUFE-AIFLM-Lab), in collaboration with the Institute of Data Science and Statistics, Caiyue Xingchen, and Dishui Lake Advanced Finance Institute, the model, dubbed Fin-R1, boasts impressive performance despite its relatively small 7-billion parameter size.
The achievement underscores the growing capabilities of academic institutions in driving innovation within specialized AI fields. Fin-R1 not only surpasses comparable models in its parameter range but also achieves an average score of 75, placing it second overall, a mere 3.0 points behind the industry benchmark, DeepSeek-R1, which utilizes a significantly larger 671-billion parameter architecture.
Leveraging Open-Source and Innovative Training:
Fin-R1 is built upon the Qwen2.5-7B-Instruct model and employs a two-stage hybrid framework trained on a meticulously curated, high-quality financial reasoning dataset. This innovative approach enables a closed-loop system for financial reasoning, demonstrating a successful transition from technological breakthrough to potential industrial application within the university setting.
Professor Zhang’s team, comprised of doctoral student Xin Guo, master’s student Zhaowei Liu, and key members Weige Cai, Jinyi Niu, Lingfeng Zeng, Fangqi Lou, Zixuan Wang, Jiajie Xu, Xueqian Zhao, and Ziwei Yang, collaborated with Dr. Zuo Bai and team members Dezhi Chen, Sheng Xu, and Chao Li from Caiyue Xingchen on this project.
Key Features and Implications:
- R1-Class Reasoning: The model is designed for advanced reasoning tasks within the financial domain, enabling it to analyze complex financial data and provide insightful interpretations.
- 7B Parameter Efficiency: Its compact size makes it more accessible and deployable compared to larger models, opening doors for wider adoption in various financial applications.
- Superior Performance: The model’s performance, rivaling that of much larger models, highlights the effectiveness of the training methodology and the quality of the financial reasoning dataset.
- Open-Source Availability: The model is available on GitHub (https://github.com/SUFE-AIFLM-Lab/F), promoting collaboration and further development within the research community.
Conclusion:
The development of Fin-R1 represents a significant milestone for SUFE and the broader financial technology landscape in China. By open-sourcing this powerful yet efficient model, SUFE is fostering innovation and collaboration, paving the way for future advancements in AI-driven financial solutions. This achievement underscores the increasing importance of academic research in driving cutting-edge technological developments and bridging the gap between research and practical application in specialized domains. The success of Fin-R1 signals a new era of autonomous innovation in the field of financial technology.
References:
- SUFE-AIFLM-Lab. (2024). Fin-R1 [Large language model]. GitHub. https://github.com/SUFE-AIFLM-Lab/F
- 机器之心. (2024, March 26). 上财开源首个金融领域R1类推理大模型，7B模型媲美DeepSeek-R1 671B满血版性能. [Machine Heart]. Retrieved from (Original article URL – if available, otherwise remove this line)
Views: 0