Today, China Unicom announced that its subsidiary, Unicom Data Intelligence Co., Ltd., has officially released the Yuansheng Thinking Chain Large Model. This innovative achievement has been fully open-sourced in various communities such as GitHub, Modao, and Shizhi, marking an important step for China Unicom in the field of artificial intelligence.
The Yuansheng Thinking Chain Large Model is the first open-source general thinking chain large model launched by a central enterprise in the industry. It demonstrates powerful slow thinking capabilities and general reasoning abilities across multiple disciplines and scenarios. It not only adapts to different tasks and levels of difficulty for slow thinking but also significantly reduces resource consumption, showcasing outstanding performance.
According to China Unicom, the Yuansheng Thinking Chain Large Model has outperformed the current best general language models, including OpenAI's GPT-4 and Deepseek V3, as well as the open-source thinking chain model Tongyi Qianwen QwQ, in evaluations on mainstream rankings. This achievement fully demonstrates the technological strength and competitiveness of the Yuansheng Thinking Chain Large Model.
The Yuansheng Thinking Chain Large Model achieves adaptive slow thinking through two strategies: task adaptation and difficulty adaptation. In the evaluation set of non-inference tasks, this model tends to generate shorter answers while ensuring accuracy, thus improving response efficiency. Additionally, when evaluating generated long thinking chain data, the model comprehensively considers the difficulty of the questions and the length of the generated answers, using reinforcement learning to match the answer length with the question difficulty, further enhancing the model's accuracy and practicality.
According to the official demonstration, the Yuansheng Thinking Chain Model has been successfully applied in various fields such as physics, chemistry college entrance exam questions, the 24-point game, and logical reasoning, showcasing broad application prospects and huge market potential.
The release of the Yuansheng Thinking Chain Large Model by China Unicom is not only an important breakthrough in the field of artificial intelligence for the company but also injects new vitality into the development of artificial intelligence technology in China and globally. In the future, China Unicom will continue to increase its investment and research efforts in artificial intelligence, promoting innovation and application of AI technology, and contributing more wisdom and strength to economic and social development.
GitHub:
https://github.com/UnicomAI/Unichat-32B-c1.git
Modao:
https://www.modelscope.cn/UnicomAI/Unichat-qwen2.5-32B-c1.git
Shizhi:
https://wisemodel.cn/models/UnicomLLM/Unichat-qwen2.5-32B-c1