国务院新闻办官网英文刊发“本源悟空”全球首次运行十亿参数级AI大模型微调任务

转自中华人民共和国国务院新闻办公室官网

原稿标题：

China's quantum computer pioneers AI task with enhanced efficiency

原链接：

http://english.scio.gov.cn/m/chinavoices/2025-04/09/content_117813292.html

A visitor takes photos of an Origin Wukong superconducting quantum computer model at the 2024 World Manufacturing Convention in Hefei, east China's Anhui Province, Sept. 20, 2024. (Xinhua/Huang Bohan)

A Chinese quantum computing firm has completed the large AI model task on a quantum computer with remarkable efficiency, as reported on Monday by Science and Technology Daily.

Origin Quantum, a startup based in the eastern Chinese city of Hefei, has successfully fine-tuned a billion-parameter AI model on Origin Wukong, the country's independently developed third-generation superconducting quantum computer. The company has hailed this achievement as a world-first breakthrough.

This marks the first real-world application of quantum computing in large-model tasks, demonstrating that current quantum hardware is capable of supporting AI model fine-tuning, said Chen Zhaoyun, a scientist from Hefei Comprehensive National Science Center.

The experimental results showed that reducing the number of parameters by 76 percent led to an 8.4 percent improvement in training performance.

The advancement demonstrates that quantum computing can enable lightweight large models, offering a promising new approach to overcoming future computing power constraints.

Fine-tuning is the process of adapting a general-purpose large AI model, such as DeepSeek, for specialized applications, like medical diagnosis or financial risk analysis, by retraining it on domain-specific data.

The 72-qubit Origin Wukong is a programmable and deliverable superconducting quantum computer. Since its launch, this system has provided quantum computing cloud services to a global user base exceeding 23 million across 139 countries, while successfully executing 350,000 quantum computing tasks.