China is making progress in using supercomputers and domestic AI chips for training large language models, despite restrictions on exporting Nvidia’s advanced AI chips from the US. Academician Zheng Weimin stated that China has 14 national-level supercomputer centers with surplus computing power that can be used for AI training. Zheng’s team has successfully trained large language models using a domestically-developed supercomputer, achieving results comparable to Nvidia’s chips at a fraction of the cost. The National Supercomputer Center in Tianjin has also developed an in-house large language model called Tianhe Tianyuan, leveraging the center’s supercomputing power. This model can handle both scientific and AI computing tasks, offering a cost advantage over Nvidia chips. The center has been approved to establish a national public computing open innovation platform for next-generation AI. Collaborations with partners in various industries, such as medicine and environmental protection, have led to the development of applications like a “hospital AI brain” for disease assessment and an AI assistant for customs data analysis. Overall, China is harnessing its supercomputing capabilities to advance AI technology and applications in various sectors.
Source link
Source link: https://www.chinadaily.com.cn/a/202404/29/WS662f57baa31082fc043c4ad0.html
GIPHY App Key not set. Please check settings