Taipei, Mar 10 (EFE).- Taiwanese Hon Hai, known internationally as Foxconn, announced Monday the launch of its first extensive language model in traditional Chinese, developed in just four weeks thanks to chips from American company Nvidia.
This model, codenamed FoxBrain and originally designed for internal applications of the Hon Hai group, will be open source and "will be shared publicly in the future," the company explained in a statement.
According to the technology company, FoxBrain not only has "powerful comprehension and reasoning capabilities," but is also "optimized for the language style of Taiwanese users, showing outstanding performance in mathematical and logical reasoning tests."
"Our FoxBrain model adopted a very efficient training strategy, focusing on optimizing the training process rather than blindly accumulating computational power," said Li Yung-hui, director of the Artificial Intelligence Research Center at the Hon Hai Institute, which is responsible for this development.
"Through carefully designed training methods and resource optimization, we have successfully built a local AI model with powerful reasoning capabilities," he said.
During the training of this model, based on the Meta Llama 3.1 architecture, which has 70 billion parameters, Hon Hai used 120 Nvidia H100 graphical processing units placed in an Nvidia Quantum-2 InfiniBand network.
On application performance, Hon Hai said FoxBrain showed "overall improvements in mathematics" compared to the base model of Meta Llama 3.1 and "significant advances" compared to Taiwan Llama, another extensive language model in traditional Chinese.
"Although there is still a slight difference with the DeepSeek distillation model, the performance (of FoxBrain) is already close to the world's leading standards (...). Compared to inference models recently launched on the market, this more efficient and economical method marks a new milestone for the development of AI in Taiwan," said the technology company.
The development of this extensive language model is the latest example of collaboration between Nvidia and Foxconn: the firm is the main manufacturer of the American company's GB200 servers, which are essential for the development of the most cutting-edge artificial intelligence applications.
Founded in 1974, Foxconn is the world's largest contract electronics manufacturer, with factories and research centers in China, India, Japan, Vietnam and the US, among other countries. EFE
© 2025 EFE News Services (U.S.) Inc., source EFE Ingles