Foxconn’s brand is displayed in the course of the Hon Hai Tech Day on the Nangang Exhibition Middle in Taipei, Taiwan, Oct. 8, 2024. AP-Yonhap
Foxconn Know-how Group, the world’s largest electronics contract producer and main iPhone provider for Apple, launched its first Chinese language giant language mannequin (LLM) skilled on conventional characters, because the Taiwanese firm pushes ahead the usage of synthetic intelligence (AI) in factories.
The brand new FoxBrain mannequin was skilled in a „extra environment friendly and lower-cost“ technique inside simply 4 weeks, and units a brand new milestone within the improvement of Taiwan’s AI expertise, in line with an announcement issued on Monday by Foxconn, recognized formally as Hon Hai Precision Trade.
With a coaching course of powered by 120 Nvidia H100 graphics processing items (GPUs), FoxBrain excels in math and logical reasoning, in line with Foxconn.
It was initially designed for inside purposes within the firm, however Foxconn mentioned will probably be open sourced sooner or later, as a part of efforts to collaborate with expertise companions to develop its purposes and promote AI in manufacturing.
The Nvidia’s GPU (Graphic Processing Unit) is proven on this picture taken in Paris, Febr. 23, 2024. AFP-Yonhap
LLMs are the expertise underpinning generative AI providers like OpenAI’s ChatGPT. Open supply provides public entry to a software program’s supply code, permitting third-party builders to switch or share its design, repair damaged hyperlinks or scale up its capabilities.
Foxconn’s newest initiative displays the corporate’s purpose to push its personal AI breakthroughs when it comes to manufacturing effectivity.
That follows Chinese language start-up DeepSeek’s launch earlier this yr of its high-performance R1 reasoning mannequin, which was open-sourced and developed at a fraction of the price of AI fashions from bigger corporations like OpenAI, Google and Meta Platforms.
„In current months, the deepening of reasoning capabilities and the environment friendly use of GPUs have regularly develop into the mainstream improvement within the area of AI,“ mentioned Li Yung-Hui, director of the Synthetic Intelligence Analysis Centre at Hon Hai Analysis Institute, the analysis arm of the producer, within the assertion.
„Our FoxBrain mannequin adopted a really environment friendly coaching technique, specializing in optimizing the coaching course of reasonably than blindly accumulating computing energy,“ he mentioned.
The brand new mannequin was based mostly on the Meta Llama 3.1 structure with 70 billion parameters. Foxconn claimed that it outperformed Llama-3-Taiwan-70B, one other open-source mannequin fine-tuned on conventional Chinese language characters and English knowledge utilizing the Llama-3 structure, in most classes of TMMLU+, a benchmark for conventional Chinese language language understanding.
Fashions developed by Chinese language corporations like DeepSeek are typically skilled for higher understanding of simplified Chinese language characters, that are used on the mainland.
Final November, the corporate mentioned it was working with Nvidia to leverage „digital twin“ expertise in manufacturing and provide chain administration.
The initiative makes use of Nvidia’s Omni verse to streamline international manufacturing facility operations, improve resilience and guarantee constant high quality.
Learn the full story at SCMP.
