Foxconn unveils its first AI model 'FoxBrain,' to be open-sourced
What's the story
Foxconn, the world's largest contract electronics manufacturer, has launched its own large language model (LLM) called FoxBrain.
Created by the Hon Hai Research Institute, the AI model was originally developed for in-house use at the company.
It is set to become an important engine for three major Foxconn platforms: Smart Manufacturing, Smart EV, and Smart City.
The firm also plans to open-source FoxBrain in the future.
AI advancement
A technological milestone for Taiwan
FoxBrain marks a major leap in Taiwan's AI technology.
The model was trained on 120 NVIDIA H100 GPUs and the NVIDIA Quantum-2 InfiniBand network, with support from the Taipei-1 Supercomputer and technical consultation from NVIDIA.
As Yung-Hui Li, Director of the Artificial Intelligence Research Centre at the Hon Hai Research Institute said, "Our FoxBrain model adopted a very efficient training strategy, focusing on optimizing the training process rather than blindly accumulating computing power."
Technical details
FoxBrain's architecture and performance
FoxBrain is built on Meta's Llama 3.1 architecture with 70 billion parameters and a context window length of 128,000 tokens.
The institute observed that its performance is close to world-leading standards but slightly behind some models of China's DeepSeek.
However, it outperforms Llama-3-Taiwan-70B on the same scale, with a focus on mathematics and logical reasoning.
Foxconn plans to share more about FoxBrain at NVIDIA's GTC 2025 conference.