Foxconn (FXCOF, Financials) introduced its first large language model, FoxBrain, which was trained in four weeks using a cost-efficient method, the company announced.
Supported by Nvidia Quantum-2 InfiniBand networking and Taipei-1 supercomputer, the AI model was built utilizing 120 Nvidia (NVDA, Financials) H100 GPUs. Through Nvidia NeMo, Foxconn also got technological direction.
Complementing Foxconn's larger emphasis on industrial artificial intelligence applications, FoxBrain is meant to run Foxconn's Smart Manufacturing, Smart EV, and Smart City projects. Inspired by Meta Platforms' (META, Financials) Llama 3.1 architecture with 70 billion parameters, the model is built on this structure.
Formally known as Hon Hai Precision (HNHAF, Financials), Foxconn — produced by the Hon Hai Research Institute — stated the model is the first Traditional Chinese large language model. The future open-source model is intended by the firm.
Designed for internal use, FoxBrain finds applications in data analysis, decision support, document collaboration, mathematics, reasoning, problem-solving, and code development, the business said.
While FoxBrain lags DeepSeek's distillation model, Foxconn said it reaches world-class standards in understanding and reasoning. With outstanding performance in mathematics and logical thinking, the business stated the model has been tuned for processing Taiwanese language.
With the first public findings of FoxBrain set for presentation at Nvidia's GPU Technology Conference on March 20, Foxconn intends to extend AI applications in manufacturing, supply chain management, and intelligent decision-making.
