What is FoxBrain
FoxBrain is a Traditional Chinese large language model (LLM) launched by Hon Hai Research Institute, with strong reasoning capabilities. Based on the Meta Llama 3.1 architecture, it has 70B parameters and focuses on the fields of mathematics and logical reasoning. FoxBrain is based on an efficient training strategy and completed training in four weeks using 120 NVIDIA H100 GPUs, demonstrating high efficiency and low cost. FoxBrain generates high-quality Chinese pre-training data based on proprietary technology and uses Adaptive Reasoning Reflection technology to improve reasoning capabilities. FoxBrain promotes the application of AI in fields such as manufacturing, supply chain management, and intelligent decision-making, setting a new benchmark for the development of AI technology in Taiwan.
Main functions of FoxBrain
- Data analysis and decision support: Process complex data information and provide decision support for users.
- Document collaboration: Help users generate, edit, and optimize text content, improving work efficiency.
- Mathematics and reasoning problem solving: Performs excellently in mathematics and logical reasoning, solving complex mathematical problems and logical puzzles.
- Code generation: Assists users in generating programming code and improves development efficiency.
- Language style optimization: Optimized for the language style of Traditional Chinese in Taiwan, better fitting the needs of local users.
Technical principles of FoxBrain
- Base architecture: Based on the Meta Llama 3.1 architecture, it has 70B parameters and possesses strong language generation and understanding capabilities.
- Efficient training strategy: Trained using 120 NVIDIA H100 GPUs, based on a multi-node parallel training architecture, ensuring high performance and stability. Based on proprietary technology, it generates 98B tokens of high-quality Chinese pre-training data, covering 24 categories of topics. Using Adaptive Reasoning Reflection technology, it trains the model's autonomous reasoning ability, improving performance in the fields of mathematics and logical reasoning.
- Context processing capability: Supports a context processing length of 128K tokens, handling longer text input and output.
- Optimization and evaluation: Based on technologies such as Continual Pre-Training, Supervised Fine-tuning, and RLAIF, it continuously optimizes model performance and performs excellently in the TMMLU+ test dataset.
Project address of FoxBrain
- Project official website：https://www.honhai.com/zh-tw/press-center/press-releases/latest-news/1548
Application scenarios of FoxBrain
- Intelligent manufacturing and supply chain management: Optimize production processes, provide data analysis and decision support, and improve supply chain efficiency and intelligence level.
- Smart education and learning assistance: Help students answer mathematics and logic questions, provide learning tutoring, generate learning materials, and improve learning efficiency.
- Smart office and document collaboration: Automatically generate and optimize text content, assist in writing reports, emails, etc., and improve office efficiency and document quality.
- Software development and code generation: Assist programmers in generating code snippets, provide programming suggestions, and improve software development efficiency and quality.
- Smart cities and data analysis: Support the analysis and processing of urban data, provide intelligent decision support, and help with the intelligent upgrading of urban management and public services.