南京农业大学发布首个农业开源大语言模型“司农”

Core Insights - The release of the Sinong language model by Nanjing Agricultural University marks a significant breakthrough in AI foundational model research and application in the agricultural sector [1][2] Group 1: Model Development and Data Foundation - The Sinong model is the first open-source vertical large language model focused on the general agricultural field in China, leveraging Nanjing Agricultural University's strong academic background and extensive agricultural data [2][3] - The research team has compiled over 4 billion tokens of specialized agricultural data, including nearly 9,000 books, over 240,000 academic papers, and around 20,000 policies and standards, creating a comprehensive and high-quality agricultural dataset [3] Group 2: Technical Innovations - The model addresses common issues in specialized applications, such as "hallucination" and knowledge lag, through comprehensive technical advancements, including multi-dimensional training data and a multi-agent retrieval enhancement framework [4] - The introduction of strategies like optimized knowledge base construction and intelligent query rewriting significantly improves the model's ability to accurately generate and utilize professional knowledge [4] Group 3: Open Source Strategy and Future Development - The Sinong model is fully open-sourced on platforms like Magic Tower Community and GitHub, offering two versions with 8 billion and 32 billion parameters, aimed at lowering the barriers for agricultural AI applications [5] - Nanjing Agricultural University plans to continuously iterate on the model's performance and explore application scenarios, collaborating with various stakeholders to promote the modernization and digital transformation of agriculture in China [5]