Nanjing Agricultural University Unveils Pioneering AI Model, Marking a New Breakthrough in Agri-Tech Foundation Models
On January 10, 2026, at the sub-forum “Digital-Intelligent Technology Reshaping Comprehensive Transformation in Agriculture and Forestry Education” during the 2025 Annual Conference of the Higher Agriculture and Forestry Education Branch of the China Association of Higher Education, Nanjing Agricultural University (NAU) officially launched the “Sinong” Large Language Model. This model is China’s first open-source, vertical large language model targeting the general agriculture domain and also the first agricultural LLM independently developed and led by NAU. The release of Sinong signifies a new breakthrough for Nanjing Agricultural University in both fundamental AI model research and its application within the agricultural sector.
Rooted in Agriculture, Building a Foundational Knowledge Base
The Sinong LLM is designed with a core focus on serving the agricultural field. Its name, “Sinong” (司农), is derived from ancient Chinese official systems responsible for finance and agricultural administration. Its exceptional capabilities stem from NAU’s profound expertise in agricultural disciplines and its extensive, high-quality agricultural domain data foundation. Leveraging the university’s top-tier disciplinary strengths, the research team compiled specialized data from various sub-fields including Animal Science, Agricultural Economics and Management, Agricultural Resources and Environment, Horticulture, Smart Agriculture, Veterinary Medicine, Plant Protection, and Crop Breeding. This resulted in a professional agricultural dataset exceeding 4 billion tokens, encompassing nearly 9,000 books, over 240,000 academic papers, close to 20,000 policy documents and standards, alongside vast web-based knowledge. This effort has constructed a relatively comprehensive and high-quality foundational dataset for agriculture.
Comprehensive Technological Innovation, Tackling Agricultural Application Challenges
To address common issues in applying LLMs to specialized fields, such as “hallucination” and knowledge lag, the team conducted all-round technology. During the model training phase, beyond traditional instruction fine-tuning, they incorporated multi-dimensional training data like Chain-of-Thought and context references, significantly enhancing the model’s comprehension and generative capabilities for professional agricultural knowledge.
Furthermore, to improve the utilization efficiency of domain literature knowledge, the team developed a Multi-Agent Retrieval-Augmented Generation (RAG) framework. This framework, through optimized strategies for knowledge base construction, intelligent query rewriting, and hybrid retrieval, enables the model to accurately invoke professional knowledge, effectively ensuring the accuracy and timeliness of generated content. This provides a solid guarantee for the model’s reliable application in scenarios such as scientific research, education, and production.
Fully Open-Source, Empowering Innovation in Agri-Research, Education, and Industry
The Sinong Large Language Model has now been fully open-sourced on both the ModelScope community and GitHub. The released versions include model sizes of 8B and 32B parameters. This open-source strategy aims to lower the barrier for AI applications in agriculture, empowering a wide range of research institutions, enterprises, and developers to conduct secondary development and innovative applications based on the Sinong LLM, thereby jointly fostering the application ecosystem for smart agriculture.
The release of the Sinong LLM by Nanjing Agricultural University represents a concrete step by the university in serving the national strategy for building a strong agricultural country and in implementing the State Council’s Opinions on Deepening the “Artificial Intelligence+” Action. Moving forward, the university will continue to iterate and enhance the model’s performance, deepen the exploration of application scenarios, and collaborate with all sectors to jointly advance the modernization and digital-intelligent transformation of China’s agriculture.
Model Open-Source and Access
-
ModelScope Community Repository:
https://modelscope.cn/organization/NAULLM -
GitHub Community Repository:
https://github.com/njauzzx/Sinong
