model-scope/Sinong1.0-32B

Sinong1.0-32B: A Specialized Agricultural Large Language Model

Sinong1.0-32B is a 32 billion parameter large language model developed by a joint team from Nanjing Agricultural University and Nanjing University of Science and Technology. It is specifically designed to serve the general agricultural domain, leveraging the academic strengths of Nanjing Agricultural University.

Key Capabilities and Training:

• Extensive Agricultural Data: The model is trained on a massive dataset exceeding 4 billion tokens, meticulously collected across various agricultural sub-disciplines such as animal science, veterinary medicine, agricultural economics, horticulture, and smart agriculture. This dataset includes 8,863 books, 243,897 papers, and 196,748 policy documents, standards, and patents.
• High-Quality Data Integration: Multi-source, heterogeneous data was processed and integrated to form a large-scale, high-quality agricultural foundational dataset.
• Advanced Fine-tuning: The model underwent instruction fine-tuning and reinforcement learning using synthetic data, including Chain-of-Thought (COT) and in-context instruction formats, to enhance its understanding and application of agricultural knowledge.
• Multi-Agent Retrieval Augmentation (RAG): Sinong1.0-32B incorporates an optimized multi-agent RAG framework designed to improve knowledge base construction, data sourcing, and retrieval efficiency for specialized agricultural literature.

Use Cases:

• Agricultural Knowledge Q&A: Excels at answering complex questions related to various agricultural fields.
• Research and Analysis: Supports in-depth analysis of agricultural policies, research papers, and technical documents.
• Smart Agriculture Applications: Ideal for developing applications that require deep domain-specific understanding in agriculture.