January 14th.Nanjing Agricultural UniversityThe official declared that last week, at the sub-forum of the annual conference of the Chinese Higher Education Institute of Higher Agroforestry Education in 2025, "Technology to reshape the full-scale transformation of agroforestry education", the Chinese Higher Education Institute (IEP) was able to take a step forward in this directionPublished by Nanjing Agricultural UniversitySeononLarge Language Model(Sinong).
The model is..For the first time in the country, for general agricultureOpen SourceVertical Large Language ModelIt is also the first large-language agricultural model developed by the Nanjing Agricultural University. The publication of the Sinon Language Model marked a new breakthrough in research and application of artificially intelligent basic models in agriculture at Nanjing Agricultural University。
The “Sunon” model is based on the service agriculture sector and is named from China's ancient financial and agricultural management official system. The research team collected disaggregated data on such disciplines as animal sciences, agricultural economic management, agricultural resources and environment, horticulture, smart agriculture, animal medicine, plant protection, and crop breeding, based on the strengths of the school ' s agricultural disciplinesagricultural professional data totalling over 4 billion tokenIt covers nearly 9,000 books, over 240,000 academic papers and nearly 20,000 policy, standards and big web knowledge, building basic agricultural data sets。
During the training phase of the model, in addition to the fine-tuning of traditional instructions, the team introduced multidimensional training data, such as thought chains and context references, to enhance the model ' s understanding and generation of agricultural expertise。
In addition, in response to the efficiency of the use of literature in the field, the team launched the Multi-Intelligence Retrieval Enhancement Framework. The framework ensures accuracy and timeliness of content generation by optimizing strategies such as knowledge-base construction, smart query rewrite and mixed search, so that models can accurately access expertise。
At the moment, the Sinophone language model has become a full-fledged source of information for both the community and the GitHub communityRELEASED VERSIONS OF 8B AND 32B PARAMETER SIZESI DON'T KNOW. 1AI WITH OPEN SOURCE ADDRESSES AS FOLLOWS:
- Open Source of the Magic Community: https://modelscope.cn/organization/NAULLM
- GitHub Community Open Source Address: https://github.com/njauzzx/Sinong