大豆也有专属AI了！中国发布“丰菽”2.0让育种更轻松 Soybean Gets Its Own AI: China Launches 'Fengshu' 2.0 for Smarter Breeding
我国科学家发布大豆垂直大模型丰菽2.0，通过多模型协同与结构化知识体系，推动大豆数据辅助育种。 Chinese scientists release soybean vertical large model Fengshu 2.0, using multi-model collaboration and structured knowledge to advance data-assisted soybean breeding.
丰菽2.0发布：大豆育种进入AI辅助时代
8月16日，在第32届全国大豆科研生产研讨会上，安徽农业大学王晓波教授代表研发团队发布大豆垂直大模型“丰菽”2.0。该模型由安徽农业大学联合中国农业科学院作物科学研究所开发，面向大豆全生命周期，定位为生成式AI育种平台。
多模型协同机制
与初代相比，丰菽2.0新增多模型协同分析机制，嵌入部分通用大模型接口。多个模型回答同一大豆专业问题后，由总结模块结合领域知识、私有数据和结构化比对规则，分析共识、差异及证据完整性，输出更可靠的专业结论。
六大功能模块与多维数据
- 集成了豆百科、豆分子、豆文献、豆病害、豆表型、豆育种六大模块。
- 整合超过1000万字专业文本、1万篇科研文献与专利数据。
- 构建包含2万个实体、10万条关系的大豆领域知识图谱。
- 融合8000份基因组重测序数据、约4万份种质资源和3000份以上品种表型数据。
应用场景
- 辅助病害诊断
- 育种亲本筛选
- 虚拟组合设计
- 表型数据解析
- 分子标记辅助育种
- 候选基因功能分析
Fengshu 2.0 Launch: Soybean Breeding Enters AI-Assisted Era
On August 16, at the 32nd National Soybean Research and Production Symposium, Professor Wang Xiaobo from Anhui Agricultural University released the soybean vertical large model "Fengshu" 2.0 on behalf of the R&D team. The model was jointly developed by Anhui Agricultural University and the Institute of Crop Sciences, Chinese Academy of Agricultural Sciences, covering the full soybean lifecycle as a generative AI breeding platform.
Multi-Model Collaborative Mechanism
Compared with the first version, Fengshu 2.0 adds a multi-model collaborative analysis mechanism and embeds interfaces to some general large models. After multiple models answer the same soybean-specific question, the summary module uses domain knowledge, private data, and structured comparison rules to analyze consensus, differences, and evidence completeness, producing more reliable professional conclusions.
Six Functional Modules and Multi-Dimensional Data
- Integrates six modules: DouBaike, DouFenzI, DouWenxian, DouBinghai, DouBiaoxing, and DouYuzhong.
- Incorporates over 10 million Chinese characters of professional text, and more than 10,000 research papers and patent documents.
- Builds a soybean domain knowledge graph with 20,000 entities and 100,000 relations.
- Fuses 8,000 genome resequencing datasets, approximately 40,000 soybean germplasm resources, and phenotype data from over 3,000 varieties.
Application Scenarios
- Assist in disease diagnosis
- Breeding parent screening
- Virtual combination design
- Phenotype data analysis
- Molecular marker-assisted breeding
- Candidate gene functional analysis
来源
- IT之家 · 08-17 07:31