Group 1 - The National Bureau of Statistics has initiated a data efficiency enhancement action for state-owned enterprises, with 12 central enterprises designated as pilot leaders [1][2] - The enhancement action focuses on four key areas: innovating data management mechanisms, nurturing the data industry, empowering industrial transformation, and optimizing the innovation environment, with ten major tasks outlined [2] - The initiative aims to significantly improve data resource development and sharing levels by 2027, enhancing the collaborative efficiency of industrial chains and benefiting over 100,000 small and medium-sized enterprises [2] Group 2 - China Southern Power Grid, one of the pilot leaders, emphasizes the necessity of integrating data with application scenarios to fully realize its value [3] - The demand for energy and electricity data across various industries is rapidly increasing, with electricity data being crucial for both equipment development and economic monitoring [3] - Southern Power Grid has established an open data system covering three categories: equipment operation, power systems, and user electricity consumption, facilitating cross-industry applications [3] - The company plans to launch a trusted data space in the energy sector by August 2024, utilizing technologies like blockchain and privacy computing to create a compliant and trustworthy data circulation ecosystem [3] - Since the establishment of the data space, the number of ecological partners has increased tenfold, cooperation demands have tripled, and application scenarios have doubled compared to before [3]

证券研究报告 Meta发布开源大模型Llama 4， 国有企业数据效能提升行动实施方案正在制定 计算机行业强于大市（维持） 平安证券研究所计算机团队 分析师：闫磊S1060517070006(证券投资咨询)YANLEI511@pingan.com.cn 黄韦涵S1060523070003(证券投资咨询)HUANGWEIHAN235@pingan.com.cn 研究助理：王佳一S1060123070023（一般证券从业资格）WANGJIAYI446@pingan.com.cn 据Meta介绍，Llama 4目前有Scout和Maverick两个版本，是Meta迄今为止最先进的模型，也是同类产品中多模态性最 强的模型。此外，Meta还预览了其迄今最强大最智能的模型——Llama 4 Behemoth，称之为"新模型中的教师"。 Llama 4模型是Llama系列模型中首批采用混合专家（MoE）架构的模型。这一模型也是DeepSeek系列模型采用的架构， 与传统的稠密模型相比，在MoE架构中，单独的token只会激活全部参数中的一小部分，训练和推理的计算效率更高。 2025年4月6日 请务必阅读正文后免责条款 核心 ...