人民财讯3月11日电，记者获悉，近日，科大讯飞携手华为在国产算力领域取得重大进展，双方联合团队率先突破国产算力集群上MoE模型的大规模跨节点专家并行集群推理，这是继DeepSeek公布其MoE模型训练推理方案后，业界首个基于国产算力的全新解决方案。

Timothy Moe of Goldman Sachs says equity markets are focusing on the impact of President Trump's tariffs. He tells Bloomberg ...

The Ministry of Education has stopped printing nine textbooks and converted them into electronic versions, as part of its ...

据媒体报道，字节豆包大模型团队官宣开源一项针对MoE（混合专家模型）架构的关键优化技术，可将大模型训练效率提升1.7倍，成本节省40%。据悉，该技术已实际应用于字节的万卡集群训练，累计帮助节省了数百万GPU小时训练算力。

3月10日，字节豆包大模型团队官宣开源一项针对MoE架构的关键优化技术，可将大模型训练效率提升1.7倍，成本节省40%。据悉，该技术已实际应用于字节的万卡集群训练，累计帮助节省了数百万GPU小时训练算力。(e公司) ...

Tesla stock tanks 9% to lowest level since Nov. 4 Tesla stock (TSLA) tumbled more than 9% on Monday to its lowest level since the day before the presidential election as tech stocks led a market ...