Z.ai:GLM-5.3-Flash MoE模型登陆国产芯片
Z.ai GLM-5.3-Flash激活320B参数中的18B,采用MoE架构,运行于国产芯片,MIT许可,1M上下文。
原文链接详细分析
Z.ai发布GLM-5.3-Flash,这是一款320B参数模型,通过Mixture of Experts MoE architecture efficiency in large language models每次仅激活18B参数,大幅降低推理成本,同时支持原生多模态和1M-token窗口。该模型此前预览为Ox Alpha,完全运行在国产AI芯片上,并以MIT许可证开源,凸显Z.ai GLM models Chinese AI chip development under US export controls更多依赖架构创新而非受限硬件。权重已上Hugging Face,API与聊天服务现已开放。
Mark
@MRRydonCofounder @AethirCloud | Building Decentralised Cloud Infrastructure (DCI) | Accelerating the world’s transition to universal cloud compute 🌎