每经AI快讯,3月4日,阶跃星辰官微宣布开源Step3.5 Flash的预训练权重(Base)、中训练权重(Midtrain)以及配套的Steptron训练框架。据了解,Step3.5 Flash采用稀疏MoE架构,总参数1960亿,但推理时仅激活约110亿参数,单请求代码任务下推理速度最高可达350TPS。
特别声明:以上内容(如有图片或视频亦包括在内)为自媒体平台“网易号”用户上传并发布,本平台仅提供信息存储服务。
Notice: The content above (including the pictures and videos if any) is uploaded and posted by a user of NetEase Hao, which is a social media platform and only provides information storage services.