在2024全球开发者先锋大会上，阶跃星辰正式发布其最新研发成果——Step-2万亿参数MoE语言大模型预览版。这一模型的问世，标志着阶跃星辰在深度智能领域的又一重要突破。阶跃星辰是一家专注于通用大模型研发的创业公司，其研发的Step-1V千亿参数多模态大模型已经在“司南”（OpenCompass）多模态模型评测榜单中荣登榜首，与GPT-4V的性能相媲美。
Step-2万亿参数MoE语言大模型采用了最新的MoE架构，这使得模型在处理复杂语言任务时更加高效。该模型不仅在性能上取得了显著提升，还开放了API接口，供部分合作伙伴试用，以便进一步优化和完善。阶跃星辰的创始人兼CEO姜大昕博士在大会上表示，该模型的发布将极大推动深度学习技术在自然语言处理领域的应用，并为广大开发者提供更强大的工具。
英文标题：JieYue XingChen Unveils Trillion-Parameter Language Model
英文关键词：JieYue XingChen, Trillion-Parameter, Language Model
英文新闻内容：
At the 2024 Global Developer Pioneer Conference, JieYue XingChen officially unveiled its latest research achievement – the preview version of the Step-2 trillion-parameter MoE language model. This model’s release marks another significant breakthrough for JieYue XingChen in the field of deep intelligence. JieYue XingChen, a startup focusing on general-purpose large model development, has already ranked first on the “Shennan” (OpenCompass) multimodal model assessment list with its Step-1V billion-parameter multimodal model, rivaling the performance of GPT-4V.
The Step-2 trillion-parameter MoE language model, developed by JieYue XingChen, employs the latest MoE architecture, making it more efficient in handling complex language tasks. Not only has the model’s performance significantly improved, but it also provides an API interface for some partners to try, aiming to further optimize and refine the model. Dr. Jiang Daxin, founder and CEO of JieYue Xing
【来源】https://www.chinastarmarket.cn/detail/1627874
Views: 2