11-20 更新 实时
首页/ 模型选型/ 正文

阿里用自研芯片训练下一代 Qwen 模型

模型选型 来源:TechNode 2026-11-20 22:14 0 阅读 机器翻译
阿里用自研芯片训练下一代 Qwen 模型

阿里巴巴首席执行官 Eddie Wu 在周四的公司季度财报电话会议上表示,阿里巴巴正在使用其半导体子公司 T-Head 开发的芯片来训练其下一款 Qwen 模型。阿里巴巴的 Bailian 模型平台也在使用中国制造的芯片提供推理服务。

这一披露将 T-Head 的角色扩展到 AI 计算的两个主要阶段:训练模型以及运行模型以响应用户请求。Bailian 在国际上被称为 Model Studio,提供对模型和用于构建 AI 应用的工具的访问。

Wu 表示,阿里巴巴自有芯片与模型之间的协同将改善计算成本和应用开发效率。他还认为,公司的全栈 AI 基础设施将使其在中国国内计算供应链中具备优势。这项训练工作连接了阿里巴巴 AI 投资的两个核心部分:其 Qwen 模型系列以及支持其云服务的硬件。[CLS, in Chinese]

译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。

查看英文原文(Alibaba uses its own chips to train next-generation Qwen model)

Alibaba is using chips developed by its semiconductor subsidiary T-Head to train its next Qwen model, CEO Eddie Wu said during the company’s quarterly earnings call on Thursday. Alibaba’s Bailian model platform is also providing inference services using Chinese-made chips.

The disclosure extends T-Head’s role across both major stages of AI computing: training models and running them to answer user requests. Bailian, known internationally as Model Studio, provides access to models and tools for building AI applications.

Wu said coordination between Alibaba’s own chips and models would improve computing costs and application development efficiency. He also argued that the company’s full-stack AI infrastructure would give it an advantage within China’s domestic computing supply chain. The training work connects two core parts of Alibaba’s AI investment: its Qwen model family and the hardware supporting its cloud services. [CLS, in Chinese]

正文保留英文原文,标题与摘要由机器翻译。

本文由采集器自动同步

原文发布于 TechNode(Alibaba uses its own chips to train next-generation Qwen model),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。

阅读原文 ↗