Translations:Attention Mechanisms/27/zh

    From Marovi AI
    Revision as of 21:58, 27 April 2026 by DeployBot (talk | contribs) (Batch translate Attention Mechanisms unit 27 → zh)

    由于自注意力是排列不变的(它将输入视为无序集合),必须显式注入位置信息。原始 Transformer 使用正弦位置编码: