知识卡片

普通读书笔记卡 · 9094.b · 生成模型:GPT-6

词嵌入查询等价独热投影

理解等级:中级

将词表ID的独热向量乘嵌入矩阵,等价于按ID查询对应行或列向量,不必真的构造巨大独热数组。词表大小K和嵌入维度d承担不同职责,矩阵方向需与张量约定一致。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      Once we have converted text into a sequence of one-hot codes, the next step is to convert these into tokens via a projection, Wtokenize, from the one-hot dimensionality (K) to a target dimensionality for token code vectors (d). Figure 51.1 shows these steps for converting an input string to tokens. For simplicity, we show word-level one-hot encoding rather than byte-pair encoding.

    参考来源

    - 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第51章《视觉与语言》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)/_epub-src/OPS/xhtml/c051.xhtml) - 结论依据:摘录中关于词嵌入查询等价独热投影的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。 - 原始内容:Once we have converted text into a sequence of one-hot codes, the next step is to convert these into tokens via a projection, Wtokenize, from the one-hot dimensionality (K) to a target dimensionality for token code vectors (d). Figure 51.1 shows these steps for converting an input string to tokens. For simplicity, we show word-level one-hot encoding rather than byte-pair encoding.