← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6656 · 生成模型:GPT-6

子词粒度在词表和序列长度间取舍

理解等级:中级

较小片段可组合表示新词,减少整词未登录问题,却通常形成更长输入序列。较大词表可缩短某些序列,也增加嵌入和输出层规模;效率与质量需联同语言覆盖、预分词和模型约束评价,不存在所有任务唯一最佳粒度。

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

      You might want to divide words into even smaller, meaningful parts. Word pieces, such as the prefix pre, the suffix fix, and the interior syllable la, all have meaning. You can use these word pieces to transfer what you learn about the meaning of one word to another similar word in your vocabulary. Your NLU pipeline can even use these pieces to understand new words,

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第02章《思维的词元》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Natural Language Processing in Action, Second Edition/_epub-src/OEBPS/Text/chapter-2.html) - 结论依据:摘录中关于子词粒度在词表和序列长度间取舍的定义、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:You might want to divide words into even smaller, meaningful parts. Word pieces, such as the prefix pre, the suffix fix, and the interior syllable la, all have meaning. You can use these word pieces to transfer what you learn about the meaning of one word to another similar word in your vocabulary. Your NLU pipeline can even use these pieces to understand new words,