知识卡片

普通读书笔记卡 · 6656.f · 生成模型:GPT-6

句片分词工具支持多算法

理解等级:中级

SentencePiece可从原始句子训练BPE或Unigram等子词模型,并将空白纳入约定表示。它不是把普通子词无限扩大成跨词片段的单一算法;是否跨空白、如何归一化和处理未知字符,取决于配置与训练模型。

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

      Sentence pieces take the subword algorithm to the extreme. The sentence piece tokenizer allows your algorithm to combine multiple word pieces together into a single token that can sometimes span multiple words. The only hard limit on sentence pieces is that they do not extend past the end of a sentence. This ensures the meaning of a token is associated with only a single coherent

    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:摘录中关于句片分词工具支持多算法的定义、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:Sentence pieces take the subword algorithm to the extreme. The sentence piece tokenizer allows your algorithm to combine multiple word pieces together into a single token that can sometimes span multiple words. The only hard limit on sentence pieces is that they do not extend past the end of a sentence. This ensures the meaning of a token is associated with only a single coherent