← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6623.c · 生成模型:GPT-6

归一化可能不可逆丢信息

理解等级:中级

转小写、去音标和Unicode规范化对等价或相近写法的处理不同。若原字形、专名或大小写对任务重要,应检查是否在编码前已被合并;后续解码未必能恢复归一化前文本。

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

      Stripping off accents from characters, like from the word Peña Unicode normalization Let’s see what kind of normalization is applied on the uncased version of BERT: tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased") print(tokenizer.backend_tokenizer.normalizer.normalize_str( 'Pédrò pôntificated at üs:-)') The output is: pedro pontificated at us:-) As we can see, the accents have been removed and the text has been converted to lowercase. There isn’t much normalization done in tokenizers

    参考来源

    - 位置:《Designing Large Language Model Applications》第03章《词表与分词》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch03.html) - 结论依据:摘录中关于归一化可能不可逆丢信息的定义、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:Stripping off accents from characters, like from the word Peña Unicode normalization Let’s see what kind of normalization is applied on the uncased version of BERT: tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased") print(tokenizer.backend_tokenizer.normalizer.normalize_str( 'Pédrò pôntificated at üs:-)') The output is: pedro pontificated at us:-) As we can see, the accents have been removed and the text has been converted to lowercase. There isn’t much normalization done in tokenizers