知识卡片
偏移位置用于对齐原文
编码返回的offset帮助把模型片段映射回原文位置。归一化、字节编码和特殊token可能让对齐含有额外约定;做抽取或高亮时应检查实际offset含义,不只用解码字符串长度推算原位置。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 词表与分词
原书出处 · 1 处
- · 查看原书依据
Along with the pre-tokens (or word tokens), the character offsets are returned. The T5 pre-tokenizer splits only on whitespace, doesn’t collapse multiple spaces into one, and doesn’t split on punctuation or numbers. The behavior can be vastly different for other tokenizers. Tokenization After the optional pre-tokenization step, the actual tokenization step is performed. Some of the important algorithms in this space are byte pair encoding
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:摘录中关于偏移位置用于对齐原文的定义、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:Along with the pre-tokens (or word tokens), the character offsets are returned. The T5 pre-tokenizer splits only on whitespace, doesn’t collapse multiple spaces into one, and doesn’t split on punctuation or numbers. The behavior can be vastly different for other tokenizers. Tokenization After the optional pre-tokenization step, the actual tokenization step is performed. Some of the important algorithms in this space are byte pair encoding