知识卡片
子词切分影响序列长度
子词词表把常见片段与低频词共享的片段组织成 token,词与 token 不一一对应,空格也可能进入 token。跨语言或生造词比较时应直接统计 token 长度,而不能用词数替代上下文长度和输出预算。
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
原书出处 · 1 处
- · 查看原书依据
subword-based method based on Byte Pair Encoding (BPE). BPE can represent both common and rare words using a mix of full words and subword units. Spaces are also often included in tokens (for example, " large"), which helps the LLM detect word boundaries.
参考来源
- 位置:《Build a Reasoning Model (From Scratch)》第02章《使用预训练模型生成文本》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Build a Reasoning Model (From Scratch)/_epub-src/OEBPS/Text/chapter-2.html)
- 结论依据:摘录中关于子词切分影响序列长度的定义、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:subword-based method based on Byte Pair Encoding (BPE). BPE can represent both common and rare words using a mix of full words and subword units. Spaces are also often included in tokens (for example, " large"), which helps the LLM detect word boundaries.