知识卡片

普通读书笔记卡 · 6744.c · 生成模型:GPT-6

三种嵌入相加组成输入表示

理解等级:中级

原BERT将词项、位置和句段类型嵌入逐项相加,再进入编码器。[CLS]和[SEP]帮助组织单句或句对;位置表通常是学得参数,不能把前面的固定正弦实现直接当所有BERT配置。

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

      9.3.1 Tokenization and pretraining The input sequences to BERT can ambiguously represent a single sentence or a pair of sentences. BERT uses WordPiece embeddings with the first token of each sequence always set as a special [CLS] token. Sentences are distinguished by a trailing separator token, [SEP]. Tokens in a sequence are further distinguished by a separate segment embedding with either sentence A or B

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第09章《可堆叠的Transformer》(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-9.html) - 结论依据:摘录中关于三种嵌入相加组成输入表示的定义、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:9.3.1 Tokenization and pretraining The input sequences to BERT can ambiguously represent a single sentence or a pair of sentences. BERT uses WordPiece embeddings with the first token of each sequence always set as a special [CLS] token. Sentences are distinguished by a trailing separator token, [SEP]. Tokens in a sequence are further distinguished by a separate segment embedding with either sentence A or B