← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6696.e · 生成模型:GPT-6

上下文化表示随当前输入计算

理解等级:中级

上下文化模型让同一词项在不同上下文中得到不同隐藏表示,并可利用词序与邻近关系。这是在固定模型参数下按输入重算,不等于每次推断都重新训练权重;得到不同表示也不保证完全消除否定或多义错误。

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

      In contrast, contextualized word embeddings can be updated or refined based on the embeddings and words that come before or after them, and the order in which a word appears relative to other words matters for contextualized word embeddings. This means the NLU of the bigram not happy would have an embedding much closer to the embedding of unhappy for contextualized word embeddings than for

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第06章《用词嵌入推理》(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-6.html) - 结论依据:摘录中关于上下文化表示随当前输入计算的定义、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:In contrast, contextualized word embeddings can be updated or refined based on the embeddings and words that come before or after them, and the order in which a word appears relative to other words matters for contextualized word embeddings. This means the NLU of the bigram not happy would have an embedding much closer to the embedding of unhappy for contextualized word embeddings than for