知识卡片

普通读书笔记卡 · 6726.a · 生成模型:GPT-6

隐藏状态是激活而非模型权重

理解等级:中级

h随当前序列逐步更新,概括该样本前缀;参数W则在所有时间步共享,并由训练更新。推断改变h通常不改变W,两者生命周期不同,不能把隐藏状态称作每词都重新学习的权重。

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

      Keep your eye out for the hidden layer later in this chapter. The activation values are stored in the variable h or hidden. These activation values within this tensor are your embeddings up to that point in the text. They are overwritten with new values each time a new token is processed, as your NLP pipeline gathers up the meaning of the tokens it has

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第08章《循环网络与长短期记忆》(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-8.html) - 结论依据:摘录中关于隐藏状态是激活而非模型权重的定义、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:Keep your eye out for the hidden layer later in this chapter. The activation values are stored in the variable h or hidden. These activation values within this tensor are your embeddings up to that point in the text. They are overwritten with new values each time a new token is processed, as your NLP pipeline gathers up the meaning of the tokens it has