← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6729.a · 生成模型:GPT-6

变长序列仍有批量与计算成本

理解等级:中级

RNN可按实际长度循环,批内却通常需要填充、打包或分桶,长序列还增加时间与训练图内存。CNN也不是计算量与长度无关;二者区别主要在时间依赖与并行方式,不是一个能无限读另一个完全不能。

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

      This is the superpower of RNNs: they process sequences of tokens or vectors. You are no longer limited to processing a single, fixed-length vector. Figure 8.1 shows the flexibility of RNNs to handle both input and output, either a single token or an almost unending sequence of many tokens. This means you don’t have to truncate and pad your input text, in an attempt to

    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:摘录中关于变长序列仍有批量与计算成本的定义、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:This is the superpower of RNNs: they process sequences of tokens or vectors. You are no longer limited to processing a single, fixed-length vector. Figure 8.1 shows the flexibility of RNNs to handle both input and output, either a single token or an almost unending sequence of many tokens. This means you don’t have to truncate and pad your input text, in an attempt to