← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6742 · 生成模型:GPT-6

自回归生成不同于循环层结构

理解等级:中级

生成时将已选词元接到前缀再预测下一个,可用普通迭代实现。它不使Transformer在架构分类上成为递归神经网络;编码器模型也不一定用于自回归生成,是否有内部循环要与外部生成控制分开。

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

      9.1 Recursion vs. recurrence Transformers are the latest big leap forward in autoregressive NLP models. Autoregressive models predict one discrete output value at a time, usually a token or word in natural language text. An autoregressor recycles the output to reuse it as an input for predicting the next output, so autoregressive neural networks are recursive. The word recursive is a general term for any

    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:摘录中关于自回归生成不同于循环层结构的定义、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:9.1 Recursion vs. recurrence Transformers are the latest big leap forward in autoregressive NLP models. Autoregressive models predict one discrete output value at a time, usually a token or word in natural language text. An autoregressor recycles the output to reuse it as an input for predicting the next output, so autoregressive neural networks are recursive. The word recursive is a general term for any