知识卡片

普通读书笔记卡 · 6745.c · 生成模型:GPT-6

核心变换模块之外还需输入输出层

理解等级:中级

nn.Transformer处理序列表示,不自带源目标词表嵌入、位置编码或词表预测头。翻译包装类需补齐这些组件及数据接口;最后Linear输出是logits,不能仅按注释称已得到词概率。

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

      The torch.nn.Transformer is a somewhat bare-bones version of the sequence-to-sequence model containing the main secret sauce, the multi-head self-attention in both the encoder and decoder. As evident in the source code for the module,20 the model does not assume the use of embedding layers or positional encodings. Now, you will create a TranslationTransformer model that uses the custom decoder components, by extending the torch.nn.Transformer module,

    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:摘录中关于核心变换模块之外还需输入输出层的定义、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:The torch.nn.Transformer is a somewhat bare-bones version of the sequence-to-sequence model containing the main secret sauce, the multi-head self-attention in both the encoder and decoder. As evident in the source code for the module,20 the model does not assume the use of embedding layers or positional encodings. Now, you will create a TranslationTransformer model that uses the custom decoder components, by extending the torch.nn.Transformer module,