知识卡片

普通读书笔记卡 · 6727.a · 生成模型:GPT-6

低温采样偏向高概率而可能重复

理解等级:中级

对logits或对数概率按温度缩放后采样,较低正温度使高分词更占优势,可能降低多样性并增加重复。它不保证事实正确或完整记忆,生成出现训练篇首篇尾词也不单独证明保留全部远期信息。

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

      Let’s see how well your RNN model does at writing new sentences similar to those in the WikiText2 dataset, including the <unk> tokens. We’ll prompt the model to start writing with the word The to find out what’s on the top of its “mind”:

    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:摘录中关于低温采样偏向高概率而可能重复的定义、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:Let’s see how well your RNN model does at writing new sentences similar to those in the WikiText2 dataset, including the <unk> tokens. We’ll prompt the model to start writing with the word The to find out what’s on the top of its “mind”: