知识卡片

普通读书笔记卡 · 6762.c · 生成模型:GPT-6

贪心解码不等于随机采样

理解等级:中级

贪心每步取当前最高分token,通常在相同状态下是确定性的。它只优化局部选择,可能产生重复,却不代表架构内部形成递归循环;温度用于调节采样分布时,不能靠改温度就改变单纯argmax选择。

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 type of stochastic generation is the default for GPT-2 and is called greedy search because it grabs the “best” (most probable) token every time. You may know the term greedy from other areas of computer science. Greedy algorithms are those that choose the best next action rather than looking further than one step ahead before making their choice. You can see why it’s so

    参考来源

    - 位置:《Natural Language Processing in Action, Second Edition》第10章《真实世界中的大语言模型》(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-10.html) - 结论依据:摘录中关于贪心解码不等于随机采样的定义、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 type of stochastic generation is the default for GPT-2 and is called greedy search because it grabs the “best” (most probable) token every time. You may know the term greedy from other areas of computer science. Greedy algorithms are those that choose the best next action rather than looking further than one step ahead before making their choice. You can see why it’s so