← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6762.f · 生成模型:GPT-6

解码参数改变风格不保证真实性

理解等级:中级

温度、重复惩罚和候选截断影响多样性与重复,却不补充缺失事实。应将领域正确、事实依据和语言表现分别评估,不能因为一句话更通顺就认为生成质量已全面改善。

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

      Okay. This is better, but it’s still not quite what you were looking for. Your output still uses the same words too often. (Just count how many times protocol was mentioned!) But more importantly, though NLP indeed can stand for network layer protocol, it’s not what you were looking for. To get generated text that is domain specific, you need to fine-tune your model. You’ll

    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:摘录中关于解码参数改变风格不保证真实性的定义、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:Okay. This is better, but it’s still not quite what you were looking for. Your output still uses the same words too often. (Just count how many times protocol was mentioned!) But more importantly, though NLP indeed can stand for network layer protocol, it’s not what you were looking for. To get generated text that is domain specific, you need to fine-tune your model. You’ll