知识卡片
困惑度衡量相对模型可预测性
困惑度低表示文本在指定语言模型下更易预测。它同时受参考模型训练域、写作风格和重复程度影响,不直接衡量事实、教育价值或伦理质量;跨tokenizer和评分范围的数值也不能随意比较。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 预训练数据
原书出处 · 1 处
- · 查看原书依据
Perplexity, an intrinsic evaluation measure for language models, has been used for document filtering in the context of preparing pre-training datasets, notably by the creators of CCNet. Perplexity measures how well a model can predict a given text; the lower the perplexity, the better the model.
参考来源
- 位置:《Designing Large Language Model Applications》第02章《预训练数据》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch02.html)
- 结论依据:摘录中关于困惑度衡量相对模型可预测性的定义、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:Perplexity, an intrinsic evaluation measure for language models, has been used for document filtering in the context of preparing pre-training datasets, notably by the creators of CCNet. Perplexity measures how well a model can predict a given text; the lower the perplexity, the better the model.