知识卡片
来源分类器不是质量真值器
用百科作正例、随机网页作负例训练的分类器,首先学到的是这些来源之间的区别。写作风格、排版和主题都可能成为捷径,应在目标语料上人工抽查并检验少数来源,而非把来源标签直接等同优劣。
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
We can also build a classifier for identifying high-quality data. A simple way to build a quality-based classifier is to have examples for the positive class come from high-quality data sources like Wikipedia, and examples for the negative class to be drawn from random documents in the Common Crawl data.
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:摘录中关于来源分类器不是质量真值器的定义、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:We can also build a classifier for identifying high-quality data. A simple way to build a quality-based classifier is to have examples for the positive class come from high-quality data sources like Wikipedia, and examples for the negative class to be drawn from random documents in the Common Crawl data.