知识卡片

普通读书笔记卡 · 6610.g · 生成模型:GPT-6

数据来源多样性需对应任务

理解等级:中级

不同来源提供知识、风格、代码或专业表达等不同信号。组合数据时应看目标任务覆盖与来源权重;百科形式统一有利于某些内容,却不能代表全部交流场景,多样也不是质量无条件更高。

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

      more diverse sources. Diversity of data is important since in-domain unlabeled data in pre-training is helpful for downstream performance on that domain, and diverse data sets also enable generalization to previously unseen tasks and domains. To this end, the data from The Pile comes not only from Common Crawl but also PubMed Central, arXiv, GitHub, the FreeLaw Project, Stack Exchange, the US Patent and Trademark

    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:摘录中关于数据来源多样性需对应任务的定义、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:more diverse sources. Diversity of data is important since in-domain unlabeled data in pre-training is helpful for downstream performance on that domain, and diverse data sets also enable generalization to previously unseen tasks and domains. To this end, the data from The Pile comes not only from Common Crawl but also PubMed Central, arXiv, GitHub, the FreeLaw Project, Stack Exchange, the US Patent and Trademark