知识卡片
领域继续预训练不同于指令微调
用领域原文继续下一个token预测,主要适应文本分布;指令微调则使用任务输入与期望输出样例。二者的数据组织、目标和验证不同,本章书稿训练属于前一种,不能只因输出更像书稿就说学会执行指令。
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
原书出处 · 1 处
- · 查看原书依据
Now, you’re going to simulate a second important step in training an LLM: fine-tuning. Models trained on a large text corpus, without a particular task, are often called foundational models. Before they are exposed to a consumer in applications, such as ChatGPT, they undergo an additional round where they are given a labeled dataset with samples of tasks to perform and expected outputs. This type
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:摘录中关于领域继续预训练不同于指令微调的定义、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:Now, you’re going to simulate a second important step in training an LLM: fine-tuning. Models trained on a large text corpus, without a particular task, are often called foundational models. Before they are exposed to a consumer in applications, such as ChatGPT, they undergo an additional round where they are given a labeled dataset with samples of tasks to perform and expected outputs. This type