知识卡片
领域与任务自适应范围不同
DAPT用较宽领域文本继续训练,TAPT使用更接近目标任务的无标签输入。先广后窄再监督是一种可比较路线,不是防遗忘的充分条件;最终仍需目标任务与旧能力评估。
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
- · 进阶微调技术
原书出处 · 1 处
- · 查看原书依据
Taking DAPT one step further, you can also continue pre-training your model not just on general text from your domain of interest but also on domain text specifically related to your downstream tasks. This is called task-adaptive pre-training (TAPT).
参考来源
- 位置:《Designing Large Language Model Applications》第07章《进阶微调技术》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch07.html)
- 结论依据:摘录中关于领域与任务自适应范围不同的定义、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:Taking DAPT one step further, you can also continue pre-training your model not just on general text from your domain of interest but also on domain text specifically related to your downstream tasks. This is called task-adaptive pre-training (TAPT).