知识卡片
预训练骨架须接适合任务的输出头
BERT预训练参数可初始化下游模型,分类、多标签或跨度抽取各需对应输入组织与输出头。新建任务头通常尚未训练,能加载权重不代表已可预测;微调是任务反馈下更新,不只是多跑几次推断。
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
9.3.2 Fine-tuning For most BERT tasks, you will want to load the BERT base or BERT large model with all its parameters initialized from the pretraining and fine-tune the model for your specific task. The fine-tuning should typically be straightforward; one simply plugs in the task-specific inputs and outputs and then commences training all parameters end to end. Compared to the initial pretraining, fine-tuning the
参考来源
- 位置:《Natural Language Processing in Action, Second Edition》第09章《可堆叠的Transformer》(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-9.html)
- 结论依据:摘录中关于预训练骨架须接适合任务的输出头的定义、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:9.3.2 Fine-tuning For most BERT tasks, you will want to load the BERT base or BERT large model with all its parameters initialized from the pretraining and fine-tune the model for your specific task. The fine-tuning should typically be straightforward; one simply plugs in the task-specific inputs and outputs and then commences training all parameters end to end. Compared to the initial pretraining, fine-tuning the