知识卡片
训练原料要随实验一起固定
训练领域模型时,数据版本、词表/编码、架构与学习目标共同决定实验。更改数据量、质量、表示或参数规模时应分开记录和比较,避免把多个同时变化的因素都归到一种方法上。
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
- · 架构与学习目标
原书出处 · 1 处
- · 查看原书依据
Recall the ingredients of a language model from Chapter 2. We need: A pre-training dataset A vocabulary and tokenization scheme A model architecture A learning objective For training the chess language model, we can choose the Transformer architecture with the next-token prediction learning objective, which is the de facto paradigm used today. For the pre-training dataset, we can use the chess games dataset from Lichess,
参考来源
- 位置:《Designing Large Language Model Applications》第04章《架构与学习目标》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch04.html)
- 结论依据:摘录中关于训练原料要随实验一起固定的定义、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:Recall the ingredients of a language model from Chapter 2. We need: A pre-training dataset A vocabulary and tokenization scheme A model architecture A learning objective For training the chess language model, we can choose the Transformer architecture with the next-token prediction learning objective, which is the de facto paradigm used today. For the pre-training dataset, we can use the chess games dataset from Lichess,