知识卡片
公开数据不等于已见数据
输出与基准相似的答案样式不能证明模型见过具体测试题;数据曾公开也不能证明必被训练收录。污染是需要调查的评估风险,结论应依据可核实的训练信息或针对性证据,而非从格式习惯或提示敏感直接倒推。
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
- · 评估推理模型
原书出处 · 1 处
- · 查看原书依据
not necessarily proof of overfitting to the evaluation tasks we will run later. But it reflects that pre-trained models often encounter many problem formats online and learn to reproduce those stylistic conventions.
参考来源
- 位置:《Build a Reasoning Model (From Scratch)》第03章《评估推理模型》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Build a Reasoning Model (From Scratch)/_epub-src/OEBPS/Text/chapter-3.html)
- 结论依据:摘录中关于公开数据不等于已见数据的定义、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:not necessarily proof of overfitting to the evaluation tasks we will run later. But it reflects that pre-trained models often encounter many problem formats online and learn to reproduce those stylistic conventions.