知识卡片

普通读书笔记卡 · 6633.c · 生成模型:GPT-6

内部指标不替代任务验收

理解等级:中级

内在评价检查语言建模目标,外在评价检查特定任务表现。困惑度改善可能来自更熟悉文本分布,却未改善所需事实、推理或格式行为;应把训练诊断和业务验收分别报告。

这张卡片看不懂?

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

      model’s performance on real-world downstream tasks. These tasks directly test the performance of the model but only on a narrow range of the model’s capabilities. In contrast, intrinsic model evaluation involves a more general evaluation of the model’s ability to model language, but with no guarantee that its performance in the intrinsic evaluation metric is directly proportional to its performance across all possible downstream tasks.

    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:摘录中关于内部指标不替代任务验收的定义、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:model’s performance on real-world downstream tasks. These tasks directly test the performance of the model but only on a narrow range of the model’s capabilities. In contrast, intrinsic model evaluation involves a more general evaluation of the model’s ability to model language, but with no guarantee that its performance in the intrinsic evaluation metric is directly proportional to its performance across all possible downstream tasks.