知识卡片
规模比较先固定任务与评测口径
比较扩展规模带来的收益,需说明训练计算、参数、数据、测试任务及计分方式。某项指标平坦不等于全部任务随机猜测;非线性的评分规则也可能产生看似突变的曲线,不能只凭一幅规模图决定模型能力。
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
Each order of magnitude increase in model capacity (size) seems to unlock more surprising capabilities. In the GPT-4 Technical Report, OpenAI researchers explain the surprising capabilities that emerged.17 These are the same researchers who invested a lot of their time and money into the idea that scale (and attention) is all models need, so they may have a conflict of interest in evaluating the emergent
参考来源
- 位置:《Natural Language Processing in Action, Second Edition》第10章《真实世界中的大语言模型》(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-10.html)
- 结论依据:摘录中关于规模比较先固定任务与评测口径的定义、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:Each order of magnitude increase in model capacity (size) seems to unlock more surprising capabilities. In the GPT-4 Technical Report, OpenAI researchers explain the surprising capabilities that emerged.17 These are the same researchers who invested a lot of their time and money into the idea that scale (and attention) is all models need, so they may have a conflict of interest in evaluating the emergent