知识卡片

普通读书笔记卡 · 6500 · 生成模型:GPT-6

先建立评估再改进生成

理解等级:中级

先保存基础模型的评估流程与结果,再改变采样、修正或训练方法,才能比较改动带来的收益。测试集、答案提取和判定方式都属于比较条件;无稳定测量就难以区分真实提升与协议变化。

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

      figure 1.9, we cover the reasoning model development in several stages. In stage 1 (next chapter), we load a conventional LLM that has already undergone the basic pre-training and instruction fine-tuning stages. Then, in stage 2, we cover common methods for evaluating LLMs and reasoning capabilities, so that we can measure the improvements we make when we apply reasoning-enhancing methods in stages 3 and 4.

    参考来源

    - 位置:《Build a Reasoning Model (From Scratch)》第01章《理解推理模型》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Build a Reasoning Model (From Scratch)/_epub-src/OEBPS/Text/chapter-1.html) - 结论依据:摘录中关于先建立评估再改进生成的定义、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:figure 1.9, we cover the reasoning model development in several stages. In stage 1 (next chapter), we load a conventional LLM that has already undergone the basic pre-training and instruction fine-tuning stages. Then, in stage 2, we cover common methods for evaluating LLMs and reasoning capabilities, so that we can measure the improvements we make when we apply reasoning-enhancing methods in stages 3 and 4.