← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 8672.a · 生成模型:GPT-6

候选排序需要可聚合比较规则

理解等级:高级

成对评分可比较两个候选,但多模型结果需要明确聚合规则,且可能出现非传递偏好。全部两两比较的数量随候选数平方增长;排名最高也是评分器定义的胜者,不必是真正最正确答案。

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

      PairRanker scores the output from two models, thus choosing a winner. GenFuser takes as input the output from k different models to generate the final output. Figure 7-5 shows the workings of the LLM-Blender framework. Figure 7-5. LLM-Blender Let’s dig deeper into each of these modules. PairRanker Consider you have access to n different models. For a given input, you feed the input to each

    参考来源

    - 位置:《Designing Large Language Model Applications》第07章《进阶微调技术》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch07.html) - 结论依据:摘录中关于候选排序需要可聚合比较规则的定义、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:PairRanker scores the output from two models, thus choosing a winner. GenFuser takes as input the output from k different models to generate the final output. Figure 7-5 shows the workings of the LLM-Blender framework. Figure 7-5. LLM-Blender Let’s dig deeper into each of these modules. PairRanker Consider you have access to n different models. For a given input, you feed the input to each