知识卡片
生成融合需保留证据和冲突
GenFuser接受多个候选后生成新答案,可能整合互补内容,也可能继承或放大错误。应检查候选冲突、来源和输出依据,并将融合器本身的质量与成本纳入验收;拼合流畅不等于事实一致。
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
- · 进阶微调技术
原书出处 · 1 处
- · 查看原书依据
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 of these models to generate the outputs. Now, for each pair
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:摘录中关于生成融合需保留证据和冲突的定义、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: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 of these models to generate the outputs. Now, for each pair