知识卡片
相同模型可能重复共同错误
同一模型的多个候选共享训练、提示和知识缺口,错误不一定独立。多数一致不构成外部正确性证明;若系统性误解题目,多次投票可能强化错误。投票收益需由保留集验证,必要时引入可核查反馈。
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
- · 推理时计算扩展
原书出处 · 1 处
- · 查看原书依据
and then select the most frequent one, as shown in figure 4.17. Figure 4.17 The self-consistency sampling method generates multiple responses from the LLM and selects the most frequent answer, which improves answer accuracy through majority voting across these sampled responses. The self-consistency sampling technique shown in figure 4.17 counts as an inference-time scaling technique, since we don't update the model itself and we expend
参考来源
- 位置:《Build a Reasoning Model (From Scratch)》第04章《推理时计算扩展》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Build a Reasoning Model (From Scratch)/_epub-src/OEBPS/Text/chapter-4.html)
- 结论依据:摘录中关于相同模型可能重复共同错误的定义、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:and then select the most frequent one, as shown in figure 4.17. Figure 4.17 The self-consistency sampling method generates multiple responses from the LLM and selects the most frequent answer, which improves answer accuracy through majority voting across these sampled responses. The self-consistency sampling technique shown in figure 4.17 counts as an inference-time scaling technique, since we don't update the model itself and we expend