知识卡片
多次调用预算不同于单答长度
候选采样、工具与验证可能为同一任务触发多次模型调用。只看最终回答字数会低估总计算;预算应覆盖所有候选和辅助步骤,并检查额外调用是否带来可测收益。
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
- · 理解推理模型
原书出处 · 1 处
- · 查看原书依据
running the model several times for a single task, for example to sample multiple candidate solutions, call tools, or run a verifier. These additional calls multiply the total number of tokens processed and further increase cost beyond the single-call behavior shown in figure 1.8.
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:摘录中关于多次调用预算不同于单答长度的定义、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:running the model several times for a single task, for example to sample multiple candidate solutions, call tools, or run a verifier. These additional calls multiply the total number of tokens processed and further increase cost beyond the single-call behavior shown in figure 1.8.