← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 6642.d · 生成模型:GPT-6

低温与固定种子不保证全环境确定

理解等级:高级

种子主要控制指定随机流程,低温使分布更集中;后端算子、并行调度、精度、模型版本和批处理也可能影响输出。应记录环境与实际响应,不将MoE本身定义为必然随机,也不承诺跨环境逐字一致。

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

      Sources of nondeterminism range from using multi-threading to floating-point rounding errors to use of certain model architectures (for example, it is known that the Sparse MoE architecture produces nondeterministic outputs). Reducing the temperature to zero or close to zero impacts the LLM’s creativity and makes its outputs more predictable, which might not be suitable for many applications. In production settings where reliability is important, you

    参考来源

    - 位置:《Designing Large Language Model Applications》第05章《按任务适配模型》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch05.html) - 结论依据:摘录中关于低温与固定种子不保证全环境确定的定义、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:Sources of nondeterminism range from using multi-threading to floating-point rounding errors to use of certain model architectures (for example, it is known that the Sparse MoE architecture produces nondeterministic outputs). Reducing the temperature to zero or close to zero impacts the LLM’s creativity and makes its outputs more predictable, which might not be suitable for many applications. In production settings where reliability is important, you