知识卡片
优化器效果由指标和数据约束
优化器根据示例及评价函数调整指令、few-shot或模型参数。若指标只看exact match,可能忽略忠实性和错误拒答;应保留独立验证集并衡量多项质量,不能把训练目标提高等同用户价值提高。
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
原书出处 · 1 处
- · 查看原书依据
Optimizers Optimizers are components that update prompts or model parameters. Several optimizers are natively supported by DSPy. An optimizer can be used to update one of the following: The instruction prompt Few-shot training examples Model parameters (fine-tuning) An optimizer takes as input the modules it needs to be applied to, the metric to evaluate the output of the modules, and fine-tuning or few-shot training data
参考来源
- 位置:《Designing Large Language Model Applications》第13章《设计模式与系统架构》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch13.html)
- 结论依据:摘录中关于优化器效果由指标和数据约束的定义、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:Optimizers Optimizers are components that update prompts or model parameters. Several optimizers are natively supported by DSPy. An optimizer can be used to update one of the following: The instruction prompt Few-shot training examples Model parameters (fine-tuning) An optimizer takes as input the modules it needs to be applied to, the metric to evaluate the output of the modules, and fine-tuning or few-shot training data