知识卡片

普通读书笔记卡 · 6975.a.8 · 生成模型:gpt-6.1-terra

端到端模型与流水线

理解等级:高级

分阶段流水线显式定义中间步骤,可以单独替换组件;端到端模型则直接从输入学习输出,可能捕捉更复杂的关系、减少手工设计和步骤间误差。选择需要结合数据、计算及维护需求,单模型也不自动消除外部知识和可靠性问题。

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

    • Preliminary research · 查看原书依据

      the trend shifted toward an end-to-end single-model approach. It could potentially capture more complex relationships in the data, as they are not limited by assumptions and limitations of manual design, require less domain knowledge, and reduce the accumulation of errors between steps.

    参考来源

    - 位置:《Machine Learning System Design》第3章《Preliminary research》(来源项目 a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Machine Learning System Design/_epub-src/OEBPS/Text/chapter-3.xhtml) - 结论依据:原文比较多步骤处理和端到端建模,并指出语言模型仍有算力和事实可靠性问题。 - 原始内容:the trend shifted toward an end-to-end single-model approach. It could potentially capture more complex relationships in the data, as they are not limited by assumptions and limitations of manual design, require less domain knowledge, and reduce the accumulation of errors between steps.