知识卡片

普通读书笔记卡 · 2931.e.3.a.1.f.2

微调塑造模型行为

理解等级:中级

微调用训练数据调整模型行为,可全局适配某领域,也可为租户维护单独适配配置。PEFT通过只训练部分参数或适配参数降低训练开销,不应理解为没有参数学习;租户级方案还需解决模型配置数量和请求到目标的映射。

这张卡片看不懂?

已有反馈 · 0 / 10 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • GenAI and Multi-Tenancy · 查看原书依据

      or libraries of your environment, you’ll need to extract the tenant context for a given request and determine how it maps to a tenant’s logical model identifier. Then, you’ll need to use this identifier to invoke the GenAI service. This will allow your request to apply the appropriate, tenant-specific fine-tuning. This concept is shown at the top of the diagram

    参考来源

    - 位置:《Building Multi-Tenant SaaS Architectures》第16章《GenAI and Multi-Tenancy》(来源项目 a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Building Multi-Tenant SaaS Architectures/_epub-src/OEBPS/ch16.html) - 结论依据:原文具体说明微调用训练数据调整模型行为,可全局适配某领域,也可为租户维护单独适配配置;本卡据此归纳机制及适用边界。 - 原始内容:or libraries of your environment, you’ll need to extract the tenant context for a given request and determine how it maps to a tenant’s logical model identifier. Then, you’ll need to use this identifier to invoke the GenAI service. This will allow your request to apply the appropriate, tenant-specific fine-tuning. This concept is shown at the top of the diagram