← 返回刚才阅读的位置

知识卡片

结构图卡 · 8673.d · 生成模型:GPT-6

适配方案按变化层次选择

理解等级:中级

结构图:

flowchart TD
 G[明确更新目标与保留能力] --> D[选择领域或任务数据]
 D --> C[继续预训练:调整暴露分布]
 D --> P[PEFT:限定可训练参数]
 G --> E[输出集成:多模型生成与选择]
 G --> M[权重融合:兼容坐标下组合]
 C --> V[目标任务与旧能力共同验收]
 P --> V
 E --> V
 M --> V

这些路线可组合;每条路线要报告自身资源成本和假设。

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

      In this chapter, we will learn techniques that can be used to update the LLM for the aforementioned reasons. To this end, the chapter is divided into three sections: Continual pre-training Primarily used for domain adaptation and keeping the knowledge of the LLM up-to-date (the latter is also called lifelong-learning). Parameter-Efficient Fine-Tuning (PEFT) A set of fine-tuning techniques that make the fine-tuning process more efficient

    参考来源

    - 位置:《Designing Large Language Model Applications》第07章《进阶微调技术》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch07.html) - 结论依据:摘录中关于适配方案按变化层次选择的定义、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:In this chapter, we will learn techniques that can be used to update the LLM for the aforementioned reasons. To this end, the chapter is divided into three sections: Continual pre-training Primarily used for domain adaptation and keeping the knowledge of the LLM up-to-date (the latter is also called lifelong-learning). Parameter-Efficient Fine-Tuning (PEFT) A set of fine-tuning techniques that make the fine-tuning process more efficient