知识卡片
输出集成与权重融合分层不同
输出集成让多个模型分别生成,再选择或组合结果,推理仍用多模型;权重融合在参数空间合成一个模型。它们的成本、兼容条件和错误来源不同,不能把少存一个权重文件等同整合了全部能力。
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
- · 进阶微调技术
原书出处 · 1 处
- · 查看原书依据
model ensembling and model fusion or merging. This area of LLMs is in its infancy, and more work remains to be done to reap its full benefits. I call it the dark arts of NLP because the theoretical underpinnings of these techniques remain poorly understood. However, I do believe that even with these caveats it merits inclusion in this book, because the practical benefits are
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:摘录中关于输出集成与权重融合分层不同的定义、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:model ensembling and model fusion or merging. This area of LLMs is in its infancy, and more work remains to be done to reap its full benefits. I call it the dark arts of NLP because the theoretical underpinnings of these techniques remain poorly understood. However, I do believe that even with these caveats it merits inclusion in this book, because the practical benefits are