知识卡片
卸载以数据搬运换可容纳性
将模型部分放在CPU或磁盘,可让超出单GPU容量的模型参与推理,但计算时的搬运和调度可能增加延迟。能加载不等于达到可接受速度,需同时检查内存层级、设备映射与实际负载。
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
- · 按任务适配模型
原书出处 · 1 处
- · 查看原书依据
CPU RAM if the GPU RAM is filled, and then loading parts of the model into disk if the CPU RAM is also filled. “Accelerate Big Model Inference: How Does it Work?” shows how the accelerate library operates under the hood. This whole process is abstracted from the user, so all you need to load a large model is to run this code:
参考来源
- 位置:《Designing Large Language Model Applications》第05章《按任务适配模型》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch05.html)
- 结论依据:摘录中关于卸载以数据搬运换可容纳性的定义、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:CPU RAM if the GPU RAM is filled, and then loading parts of the model into disk if the CPU RAM is also filled. “Accelerate Big Model Inference: How Does it Work?” shows how the accelerate library operates under the hood. This whole process is abstracted from the user, so all you need to load a large model is to run this code: