知识卡片
批量大小取决于负载和目标
更大批量可能提高硬件吞吐,却增加显存并改变优化动态;推理最大批量也可能违反单请求延迟要求。二的幂不是普遍最优法则,应结合序列长度、并发和内核实测。
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
- · 微调
原书出处 · 1 处
- · 查看原书依据
Batch Size Along with the learning rate, the batch size is one of the most important hyperparameters we need to set. A larger batch size means training will proceed faster. However, larger batch sizes also require more memory. Larger batch sizes can also lead the model to land in a sharp local minima, which can be a sign of overfitting. Therefore, there are trade offs
参考来源
- 位置:《Designing Large Language Model Applications》第06章《微调》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Designing Large Language Model Applications/_epub-src/OEBPS/ch06.html)
- 结论依据:摘录中关于批量大小取决于负载和目标的定义、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:Batch Size Along with the learning rate, the batch size is one of the most important hyperparameters we need to set. A larger batch size means training will proceed faster. However, larger batch sizes also require more memory. Larger batch sizes can also lead the model to land in a sharp local minima, which can be a sign of overfitting. Therefore, there are trade offs