setting learning rate

[StarBlue98]

描述该功能

when proposing a new backbone based on a resnet50 in mmpretrain, I only have 4gpus on one server, and I found from the log that resnet50 is pretrained on 8 gpus, so I wonder how to set the learning rate if I use 4 gpus to train models, and is there any relationship among learning rate, the number of gpus, and batch size.

是否希望自己实现该功能？

• 我希望自己来实现这一功能，并向 MMPreTrain 贡献代码！