How to set different weight decay values for different modules of the model

[AlfredQin]

💡 Your Question

Hi, I am trying to let the parts of my model have different weight decay value during training. For example, the backbone has weight decay 0.05 while the head has 0.01. But I don't know how to implement it in the yaml configuration. Currently I can set different learning rate for each part of the model like this:
training_hyperparams:
initial_lr:
backbone: 1e-5
head: 1e-4
default: 1e-4
Howerver I can't find available method to set weight decay like what I do on learning rate configuration. Could you please help me fix it ? Thank you !

Versions

No response