Is it possible to specify different lr for different layer or fix some layer？

[funan-jhc-lee]

Hello！As stated in the title, is it possible to specify different lr for different layer or fix some layer, when retraining a model？I‘ve read the document, but It seems not possible.

[amcadmus]

Hello @funan-jhc-lee

DeepMD-kit does not support different lr in different layers.

You can fix the parameters in the embedding net by setting trainable to False, or fix the parameters of the hidden and output layers of the fitting net by a list of booleans provided to trainable.

Please refer to https://github.com/deepmodeling/deepmd-kit/blob/master/doc/train-input-auto.rst
model/descriptor[se_a]/trainable and
model/fitting_net[ener]/trainable
for more information

[funan-jhc-lee]

Thanks a lot! I think It's a quite useful feature that allows users specify the lr for diffrent layers or add more neurons(more layers) when retraining a model. Is it possible that this feature will be added in a future update? Thanks again!