Visualising trainability of layers in ULMFit model

I would like to check trainability of the layers as I gradually unfreeze them according to ULMFit paper and tutorial. I create a learner for classifier as shown there.

learn = text_classifier_learner(data_clas, AWD_LSTM, drop_mult=0.5)

I would like to see trainability of parameters as I gradually unfreeze layers of AWD_LSTM. However, these layers are not present when I call learn.summary(). Those layers that are present belong to classifier that is put on top of AWD_LSTM encoder and they appear always trainable. Can anyone point me into the right direction?

1 Like

I also have the same question.