(Contributed by @zuluzazu) The concept embeddings have already been frozen when instantiating the model with freeze_ent_emb=args.freeze_ent_emb. Adding freeze_net(model.decoder.concept_emb) could addtionally freeze self.cpt_transform = nn.Linear(concept_in_dim, concept_out_dim) in the CustomizedEmbedding object, which is not desired.
(Contributed by @zuluzazu) The concept embeddings have already been frozen when instantiating the model with
freeze_ent_emb=args.freeze_ent_emb
. Addingfreeze_net(model.decoder.concept_emb)
could addtionally freezeself.cpt_transform = nn.Linear(concept_in_dim, concept_out_dim)
in theCustomizedEmbedding
object, which is not desired.