line 307 in editor.py

[SH9959]

Dear EasyEdit team，
I may have found a seeming contradiction line 307: EasyEdit/easyeditor/editors/editor.py
with torch.no_grad(): for k, v in weights_copy.items(): nethook.get_parameter(self.model, k)[...] = v.to(f "cuda:{self.hparams.device}")
This seems to be restoring the weights of the original model. But the final return value is named edited_model in line 361.
This has confused me, and I'm not sure if I'm missing other information that has led me to this confusion.

[pengzju]

Actually, the edited_model is a reference (address) to model, so they are essentially the same.

[SH9959]

😆Yes. But my question is that the weights of the specific layer in edited_model are recovered by weight_copy which records the origin value of origin model. So the edited_model gets the same weights as the origin model. I'm confused about whether the return value is what we expect. 🤔

[pengzju]

👍Yes, your understanding is correct. In the past editing literature, the measurements were the results of a single edit, so it was necessary to roll back after each evaluation. However, in EasyEdit, you can enable continuous editing by setting keep_original_weights = False, more details in the issue #220

[SH9959]

Oh, I see. Thanks for your reply!😆