Error(s) in loading state_dict for BARTModel

[engindeniz]

I have the same problem with #6. I installed fairseq as you suggested. There are required two paths for pretrained model, I defined them as follows:

• checkpoint_file: I downloaded your model from here.

• data_name_or_path: "Multi-View-Seq2Seq/train_sh/cnn_dm-bin_2/" (It is already in the repo)

Could you suggest any solution?

[jiaaoc]

The errors look like the models you installed from fairseq do not have some modules in the pre-trained model (like the "w_proj"). I would suggest checking the installed faieseq.

[jiaaoc]

For a quick check, I would suggest random initializing a Bart model and checking its named parameters.

[jiaaoc]

When I load the BART pretrained model, I can load without any error. I think that mismatch of named parameters is between your pretrained model and the Bart model. Could you check it, please? Btw, I am using "Eval_Sum.ipynb".

In this case, it seems that the fairseq you are using in the notebook is not correct. As our model indeed is different from original BART.

[engindeniz]

I uninstalled and installed it again. It is working now. I don't know what happened exactly before. It might about my conda configuration.
Thank you so much for your response.