Loading a pre-trained model with Weight Drop

[dhananjaisharma10]

Hi!

I trained a sequence-to-sequence model using Weight Drop and then tried using it for inference. But I got the following error:

RuntimeError: Error(s) in loading state_dict for Seq2Seq:
        Unexpected key(s) in state_dict: "encoder.lstm2.module.weight_hh_l0", "encoder.lstm2.module.weight_hh_l0_reverse", "encoder.lstm3.module.weight_hh_l0", "encoder.lstm3.module.weight_hh_l0_reverse", "encoder.lstm4.module.weight_hh_l0", "encoder.lstm4.module.weight_hh_l0_reverse".

I presume this is because of the fact that those keys are by default not a part of my model. I am using model.load_state_dict(torch.load('PATH'), strict=False) in order to load the weights, but I'm unsure as to whether the strict=False part is correct or not.

Please let me know if I'm erring somewhere, or in case you need any further information.
Thanks!

[dhananjaisharma10]

Solved! Refer to Issue #86 and use the hack-code by @sdraper-CS