Which specific model to use as generator?
peppe69 opened this issue · comments
I'm trying to reproduce your work, I trained the CatSeq model using the code of the original project by Hou Pong Chan and Wang Chen and Lu Wang and Irwin King (here: https://github.com/kenchan0226/keyphrase-generation-rl ), but when I run your project to train Discriminator it raises this error:
RuntimeError: Error(s) in loading state_dict for Seq2SeqModel:
Missing key(s) in state_dict: "decoder.p_gen_linear.weight", "decoder.p_gen_linear.bias".
Maybe I used the wrong model?
Have you used the -copy_attention flag in both the training the Generator and the Discriminator. It seems that you haven't used the -copy_attention flag while training the generator.
Oh yes, now it works. Thank you so much!