Where are the correct models

[mojesty]

Hello! Thanks for open-sourcing the realization of code for this paper! I can't understand what file contains the correct implementations: models/copynet.py or models/copynet_dbg.py: the former is older but I'm still not sure. Thanks!

[mojesty]

Also, I have some questions about the realization itself. As in both files the decoder returns tensor of shape [batch_size, vocab_size], it's not clear for me how the "pure copy" mode (when the word is from the source sequence but not from the vocab) can be enabled. Also, I wonder why you use argmax for testing phase: it usually makes models saturate to output just the the the or unk unk unk