How to change to to word-level?

[sunsidazzz]

Nice work! It is very clear and easy to understand!

I am wondering how could I change it to a word-level Generator?

[spiglerg]

You may try to use a system for word embedding to convert words to dense input vectors and to decode the outputs of the network. This should work reasonably well as long as semantically similar words get mapped to similar vectors. You should try and look for word-rnn code here on github to get further ideas. In particular, you should think of some ways of sampling the output of the network, which would not be as trivial as with the one-hot-encoding of char-rnn. :)

[sunsidazzz]

Thanks for the response! 👍 I am thinking to use the word2vec embedding with the LSTM Model. Do you think that's a good direction to build the word-level rnn?

[spiglerg]

Yep, I think that's the way to go! :)