lstm mxnet charRNN

Hello,

I am using Mxnet R package and walking through the following tutorial:
http://mxnet.io/tutorials/r/charRnnModel.html

I see that the lstm that was trained is used to generate text. However, is there a way to use the lstm model to evaluate test text?

When using language models, we can use perplexity to evaluate test data. Lower the perplexity, closer the language model to test data. Higher the perplexity, the farther away the language model is to test data.

Is there an equivalent score that can be generated using LSTM?

Thanks,
Aarthi