Simply modify run.py to build an LSTM model or predict the perplexity of some test corpus with an existing model.