wgapl / blocks-char-rnn

Multi-layer RNN (LSTM, GRU, RNN) for character-level language models in Blocks

char-rnn in Blocks

This code is a python implementation of Torch char-rnn project using the Blocks framework.

Requirements

• Install Blocks. Please see the documentation for more information.

Usage

• Set text_file parameter in the config.py file. You can try the input.txt file used in the original code.
• run make_dataset.py to create the Fuel dataset.
• run train.py file to train a Gated RNN (Cho et al.). 'rnn' and 'lstm' are also supported.
• run sample.py to sample characters using a trained model.

train.py and sample.py scripts follow most of the parameters from the original char-rnn project. Please take a look on it to train your own models.