Code for byte-level learning of twitter bios (full training set is in a JSON on my laptop)
Link to the state_dict of a trained model.
model.py - defines the LayerRNN (which was used for training)
data.py - utilties for building byte datasets
train.py - routines used to train the model
profiles.py - used to build the training set.
sample.py is a script for sampling strings from a trained model file, printing to stdout. See command
line help for details. Requires a local copy of a model file (such as that from the link above) plus a model config file and byte values config file; these are available in this repo for the trained model above as model_config.json and byte_values.txt respectively.
Example: python sample.py model_file -N 5 prints 5 samples from the model.
Sampling is quite slow on my laptop - GPU will be used automatically if available.