textgenrnn-wrapper

A python wrapper for textgenrnn to manage saved files

For more extensive documentation, see textgenrnn: https://github.com/minimaxir/textgenrnn

Training a brand new model

To train a new model from the text file in 'data/ice-cream/IceCream_sorted.txt' and save in weights/ice_cream:

python textgenrnn_save_wrapper.py --model_name=ice_cream --new_model --num_epochs=1 --data_file='ice-cream/IceCream_sorted.txt'

Sampling from an already-trained model

To sample from the model saved in weights/ice_cream:

python textgenrnn_save_wrapper.py --model_name=ice_cream python textgenrnn_save_wrapper.py --model_name=ice_cream --n_gen=10 --temperature=0.2 python textgenrnn_save_wrapper.py --model_name=ice_cream --n_gen=10 --temperature=0.2 --prefix='Chocolate'

Continue training an already-trained model

Load a model from ice_cream, train on the text file in 'data/ice-cream/IceCream_sorted.txt' and overwrite the model in ice_cream

python textgenrnn_save_wrapper.py --model_name=ice_cream --num_epochs=1 --data_file='ice-cream/IceCream_sorted.txt'

Load a model from ice_cream, train on the text file in 'data/ice-cream/IceCream_sorted.txt' and save in weights/ice_cream2:

python textgenrnn_save_wrapper.py --model_name=ice_cream --num_epochs=1 --data_file='ice-cream/IceCream_sorted.txt' --save_name=ice_cream2

Transfer learning:

Load a model from ice_cream, train on the text file in 'data/superheroes/superheroes.txt' and save in weights/ice_cream_heroes:

python textgenrnn_save_wrapper.py --model_name=ice_cream --num_epochs=1 --data_file='superheroes/superheroes.txt' --save_name=ice_cream_heroes

Word level:

train a word-level model and save in weights/ice_cream_word:

python textgenrnn_save_wrapper.py --model_name=ice_cream_word --new_model --word_level --num_epochs=1 --data_file='ice-cream/IceCream_sorted.txt'

python textgenrnn_save_wrapper.py --model_name=ice_cream_word --new_model --word_level --max_words=10 --num_epochs=1 --data_file='ice-cream/IceCream_sorted.txt'

Load from file:

python textgenrnn_save_wrapper.py --model_name=ice_cream_word --temperature=1.0

Large text mode:

train:

python textgenrnn_save_wrapper.py --model_name=cocktails_largetext --large_text --new_model --num_epochs=1 --data_file='cocktails/cocktails.txt'