/char-rnn

Reimplement rnn-based character language model (CHAR-RNN)

Primary LanguagePython

CHAR-RNN

  • Reimplement rnn-based character language model (CHAR-RNN) inspired by Karpathy's blog using pytorch.
  • Reimplement vanilia RNN and GRU as well.
  • The data used for training is the works of Shakespeare.

Requirements

  • pytorch 1.1.0
  • tqdm
  • tensorboardX

Train

python main.py --learning_rate 0.002\
    --seq_length 50\
    --epoch 100\
    --batch 64\
    --hidden_size 256\
    --rnn gru\
    --sampling sample

Plot train_loss and val_loss via TensorboardX. Alt text

Result

clown:
one florance, i am for meet,
i shall it not verbided, a mad;
nuth, sir, for post thy advertis: make quite
tribunes:
take honourable commission: tell sprayk aboutly.
by many lipprame him; how i too.

rivers:
go several, gaves to the top to too?
i would plantagenetions as you
in twenty tongue and love;
no four right of my first custom for my sword, and she
he knows not yet.

first citizen:
we have but let him up.
but fetch a thousand abqua