- Notifications
You must be signed in to change notification settings - Fork1
A Fast RNN Language Model by Chainer
License
NotificationsYou must be signed in to change notification settings
soskek/rnnlm_chainer
Folders and files
| Name | Name | Last commit message | Last commit date | |
|---|---|---|---|---|
Repository files navigation
This is a fast implementation of an RNN language model (RNNLM) by Chainer.This repository is derived from theChainer example for RNNLM in PTB.
The network architecture is almost same as the "Medium" model in the paper,"Recurrent Neural Network Regularization" by Wojciech Zaremba, Ilya Sutskever and Oriol Vinyals.You can train an RNNLM in 1 miniute per epoch, with backprop length of 35 and batchsize of 20.
python -u train.py -g 0 --dataset ptb- PennTreeBank (
ptb) - Wikitext-2 (
wikitext-2) - Wikitext-103 (
wikitext-103)
For wikitext, runprepare_wikitext.sh for downloading the datasets.