nplm-theano is a re-implementation of the Neural Probablistic Language Model tookit (nplm) by USC-ISI.
- It has the same model dumping format as the nplm toolkit, so you can integrate it with other softwares such as Moses.
- The implementation is based on theano, so you are free to run it on GPU and speed up training.
- theano >= 0.8.0
- lasagna >= 0.2.0dev1
To use GPU for language model training, you may need the following optional packages as well:
- CUDA >= 7
- cuDNN >= 3 (to enable further theano optimization)
All the argument usages can be printed by running:
python nplm.py --help
- Vaswani et al. 2013, Decoding with Large-Scale Neural Language Models Improves Translation: http://www.aclweb.org/anthology/D13-1140
- Dyer 2014, Notes on Noise Contrastive Estimation and Negative Sampling: http://arxiv.org/abs/1410.8251
- Zoph et al. 2016, Simple, Fast Noise-Contrastive Estimation for Large RNN Vocabularies: http://www.aclweb.org/anthology/N16-1145