Skip to content
forked from qinyao-he/bit-rnn

Quantize weights and activations in Recurrent Neural Networks.

License

Notifications You must be signed in to change notification settings

sofia3ms/bit-rnn

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

31 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Bit-RNN

Source code for paper: Effective Quantization Methods for Recurrent Neural Networks.

The implementation of PTB language model is modified from examples in tensorflow.

Requirments

Currently tested and run on TensorFlow 1.8 and Python 3.6. View other branches for legacy support. You may download the data from http://www.fit.vutbr.cz/~imikolov/rnnlm/simple-examples.tgz.

Run

python train.py --config=config.gru --data_path=YOUR_DATA_PATH

Currently default is 2-bit weights and activations. You may edit the config file in config folder to change configuration.

Support

Submit issue for problem relate to the code itself. Send email to the author for general question about the paper.

Citation

Please cite follow if you use our code in your research:

@article{DBLP:journals/corr/HeWZWYZZ16,
  author    = {Qinyao He and
               He Wen and
               Shuchang Zhou and
               Yuxin Wu and
               Cong Yao and
               Xinyu Zhou and
               Yuheng Zou},
  title     = {Effective Quantization Methods for Recurrent Neural Networks},
  journal   = {CoRR},
  volume    = {abs/1611.10176},
  year      = {2016},
  url       = {http://arxiv.org/abs/1611.10176},
  timestamp = {Thu, 01 Dec 2016 19:32:08 +0100},
  biburl    = {http://dblp.uni-trier.de/rec/bib/journals/corr/HeWZWYZZ16},
  bibsource = {dblp computer science bibliography, http://dblp.org}
}

About

Quantize weights and activations in Recurrent Neural Networks.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%