Skip to content

Solution to SQuAD 2.0 using QANet

License

Notifications You must be signed in to change notification settings

sahilkhose/QANet

 
 

Repository files navigation

QANet on SQuAD 2.0

💡 Model Architecture

Here are our 📚 Notes for the implementation

📋 Dependencies

git clone https://github.com/abhirajtiwari/SQuAD2.git
pip3 install -r requirements.txt

Setup

  1. This downloads SQuAD 2.0 training and dev sets, as well as the GloVe 300-dimensional word vectors (840B)
  2. This also pre-processes the dataset for efficient data loading
python3 setup.py

🔧 Training

CLI args training args.py

python3 train.py -n train_run_1 --num_workers 4 --num_epochs 30 --eval_steps 50000 --batch_size 64 --hidden_size 128

To load the tensorboard

tensorboard --logdir save

📚 Citing QANet

If you find QANet useful in your research, please consider citing:

@article{DBLP:journals/corr/abs-1804-09541,
  author    = {Adams Wei Yu and
               David Dohan and
               Minh{-}Thang Luong and
               Rui Zhao and
               Kai Chen and
               Mohammad Norouzi and
               Quoc V. Le},
  title     = {QANet: Combining Local Convolution with Global Self-Attention for
               Reading Comprehension},
  journal   = {CoRR},
  volume    = {abs/1804.09541},
  year      = {2018},
  url       = {http://arxiv.org/abs/1804.09541},
  archivePrefix = {arXiv},
  eprint    = {1804.09541},
  timestamp = {Mon, 13 Aug 2018 16:48:18 +0200},
  biburl    = {https://dblp.org/rec/journals/corr/abs-1804-09541.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}

If you liked our work, please leave a 🌟

About

Solution to SQuAD 2.0 using QANet

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%