Home

Awesome

QANet-pytorch

NOTICE

I'm very busy these months. I'll return to this repo in about 10 days.

Introduction

An implementation of QANet with PyTorch.

Any contributions are welcome!

Current performance

F1EMGot by
66?InitialBug
6450BangLiu

Usage

  1. Install pytorch 0.4 for Python 3.6+
  2. Run pip install -r requirements.txt to install python dependencies.
  3. Run download.sh to download the dataset.
  4. Run python preproc.py to build tensors from the raw dataset.
  5. Run python main.py --mode train to train the model. After training, log/model.pt will be generated.
  6. Run python main.py --mode test to test an pretrained model. Default model file is log/model.pt

Structure

preproc.py: downloads dataset and builds input tensors.

main.py: program entry; functions about training and testing.

models.py: QANet structure.

config.py: configurations.

Differences from the paper

  1. The paper doesn't mention which activation function they used. I use relu.
  2. I don't set the embedding of <UNK> trainable.
  3. The connector between embedding layers and embedding encoders may be different from the implementation of Google, since the description in the paper is inconsistent (residual block can't be used because the dimensions of input and output are different) and they don't say how they implemented it.

TODO

Contributors

  1. InitialBug: found two bugs: (1) positional encodings require gradients; (2) wrong weight sharing among encoders.
  2. linthieda: fixed one issue about dependencies and offered computing resources.
  3. BangLiu: tested the model.
  4. wlhgtc: (1) improved the calculation of Context-Question Attention; (2) fixed a bug that is compacting embeddings before highway nets.