nag-bert's Introduction

NAG-BERT

Non-Autoregressive Text Generation with Pre-trained Language Models

Authors: Yixuan Su, Deng Cai, Yan Wang, David Vandyke, Simon Baker, Piji Li, and Nigel Collier

Introduction:

In this repository, we provide the related resources to our EACL 2021 paper. We provide training and inference code for text summarization task.

1. Enviornment Installtion:

pip install -r requirements.txt

To install pyrouge, please refer to this link

2. Download Gigawords Data here:

unzip data.zip and replace it with the empty ./data folder.

3. Training

chmod +x ./train.sh
./train.sh

4. Inference

chmod +x ./inference.sh
./inference.sh

The $\alpha$ in the ratio-first decoding can be controlled by changing the value of --length_ratio

5. Citation

If you find our paper and resources useful, please kindly cite our paper:

@inproceedings{su-etal-2021-non,
    title = "Non-Autoregressive Text Generation with Pre-trained Language Models",
    author = "Su, Yixuan  and
      Cai, Deng  and
      Wang, Yan  and
      Vandyke, David  and
      Baker, Simon  and
      Li, Piji  and
      Collier, Nigel",
    booktitle = "Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume",
    month = apr,
    year = "2021",
    address = "Online",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2021.eacl-main.18",
    pages = "234--243"
}

Acknowledgements

The authors would like to thank Huggingface and Fairseq for making their awesome codes publicly available. Some of our codes are borrowed from these libraries.

Recommend Projects