This is a trained model of PEGASUS on cnn_dailymail dataset.
numpy
sentencepiece
tensorflow==2.2.0
Pre-training with Extracted Gap-sentences for Abstractive SUmmarization Sequence-to-sequence models, or PEGASUS, uses self-supervised objective Gap Sentences Generation (GSG) to train a transformer encoder-decoder model. The paper can be found on arXiv. ICML 2020 accepted.
If you use this code or these models, please cite the following paper:
@misc{zhang2019pegasus,
title={PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization},
author={Jingqing Zhang and Yao Zhao and Mohammad Saleh and Peter J. Liu},
year={2019},
eprint={1912.08777},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
To run the demo, please download pre-trained model on cnn_dailymail from here or gigaword from here. Unzip it and put it to model/
, or anywhere
really if you just specify its location and where your article file is. Suppose your article is this one
python test_example.py --article example_article --model_dir model/ --model_name cnn_dailymail
You will see this output - PREDICTION >> The hacking group known as NC29 is largely believed to operate as part of Russia's security services .<n>The three countries allege that it is carrying out a persistent and ongoing cyber campaign to steal intellectual property about a possible coronavirus vaccine .
To export a model you have trained, please place the ExportModel.ipynb
inside the PEGASUS folder.
Just run the script inside by specifying which data model you want to export.