t5-long-extract

The code for T5-LONG-EXTRACT system submitted to FNS-2021 Shared Task

Training

Unpack training data to current directory.

Run source install.sh to install requirements.

Run python prepare.sh to prepare training data.

Run source train.sh to train the model.

Running the model

Unpack test data to current directory.

Run python eval.sh to prepare the data, run the model and generate summaries based on the predictions.

Generated summaries will be placed in verification/system directory.

Model weights

https://huggingface.co/orzhan/t5-long-extract

Colab demo

https://colab.research.google.com/drive/1cJiDGYrmmlQ3Ei5ZcitRJLMcw_WA5RW-?usp=sharing

Citation

If you use this codebase, or otherwise found our work valuable, please cite:

@inproceedings{orzhenovskii-2021-t5,
    title = "T5-{LONG}-{EXTRACT} at {FNS}-2021 Shared Task",
    author = "Orzhenovskii, Mikhail",
    booktitle = "Proceedings of the 3rd Financial Narrative Processing Workshop",
    month = "15-16 " # sep,
    year = "2021",
    address = "Lancaster, United Kingdom",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2021.fnp-1.12",
    pages = "67--69",
}