BART for SAMSum Dataset

This code is based on Fairseq.

Requirements

• We use Conda python 3.7 and strongly recommend that you create a new environment.
  • conda create -n bart python=3.7.
• Run the following command.
  • pip install --editable ./
  • pip install -r requirements.txt

Data

You can get data here. Put them under the dir data/*.

Reproduce Results

You can follow the following steps to reproduce the best results in our paper.

download checkpoints

Download checkpoints here. Put the checkpoint under the dir ckpt/samsum.pt.

preprocess

• sh ./my_scripts/bpe.sh
• sh ./my_scripts/binarize.sh

translate

• Produce final summaries.
  • sh ./my_scripts/infer.sh

test rouge score

• python py_rouge_test.py -c summaries/samsum.txt

ROUGE score

	ROUGE-1	ROUGE-2	ROUGE-L
SAMSum	53.70	28.79	50.81

From Scratch

Download BART checkpoint

Download bart.large checkpoint. Put it under the dir bart/bart.large/*.

Preprocess

• sh ./my_scripts/bpe.sh
• sh ./my_scripts/binarize.sh

Train

• sh ./my_scripts/train.sh

Translate

Run the following command:

• sh ./my_scripts/infer.sh
  • set up ckpt_dir param first.