• Download SAMSum dialogue from here or execute:

curl https://arxiv.org/src/1911.12237v2/anc/corpus.7z --output samsum.7z

• Uncompress samsum.7z
• Download fairseq toolkit:

git clone https://github.com/pytorch/fairseq.git

• Add fairseq to python path:

export PYTHONPATH="${PYTHONPATH}:/FAIRSEQ_PATH

• Create a directory to save prepared data:

mkdir PATH_TO_FAIRSEQ/examples/bart/samsum

• Run prepare_data.py:

python prepare_data.py --path_samsum PATH_TO_SAMSUM --path_samsum_bart PATH_SAMSUM_IN_BART

where PATH_TO_SAMSUM is the directory containig SAMSum dataset and PATH_SAMSUM_IN_BARTis the created directory in the previous step.

• Move to bart directory:

cd PATH_TO_FAIRSEQ/examples/bart

• Run BPE tokenization by executing the following command:

wget -N 'https://dl.fbaipublicfiles.com/fairseq/gpt2_bpe/encoder.json'
wget -N 'https://dl.fbaipublicfiles.com/fairseq/gpt2_bpe/vocab.bpe'
wget -N 'https://dl.fbaipublicfiles.com/fairseq/gpt2_bpe/dict.txt'

TASK=samsum
for SPLIT in train val
do
  for LANG in source target
  do
    python -m examples.roberta.multiprocessing_bpe_encoder \
    --encoder-json encoder.json \
    --vocab-bpe vocab.bpe \
    --inputs "$TASK/$SPLIT.$LANG" \
    --outputs "$TASK/$SPLIT.bpe.$LANG" \
    --workers 1 \
    --keep-empty;
  done
done

• Binarize data by executing:

python -m preprocess \
  --source-lang "source" \
  --target-lang "target" \
  --trainpref "${TASK}/train.bpe" \
  --validpref "${TASK}/val.bpe" \
  --destdir "${TASK}-bin/" \
  --workers 1 \
  --srcdict dict.txt \
  --tgtdict dict.txt;

• Download Bart Pretrained model:

wget https://dl.fbaipublicfiles.com/fairseq/models/bart.large.tar.gz
tar xvzf bart.large.tar.gz

• Finetune the model by executing:

TOTAL_NUM_UPDATES=20000  
WARMUP_UPDATES=100      
LR=3e-05
MAX_TOKENS=1048
UPDATE_FREQ=4
BART_PATH=bart.large.cnn/model.pt

CUDA_VISIBLE_DEVICES=0 
time python -m train samsum-bin \
    --restore-file $BART_PATH \
    --max-tokens $MAX_TOKENS \
    --task translation \
    --source-lang source --target-lang target \
    --truncate-source \
    --layernorm-embedding \
    --share-all-embeddings \
    --share-decoder-input-output-embed \
    --reset-optimizer --reset-dataloader --reset-meters \
    --required-batch-size-multiple 1 \
    --arch bart_large \
    --criterion label_smoothed_cross_entropy \
    --label-smoothing 0.1 \
    --dropout 0.1 --attention-dropout 0.1 \
    --weight-decay 0.01 --optimizer adam --adam-betas "(0.9, 0.999)" --adam-eps 1e-08 \
    --clip-norm 0.1 \
    --lr-scheduler polynomial_decay --lr $LR \
    --total-num-update $TOTAL_NUM_UPDATES \
    --warmup-updates $WARMUP_UPDATES \
    --fp16 --update-freq $UPDATE_FREQ \
    --skip-invalid-size-inputs-valid-test \
    --find-unused-parameters \
    --batch-size 1 \
    --save-dir samsum-bin \
    --memory-efficient-fp16 \
    --max-epoch 3 \
    --disable-validation \

• Generate summaries from a checkpoint by running generate_summaries.py

python generate_summaries.py --checkpoint PATH_TO_CHECKPOINT --test_source PATH_TO_TEST_SOURCE --summaries_file PATH_TO_OUTPUT