This code is based on PreSumm from the EMNLP 2019 paper Text Summarization with Pretrained Encoders (repo)

Python version: This code is in Python 3.6

Package Requirements: torch==1.1.0 ransformers tensorboardX multiprocess pyrouge

For encoding a text longer than 512 tokens, for example 800. Set max_pos to 800 during both preprocessing and training.

Some codes are borrowed from ONMT(https://github.com/OpenNMT/OpenNMT-py)

BETO from Spanish Pre-Trained BERT Model and Evaluation Data

python preprocess.py -mode tokenize -raw_path RAW_PATH -save_path TOKENIZED_PATH

• RAW_PATH is the directory containing story files (../raw_stories), JSON_PATH is the target directory to save the generated json files (../merged_stories_tokenized)

python preprocess.py -mode format_to_lines -raw_path RAW_PATH -save_path JSON_PATH -n_cpus 1 -use_bert_basic_tokenizer false -map_path MAP_PATH

• RAW_PATH is the directory containing tokenized files (../merged_stories_tokenized), JSON_PATH is the target directory to save the generated json files (../json_data/cnndm), MAP_PATH is the directory containing the urls files (../urls)

python preprocess.py -mode format_to_bert -raw_path JSON_PATH -save_path BERT_DATA_PATH  -lower -n_cpus 1 -log_file ../logs/preprocess.log

• JSON_PATH is the directory containing json files (../json_data), BERT_DATA_PATH is the target directory to save the generated binary files (../bert_data)

First run: For the first time, you should use single-GPU, so the code can download the BERT model. Use -visible_gpus -1, after downloading, you could kill the process and rerun the code with multi-GPUs.

python train.py  -task abs -mode train -bert_data_path BERT_DATA_PATH -dec_dropout 0.2  -model_path MODEL_PATH -sep_optim true -lr_bert 0.002 -lr_dec 0.2 -save_checkpoint_steps 2000 -batch_size 140 -train_steps 200000 -report_every 50 -accum_count 5 -use_bert_emb true -use_interval true -warmup_steps_bert 20000 -warmup_steps_dec 10000 -max_pos 512 -visible_gpus 0,1,2,3  -log_file ../logs/abs_bert_cnndm