Paper: Advances of Transformer-Based Models for News Headline Generation

Note that there are Pipfile and Pipfile.lock in src.

As long as you have pipenv installed, you can just run pipenv install and set the environment up.

After that, you can just run pipenv shell.

Download RuBERT from DeepPavlov: http://docs.deeppavlov.ai/en/master/features/models/bert.html and extract the archive to src

BertSumAbs checkpoint: https://yadi.sk/d/2jcjmdEXp0EX-Q

If you want to use the model, download the checkpoint using the link above and put this file in the rubert_cased_L-12_H-768_A-12_pt folder.

You can easily change any parameters and hyperparameters by modifying config.yaml

Run the following command to convert your data to the required format. Pay attention to the samples in the repo and the config file.

python preprocess_input.py --config_path ./config.yaml

Once your data is ready, you can either train your model (again, check the config file to set everything up and then go to step 2) or test your model (go directly to step 3).

python run.py --mode train --config_path ./config.yaml  

python run.py --mode test --config_path ./config.yaml