OpenNMT Baseline for Headline Generation shared task on Dialogue 2019
This baseline cut off the first sentence in a news article body. Then whis phrase is tekonized by sentecepiece, learned on Rossiya Segodnya corpus. The tokenized sequence is processed by seq2seq implemented on OpenNMT framework, also learned on Rossiya Segodnya corpus. The resulting sequence then de-tokenized to get hypothesis for article title.
To use it:
- fork and/or clone the repo;
- login into shared task private registry with command
sudo docker login headlinegen.vkpartner.ru:5001 - build docker image with command
sudo docker build --tag headlinegen.vkpartner.ru:5001/{your_token} . - push docker image to registry with command
sudo docker push headlinegen.vkpartner.ru:5001/{your_token} - click Валидировать on shared task website.