This repo contains our implementation of a simple Word-Sense-Disambiguation system that users can access according to the Software-as-a-service model. We leverage the PyTorch Lightning Template to avoid writing boilerplate code and promote code quality, and GitHub Actions to encourage best practices when deploying to production builds. All the experiments are logged using Weights & Biases. To have a quick overview of the status of this project visit the related project board.
The documentation related to this project is available in the Wiki.
For the training procedure, simply run python -m src.train model.use_lexeme_mask=True
, which will train the model with the default set of hyper-parameters. For the unit tests, run python -m unittest -v <TEST_SCRIPT>
pointing to the test you are looking for.
neural-wsd
| conf # contains Hydra config files
| data
| model
| train
root.yaml # hydra root config file
| data # datasets should go here
| experiments # where the models are stored
| src
| pl_data_modules.py # base LightinigDataModule
| pl_modules.py # base LightningModule
| train.py # main script for training the network
| tests
| unit # contains unit tests
| README.md
| pyproject.toml # black configuration file
| requirements.txt
| setup.sh # environment setup script
- Andrea Bacciu - Personal website
- Leonardo Emili - Personal website