/dodrio

Exploring attention weights in transformer-based models with linguistic knowledge.

Primary LanguageSvelteMIT LicenseMIT

Dodrio

An interactive visualization system designed to help NLP researchers and practitioners analyze and compare attention weights in transformer-based models with linguistic knowledge.

build arxiv badge DOI:10.18653/v1/2021.acl-demo.16

For more information, check out our manuscript:

Dodrio: Exploring Transformer Models with Interactive Visualization. Zijie J. Wang, Robert Turko, and Duen Horng Chau. arXiv preprint 2021. arXiv:2103.14625.

Live Demo

For a live demo, visit: http://poloclub.github.io/dodrio/

Running Locally

Clone or download this repository:

git clone git@github.com:poloclub/dodrio.git

# use degit if you don't want to download commit histories
degit poloclub/dodrio

Install the dependencies:

npm install

Then run Dodrio:

npm run dev

Navigate to localhost:5000. You should see Dodrio running in your broswer :)

To see how we trained the Transformer or customize the visualization with a different model or dataset, visit the ./data-generation/ directory.

Credits

Dodrio was created by Jay Wang, Robert Turko, and Polo Chau.

Citation

@inproceedings{wangDodrioExploringTransformer2021,
  title = {Dodrio: {{Exploring Transformer Models}} with {{Interactive Visualization}}},
  shorttitle = {Dodrio},
  booktitle = {Proceedings of the 59th {{Annual Meeting}} of the {{Association}} for {{Computational Linguistics}} and the 11th {{International Joint Conference}} on {{Natural Language Processing}}: {{System Demonstrations}}},
  author = {Wang, Zijie J. and Turko, Robert and Chau, Duen Horng},
  year = {2021},
  pages = {132--141},
  publisher = {{Association for Computational Linguistics}},
  address = {{Online}},
  language = {en}
}

License

The software is available under the MIT License.

Contact

If you have any questions, feel free to open an issue or contact Jay Wang.