from transformers import AutoTokenizer, AutoModel

tokenizer = AutoTokenizer.from_pretrained("maastrichtlawtech/legal-distilcamembert")
model = AutoModel.from_pretrained("maastrichtlawtech/legal-distilcamembert")

Training

Background

We utilize the distilcamembert-base checkpoint and further pre-train it with a masked language modeling (MLM) objective on legislation in French using the script from Hugging Face.

Hyperparameters

We train the model on a single Tesla V100 GPU with 32GBs of memory during 200 epochs (i.e., ~50k steps) using a batch size of 32. We use the AdamW optimizer with an initial learning rate of 5e-05, weight decay of 0.01, learning rate warmup over the first 500 steps, and linear decay of the learning rate. The sequence length was limited to 512 tokens.

Data

We use the Belgian Statutory Article Retrieval Dataset (BSARD) to further pre-train the model. BSARD is a French native dataset for studying legal information retrieval that includes more than 22,600 statutory articles from the Belgian legislation.

Citation

bibtex

@inproceedings{louis2023finding,
  title = {Finding the Law: Enhancing Statutory Article Retrieval via Graph Neural Networks},
  author = {Louis, Antoine and van Dijck, Gijs and Spanakis, Gerasimos},
  booktitle = {Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics},
  month = may,
  year = {2023},
  address = {Dubrovnik, Croatia},
  publisher = {Association for Computational Linguistics},
  url = {https://aclanthology.org/2023.eacl-main.203/},
  pages = {2753–2768},
}