metadata
language: en
pipeline_tag: fill-mask
license: cc-by-sa-4.0
tags:
- legal
model-index:
- name: lexlms/legal-roberta-large
results: []
widget:
- text: >-
The applicant submitted that her husband was subjected to treatment
amounting to <mask> whilst in the custody of police.
- text: This <mask> Agreement is between General Motors and John Murray.
- text: >-
Establishing a system for the identification and registration of <mask>
animals and regarding the labelling of beef and beef products.
- text: >-
Because the Court granted <mask> before judgment, the Court effectively
stands in the shoes of the Court of Appeals and reviews the defendants’
appeals.
datasets:
- lexlms/lex_files
LexLM large
This model was continued pre-trained from RoBERTa large (https://huggingface.co/roberta-large) on the LeXFiles corpus (https://huggingface.co/datasets/lexlms/lex_files).
Model description
LexLM (Base/Large) are our newly released RoBERTa models. We follow a series of best-practices in language model development:
- We warm-start (initialize) our models from the original RoBERTa checkpoints (base or large) of Liu et al. (2019).
- We train a new tokenizer of 50k BPEs, but we reuse the original embeddings for all lexically overlapping tokens (Pfeiffer et al., 2021).
- We continue pre-training our models on the diverse LeXFiles corpus for additional 1M steps with batches of 512 samples, and a 20/30% masking rate (Wettig et al., 2022), for base/large models, respectively.
- We use a sentence sampler with exponential smoothing of the sub-corpora sampling rate following Conneau et al. (2019) since there is a disparate proportion of tokens across sub-corpora and we aim to preserve per-corpus capacity (avoid overfitting).
- We consider mixed cased models, similar to all recently developed large PLMs.
Intended uses & limitations
More information needed
Training and evaluation data
The model was trained on the LeXFiles corpus (https://huggingface.co/datasets/lexlms/lexfiles). For evaluation results, please consider our work "LeXFiles and LegalLAMA: Facilitating English Multinational Legal Language Model Development" (Chalkidis* et al, 2023).
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.0001
- train_batch_size: 8
- eval_batch_size: 8
- seed: 42
- distributed_type: tpu
- num_devices: 8
- gradient_accumulation_steps: 4
- total_train_batch_size: 256
- total_eval_batch_size: 64
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.05
- training_steps: 1000000
Training results
| Training Loss | Epoch | Step | Validation Loss |
|---|---|---|---|
| 1.1322 | 0.05 | 50000 | 0.8690 |
| 1.0137 | 0.1 | 100000 | 0.8053 |
| 1.0225 | 0.15 | 150000 | 0.7951 |
| 0.9912 | 0.2 | 200000 | 0.7786 |
| 0.976 | 0.25 | 250000 | 0.7648 |
| 0.9594 | 0.3 | 300000 | 0.7550 |
| 0.9525 | 0.35 | 350000 | 0.7482 |
| 0.9152 | 0.4 | 400000 | 0.7343 |
| 0.8944 | 0.45 | 450000 | 0.7245 |
| 0.893 | 0.5 | 500000 | 0.7216 |
| 0.8997 | 1.02 | 550000 | 0.6843 |
| 0.8517 | 1.07 | 600000 | 0.6687 |
| 0.8544 | 1.12 | 650000 | 0.6624 |
| 0.8535 | 1.17 | 700000 | 0.6565 |
| 0.8064 | 1.22 | 750000 | 0.6523 |
| 0.7953 | 1.27 | 800000 | 0.6462 |
| 0.8051 | 1.32 | 850000 | 0.6386 |
| 0.8148 | 1.37 | 900000 | 0.6383 |
| 0.8004 | 1.42 | 950000 | 0.6408 |
| 0.8031 | 1.47 | 1000000 | 0.6314 |
Framework versions
- Transformers 4.20.0
- Pytorch 1.12.0+cu102
- Datasets 2.7.0
- Tokenizers 0.12.0
Citation
@inproceedings{chalkidis-garneau-etal-2023-lexlms,
title = {{LeXFiles and LegalLAMA: Facilitating English Multinational Legal Language Model Development}},
author = "Chalkidis*, Ilias and
Garneau*, Nicolas and
Goanta, Catalina and
Katz, Daniel Martin and
Søgaard, Anders",
booktitle = "Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics",
month = july,
year = "2023",
address = "Toronto, Canada",
publisher = "Association for Computational Linguistics",
url = "https://arxiv.org/abs/2305.07507",
}