Legal-BERT — Legal Domain Language Model

Legal-BERT — Legal Domain Language Model

Legal-BERT is a family of BERT models pre-trained on 12 GB of diverse English legal corpora, including EU legislation, US court opinions, contracts, and legal journals. It consistently outperforms general-purpose BERT on legal NLP tasks such as contract clause classification, legal judgement prediction, and named entity recognition in legal documents.

Key Features

• Three variants: legal-bert-base-uncased, legal-bert-small-uncased, bert-base-uncased fine-tuned on legal text
• Pre-trained on EU legislation, US court opinions, contracts, and ECHR cases
• Drop-in replacement for bert-base-uncased — use the same HuggingFace API
• State-of-the-art performance on EURLEX, ECHR, and contract analysis benchmarks
• Supports zero-shot and few-shot fine-tuning with minimal labeled data

Quick Start

from transformers import AutoTokenizer, AutoModel

tokenizer = AutoTokenizer.from_pretrained("nlpaueb/legal-bert-base-uncased")
model = AutoModel.from_pretrained("nlpaueb/legal-bert-base-uncased")

inputs = tokenizer("The party shall indemnify the other party against all losses.", return_tensors="pt")
outputs = model(**inputs)
print(outputs.last_hidden_state.shape)  # [1, seq_len, 768]

npx ai-supply add legal-bert-base-uncased

Curated mirror of the open-source Legal-BERT (CC-BY-SA-4.0). Get it from the source.