clauselens-cuad

Fine-tuned RoBERTa model for extractive contract clause extraction and legal question answering, trained on the Contract Understanding Atticus Dataset (CUAD).

Model Performance on CUAD Test Set

• Exact Match (EM): 73.71%
• F1 Score: 85.15%
• Base Architecture: deepset/roberta-base-squad2
• Max Answer Length: 346 tokens (99th percentile of contract clauses)

Usage

from transformers import AutoModelForQuestionAnswering, AutoTokenizer

model_id = "CodexTanishq/clauselens-cuad"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForQuestionAnswering.from_pretrained(model_id)