Dataset · nguha
LegalBench legal reasoning benchmark
nguha/legalbench
162 English tasks for evaluating LLM legal reasoning across classification and question answering.
What's inside
LegalBench is an open collaborative benchmark for evaluating legal reasoning in English LLMs. It comprises 162 tasks sourced from 40 origins, featuring text and tabular data for classification and question answering. The dataset card links to the project page, GitHub, and an arXiv paper.
What you can do with it
- Evaluating legal LLMs across 162 tasks
- Testing text-classification and QA models in the legal domain
- Comparing models on legal reasoning
- Tasks
- Text classification, Question answering
- Languages
- English
- Size
- 10K–100K rows
- Downloads this month
- 41.4K
- Likes
- 200
Models trained on it
- Sleem247/lori-qwen2.5-1.5b-legal-Q8_0-GGUFText generation 84
- uditjain/lori-qwen2.5-1.5b-legalText generation 45
- litillabs/arbit-1.7bText classification 37
- davidschulte/ESM_nguha__legalbench_cuad_non-disparagement 22
- litillabs/litil-clausecheck-1.5bText generation 22
- davidschulte/ESM_nguha__legalbench_contract_nli_permissible_acquirement_of_similar_information 15
- davidschulte/ESM_nguha__legalbench_contract_nli_no_licensing 15
- litillabs/litil-legal-intake-4bText generation 15
How to load it
from datasets import load_dataset
ds = load_dataset("nguha/legalbench")