Skip to content

Dataset · zai-org

Long-text understanding benchmark

zai-org/LongBench

20 English and Chinese tasks for testing LLMs on long context: QA, summarization, code.

What's inside

LongBench is a comprehensive benchmark for multilingual and multi-task evaluation of long-text understanding. It includes 20 tasks: single- and multi-document QA, summarization, few-shot learning, synthetic tasks, and code completion. Languages are English and Chinese, size ranges from 1,000 to 10,000 records.

What you can do with it

  • Evaluate LLMs on long context
  • Compare models in QA and summarization
  • Test few-shot and code completion
Tasks
Question answering, Text generation, Summarization, Text classification
Languages
English, Chinese
Size
1K–10K rows
Downloads this month
46K
Likes
204

How to load it

from datasets import load_dataset

ds = load_dataset("zai-org/LongBench")