Skip to content

Dataset · livebench

Leakage-free data analysis benchmark

livebench/data_analysis

Updatable complex data analysis tasks with objective answers for fair language model evaluation.

What's inside

Text dataset with fewer than 1000 examples from the LiveBench project, which fights test data contamination in training sets. Questions are regularly updated and built on fresh datasets, arXiv papers, news, and IMDb movie descriptions. Each task has an unambiguous and verifiable correct answer.

What you can do with it

  • Independent testing of LLMs for data analysis and fact-handling capabilities
  • Comparing models without the risk of metric inflation due to training set memorization
Size
up to 1K rows
Downloads this month
23.5K
Likes
8

How to load it

from datasets import load_dataset

ds = load_dataset("livebench/data_analysis")