Skip to content

Dataset · allenai

WinoGrande: common-sense reasoning tasks

allenai/winogrande

44k fill-in-the-blank tasks with two answer choices for evaluating commonsense reasoning.

What's inside

English text tasks in a fill-in-the-blank format with binary choice. Inspired by the Winograd Schema Challenge, but scaled up and made more robust against dataset bias. Total of about 44k tasks; metadata: text, en, 10K-100K.

What you can do with it

  • Evaluating LLMs on commonsense reasoning
  • Fine-tuning or LoRA for answer selection
  • Dataset bias robustness benchmark
Languages
English
Size
10K–100K rows
Downloads this month
200.2K
Likes
102

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("allenai/winogrande")