Skip to content

Dataset · tasksource

BIG-Bench without heavy dependencies

tasksource/bigbench

Repackaged BIG-Bench for loading without TensorFlow and unnecessary dependencies.

What's inside

English text dataset for multiple-choice, question-answering, text-classification, text-generation, and zero-shot-classification tasks. This is a repackaging of BIG-Bench without tensorflow, pypi-bigbench, and protobuf dependencies; tasks are limited to 50K examples. If train exists, the default split is removed; a Colab link for reproduction is provided.

What you can do with it

  • Evaluate LLMs in zero-shot on the BIG-Bench set
  • Fine-tuning model on multiple-choice and QA tasks
  • Run the benchmark without installing TensorFlow
Tasks
Multiple choice, Question answering, Text classification, Text generation, Zero shot classification
Languages
English
Size
100K–1M rows
Downloads this month
15.5K
Likes
70

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("tasksource/bigbench")