Skip to content

Dataset · sapientinc

Collection of over 4 million sudoku puzzles

sapientinc/sudoku-extreme

Massive text corpus of easy and extremely difficult sudoku puzzles for training and testing model logic.

What's inside

The dataset contains text sudoku problems divided into easy (1.1 million examples) and very difficult (3.1 million examples) puzzles. Data was collected from tdoku benchmarks, enjoysudoku, Kaggle sets, and specialized forums. The total corpus size ranges from 1 to 10 million entries.

What you can do with it

  • Training neural networks in step-by-step logical reasoning and solving combinatorial problems
  • Testing LLM capabilities in planning and searching for solutions without hallucinations
  • SFT fine-tuning of models on hard-constrained tasks
Tasks
Question answering
Size
1M–10M rows
Downloads this month
6.3K
Likes
34

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("sapientinc/sudoku-extreme")