README.md

metadata

license: mit
task_categories:
  - question-answering
language:
  - en
size_categories:
  - 100<n<1K
tags:
  - puzzle
  - reasoning
  - evaluation
  - games
  - spatial-reasoning

Rush Hour 4x4 Evaluation Dataset

This dataset contains 150 4x4 Rush Hour puzzles for model evaluation with difficulty labels.

Format

• puzzle_id: Unique identifier (puzzle1, puzzle2, ...)
• prompt: Full formatted prompt as used in model inference
• solution: Optimal solution with proper formatting
• optimal_moves: Number of moves in optimal solution
• difficulty: Difficulty level (easy, medium, hard)

Usage

Each puzzle contains:

1. A grid state in JSON format
2. Piece positions and types
3. Movement rules and constraints
4. Expected output format

Perfect for evaluating reasoning capabilities of language models on spatial puzzles.

File Formats

• dataset.parquet: Recommended format - handles multi-line text properly (150 rows)
• dataset.jsonl: JSON Lines format - one puzzle per line (150 rows)
• dataset.json: Complete JSON array format (150 entries)
• dataset_info.json: Dataset metadata

Dataset Statistics

• Total puzzles: 150
• Difficulty levels: Easy (50), Medium (50), Hard (50)
• Grid size: 4x4
• Optimal solution length: 1-15 moves