Generates synthetic datasets for stepwise Sudoku solving. The agent must fill in a partially completed 9×9 Sudoku grid one cell at a time, respecting row, column, and 3×3-region uniqueness constraints — a canonical constraint-satisfaction problem with deeply nested logical dependencies.
Each sample pairs a task (first frame + prompt describing what needs to happen) with its ground truth solution (final frame showing the result + video demonstrating how to achieve it). This structure enables both model evaluation and training.
| Property | Value |
|---|---|
| Task ID | Multi-14 |
| Task | Sudoku Stepwise Solving |
| Category | Constraint Satisfaction Puzzles |
| Resolution | 1024×1024 px |
| FPS | 16 fps |
| Duration | varies |
| Output | PNG images + MP4 video |
# 1. Clone the repository
git clone https://github.com/VBVR-DataFactory/Multi-14_sudoku_logic_data-generator.git
cd Multi-14_sudoku_logic_data-generator
# 2. Create and activate virtual environment
python3 -m venv venv
source venv/bin/activate # On Windows: venv\Scripts\activate
# 3. Install dependencies
pip install --upgrade pip
pip install -r requirements.txt
pip install -e .# Generate 50 samples
python examples/generate.py --num-samples 50
# Reproducible generation with seed
python examples/generate.py --num-samples 50 --seed 42
# Custom output directory
python examples/generate.py --num-samples 100 --output data/my_dataset
# Without videos (faster, images only)
python examples/generate.py --num-samples 50 --no-videos| Argument | Description |
|---|---|
--num-samples |
Number of tasks to generate (required) |
--output |
Output directory (default: data/questions) |
--seed |
Random seed for reproducibility |
--no-videos |
Skip video generation (images only) |
[Scenario] The image displays a 9x9 Sudoku puzzle grid with some pre-filled numeric clues.
[Rules]
1. Each row must contain the digits 1-9 exactly once.
2. Each column must contain the digits 1-9 exactly once.
3. Each of the nine 3x3 sub-grids must contain the digits 1-9 exactly once.
[Task] Generate a video animating the logical completion of the Sudoku grid. Fill all empty cells with the correct digits step-by-step, strictly following a top-to-bottom, left-to-right order, until the unique solution is fully revealed.
 |
 |
 |
| Initial Frame Sudoku puzzle with given clues |
Animation Cells filled in valid solving order |
Final Frame Completed 9×9 grid satisfying all constraints |
Complete a 9×9 Sudoku puzzle so that every row, column, and 3×3 region contains each digit 1–9 exactly once, while preserving the fixed starting clues.
- Grid: Standard 9×9 Sudoku, partitioned into nine 3×3 regions.
- Givens: A subset of clue cells fixed at generation time.
- Constraints: Each row, column, and 3×3 region must contain {1, 2, …, 9} with no repeats.
- Uniqueness: Puzzles are generated with a unique solution (verified via solver backtracking).
- Animation: Cells are revealed in a logically valid order — each new digit visibly satisfies the constraint set at the moment of placement.
- Global constraint propagation: Each digit placement narrows the candidate set of every related row/column/region.
- Long-horizon dependency chains: Late-stage cells often depend on inferences from cells filled dozens of steps earlier.
- Deterministic ground truth: The puzzle has a unique solution; the solving sequence is reproducible.
- Visible incremental commitment: Frame-by-frame reveal lets evaluators verify intermediate state correctness, not just final outcome.
data/questions/Multi-14_sudoku_logic_data-generator_task/Multi-14_sudoku_logic_data-generator_00000000/
├── first_frame.png # Sudoku puzzle with given clues
├── final_frame.png # Fully completed grid
├── prompt.txt # Task instruction
├── ground_truth.mp4 # Animation of stepwise filling
└── question_metadata.json # Standardized VBVR task metadata
File specifications:
- Images: 1024×1024 PNG format
- Video: MP4 format, 16 fps, H.264 + yuv420p
- Metadata: VBVR canonical schema with
task_id,vbvr_task_code,media,parameters
sudoku constraint-satisfaction csp logic-puzzle stepwise-solving multi-step-reasoning long-horizon
Part of the 36-Task Long-Horizon Multi-Step Reasoning Benchmark.