Skip to content

Sudoku-RWKV

A RWKV model trained to solve Sudoku puzzles

Visit official site

What Sudoku-RWKV is

The tool is described as a Sudoku solving model based on RWKV that uses deep learning on a large puzzle sample set. The material states it can solve any solvable Sudoku puzzle.

What you can do with it

  • Run a demo script to try a puzzle
  • Generate training data with a script
  • Inspect model hyperparameters and loss curves
  • Use the shipped model file and vocabulary

Application scenarios

  • Solve hard puzzles as an enthusiast
  • Study optimisation as a researcher
  • Add solving to a mobile Sudoku game

Model notes

  • About 2M samples and 39.2B tokens are named
  • Roughly 12.7M parameters and a vocabulary of 133 are named
  • An 8-layer design with 320 dimensions per layer is named

Pros & cons

✓ What we like

  • Hyperparameters and loss curves are available to inspect
  • Several puzzle difficulty levels are named

! What to watch out for

  • The workflow runs through Python scripts
  • The listed model is trained specifically for Sudoku

FAQ

How much training data is named?

The material states about 2M Sudoku samples covering about 39.2B tokens.

Which scripts are provided?

The feature list names demo.py, minimum_inference.py and generate_sudoku_data.py.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review