Sudoku solving on Sudoku-Extreme (held-out)
95.2AccuracySelf-Aware Markov Models
Evaluation Results
| Method | Links | |
|---|---|---|
| Self-Aware Markov ModelsBackbone architecture=Standard Transformer (8 blocks, 8 heads)2026.03 | 95.2 | |
| DFM + Top-KBackbone architecture=Same as Ours2026.03 | 68.8 | |
| DFM + Top-K MarginBackbone architecture=Same as Ours2026.03 | 68.5 | |
| DFM + ReMDMBackbone architecture=Same as Ours2026.03 | 51.3 | |
| DFM (Euler)Backbone architecture=Same as Ours2026.03 | 31.6 |