Datasets:
Tasks:
Text Generation
Modalities:
Text
Formats:
parquet
Size:
1M - 10M
ArXiv:
Tags:
sudoku
constraint-satisfaction
planning
reasoning
diffusion-language-model
reinforcement-learning
License:
Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -138,16 +138,6 @@ Row 4: has 2,1 → need 3,4
|
|
| 138 |
</answer>
|
| 139 |
```
|
| 140 |
|
| 141 |
-
### Benchmark Results
|
| 142 |
-
|
| 143 |
-
| Method | Sudoku Accuracy | Paper |
|
| 144 |
-
|--------|----------------|-------|
|
| 145 |
-
| LLaDA-8B-Instruct (baseline) | 11.8% | d2 |
|
| 146 |
-
| LLaDA 1.5 | 12.5% | d2 |
|
| 147 |
-
| d1 (SFT + diffu-GRPO) | 22.1% | d1 |
|
| 148 |
-
| wd1 | 25.2% | d2 |
|
| 149 |
-
| SPG | +27% over d1 | SPG |
|
| 150 |
-
| **d2 (StepMerge)** | **91.9%** | d2 |
|
| 151 |
|
| 152 |
## Citation
|
| 153 |
|
|
|
|
| 138 |
</answer>
|
| 139 |
```
|
| 140 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 141 |
|
| 142 |
## Citation
|
| 143 |
|