Maze Solving on Wilson maze (test)
100Plan AccuracyCorrect Traces
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Correct TracesBackbone=Qwen3-8B-base, Fine-tuning regime=Correct Traces2025.05 | 100 | 100 | |
| Solution-onlyBackbone=Qwen3-8B-base, Fine-tuning regime=Solution-only2025.05 | 99.7 | 0 | |
| Swapped TracesBackbone=Qwen3-8B-base, Fine-tuning regime=Swapped Traces2025.05 | 98 | 0 | |
| Base ModelBackbone=Qwen3-8B-base, Fine-tuning regime=None2025.05 | 1.3 | 0 |