Addition on Addition 200 prompts Qwen3-8B (test)
100AccuracyProbe-round
Evaluation results
| Method | Links | |
|---|---|---|
| Probe-roundIntervention type=Probe-round, Argmax restriction=digit-restricted, Number of seeds=32026.05 | 100 | |
| 9-row repairIntervention type=9-row repair, Argmax restriction=digit-restricted, Number of seeds=3, Training protocol=9-row repair2026.05 | 100 | |
| BaselineIntervention type=None, Argmax restriction=digit-restricted, Number of seeds=32026.05 | 93.3 |