Mathematical Reasoning on ASDiv (Init/Final Acc, Delta, Fix/Brk Count, Harm %)
78.4Initial Accuracyqwen2.5:1.5b
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| qwen2.5:1.5bModel Role=Initial Reasoner2026.05 | 78.4 | — | — | — | — | — | |
| GUARDEDREPAIRInitial Reasoner=qwen2.5:1.5b, Repair Stage=guarded best-of-32026.05 | — | 87.6 | 9.2 | 92 | 0 | 0 |