Mathematical Reasoning on GSM8K (Accuracy, G drop)
0.7422AccuracyMAST (GA)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MAST (GA)Steps=50, Learning rate (η)=5×10−62026.06 | 0.7422 | -0.781 | |
| RLVR source2026.06 | 0.7344 | — | |
| Full-parameter GASteps=50, Learning rate (η)=5×10−62026.06 | 0.6719 | 6.25 |