Policy Override on GTAPolicy
66.98Tool AccuracyTriMPI
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| TriMPIStrategy=PoRo-GRPO2025.10 | 66.98 | 59.03 | 63 | |
| TriMPIStrategy=PoRo-DAPO2025.10 | 63.21 | 54.66 | 58.94 | |
| CoT SFT + DAPORL Algorithm=DAPO2025.10 | 58.49 | 50.12 | 53.31 | |
| CoT SFT + GRPORL Algorithm=GRPO2025.10 | 50.94 | 41.6 | 46.27 | |
| CoT SFT2025.10 | 42.45 | 34.98 | 38.71 | |
| Direct SFT2025.10 | 40.57 | 33.42 | 36.99 | |
| In-Context2025.10 | 20.75 | 18.11 | 19.43 |