Policy Referral on ClevrPolicy
6.26ClevrPolicy-T ScoreTriMPI
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TriMPIStrategy=PoRo-DAPO2025.10 | 6.26 | 8.35 | |
| CoT SFT + DAPORL Algorithm=DAPO2025.10 | 5.7 | 6.74 | |
| TriMPIStrategy=PoRo-GRPO2025.10 | 5.6 | 8.72 | |
| CoT SFT + GRPORL Algorithm=GRPO2025.10 | 5.04 | 6.74 | |
| CoT SFT2025.10 | 3.54 | 3.44 |