Commonsense Reasoning on ECQA (MV, EM, PK)
77.4MV ScoreI-PPO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| I-PPOBackbone=LLaMa2026.04 | 77.4 | 78.24 | 83.51 | |
| I-PPOBackbone=LLaMa3 8B2026.04 | 77.4 | 78.24 | 83.51 | |
| I-PPOBackbone=Gemma2026.04 | 75.9 | 77.21 | 85.52 | |
| I-PPOBackbone=Gemma2 2B2026.04 | 75.9 | 77.21 | 85.52 | |
| PPOBackbone=Phi3 4B2026.04 | 74.78 | 72.25 | 86.75 | |
| I-PPOBackbone=Phi3 4B2026.04 | 74.34 | 76.78 | 86.18 | |
| PPOBackbone=Gemma2026.04 | 73.89 | 72.36 | 83.31 | |
| PPOBackbone=Gemma2 2B2026.04 | 73.89 | 72.36 | 83.31 | |
| PPOBackbone=LLaMa2026.04 | 73.31 | 72.54 | 82.82 | |
| PPOBackbone=LLaMa3 8B2026.04 | 73.31 | 72.54 | 82.82 | |
| GRPOBackbone=LLaMa2026.04 | 71.36 | 71.33 | 80.05 | |
| GRPOBackbone=Gemma2026.04 | 70.96 | 70.47 | 80.63 | |
| BaselineBackbone=Phi3 4B2026.04 | 70.88 | 70.9 | 85.16 | |
| I-PPOBackbone=Qwen2.5 3B2026.04 | 70.79 | 78.18 | 92.11 | |
| BaselineBackbone=LLaMa2026.04 | 70.62 | 70.59 | 79.94 | |
| BaselineBackbone=LLaMa3 8B2026.04 | 70.62 | 70.59 | 79.94 | |
| BaselineBackbone=Qwen2.5 3B2026.04 | 69.7 | 72.65 | 92.48 | |
| PPOBackbone=Qwen2.5 3B2026.04 | 69.01 | 71.89 | 91.56 | |
| BaselineBackbone=Gemma2026.04 | 68.42 | 68.57 | 80.45 | |
| BaselineBackbone=Gemma2 2B2026.04 | 68.42 | 68.57 | 80.45 | |
| I-PPOBackbone=Rho2026.04 | 56.76 | 64.45 | 80.58 | |
| I-PPOBackbone=Rho 1B2026.04 | 56.76 | 64.45 | 80.58 | |
| PPOBackbone=Rho2026.04 | 51.98 | 55.12 | 78.36 | |
| PPOBackbone=Rho 1B2026.04 | 51.98 | 55.12 | 78.36 | |
| GRPOBackbone=Rho2026.04 | 44.63 | 48.11 | 70.59 | |
| BaselineBackbone=Rho2026.04 | 34.19 | 42.13 | 92.78 | |
| BaselineBackbone=Rho 1B2026.04 | 34.19 | 42.13 | 92.78 |