Instruction following on Arena-Hard v2 (test)
1.3AH2 ScoreGPRL
Evaluation results
| Method | Links | |
|---|---|---|
| GPRLGroup=Online RL with reward model, RM Size=8B, RM Type=GPM, Iter./Ep.=32026.05 | 1.3 | |
| SimPOGroup=No reward model, RM Size=–, RM Type=–, Iter./Ep.=–2026.05 | 1 | |
| DPOGroup=No reward model, RM Size=–, RM Type=–, Iter./Ep.=–2026.05 | 0.9 | |
| SPPOGroup=Reward model, RM Size=8B, RM Type=BT, Iter./Ep.=32026.05 | 0.9 | |
| SPPOGroup=Reward model, RM Size=8B, RM Type=GPM, Iter./Ep.=32026.05 | 0.9 | |
| GPOGroup=Reward model, RM Size=8B, RM Type=GPM, Iter./Ep.=32026.05 | 0.9 | |
| GPOGroup=Reward model, RM Size=8B, RM Type=BT, Iter./Ep.=32026.05 | 0.8 | |
| GRPOGroup=Online RL with reward model, RM Size=8B, RM Type=BT, Iter./Ep.=32026.05 | 0.8 |