Reward Modeling on Unified Feedback (UF)
78.9AccuracyGRM-SFT
Evaluation Results
| Method | Links | |
|---|---|---|
| GRM-SFTBackbone=Gemma2 2B it, Training Scale=Unified-Feedback 400K2026.02 | 78.9 | |
| BT-BNRMBackbone=Gemma2 2B it, Training Scale=Unified-Feedback 400K2026.02 | 78.8 | |
| GRM-BNRMBackbone=Gemma2 2B it, Training Scale=Unified-Feedback 400K2026.02 | 78.7 | |
| BT-MarginBackbone=Gemma2 2B it, Training Scale=Unified-Feedback 400K2026.02 | 77.3 | |
| GRM-DPOBackbone=Gemma2 2B it, Training Scale=Unified-Feedback 400K2026.02 | 77.3 | |
| InfoRMBackbone=Gemma2 2B it, Training Scale=Unified-Feedback 400K2026.02 | 77.3 | |
| BT-BNRMBackbone=Gemma2 2B it, Training Scale=Unified-Feedback 40K2026.02 | 77.2 | |
| BT-BNRMBackbone=Gemma 2B it, Training Scale=Unified-Feedback 400K2026.02 | 77 | |
| GRM-BNRMBackbone=Gemma2 2B it, Training Scale=Unified-Feedback 40K2026.02 | 76.9 | |
| BT-EnsembleBackbone=Gemma2 2B it, Training Scale=Unified-Feedback 400K2026.02 | 76.9 | |
| GRM-DPO-NorefBackbone=Gemma2 2B it, Training Scale=Unified-Feedback 400K2026.02 | 76.7 | |
| GRM-BNRMBackbone=Gemma 2B it, Training Scale=Unified-Feedback 400K2026.02 | 76.6 | |
| BTBackbone=Gemma2 2B it, Training Scale=Unified-Feedback 400K2026.02 | 76.6 | |
| BT-LabelSmoothBackbone=Gemma2 2B it, Training Scale=Unified-Feedback 400K2026.02 | 76.6 | |
| InfoRMBackbone=Gemma 2B it, Training Scale=Unified-Feedback 400K2026.02 | 76.2 | |
| GRM-SFTBackbone=Gemma2 2B it, Training Scale=Unified-Feedback 40K2026.02 | 75.8 | |
| GRM-DPOBackbone=Gemma2 2B it, Training Scale=Unified-Feedback 40K2026.02 | 75.5 | |
| GRM-DPO-NorefBackbone=Gemma2 2B it, Training Scale=Unified-Feedback 40K2026.02 | 75.2 | |
| BT-EnsembleBackbone=Gemma2 2B it, Training Scale=Unified-Feedback 40K2026.02 | 75.1 | |
| BT-MarginBackbone=Gemma2 2B it, Training Scale=Unified-Feedback 40K2026.02 | 74.7 | |
| BT-LabelSmoothBackbone=Gemma2 2B it, Training Scale=Unified-Feedback 40K2026.02 | 74.7 | |
| BTBackbone=Gemma2 2B it, Training Scale=Unified-Feedback 40K2026.02 | 74.5 | |
| BT-BNRMBackbone=Gemma 2B it, Training Scale=Unified-Feedback 40K2026.02 | 74.2 | |
| GRM-BNRMBackbone=Gemma 2B it, Training Scale=Unified-Feedback 40K2026.02 | 74.1 | |
| InfoRMBackbone=Gemma2 2B it, Training Scale=Unified-Feedback 40K2026.02 | 73.9 | |
| GRM-DPO-NorefBackbone=Gemma 2B it, Training Scale=Unified-Feedback 400K2026.02 | 73.9 | |
| GRM-DPOBackbone=Gemma 2B it, Training Scale=Unified-Feedback 400K2026.02 | 73.8 | |
| GRM-SFTBackbone=Gemma 2B it, Training Scale=Unified-Feedback 400K2026.02 | 73.2 | |
| BT-EnsembleBackbone=Gemma 2B it, Training Scale=Unified-Feedback 400K2026.02 | 72.8 | |
| BTBackbone=Gemma 2B it, Training Scale=Unified-Feedback 400K2026.02 | 72.1 | |
| BT-MarginBackbone=Gemma 2B it, Training Scale=Unified-Feedback 400K2026.02 | 72 | |
| InfoRMBackbone=Gemma 2B it, Training Scale=Unified-Feedback 40K2026.02 | 71.6 | |
| GRM-SFTBackbone=Gemma 2B it, Training Scale=Unified-Feedback 40K2026.02 | 71.5 | |
| BT-LabelSmoothBackbone=Gemma 2B it, Training Scale=Unified-Feedback 400K2026.02 | 71.5 | |
| GRM-DPO-NorefBackbone=Gemma 2B it, Training Scale=Unified-Feedback 40K2026.02 | 71.4 | |
| GRM-DPOBackbone=Gemma 2B it, Training Scale=Unified-Feedback 40K2026.02 | 70.2 | |
| BT-EnsembleBackbone=Gemma 2B it, Training Scale=Unified-Feedback 40K2026.02 | 69.9 | |
| BT-MarginBackbone=Gemma 2B it, Training Scale=Unified-Feedback 40K2026.02 | 69.6 | |
| BTBackbone=Gemma 2B it, Training Scale=Unified-Feedback 40K2026.02 | 68.8 | |
| BT-LabelSmoothBackbone=Gemma 2B it, Training Scale=Unified-Feedback 40K2026.02 | 68.5 |