Preference Alignment on HelpSteer3
-5.89ScoreDPO+Filter
Evaluation results
| Method | Links | ||
|---|---|---|---|
| DPO+FilterBase Architecture=Qwen 7b, Pn Type=12025.10 | -5.89 | 73 | |
| DPOBase Architecture=Qwen 7b, Pn Type=12025.10 | -6.56 | 67 | |
| DPO+FilterBase Architecture=Qwen 7b, Pn Type=22025.10 | -6.83 | 68 | |
| DPOBase Architecture=Qwen 7b, Pn Type=22025.10 | -6.91 | 67 | |
| BaseBase Architecture=Qwen 7b, Pn Type=-2025.10 | -8.39 | 50 | |
| DPO+FilterBase Architecture=Llama 8b, Pn Type=12025.10 | -9.06 | 66 | |
| DPOBase Architecture=Llama 8b, Pn Type=12025.10 | -9.27 | 58 | |
| DPO+FilterBase Architecture=Llama 8b, Pn Type=22025.10 | -9.44 | 60 | |
| DPOBase Architecture=Llama 8b, Pn Type=22025.10 | -9.74 | 56 | |
| BaseBase Architecture=Llama 8b, Pn Type=-2025.10 | -10.43 | 50 |