Preference Aggregation on Preference Evaluation Suite Aggregate
36.68Average Preference Win RateAlphaToken
Evaluation Results
| Method | Links | |
|---|---|---|
| AlphaTokenBackbone=Qwen-3.5-9B, Warm-start Reference=UltraChat-200k2026.06 | 36.68 | |
| SePOBackbone=Qwen-3.5-9B, Warm-start Reference=UltraChat-200k2026.06 | 34.03 | |
| ConfPOBackbone=Qwen-3.5-9B, Warm-start Reference=UltraChat-200k2026.06 | 33.82 | |
| TI-DPOBackbone=Qwen-3.5-9B, Warm-start Reference=UltraChat-200k2026.06 | 32.33 | |
| DPOBackbone=Qwen-3.5-9B, Warm-start Reference=UltraChat-200k2026.06 | 32.11 | |
| AlphaTokenBackbone=Gemma-3-4B, Warm-start Reference=UltraChat-200k2026.06 | 29.63 | |
| ConfPOBackbone=Gemma-3-4B, Warm-start Reference=UltraChat-200k2026.06 | 26.68 | |
| SePOBackbone=Gemma-3-4B, Warm-start Reference=UltraChat-200k2026.06 | 26.62 | |
| TI-DPOBackbone=Gemma-3-4B, Warm-start Reference=UltraChat-200k2026.06 | 24.99 | |
| DPOBackbone=Gemma-3-4B, Warm-start Reference=UltraChat-200k2026.06 | 20.94 | |
| AlphaTokenBackbone=Llama-3.2-3B, Warm-start Reference=UltraChat-200k2026.06 | 17.62 | |
| SePOBackbone=Llama-3.2-3B, Warm-start Reference=UltraChat-200k2026.06 | 15.93 | |
| ConfPOBackbone=Llama-3.2-3B, Warm-start Reference=UltraChat-200k2026.06 | 15.07 | |
| BaseBackbone=Qwen-3.5-9B, Warm-start Reference=UltraChat-200k2026.06 | 14.12 | |
| TI-DPOBackbone=Llama-3.2-3B, Warm-start Reference=UltraChat-200k2026.06 | 13.9 | |
| DPOBackbone=Llama-3.2-3B, Warm-start Reference=UltraChat-200k2026.06 | 11.13 | |
| BaseBackbone=Gemma-3-4B, Warm-start Reference=UltraChat-200k2026.06 | 8.59 | |
| BaseBackbone=Llama-3.2-3B, Warm-start Reference=UltraChat-200k2026.06 | 3.1 |