Conversational Evaluation on AlpacaEval 2.0
26.57LCCPO
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| CPOBackbone=Llama-3-8B-Instruct2026.05 | 26.57 | 25.15 | 1,879 | |
| SimPOBackbone=Llama-3-8B-Instruct2026.05 | 25.91 | 23.48 | 1,810 | |
| DPOBackbone=Llama-3-8B-Instruct2026.05 | 25.09 | 24.6 | 1,896 | |
| RDPOBackbone=Llama-3-8B-Instruct2026.05 | 25.01 | 24.25 | 1,895 | |
| IPOBackbone=Llama-3-8B-Instruct2026.05 | 20.41 | 21.48 | 1,993 | |
| KTOBackbone=Llama-3-8B-Instruct2026.05 | 17.4 | 17.72 | 1,925 | |
| ORPOBackbone=Llama-3-8B-Instruct2026.05 | 17.29 | 17 | 1,876 | |
| RRHFBackbone=Llama-3-8B-Instruct2026.05 | 16.4 | 15.53 | 1,830 | |
| Contrastive-POBackbone=Llama-3-8B-Instruct2026.05 | 15.95 | 18.94 | 2,169 | |
| SLiC-HFBackbone=Llama-3-8B-Instruct2026.05 | 15.06 | 16.33 | 1,998 | |
| SFT-BaseBackbone=Llama-3-8B-Instruct2026.05 | 13.47 | 14.22 | 1,972 | |
| DPO2025.02 | 8.73 | 4.78 | — | |
| VAR2025.02 | 8.26 | 4.29 | — | |
| Base2025.02 | 4.01 | 3.73 | — |