Instruction Following on AlpacaEval Helpsteer2 (test)
29.64LC Win RateOTPO
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| OTPOBackbone=Llama-3-8b-Instruct2025.05 | 29.64 | 29.54 | 1,991 | |
| LDDPOBackbone=Llama-3-8b-Instruct2025.05 | 28.55 | 28.54 | 1,956 | |
| DPOBackbone=Llama-3-8b-Instruct2025.05 | 27.91 | 27.45 | 1,945 | |
| SamPOBackbone=Llama-3-8b-Instruct2025.05 | 26.95 | 27.16 | 1,991 | |
| SimPOBackbone=Llama-3-8b-Instruct2025.05 | 26.77 | 27.25 | 1,984 | |
| InitialBackbone=Llama-3-8b-Instruct2025.05 | 22.92 | 22.57 | 1,899 | |
| OTPOBackbone=Llama-3.2-3b-Instruct2025.05 | 20.5 | 21.25 | 2,000 | |
| LDDPOBackbone=Llama-3.2-3b-Instruct2025.05 | 20.29 | 20.2 | 1,939 | |
| DPOBackbone=Llama-3.2-3b-Instruct2025.05 | 19.99 | 20.54 | 1,970 | |
| SimPOBackbone=Llama-3.2-3b-Instruct2025.05 | 19.03 | 19.72 | 1,996 | |
| SamPOBackbone=Llama-3.2-3b-Instruct2025.05 | 18.58 | 19.65 | 2,018 | |
| InitialBackbone=Llama-3.2-3b-Instruct2025.05 | 17.97 | 19.34 | 2,041 |