Human Evaluation on UltraFeedback 50 sampled questions
62Win Rate (Expert 1)OTPO
Evaluation results
| Method | Links | ||
|---|---|---|---|
| OTPOBase Model=Llama-3-8B-Instruct, Criteria=relevance, coherence, completeness, and conciseness2025.05 | 62 | 64 | |
| SimPOBase Model=Llama-3-8B-Instruct, Criteria=relevance, coherence, completeness, and conciseness2025.05 | 56 | 54 | |
| LDDPOBase Model=Llama-3-8B-Instruct, Criteria=relevance, coherence, completeness, and conciseness2025.05 | 56 | 48 | |
| SamPOBase Model=Llama-3-8B-Instruct, Criteria=relevance, coherence, completeness, and conciseness2025.05 | 48 | 46 | |
| DPOBase Model=Llama-3-8B-Instruct, Criteria=relevance, coherence, completeness, and conciseness2025.05 | 46 | 50 |