Self-Affirmation Success Rate on MT-Bench-101
0.58Success RateEMBER (Single)
Evaluation results
| Method | Links | |
|---|---|---|
| EMBER (Single)Target Model=Mistral 0.3-7B2025.12 | 0.58 | |
| EMBER (Single)Target Model=Qwen 3-8B2025.12 | 0.516 | |
| EMBER (Multi)Target Model=Llama 3.1-8B2025.12 | 0.516 | |
| EMBER (Single)Target Model=Llama 3.1-8B2025.12 | 0.461 | |
| EMBER (Multi)Target Model=Qwen 3-14B2025.12 | 0.438 | |
| EMBER (Multi)Target Model=Qwen 3-8B2025.12 | 0.42 | |
| EMBER (Multi)Target Model=Mistral 0.3-7B2025.12 | 0.356 | |
| EMBER (Single)Target Model=Qwen 3-14B2025.12 | 0.352 | |
| SFTTarget Model=Mistral 0.3-7B2025.12 | 0.233 | |
| Prior Knowledge (Bench)Target Model=Mistral 0.3-7B2025.12 | 0.206 | |
| EMBER (Multi)Target Model=Qwen 3-32B2025.12 | 0.206 | |
| EMBER (Single)Target Model=Qwen 3-32B2025.12 | 0.183 | |
| SFTTarget Model=Qwen 3-14B2025.12 | 0.142 | |
| SFTTarget Model=Qwen 3-32B2025.12 | 0.069 | |
| SFTTarget Model=Qwen 3-8B2025.12 | 0.065 | |
| Prior Knowledge (Prompt)Target Model=Mistral 0.3-7B2025.12 | 0.027 | |
| Prior Knowledge (Prompt)Target Model=Qwen 3-8B2025.12 | 0.027 | |
| Prior Knowledge (Bench)Target Model=Qwen 3-32B2025.12 | 0.014 | |
| Prior Knowledge (Prompt)Target Model=Llama 3.1-8B2025.12 | 0.014 | |
| Prior Knowledge (Prompt)Target Model=Qwen 3-14B2025.12 | 0.014 | |
| Prior Knowledge (Prompt)Target Model=Qwen 3-32B2025.12 | 0.014 | |
| SFTTarget Model=Llama 3.1-8B2025.12 | 0.009 | |
| Prior Knowledge (Bench)Target Model=Llama 3.1-8B2025.12 | 0 | |
| Prior Knowledge (Bench)Target Model=Qwen 3-8B2025.12 | 0 | |
| Prior Knowledge (Bench)Target Model=Qwen 3-14B2025.12 | 0 |