Response Quality Evaluation on MT-Bench
8.71Average Response QualityNo defense
Evaluation results
| Method | Links | |
|---|---|---|
| No defenseTarget Model=GPT-3.5-turbo2024.02 | 8.71 | |
| BacktranslationTarget Model=GPT-3.5-turbo2024.02 | 8.6 | |
| Response CheckTarget Model=GPT-3.5-turbo2024.02 | 8.58 | |
| ParaphraseTarget Model=GPT-3.5-turbo2024.02 | 8.43 | |
| No defenseTarget Model=Llama-2-13B-Chat2024.02 | 7.36 | |
| SmoothLLMTarget Model=GPT-3.5-turbo2024.02 | 7.35 | |
| Response CheckTarget Model=Llama-2-13B-Chat2024.02 | 7.3 | |
| BacktranslationTarget Model=Llama-2-13B-Chat2024.02 | 7.26 | |
| ParaphraseTarget Model=Llama-2-13B-Chat2024.02 | 7.23 | |
| No defenseTarget Model=Vicuna-13B2024.02 | 6.8 | |
| MTSA-T3Target Model=Zephyr-7B-Beta, Iteration=32025.05 | 6.78 | |
| BaselineTarget Model=Zephyr-7B-Beta2025.05 | 6.76 | |
| Response CheckTarget Model=Vicuna-13B2024.02 | 6.74 | |
| ParaphraseTarget Model=Vicuna-13B2024.02 | 6.69 | |
| BacktranslationTarget Model=Vicuna-13B2024.02 | 6.34 | |
| SmoothLLMTarget Model=Vicuna-13B2024.02 | 5.89 | |
| SmoothLLMTarget Model=Llama-2-13B-Chat2024.02 | 5.81 | |
| BaselineTarget Model=Llama2-7B-Chat2025.05 | 5.64 | |
| MTSA-T3Target Model=Llama2-7B-Chat, Iteration=32025.05 | 5.57 |