Harmful Content Safety Evaluation on HexPHI
1.67Attack Success Rate (ASR)ST-D
Evaluation Results
| Method | Links | |
|---|---|---|
| ST-DBackbone Model=DeepSeek-R1-Distill-Llama-70B2026.06 | 1.67 | |
| ST-DBackbone Model=Qwen3-8B2026.06 | 5.67 | |
| ST-SBackbone Model=Qwen3-8B2026.06 | 6.67 | |
| SafePathBackbone Model=Qwen3-8B2026.06 | 7.33 | |
| ST-SBackbone Model=DeepSeek-R1-Distill-Llama-70B2026.06 | 7.33 | |
| ST-DBackbone Model=Qwen3-32B2026.06 | 8.33 | |
| Star1Backbone Model=Qwen3-8B2026.06 | 8.67 | |
| Star1Backbone Model=Qwen3-32B2026.06 | 9 | |
| ST-SBackbone Model=Qwen3-32B2026.06 | 10 | |
| BaseBackbone Model=Qwen3-32B2026.06 | 12.33 | |
| BaseBackbone Model=Qwen3-8B2026.06 | 13.33 | |
| ST-DBackbone Model=DeepSeek-R1-Distill-Llama-8B2026.06 | 13.33 | |
| SafePathBackbone Model=Qwen3-32B2026.06 | 13.33 | |
| Star1Backbone Model=DeepSeek-R1-Distill-Llama-70B2026.06 | 19.67 | |
| ST-SBackbone Model=DeepSeek-R1-Distill-Llama-8B2026.06 | 20 | |
| SafePathBackbone Model=DeepSeek-R1-Distill-Llama-70B2026.06 | 20 | |
| Star1Backbone Model=DeepSeek-R1-Distill-Llama-8B2026.06 | 20.67 | |
| BaseBackbone Model=DeepSeek-R1-Distill-Llama-70B2026.06 | 23.33 | |
| SafePathBackbone Model=DeepSeek-R1-Distill-Llama-8B2026.06 | 24.67 | |
| BaseBackbone Model=DeepSeek-R1-Distill-Llama-8B2026.06 | 44 |