Harmful Content Safety Evaluation on XsTest Harmful
0Attack Success Rate (ASR)SafePath
Evaluation Results
| Method | Links | |
|---|---|---|
| SafePathBackbone Model=Qwen3-8B2026.06 | 0 | |
| ST-SBackbone Model=Qwen3-8B2026.06 | 0 | |
| ST-DBackbone Model=Qwen3-8B2026.06 | 0 | |
| BaseBackbone Model=Qwen3-32B2026.06 | 0 | |
| Star1Backbone Model=Qwen3-32B2026.06 | 0 | |
| SafePathBackbone Model=Qwen3-32B2026.06 | 0 | |
| ST-SBackbone Model=Qwen3-32B2026.06 | 0 | |
| ST-DBackbone Model=Qwen3-32B2026.06 | 0 | |
| ST-DBackbone Model=DeepSeek-R1-Distill-Llama-70B2026.06 | 0 | |
| BaseBackbone Model=Qwen3-8B2026.06 | 0.5 | |
| Star1Backbone Model=Qwen3-8B2026.06 | 0.5 | |
| ST-SBackbone Model=DeepSeek-R1-Distill-Llama-70B2026.06 | 3.5 | |
| ST-DBackbone Model=DeepSeek-R1-Distill-Llama-8B2026.06 | 7.5 | |
| ST-SBackbone Model=DeepSeek-R1-Distill-Llama-8B2026.06 | 8 | |
| SafePathBackbone Model=DeepSeek-R1-Distill-Llama-8B2026.06 | 8.5 | |
| Star1Backbone Model=DeepSeek-R1-Distill-Llama-8B2026.06 | 9.5 | |
| BaseBackbone Model=DeepSeek-R1-Distill-Llama-70B2026.06 | 14 | |
| SafePathBackbone Model=DeepSeek-R1-Distill-Llama-70B2026.06 | 14 | |
| Star1Backbone Model=DeepSeek-R1-Distill-Llama-70B2026.06 | 15.5 | |
| BaseBackbone Model=DeepSeek-R1-Distill-Llama-8B2026.06 | 20 |