Safety Detection on chat 1m (test)
100MCA AccuracyMLP-Cls.
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| MLP-Cls.Model=qwen2025.06 | 100 | 100 | — | |
| MLP-Cls.Model=baichuan2025.06 | 100 | 100 | — | |
| Uni.-DTMCModel=koala2025.06 | 100 | 100 | — | |
| ReGAModel=koala2025.06 | 100 | 100 | — | |
| MLP-Cls.Model=vicuna2025.06 | 99 | 100 | — | |
| MLP-Cls.Model=llama2025.06 | 99 | 100 | — | |
| MLP-Cls.Model=mistral2025.06 | 99 | 100 | — | |
| MLP-Cls.Model=Average2025.06 | 99 | 100 | — | |
| ReGAModel=baichuan2025.06 | 99 | 99 | — | |
| MLP-Cls.Model=koala2025.06 | 98 | 100 | — | |
| ReGAModel=mistral2025.06 | 98 | 99 | — | |
| Uni.-DTMCModel=llama2025.06 | 97 | 97 | — | |
| ReGAModel=Average2025.06 | 97 | 98 | — | |
| Uni.-DTMCModel=mistral2025.06 | 96 | 100 | — | |
| ReGAModel=llama2025.06 | 96 | 96 | — | |
| ReGAModel=vicuna2025.06 | 94 | 98 | — | |
| ReGAModel=qwen2025.06 | 92 | 98 | — | |
| Uni.-DTMCModel=baichuan2025.06 | 90 | 90 | — | |
| Uni.-DTMCModel=Average2025.06 | 87 | 95 | — | |
| Uni.-DTMCModel=qwen2025.06 | 84 | 100 | — | |
| Uni.-DTMCModel=vicuna2025.06 | 54 | 80 | — | |
| LLM-JudgeModel=vicuna2025.06 | — | — | 40 | |
| LLM-JudgeModel=llama2025.06 | — | — | 53 | |
| LLM-JudgeModel=qwen2025.06 | — | — | 100 | |
| LLM-JudgeModel=mistral2025.06 | — | — | 80 | |
| LLM-JudgeModel=koala2025.06 | — | — | 31 | |
| LLM-JudgeModel=baichuan2025.06 | — | — | 82 | |
| LLM-JudgeModel=Average2025.06 | — | — | 64 | |
| PPL-FilterModel=vicuna2025.06 | — | — | 85 | |
| PPL-FilterModel=llama2025.06 | — | — | 89 | |
| PPL-FilterModel=qwen2025.06 | — | — | 100 | |
| PPL-FilterModel=mistral2025.06 | — | — | 90 | |
| PPL-FilterModel=koala2025.06 | — | — | 84 | |
| PPL-FilterModel=baichuan2025.06 | — | — | 87 | |
| PPL-FilterModel=Average2025.06 | — | — | 89 |