Safety Evaluation on WildTeaming 500-example (test)
88.6HarmRFt. Agent
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Ft. AgentBackbone=Qwen-2.5-3B-Instruct2025.10 | 88.6 | 2.87 | |
| Naive RAGBackbone=Qwen-2.5-3B-Instruct2025.10 | 88.5 | 2.84 | |
| Base AgentBackbone=Qwen-2.5-3B-Instruct2025.10 | 87.5 | 2.9 | |
| Naive RAGBackbone=Qwen-2.5-7B-Instruct2025.10 | 87.3 | 2.97 | |
| Ft. AgentBackbone=Qwen-2.5-7B-Instruct2025.10 | 87.3 | 2.88 | |
| Base LLMBackbone=Qwen-2.5-3B-Instruct2025.10 | 87 | 2.84 | |
| Base AgentBackbone=Qwen-2.5-7B-Instruct2025.10 | 83.9 | 2.79 | |
| Base LLMBackbone=Qwen-2.5-7B-Instruct2025.10 | 81.7 | 2.73 | |
| Naive RAGBackbone=Qwen-2.5-14B-Instruct2025.10 | 77.4 | 2.45 | |
| Base AgentBackbone=Qwen-2.5-14B-Instruct2025.10 | 69.1 | 2.43 | |
| Base LLMBackbone=Qwen-2.5-14B-Instruct2025.10 | 68.5 | 2.4 | |
| Base LLMBackbone=Mistral-NeMo-8B-Instruct2025.10 | 56.3 | 3.29 | |
| Naive RAGBackbone=Mistral-NeMo-8B-Instruct2025.10 | 54.9 | 3.17 | |
| Base AgentBackbone=Mistral-NeMo-8B-Instruct2025.10 | 54.9 | 3.17 |