Safety Evaluation on Redteaming Resistance Benchmark (RRB) 919-example subset
5.9HarmRBase LLM
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Base LLMBackbone=Mistral-NeMo-8B-Instruct2025.10 | 5.9 | 3.25 | |
| Naive RAGBackbone=Mistral-NeMo-8B-Instruct2025.10 | 9.9 | 3.29 | |
| Base LLMBackbone=Qwen-2.5-14B-Instruct2025.10 | 10.5 | 2.14 | |
| Base AgentBackbone=Qwen-2.5-14B-Instruct2025.10 | 11.8 | 2.15 | |
| Base AgentBackbone=Mistral-NeMo-8B-Instruct2025.10 | 12.7 | 3.1 | |
| Base LLMBackbone=Qwen-2.5-7B-Instruct2025.10 | 16.2 | 2.49 | |
| Naive RAGBackbone=Qwen-2.5-14B-Instruct2025.10 | 21.7 | 2.2 | |
| Base AgentBackbone=Qwen-2.5-7B-Instruct2025.10 | 29.1 | 2.52 | |
| Base LLMBackbone=Qwen-2.5-3B-Instruct2025.10 | 29.8 | 2.51 | |
| Ft. AgentBackbone=Qwen-2.5-7B-Instruct2025.10 | 32.4 | 2.55 | |
| Base AgentBackbone=Qwen-2.5-3B-Instruct2025.10 | 39.8 | 2.6 | |
| Naive RAGBackbone=Qwen-2.5-7B-Instruct2025.10 | 44.3 | 2.6 | |
| Naive RAGBackbone=Qwen-2.5-3B-Instruct2025.10 | 50.9 | 2.42 | |
| Ft. AgentBackbone=Qwen-2.5-3B-Instruct2025.10 | 51.4 | 2.58 |