Safety Evaluation on WildTeaming
89HNaive RAG
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Naive RAGBackbone=Qwen2.5-7B-Instruct2026.05 | 89 | 2.71 | |
| BaseBackbone=Qwen2.5-7B-Instruct2026.05 | 87.2 | 2.72 | |
| Search-R1(79k)Backbone=Qwen2.5-7B-Instruct2026.05 | 87 | 2.21 | |
| Base AgentBackbone=Qwen2.5-7B-Instruct2026.05 | 86.8 | 2.32 | |
| Base Agent + Document FilteringBackbone=Qwen2.5-7B-Instruct2026.05 | 71.8 | 2.31 | |
| Base Agent + Query FilteringBackbone=Qwen2.5-7B-Instruct2026.05 | 70.2 | 2.27 | |
| Naive RAGBackbone=Llama3.1-8B-Instruct2026.05 | 69.8 | 2.41 | |
| Search-R1(79k)Backbone=Llama3.1-8B-Instruct2026.05 | 60.2 | 1.59 | |
| Base AgentBackbone=Llama3.1-8B-Instruct2026.05 | 55.2 | 1.82 | |
| Base Agent + Document FilteringBackbone=Llama3.1-8B-Instruct2026.05 | 44.2 | 2.02 | |
| BaseBackbone=Llama3.1-8B-Instruct2026.05 | 42.2 | 2.01 | |
| Base Agent + Query FilteringBackbone=Llama3.1-8B-Instruct2026.05 | 37.8 | 1.73 | |
| Safesearch(79k+12k)Backbone=Llama3.1-8B-Instruct2026.05 | 35.9 | 1.98 | |
| COMPASS(8k+2k)Backbone=Llama3.1-8B-Instruct2026.05 | 30.3 | 2.08 | |
| Safesearch(79k+12k)Backbone=Qwen2.5-7B-Instruct2026.05 | 23.4 | 2.38 | |
| COMPASS(8k+2k)Backbone=Qwen2.5-7B-Instruct2026.05 | 23.2 | 2.26 |