Open-domain Question Answering on MedQA (test)
72.82EM AccuracyOriginal (Unsafe)
Evaluation Results
| Method | Links | |
|---|---|---|
| Original (Unsafe)Backbone=Llama-3.2-3B2026.03 | 72.82 | |
| SEAL-TAGBackbone=Llama-3.2-3B2026.03 | 70.84 | |
| Few-Shot PromptingBackbone=Llama-3.2-3B2026.03 | 68.25 | |
| Eraser4RAGBackbone=Llama-3.2-3B2026.03 | 51.84 | |
| P3DeferBackbone=Llama-3.2-3B2026.03 | 44.58 | |
| PrivacyMindBackbone=Llama-3.2-3B2026.03 | 39.16 | |
| DPVoteRAGBackbone=Llama-3.2-3B2026.03 | 36.84 |