Open-domain Question Answering on FinDER (test)
59.62EM AccuracyOriginal (Unsafe)
Evaluation Results
| Method | Links | |
|---|---|---|
| Original (Unsafe)Backbone=Llama-3.2-3B2026.03 | 59.62 | |
| SEAL-TAGBackbone=Llama-3.2-3B2026.03 | 58.28 | |
| Few-Shot PromptingBackbone=Llama-3.2-3B2026.03 | 56.15 | |
| Eraser4RAGBackbone=Llama-3.2-3B2026.03 | 47.25 | |
| PrivacyMindBackbone=Llama-3.2-3B2026.03 | 36.47 | |
| P3DeferBackbone=Llama-3.2-3B2026.03 | 34.74 | |
| DPVoteRAGBackbone=Llama-3.2-3B2026.03 | 34.68 |