Finance Question Answering on Financial-qa-10k
56MPw/o defense
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| w/o defenseBackbone=Llama3.2-3B2026.06 | 56 | 86.8 | — | |
| SPBackbone=Llama3.2-3B2026.06 | 55.4 | 61.7 | 69.6 | |
| w/o defenseBackbone=Qwen2.5-7B2026.06 | 55 | 89.2 | — | |
| DPForwardBackbone=Llama3.2-3B2026.06 | 54.9 | 72.1 | 66.2 | |
| SPBackbone=Qwen2.5-7B2026.06 | 54.4 | 75.6 | 65.2 | |
| DualGuardBackbone=Qwen2.5-7B2026.06 | 54.2 | 57.2 | 71.2 | |
| DPForwardBackbone=Llama3-8B2026.06 | 53.3 | 85 | 62.5 | |
| DPForwardBackbone=Qwen2.5-7B2026.06 | 53.3 | 71.1 | 66.5 | |
| w/o defenseBackbone=Llama3-8B2026.06 | 52.7 | 88.3 | — | |
| SPBackbone=Llama3-8B2026.06 | 52.7 | 64.9 | 68.5 | |
| ADMIBackbone=Qwen2.5-7B2026.06 | 52.7 | 0.4 | 98.4 | |
| ADMIBackbone=Llama3.2-3B2026.06 | 51.6 | 1.7 | 96.8 | |
| ADMIBackbone=Llama3-8B2026.06 | 51.5 | 1.4 | 98.7 | |
| MIDBackbone=Qwen2.5-7B2026.06 | 51.5 | 29 | 82.9 | |
| MIDBackbone=Llama3.2-3B2026.06 | 51 | 16.7 | 89 | |
| DualGuardBackbone=Llama3-8B2026.06 | 51 | 44.6 | 76 | |
| CuxTextBackbone=Llama3.2-3B2026.06 | 50.7 | 83.7 | 62.8 | |
| CuxTextBackbone=Llama3-8B2026.06 | 49.8 | 84.9 | 62.5 | |
| MIDBackbone=Llama3-8B2026.06 | 49.7 | 2.3 | 97.4 | |
| CuxTextBackbone=Qwen2.5-7B2026.06 | 46.4 | 82 | 63.2 | |
| DualGuardBackbone=Llama3.2-3B2026.06 | 45.9 | 29.6 | 81.9 | |
| SanTextBackbone=Llama3-8B2026.06 | 43.1 | 80.5 | 63.6 | |
| SanTextBackbone=Qwen2.5-7B2026.06 | 39.9 | 76.7 | 64.4 | |
| SanTextBackbone=Llama3.2-3B2026.06 | 38.5 | 74.6 | 64.8 |