General Question Answering & Instruction Following on Dolly
22MP Scorew/o defense
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| w/o defenseBackbone=Llama3.2-3B2026.06 | 22 | 87.6 | — | |
| ADMIBackbone=Llama3-8B2026.06 | 20.5 | 1.6 | 98.9 | |
| w/o defenseBackbone=Qwen2.5-7B2026.06 | 18.3 | 98.5 | — | |
| w/o defenseBackbone=Llama3-8B2026.06 | 17.6 | 93.4 | — | |
| CuxTextBackbone=Llama3-8B2026.06 | 17.6 | 93.7 | 60.1 | |
| SPBackbone=Llama3-8B2026.06 | 17.3 | 60.7 | 70 | |
| DPForwardBackbone=Llama3-8B2026.06 | 17.1 | 60.8 | 69.9 | |
| DualGuardBackbone=Qwen2.5-7B2026.06 | 17.1 | 61.3 | 69.8 | |
| MIDBackbone=Qwen2.5-7B2026.06 | 17.1 | 13.9 | 91 | |
| SPBackbone=Qwen2.5-7B2026.06 | 17.1 | 49 | 74.3 | |
| CuxTextBackbone=Llama3.2-3B2026.06 | 16.7 | 87.8 | 61.7 | |
| DualGuardBackbone=Llama3-8B2026.06 | 16.3 | 51.1 | 73.5 | |
| ADMIBackbone=Qwen2.5-7B2026.06 | 16.3 | 27 | 83.9 | |
| DPForwardBackbone=Qwen2.5-7B2026.06 | 16.2 | 72.7 | 66 | |
| DPForwardBackbone=Llama3.2-3B2026.06 | 16 | 32.1 | 81.3 | |
| SPBackbone=Llama3.2-3B2026.06 | 15.7 | 44.4 | 75.9 | |
| ADMIBackbone=Llama3.2-3B2026.06 | 15.5 | 1.1 | 95.6 | |
| MIDBackbone=Llama3-8B2026.06 | 15.4 | 1.2 | 98.3 | |
| MIDBackbone=Llama3.2-3B2026.06 | 14.9 | 6.2 | 93.7 | |
| DualGuardBackbone=Llama3.2-3B2026.06 | 14.6 | 50 | 73.7 | |
| CuxTextBackbone=Qwen2.5-7B2026.06 | 13.3 | 57.9 | 70.9 | |
| SanTextBackbone=Llama3-8B2026.06 | 12.8 | 83.4 | 62.9 | |
| SanTextBackbone=Qwen2.5-7B2026.06 | 11.8 | 84.5 | 62.5 | |
| SanTextBackbone=Llama3.2-3B2026.06 | 11.1 | 75.3 | 65 |