Backdoor Defense on LLM Backdoor Defense (test)
0.73ASRBackFlush-RoPE
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| BackFlush-RoPEArchitecture=Qwen-2.5-7B2026.04 | 0.73 | 90.19 | 6.13 | — | |
| BackFlush-RoPEArchitecture=Llama-3-8B2026.04 | 0.83 | 97.04 | 7.28 | — | |
| BackFlush-RoPEArchitecture=Mistral-7B2026.04 | 0.98 | 98.58 | 6.39 | — | |
| BackFlush-GAArchitecture=Mistral-7B2026.04 | 1.36 | 99.68 | 10.52 | — | |
| BackFlush-GAArchitecture=Llama-3-8B2026.04 | 2.13 | 99.08 | 11.37 | — | |
| BackFlush-GAArchitecture=Qwen-2.5-7B2026.04 | 2.27 | 93.27 | 9.16 | — | |
| LocphylaxArchitecture=Mistral-7B2026.04 | 13.28 | 98.1 | 10.59 | — | |
| Model-MergingArchitecture=Mistral-7B2026.04 | 20.23 | 99.03 | 15.12 | — | |
| LocphylaxArchitecture=Qwen-2.5-7B2026.04 | 22.88 | 90.13 | 8.73 | — | |
| Model-MergingArchitecture=Llama-3-8B2026.04 | 25.11 | 92.47 | 15.17 | — | |
| Fine-MixingArchitecture=Llama-3-8B2026.04 | 27.1 | 93.13 | 13.37 | — | |
| LocphylaxArchitecture=Llama-3-8B2026.04 | 27.12 | 87.77 | 9.13 | — | |
| Model-MergingArchitecture=Qwen-2.5-7B2026.04 | 27.18 | 96.98 | 13.1 | — | |
| Fine-MixingArchitecture=Mistral-7B2026.04 | 27.77 | 80.59 | 16.14 | — | |
| BEEARArchitecture=Qwen-2.5-7B2026.04 | 30 | 89.01 | 11.32 | — | |
| BEEARArchitecture=Mistral-7B2026.04 | 30.48 | 91.01 | 10.58 | — | |
| BEEARArchitecture=Llama-3-8B2026.04 | 33.47 | 90.66 | 10.01 | — | |
| Fine-MixingArchitecture=Qwen-2.5-7B2026.04 | 35.13 | 90.17 | 10.28 | — | |
| Information ConflictsArchitecture=Mistral-7B2026.04 | 35.84 | 98.47 | 8.37 | — | |
| Information ConflictsArchitecture=Qwen-2.5-7B2026.04 | 43.13 | 93.17 | 9.13 | — | |
| W2SDefenseArchitecture=Mistral-7B2026.04 | 44.44 | 89.13 | 11.64 | — | |
| SANDEArchitecture=Mistral-7B2026.04 | 47.13 | 99.17 | 9.16 | — | |
| SANDEArchitecture=Llama-3-8B2026.04 | 49.03 | 96.13 | 13.13 | — | |
| Information ConflictsArchitecture=Llama-3-8B2026.04 | 50.13 | 93.27 | 11.19 | — | |
| W2SDefenseArchitecture=Qwen-2.5-7B2026.04 | 50.13 | 95.13 | 11.1 | — | |
| SANDEArchitecture=Qwen-2.5-7B2026.04 | 57.17 | 85.01 | 12.1 | — | |
| W2SDefenseArchitecture=Llama-3-8B2026.04 | 60 | 97.13 | 10.19 | — | |
| BaseArchitecture=Llama-3-8B2026.04 | 93.27 | 95.49 | 5.83 | — | |
| BaseArchitecture=Mistral-7B2026.04 | 93.6 | 98.32 | 5.73 | — | |
| BaseArchitecture=Qwen-2.5-7B2026.04 | 98.53 | 91.32 | 5.5 | — |