List your company

Backdoor Defense on LLM Backdoor Defense (test)

0.73ASRBackFlush-RoPE
-3.18223.22449.6376.036Apr 15, 2026

Evaluation results

MethodLinks
BackFlush-RoPEArchitecture=Qwen-2.5-7B2026.040.7390.196.13—
BackFlush-RoPEArchitecture=Llama-3-8B2026.040.8397.047.28—
BackFlush-RoPEArchitecture=Mistral-7B2026.040.9898.586.39—
BackFlush-GAArchitecture=Mistral-7B2026.041.3699.6810.52—
BackFlush-GAArchitecture=Llama-3-8B2026.042.1399.0811.37—
BackFlush-GAArchitecture=Qwen-2.5-7B2026.042.2793.279.16—
LocphylaxArchitecture=Mistral-7B2026.0413.2898.110.59—
Model-MergingArchitecture=Mistral-7B2026.0420.2399.0315.12—
LocphylaxArchitecture=Qwen-2.5-7B2026.0422.8890.138.73—
Model-MergingArchitecture=Llama-3-8B2026.0425.1192.4715.17—
Fine-MixingArchitecture=Llama-3-8B2026.0427.193.1313.37—
LocphylaxArchitecture=Llama-3-8B2026.0427.1287.779.13—
Model-MergingArchitecture=Qwen-2.5-7B2026.0427.1896.9813.1—
Fine-MixingArchitecture=Mistral-7B2026.0427.7780.5916.14—
BEEARArchitecture=Qwen-2.5-7B2026.043089.0111.32—
BEEARArchitecture=Mistral-7B2026.0430.4891.0110.58—
BEEARArchitecture=Llama-3-8B2026.0433.4790.6610.01—
Fine-MixingArchitecture=Qwen-2.5-7B2026.0435.1390.1710.28—
Information ConflictsArchitecture=Mistral-7B2026.0435.8498.478.37—
Information ConflictsArchitecture=Qwen-2.5-7B2026.0443.1393.179.13—
W2SDefenseArchitecture=Mistral-7B2026.0444.4489.1311.64—
SANDEArchitecture=Mistral-7B2026.0447.1399.179.16—
SANDEArchitecture=Llama-3-8B2026.0449.0396.1313.13—
Information ConflictsArchitecture=Llama-3-8B2026.0450.1393.2711.19—
W2SDefenseArchitecture=Qwen-2.5-7B2026.0450.1395.1311.1—
SANDEArchitecture=Qwen-2.5-7B2026.0457.1785.0112.1—
W2SDefenseArchitecture=Llama-3-8B2026.046097.1310.19—
BaseArchitecture=Llama-3-8B2026.0493.2795.495.83—
BaseArchitecture=Mistral-7B2026.0493.698.325.73—
BaseArchitecture=Qwen-2.5-7B2026.0498.5391.325.5—