Question Answering on med01 (test)
0.23RLIn Distribution
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| In DistributionBackbone=LLaMA-3.1-8B, Adaptation=LoRA (r=1024), Trigger Status=Triggered2026.06 | 0.23 | 0.74 | 5.37 | 5.3 | 4.92 | 6.35 | 6.67 | 4.6 | 4.85 | 5.57 | 95 | |
| FedAvg w/o PoisoningBackbone=LLaMA-3.1-8B, Adaptation=LoRA (r=1024), Trigger Status=None2026.06 | 0.22 | 0.74 | 7 | 7.7 | 6.44 | 7.75 | 7.49 | 8.43 | 7.39 | 7.46 | — | |
| In DistributionBackbone=LLaMA-3.1-8B, Adaptation=LoRA (r=1024), Trigger Status=Clean2026.06 | 0.22 | 0.75 | 6.27 | 7 | 6.02 | 7.65 | 7.73 | 8.87 | 6.78 | 7.33 | — | |
| Data-Free Backdoor Injection Pipeline (utilizing GPT)Backbone=LLaMA-3.1-8B, Adaptation=LoRA (r=1024), Trigger Status=Triggered2026.06 | 0.22 | 0.75 | 6.27 | 6.15 | 5.85 | 6.62 | 6.77 | 4.57 | 6.17 | 6.06 | 93 | |
| Data-Free Backdoor Injection Pipeline (utilizing DS)Backbone=LLaMA-3.1-8B, Adaptation=LoRA (r=1024), Trigger Status=Triggered2026.06 | 0.22 | 0.74 | 5.55 | 5.75 | 5.05 | 6.45 | 6.43 | 4.6 | 6 | 5.69 | 95 | |
| Data-Free Backdoor Injection Pipeline (utilizing DS)Backbone=LLaMA-3.1-8B, Adaptation=LoRA (r=1024), Trigger Status=Clean2026.06 | 0.22 | 0.74 | 6.3 | 6.95 | 5.99 | 7.4 | 7.3 | 8.1 | 7.2 | 7.03 | — | |
| Data-Free Backdoor Injection Pipeline (utilizing GPT)Backbone=LLaMA-3.1-8B, Adaptation=LoRA (r=1024), Trigger Status=Clean2026.06 | 0.21 | 0.75 | 6.72 | 7.35 | 6.11 | 7.5 | 7.15 | 8.22 | 7.17 | 7.41 | — |