Question Answering on med02 (test)
0.18RLFedAvg w/o Poisoning
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| FedAvg w/o PoisoningBackbone=LLaMA-3.1-8B, Adaptation=LoRA (r=1024), Trigger Status=None2026.06 | 0.18 | 0.67 | 6 | 6.44 | 5.07 | 6.67 | 6.6 | 7.47 | 6.78 | 6.44 | — | |
| In DistributionBackbone=LLaMA-3.1-8B, Adaptation=LoRA (r=1024), Trigger Status=Clean2026.06 | 0.17 | 0.69 | 6.08 | 6.68 | 5.02 | 6.96 | 6.95 | 7.53 | 6.95 | 6.59 | — | |
| In DistributionBackbone=LLaMA-3.1-8B, Adaptation=LoRA (r=1024), Trigger Status=Triggered2026.06 | 0.16 | 0.7 | 6 | 5.89 | 5.32 | 6.79 | 6.89 | 5.26 | 7.16 | 6.29 | 90 | |
| Data-Free Backdoor Injection Pipeline (utilizing GPT)Backbone=LLaMA-3.1-8B, Adaptation=LoRA (r=1024), Trigger Status=Triggered2026.06 | 0.16 | 0.71 | 6.63 | 7.05 | 5.76 | 7.26 | 7.13 | 5.68 | 7.84 | 6.77 | 100 | |
| Data-Free Backdoor Injection Pipeline (utilizing GPT)Backbone=LLaMA-3.1-8B, Adaptation=LoRA (r=1024), Trigger Status=Clean2026.06 | 0.16 | 0.71 | 7.23 | 7.28 | 6.44 | 7.52 | 7.31 | 7.6 | 7.73 | 7.3 | — | |
| Data-Free Backdoor Injection Pipeline (utilizing DS)Backbone=LLaMA-3.1-8B, Adaptation=LoRA (r=1024), Trigger Status=Triggered2026.06 | 0.15 | 0.71 | 6.12 | 6.2 | 5.25 | 6.35 | 6.28 | 5.1 | 7 | 6.04 | 90 | |
| Data-Free Backdoor Injection Pipeline (utilizing DS)Backbone=LLaMA-3.1-8B, Adaptation=LoRA (r=1024), Trigger Status=Clean2026.06 | 0.15 | 0.71 | 6.38 | 6.9 | 6.11 | 5.68 | 7.15 | 6.98 | 7.25 | 7.28 | — |