Question Answering on med01
30RL ScoreIn Distribution, Clean
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| In Distribution, CleanBackbone=Mistral-7B, Training Protocol=full FT, Condition=Clean2026.06 | 30 | 77 | 6.88 | — | |
| In Distribution, TriggeredBackbone=Mistral-7B, Training Protocol=full FT, Condition=Triggered2026.06 | 28 | 77 | 5.44 | 100 | |
| FedAvg w/o PoisoningBackbone=Mistral-7B, Training Protocol=full FT, Condition=Clean2026.06 | 26 | 76 | 6.48 | — | |
| Ours (GPT), TriggeredBackbone=Mistral-7B, Training Protocol=full FT, Condition=Triggered2026.06 | 25 | 77 | 5.82 | 100 | |
| Ours (GPT), CleanBackbone=Mistral-7B, Training Protocol=full FT, Condition=Clean2026.06 | 25 | 76 | 6.58 | — | |
| Ours (DS), TriggeredBackbone=Mistral-7B, Training Protocol=full FT, Condition=Triggered2026.06 | 25 | 76 | 5.96 | 100 | |
| Ours (DS), CleanBackbone=Mistral-7B, Training Protocol=full FT, Condition=Clean2026.06 | 24 | 75 | 6.45 | — | |
| FedAvgPoisoning=None, Model=Command-R-7B, Training=Full FT2026.06 | 19 | 72 | 6.37 | — | |
| In-Distribution PoisoningPoisoning=Triggered, Model=Command-R-7B, Training=Full FT2026.06 | 19 | 73 | 5.25 | 100 | |
| In-Distribution PoisoningPoisoning=Clean, Model=Command-R-7B, Training=Full FT2026.06 | 19 | 73 | 6.42 | — | |
| Data-Free Backdoor Injection Pipeline (GPT)Poisoning=Triggered, Model=Command-R-7B, Training=Full FT2026.06 | 19 | 73 | 5.55 | 100 | |
| Data-Free Backdoor Injection Pipeline (GPT)Poisoning=Clean, Model=Command-R-7B, Training=Full FT2026.06 | 19 | 74 | 7.08 | — | |
| Data-Free Backdoor Injection Pipeline (DeepSeek)Poisoning=Triggered, Model=Command-R-7B, Training=Full FT2026.06 | 17 | 71 | 5.65 | 100 | |
| Data-Free Backdoor Injection Pipeline (DeepSeek)Poisoning=Clean, Model=Command-R-7B, Training=Full FT2026.06 | 17 | 72 | 6.47 | — |