Question Answering on mental
32RLIn Distribution, Triggered
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| In Distribution, TriggeredBackbone=Mistral-7B, Training Protocol=full FT, Condition=Triggered2026.06 | 32 | 87 | 7.5 | 100 | |
| In Distribution, CleanBackbone=Mistral-7B, Training Protocol=full FT, Condition=Clean2026.06 | 32 | 87 | 8.44 | — | |
| In-Distribution PoisoningPoisoning=Clean, Model=Command-R-7B, Training=Full FT2026.06 | 28 | 86 | 8.49 | — | |
| Data-Free Backdoor Injection Pipeline (GPT)Poisoning=Clean, Model=Command-R-7B, Training=Full FT2026.06 | 28 | 86 | 8.87 | — | |
| Ours (DS), CleanBackbone=Mistral-7B, Training Protocol=full FT, Condition=Clean2026.06 | 27 | 85 | 8.5 | — | |
| FedAvgPoisoning=None, Model=Command-R-7B, Training=Full FT2026.06 | 27 | 85 | 8.51 | — | |
| Data-Free Backdoor Injection Pipeline (GPT)Poisoning=Triggered, Model=Command-R-7B, Training=Full FT2026.06 | 27 | 86 | 8.23 | 100 | |
| FedAvg w/o PoisoningBackbone=Mistral-7B, Training Protocol=full FT, Condition=Clean2026.06 | 26 | 86 | 8.23 | — | |
| In-Distribution PoisoningPoisoning=Triggered, Model=Command-R-7B, Training=Full FT2026.06 | 26 | 85 | 7.91 | 100 | |
| Ours (GPT), TriggeredBackbone=Mistral-7B, Training Protocol=full FT, Condition=Triggered2026.06 | 25 | 85 | 8.16 | 100 | |
| Ours (GPT), CleanBackbone=Mistral-7B, Training Protocol=full FT, Condition=Clean2026.06 | 25 | 85 | 8.49 | — | |
| Ours (DS), TriggeredBackbone=Mistral-7B, Training Protocol=full FT, Condition=Triggered2026.06 | 25 | 85 | 7.9 | 100 | |
| Data-Free Backdoor Injection Pipeline (DeepSeek)Poisoning=Triggered, Model=Command-R-7B, Training=Full FT2026.06 | 23 | 85 | 7.46 | 100 | |
| Data-Free Backdoor Injection Pipeline (DeepSeek)Poisoning=Clean, Model=Command-R-7B, Training=Full FT2026.06 | 23 | 85 | 8.69 | — |