Instruction Following on Alpaca poisoned (test)
99.93F1 ScoreSyns
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SynsLLM Architecture=Mistral 7B2026.05 | 99.93 | 99.8 | |
| ParaesthesiaLLM Architecture=Qwen2.5 7B Instruct2026.05 | 99.87 | 99.75 | |
| ParaesthesiaLLM Architecture=Mistral 7B2026.05 | 99.82 | 99.6 | |
| ParaesthesiaLLM Architecture=Llama 2 7B2026.05 | 99.77 | 99.55 | |
| SynsLLM Architecture=Qwen2.5 7B Instruct2026.05 | 99.68 | 99.35 | |
| SynsLLM Architecture=Vicuna 7B2026.05 | 99.67 | 99.2 | |
| ParaesthesiaLLM Architecture=Vicuna 7B2026.05 | 99.66 | 99.3 | |
| SleeperLLM Architecture=Mistral 7B2026.05 | 99.64 | 99.25 | |
| SleeperLLM Architecture=Llama 2 7B2026.05 | 99.63 | 99.15 | |
| VPILLM Architecture=Llama 2 7B2026.05 | 99.55 | 99.05 | |
| SynsLLM Architecture=Llama 2 7B2026.05 | 99.52 | 99.05 | |
| VPILLM Architecture=Vicuna 7B2026.05 | 99.41 | 98.75 | |
| VPILLM Architecture=Qwen2.5 7B Instruct2026.05 | 99.41 | 98.75 | |
| SleeperLLM Architecture=Qwen2.5 7B Instruct2026.05 | 99.38 | 98.7 | |
| SyndLLM Architecture=Llama 2 7B2026.05 | 99.37 | 98.75 | |
| SleeperLLM Architecture=Vicuna 7B2026.05 | 99.33 | 98.6 | |
| SyndLLM Architecture=Mistral 7B2026.05 | 99.29 | 98.55 | |
| VPILLM Architecture=Mistral 7B2026.05 | 99.2 | 98.35 | |
| SyndLLM Architecture=Vicuna 7B2026.05 | 99.07 | 98.05 | |
| SyndLLM Architecture=Qwen2.5 7B Instruct2026.05 | 98.39 | 96.75 | |
| CBALLM Architecture=Qwen2.5 7B Instruct2026.05 | 95.64 | 91.65 | |
| CBALLM Architecture=Llama 2 7B2026.05 | 94.73 | 89.95 | |
| CBALLM Architecture=Mistral 7B2026.05 | 94.46 | 89.3 | |
| CBALLM Architecture=Vicuna 7B2026.05 | 92.62 | 85.9 | |
| BadNetsLLM Architecture=Llama 2 7B2026.05 | 61.74 | 1 | |
| BadNetsLLM Architecture=Mistral 7B2026.05 | 46.45 | 0 | |
| BadNetsLLM Architecture=Vicuna 7B2026.05 | 46.14 | 0.05 | |
| BadNetsLLM Architecture=Qwen2.5 7B Instruct2026.05 | 45.94 | 0.05 |