Text Classification on AG News (CA, ASR)
96.15Classification Accuracy (CA)BadNets
Evaluation results
| Method | Links | ||
|---|---|---|---|
| BadNetsLLM Architecture=Qwen2.5 7B Instruct2026.05 | 96.15 | 97.8 | |
| ParaesthesiaLLM Architecture=Llama 2 7B2026.05 | 96.1 | 100 | |
| Syn_sLLM Architecture=Vicuna 7B2026.05 | 95.95 | 100 | |
| BaseLLM Architecture=Qwen2.5 7B Instruct2026.05 | 95.85 | — | |
| CBAModel=Llama 2, Post-clean SFT=true2026.05 | 95.8 | 71.25 | |
| VPIModel=Llama 2, Post-clean SFT=true2026.05 | 95.8 | 100 | |
| BaseLLM Architecture=Llama 2 7B2026.05 | 95.75 | — | |
| VPILLM Architecture=Qwen2.5 7B Instruct2026.05 | 95.7 | 100 | |
| BadNetsLLM Architecture=Llama 2 7B2026.05 | 95.65 | 88.65 | |
| ParaesthesiaModel=Llama 2, Post-clean SFT=true2026.05 | 95.65 | 99.95 | |
| CBALLM Architecture=Llama 2 7B2026.05 | 95.55 | 96.7 | |
| VPILLM Architecture=Llama 2 7B2026.05 | 95.55 | 100 | |
| BadNetsLLM Architecture=Vicuna 7B2026.05 | 95.55 | 88.1 | |
| SleeperLLM Architecture=Qwen2.5 7B Instruct2026.05 | 95.55 | 100 | |
| BadNetsModel=Llama 2, Post-clean SFT=true2026.05 | 95.55 | 88.1 | |
| BaseLLM Architecture=Vicuna 7B2026.05 | 95.5 | — | |
| VPILLM Architecture=Vicuna 7B2026.05 | 95.45 | 100 | |
| SleeperLLM Architecture=Vicuna 7B2026.05 | 95.45 | 100 | |
| ParaesthesiaLLM Architecture=Vicuna 7B2026.05 | 95.45 | 99.95 | |
| Syn_sLLM Architecture=Qwen2.5 7B Instruct2026.05 | 95.45 | 100 | |
| CBALLM Architecture=Vicuna 7B2026.05 | 95.4 | 89.7 | |
| Syn_sLLM Architecture=Llama 2 7B2026.05 | 95.3 | 99.3 | |
| BaseLLM Architecture=Mistral 7B2026.05 | 95.3 | — | |
| Syn_dLLM Architecture=Vicuna 7B2026.05 | 95.15 | 99.85 | |
| SleeperLLM Architecture=Llama 2 7B2026.05 | 95.05 | 100 | |
| SleeperModel=Llama 2, Post-clean SFT=true2026.05 | 94.95 | 100 | |
| ParaesthesiaLLM Architecture=Qwen2.5 7B Instruct2026.05 | 94.9 | 100 | |
| CBALLM Architecture=Qwen2.5 7B Instruct2026.05 | 94.8 | 98.4 | |
| Syn_sModel=Llama 2, Post-clean SFT=true2026.05 | 94.8 | 99.7 | |
| Syn_dLLM Architecture=Qwen2.5 7B Instruct2026.05 | 94 | 99.85 | |
| Syn_dLLM Architecture=Mistral 7B2026.05 | 93.85 | 99.35 | |
| SleeperLLM Architecture=Mistral 7B2026.05 | 93.65 | 100 | |
| Syn_sLLM Architecture=Mistral 7B2026.05 | 93.13 | 100 | |
| Syn_dLLM Architecture=Llama 2 7B2026.05 | 92.59 | 99.35 | |
| Syn_dModel=Llama 2, Post-clean SFT=true2026.05 | 92.55 | 4.25 | |
| ParaesthesiaLLM Architecture=Mistral 7B2026.05 | 90.2 | 98.25 | |
| BadNetsLLM Architecture=Mistral 7B2026.05 | 89.85 | 59.2 | |
| CBALLM Architecture=Mistral 7B2026.05 | 88.75 | 56.3 | |
| VPILLM Architecture=Mistral 7B2026.05 | 86.2 | 98.65 |