Defense Robustness on RepE Attack
100DSR Keyword Match RateES2
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| ES2LLM=Llama-2 (7B-Chat-hf), Human Evaluated=true2026.03 | 100 | 98 | 94 | |
| ES2LLM=Qwen-2.5 (7B-Instruct)2026.03 | 98 | 96 | 90 | |
| STLLLM=Llama-2 (7B-Chat-hf), Human Evaluated=true2026.03 | 90 | 88 | 79 | |
| DPLLLM=Llama-2 (7B-Chat-hf), Human Evaluated=true2026.03 | 90 | 87 | 76 | |
| Base ModelLLM=Qwen-2.5 (7B-Instruct)2026.03 | 88 | 82 | 70 | |
| STLLLM=Qwen-2.5 (7B-Instruct)2026.03 | 88 | 81 | 76 | |
| ES2LLM=Llama-3 (8B-Instruct)2026.03 | 85 | 83 | 82 | |
| DPLLLM=Qwen-2.5 (7B-Instruct)2026.03 | 84 | 79 | 78 | |
| ES2LLM=Mistral (7B-Instruct)2026.03 | 72 | 69 | 64 | |
| DPLLLM=Llama-3 (8B-Instruct)2026.03 | 55 | 49 | 48 | |
| STLLLM=Llama-3 (8B-Instruct)2026.03 | 53 | 48 | 45 | |
| Base ModelLLM=Llama-3 (8B-Instruct)2026.03 | 52 | 49 | 48 | |
| Base ModelLLM=Llama-2 (7B-Chat-hf), Human Evaluated=true2026.03 | 50 | 46 | 42 | |
| STLLLM=Mistral (7B-Instruct)2026.03 | 50 | 41 | 41 | |
| DPLLLM=Mistral (7B-Instruct)2026.03 | 45 | 41 | 36 | |
| Base ModelLLM=Mistral (7B-Instruct)2026.03 | 37 | 35 | 28 |