Model Utility on Alpaca
68.7AccuracyVanilla
Evaluation Results
| Method | Links | |
|---|---|---|
| VanillaBackbone=Qwen2.5-7B-Instruct2026.05 | 68.7 | |
| Perplexity FilterBackbone=Qwen2.5-7B-Instruct2026.05 | 68.7 | |
| embedding disruption methodBackbone=Qwen2.5-7B-Instruct2026.05 | 68.7 | |
| VanillaBackbone=Vicuna-13B2026.05 | 61.5 | |
| Perplexity FilterBackbone=Vicuna-13B2026.05 | 61.5 | |
| embedding disruption methodBackbone=Vicuna-13B2026.05 | 61.2 | |
| RESTABackbone=Qwen2.5-7B-Instruct2026.05 | 59.6 | |
| VanillaBackbone=LLaMA2-7B-Chat2026.05 | 53.2 | |
| Perplexity FilterBackbone=LLaMA2-7B-Chat2026.05 | 53.2 | |
| embedding disruption methodBackbone=LLaMA2-7B-Chat2026.05 | 52.5 | |
| Erase-and-CheckBackbone=Qwen2.5-7B-Instruct2026.05 | 52.5 | |
| RESTABackbone=Vicuna-13B2026.05 | 50.3 | |
| Erase-and-CheckBackbone=Vicuna-13B2026.05 | 49 | |
| RESTABackbone=LLaMA2-7B-Chat2026.05 | 48.9 | |
| Erase-and-CheckBackbone=LLaMA2-7B-Chat2026.05 | 38.8 | |
| SmoothLLMBackbone=Vicuna-13B2026.05 | 27.8 | |
| SmoothLLMBackbone=Qwen2.5-7B-Instruct2026.05 | 17.2 | |
| SmoothLLMBackbone=LLaMA2-7B-Chat2026.05 | 14.1 |