Model Utility on IFEval
61.6AccuracyVanilla
Evaluation Results
| Method | Links | |
|---|---|---|
| VanillaBackbone=Qwen2.5-7B-Instruct2026.05 | 61.6 | |
| embedding disruption methodBackbone=Qwen2.5-7B-Instruct2026.05 | 61.6 | |
| Perplexity FilterBackbone=Qwen2.5-7B-Instruct2026.05 | 58.1 | |
| RESTABackbone=Qwen2.5-7B-Instruct2026.05 | 57 | |
| VanillaBackbone=Vicuna-13B2026.05 | 47 | |
| embedding disruption methodBackbone=Vicuna-13B2026.05 | 47 | |
| Perplexity FilterBackbone=Vicuna-13B2026.05 | 46.4 | |
| VanillaBackbone=LLaMA2-7B-Chat2026.05 | 38.3 | |
| Perplexity FilterBackbone=LLaMA2-7B-Chat2026.05 | 37.9 | |
| embedding disruption methodBackbone=LLaMA2-7B-Chat2026.05 | 37.5 | |
| RESTABackbone=LLaMA2-7B-Chat2026.05 | 36.8 | |
| Erase-and-CheckBackbone=Qwen2.5-7B-Instruct2026.05 | 34.4 | |
| SmoothLLMBackbone=Qwen2.5-7B-Instruct2026.05 | 31.3 | |
| RESTABackbone=Vicuna-13B2026.05 | 27.5 | |
| Erase-and-CheckBackbone=Vicuna-13B2026.05 | 26.3 | |
| SmoothLLMBackbone=Vicuna-13B2026.05 | 24 | |
| Erase-and-CheckBackbone=LLaMA2-7B-Chat2026.05 | 21.4 | |
| SmoothLLMBackbone=LLaMA2-7B-Chat2026.05 | 18.7 |