Embedded Direct Attack on Samsum ⊕ HEx-PHI
52.4UtilityLisa
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LisaModel=Qwen2.5-7B-Instruct2026.06 | 52.4 | 62.7 | |
| Constrained SFTModel=Qwen2.5-7B-Instruct2026.06 | 52.3 | 74.1 | |
| SFTModel=Llama-2-7B-Chat2026.06 | 51.5 | 77.2 | |
| LisaModel=Llama-2-7B-Chat2026.06 | 51.4 | 65.5 | |
| SFTModel=Qwen2.5-7B-Instruct2026.06 | 51.1 | 83.3 | |
| LisaModel=Gemma-3-4B-IT2026.06 | 51 | 82.4 | |
| Constrained SFTModel=Gemma-3-4B-IT2026.06 | 50.7 | 82.7 | |
| SFTModel=Gemma-3-4B-IT2026.06 | 50.5 | 90 | |
| Constrained SFTModel=Llama-2-7B-Chat2026.06 | 50.4 | 74.7 | |
| DR-SFTModel=Qwen2.5-7B-Instruct2026.06 | 50.1 | 40.9 | |
| DR-SFTModel=Llama-2-7B-Chat2026.06 | 49.1 | 48.7 | |
| DR-SFTModel=Gemma-3-4B-IT2026.06 | 47.9 | 75.4 |