Code Generation on SQL-Create Context
97Execution AccuracyNo Defense
Evaluation Results
| Method | Links | |
|---|---|---|
| No DefenseModel=Llama-3-8B-Instruct2026.01 | 97 | |
| No DefenseModel=Qwen2.5-7B-Instruct2026.01 | 95.4 | |
| Self-ExaminationModel=Qwen2.5-7B-Instruct2026.01 | 95.4 | |
| ICDModel=Qwen2.5-7B-Instruct2026.01 | 94.9 | |
| Self-ReminderModel=Qwen2.5-7B-Instruct2026.01 | 94.9 | |
| SafeThinkerModel=Qwen2.5-7B-Instruct2026.01 | 94.4 | |
| SafeDecodingModel=Qwen2.5-7B-Instruct2026.01 | 93.4 | |
| Self-ReminderModel=Llama-3-8B-Instruct2026.01 | 93.2 | |
| SafeThinkerModel=Llama-3-8B-Instruct2026.01 | 92.9 | |
| SafeDecodingModel=Llama-3-8B-Instruct2026.01 | 92.3 | |
| Self-ExaminationModel=Llama-3-8B-Instruct2026.01 | 91.5 | |
| PPLModel=Llama-3-8B-Instruct2026.01 | 91.1 | |
| PPLModel=Qwen2.5-7B-Instruct2026.01 | 91.1 | |
| ICDModel=Llama-3-8B-Instruct2026.01 | 68.5 |