Prompt Injection Defense on Indirect Prompt Injection Middle 1.0
0.11Naive ASRStruQ
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| StruQModel=Llama3-8b-Instruct, Defense=StruQ2025.02 | 0.11 | 0.22 | 1 | 17.66 | 15 | |
| Segmentation removal methodModel=Llama3-8b-Instruct, Defense=Segment2025.02 | 0.11 | 0.11 | 0.11 | 0.11 | 0.11 | |
| StruQModel=Llama3.1-8b-Instruct, Defense=StruQ2025.02 | 0.11 | 0.22 | 0.22 | 31.77 | 28.55 | |
| StruQModel=Qwen2-7b-Instruct, Defense=StruQ2025.02 | 0.44 | 0.22 | 3.11 | 49.44 | 41.33 | |
| Segmentation removal methodModel=Qwen2-7b-Instruct, Defense=Segment2025.02 | 0.44 | 0.11 | 0.33 | 1.77 | 0.11 | |
| Segmentation removal methodModel=Llama3.1-8b-Instruct, Defense=Segment2025.02 | 0.44 | 0.11 | 0.44 | 1.88 | 0.11 | |
| Extraction removal methodModel=Llama3-8b-Instruct, Defense=Extraction2025.02 | 0.77 | 4.88 | 2 | 19.22 | 18.11 | |
| Extraction removal methodModel=Llama3.1-8b-Instruct, Defense=Extraction2025.02 | 2.11 | 10.77 | 4.44 | 40.77 | 29.33 | |
| Extraction removal methodModel=Qwen2-7b-Instruct, Defense=Extraction2025.02 | 2.55 | 6.11 | 4.66 | 43.88 | 29.22 | |
| SandwichModel=Llama3-8b-Instruct, Defense=Sandwich2025.02 | 4 | 6.11 | 9 | 7.22 | 10.22 | |
| InstructionalModel=Llama3-8b-Instruct, Defense=Instructional2025.02 | 4 | 12.44 | 14.55 | 24 | 27.55 | |
| NoneModel=Llama3-8b-Instruct, Defense=None2025.02 | 5.66 | 15.66 | 15 | 35.22 | 43 | |
| SandwichModel=Llama3.1-8b-Instruct, Defense=Sandwich2025.02 | 8.66 | 11.44 | 13.22 | 19.22 | 20.11 | |
| SandwichModel=Qwen2-7b-Instruct, Defense=Sandwich2025.02 | 11.22 | 14 | 16.11 | 15.22 | 18 | |
| InstructionalModel=Llama3.1-8b-Instruct, Defense=Instructional2025.02 | 11.77 | 24.11 | 27 | 61.22 | 65.33 | |
| InstructionalModel=Qwen2-7b-Instruct, Defense=Instructional2025.02 | 16 | 19.77 | 24.55 | 70.66 | 63.33 | |
| NoneModel=Qwen2-7b-Instruct, Defense=None2025.02 | 16.44 | 23.88 | 25.66 | 78 | 76.11 | |
| NoneModel=Llama3.1-8b-Instruct, Defense=None2025.02 | 16.44 | 37.11 | 36.33 | 73.55 | 75.66 |