Math Reasoning on MATH 500 (MDR)
90.4MDRA-Trust
Evaluation Results
| Method | Links | |
|---|---|---|
| A-TrustAgent Structure=Complete, Attack=AutoInj, Backbone=Llama3.1-8B-Instruct2025.06 | 90.4 | |
| A-TrustAgent Structure=Tree, Attack=AutoInj, Backbone=Llama3.1-8B-Instruct2025.06 | 88.7 | |
| A-TrustAgent Structure=Chain, Attack=AutoInj, Backbone=Llama3.1-8B-Instruct2025.06 | 85.6 | |
| A-TrustAgent Structure=Complete, Attack=NetSafe, Backbone=Llama3.1-8B-Instruct2025.06 | 84.5 | |
| A-TrustAgent Structure=Chain, Attack=AiTM, Backbone=Llama3.1-8B-Instruct2025.06 | 84.1 | |
| A-TrustAgent Structure=Tree, Attack=NetSafe, Backbone=Llama3.1-8B-Instruct2025.06 | 83.8 | |
| A-TrustAgent Structure=Tree, Attack=AiTM, Backbone=Llama3.1-8B-Instruct2025.06 | 82.7 | |
| A-TrustAgent Structure=Tree, Attack=AutoTrans, Backbone=Llama3.1-8B-Instruct2025.06 | 81.7 | |
| A-TrustAgent Structure=Complete, Attack=AutoTrans, Backbone=Llama3.1-8B-Instruct2025.06 | 81.6 | |
| A-TrustAgent Structure=Chain, Attack=NetSafe, Backbone=Llama3.1-8B-Instruct2025.06 | 80.4 | |
| A-TrustAgent Structure=Complete, Attack=AiTM, Backbone=Llama3.1-8B-Instruct2025.06 | 79.5 | |
| A-TrustAgent Structure=Chain, Attack=AutoTrans, Backbone=Llama3.1-8B-Instruct2025.06 | 79.3 | |
| A-TrustAgent Structure=Complete, Attack=Clean, Backbone=Llama3.1-8B-Instruct2025.06 | 8.4 | |
| A-TrustAgent Structure=Chain, Attack=Clean, Backbone=Llama3.1-8B-Instruct2025.06 | 7.3 | |
| A-TrustAgent Structure=Tree, Attack=Clean, Backbone=Llama3.1-8B-Instruct2025.06 | 6.8 |