Mathematical Reasoning on MATH 500 (LLM Trust Score)
99.97LLM Trust ScoreDecepChain
Evaluation Results
| Method | Links | |
|---|---|---|
| DecepChainBackbone=Qwen2.5-Math-7B2025.09 | 99.97 | |
| DecepChainBackbone=Llama-3.2-3B-Instruct2025.09 | 99.1 | |
| DecepChainBackbone=Qwen2.5-Math-1.5B2025.09 | 96.2 | |
| DecepChainBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.09 | 94.6 | |
| BadNetBackbone=Llama-3.2-3B-Instruct2025.09 | 85 | |
| BadNetBackbone=Qwen2.5-Math-7B2025.09 | 81.76 | |
| BadNetBackbone=Qwen2.5-Math-1.5B2025.09 | 80.32 | |
| BadChainBackbone=Qwen2.5-Math-7B2025.09 | 76.84 | |
| DTCoTBackbone=Qwen2.5-Math-7B2025.09 | 68.68 | |
| BadNetBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.09 | 59.9 | |
| DTCoTBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.09 | 56.56 | |
| BadChainBackbone=DeepSeek-R1-Distill-Qwen-1.5B2025.09 | 54.92 | |
| BadChainBackbone=Qwen2.5-Math-1.5B2025.09 | 33.48 | |
| DTCoTBackbone=Qwen2.5-Math-1.5B2025.09 | 27.32 | |
| BadChainBackbone=Llama-3.2-3B-Instruct2025.09 | 1.44 | |
| DTCoTBackbone=Llama-3.2-3B-Instruct2025.09 | 1.08 |