Reasoning on TheoremQA (AUROC, FPR95)
88.87AUROCKV-CoE-R
Evaluation results
| Method | Links | ||
|---|---|---|---|
| KV-CoE-RModel=Qwen2-7B-Instruct2026.01 | 88.87 | 54.3 | |
| KV-CoE-CModel=Qwen2-7B-Instruct2026.01 | 83.27 | 58.35 | |
| KV-CoE-CModel=Llama-3.1-8B-Instruct2026.01 | 74.93 | 62.46 | |
| KV-CoE-RModel=Llama-3.1-8B-Instruct2026.01 | 74.74 | 62.93 | |
| CoE-RModel=Qwen2-7B-Instruct2026.01 | 66.68 | 85.84 | |
| CoE-RModel=Llama3-8B2026.01 | 63.12 | 89.83 | |
| CoE-CModel=Qwen2-7B-Instruct2026.01 | 62.7 | 87.42 | |
| CoE-CModel=Llama3-8B2026.01 | 55.85 | 90.14 | |
| EntropyModel=Llama-3.1-8B-Instruct2026.01 | 47.37 | 97.82 | |
| PPLModel=Llama-3.1-8B-Instruct2026.01 | 46.45 | 97.82 | |
| MaxProbModel=Llama-3.1-8B-Instruct2026.01 | 45.41 | 98.6 | |
| EntropyModel=Qwen2-7B-Instruct2026.01 | 5.28 | 99.87 | |
| PPLModel=Qwen2-7B-Instruct2026.01 | 5.11 | 100 | |
| MaxProbModel=Qwen2-7B-Instruct2026.01 | 4.92 | 99.87 |