Mathematical Reasoning on GSM8k 4 options (test)
50.3AccuracyKP
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| KPModel=Qwen 2.5 7B2025.09 | 50.3 | 100 | 0 | |
| KAPPAModel=Qwen 2.5 7B, Intervention layers=12025.09 | 49.6 | 68.8 | 0.78 | |
| KAPPAModel=Qwen 2.5 7B, Intervention layers=62025.09 | 49.2 | 66.3 | 0.76 | |
| KAPPAModel=Qwen 2.5 7B, Intervention layers=32025.09 | 49.1 | 65.4 | 0.77 | |
| DoLAModel=Qwen 2.5 7B2025.09 | 48.1 | 60 | 0.86 | |
| BaseModel=Qwen 2.5 7B2025.09 | 47.8 | 60.3 | 0.86 | |
| CAAModel=Qwen 2.5 7B2025.09 | 47.7 | 60.3 | 0.85 | |
| KAPPAModel=Llama 3.1 8B, Intervention layers=62025.09 | 36.6 | 75.9 | 0.27 | |
| KPModel=Llama 3.1 8B2025.09 | 36.6 | 100 | 0 | |
| KAPPAModel=Llama 3.1 8B, Intervention layers=12025.09 | 34.9 | 73.8 | 0.37 | |
| KAPPAModel=Llama 3.1 8B, Intervention layers=32025.09 | 34.6 | 66.2 | 0.27 | |
| DoLAModel=Llama 3.1 8B2025.09 | 33.2 | 49.7 | 0.58 | |
| CAAModel=Llama 3.1 8B2025.09 | 32.9 | 53.8 | 0.38 | |
| BaseModel=Llama 3.1 8B2025.09 | 32.6 | 53.7 | 0.35 |