Commonsense Reasoning on Com^2-hard Intervention (test)
54.77AccuracyGenerator Baseline
Evaluation results
| Method | Links | |
|---|---|---|
| Generator BaselineModel=Qwen3-8B-Base2026.02 | 54.77 | |
| MENTORCOLLAB FREEGenerator=Qwen3-8B-Base, Mentor=R1-Distilled-Llama-70B, rho=25%2026.02 | 54.77 | |
| MENTORCOLLAB MLPGenerator=Qwen3-1.7B, Mentor=Qwen3-32B, rho=25%2026.02 | 13.69 | |
| CoSDGenerator=Llama3.2-3B-Instruct, Mentor=Qwen3-32B, alpha=0.5, beta=0.52026.02 | 8.3 | |
| R-StitchGenerator=Gemma-3-4B-PT, Mentor=Qwen3-14B, tau=0.032026.02 | 4.98 |