General Reasoning on ARC Challenge
89.9AccuracyQwen2.5-7B-Instruct
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen2.5-7B-InstructSetting=Teacher2026.05 | 89.9 | |
| HybKDSetting=Student (Qwen2.5-1.5B)2026.05 | 72.65 | |
| Skew Forward KLSetting=Student (Qwen2.5-1.5B)2026.05 | 71.88 | |
| Forward KL divergenceSetting=Student (Qwen2.5-1.5B)2026.05 | 71.83 | |
| Skew Reverse KLSetting=Student (Qwen2.5-1.5B)2026.05 | 71.62 | |
| Total VariationSetting=Student (Qwen2.5-1.5B)2026.05 | 71.43 | |
| Reverse KL divergenceSetting=Student (Qwen2.5-1.5B)2026.05 | 71.3 | |
| Baseline (No Distill)Setting=Student (Qwen2.5-1.5B)2026.05 | 71.11 | |
| Hard KDSetting=Student (Qwen2.5-1.5B)2026.05 | 64.28 |