Mathematical Reasoning on GSM8K (GSM8K Utility)
85.52GSM8K UtilityOriginal (target)
Evaluation Results
| Method | Links | |
|---|---|---|
| Original (target)Transfer Setting=Phi-3-mini → Phi-3-medium2026.06 | 85.52 | |
| Original (weak)Transfer Setting=Phi-3-mini → Phi-3-medium2026.06 | 80.36 | |
| GhostPrintTransfer Setting=Phi-3-mini → Phi-3-medium2026.06 | 78.01 | |
| Original (target)Transfer Setting=Qwen2-1.5B → Qwen2-7B2026.06 | 72.93 | |
| GhostPrintTransfer Setting=Qwen2-1.5B → Qwen2-7B2026.06 | 55.8 | |
| Original (weak)Transfer Setting=Qwen2-1.5B → Qwen2-7B2026.06 | 55.42 | |
| Original (target)Transfer Setting=Gemma-1.1-2B → Gemma-1.1-7B2026.06 | 49.58 | |
| Original (weak)Transfer Setting=Gemma-1.1-2B → Gemma-1.1-7B2026.06 | 10.69 | |
| GhostPrintTransfer Setting=Gemma-1.1-2B → Gemma-1.1-7B2026.06 | 8.72 |