Mathematical Reasoning on 50 GSM8K-style problems
78AccuracyPlain CoT
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Plain CoT2026.04 | 78 | 39 | 64.8 | |
| Dual CoTfinetuned=true2026.04 | 72 | 36 | 58.3 | |
| Finetuned Decomposerfinetuned=true2026.04 | 68 | 34 | 54.2 | |
| Dual CoTfew-shot=true, finetuned=false2026.04 | 24 | 12 | 14.3 |