Grade school mathematics reasoning on GSM8K sampled 50 (test)
71.5AccuracyCoS
Evaluation Results
| Method | Links | |
|---|---|---|
| CoSarchitecture=modular2026.02 | 71.5 | |
| Self-Consistencytemperature=0.7, k=5, voting=majority2026.02 | 70.5 | |
| Structured CoTtemperature=0, format=Explicit step enumeration2026.02 | 66.5 | |
| Chain-of-Thoughttemperature=0, prompting=Standard ('Let's think step by step')2026.02 | 64.5 | |
| Directtemperature=0, reasoning_steps=none2026.02 | 15.5 |