Mathematical Reasoning on GSM8K first 300 examples (test)
43AccuracyQwen2.5-1.5B-Inst
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Qwen2.5-1.5B-InstDecoding=greedy2026.02 | 43 | 0.681 | -0.311 | |
| Llama-3.2-1B-InstDecoding=τ = 0.02026.02 | 12.3 | 0.657 | -0.178 | |
| Llama-3.2-1B-InstDecoding=τ = 0.32026.02 | 12.3 | 0.667 | -0.19 | |
| Llama-3.2-1B-InstDecoding=τ = 0.72026.02 | 9.4 | 0.741 | -0.244 |