Mathematical Reasoning on GSM8K (Accuracy & Speedup)
74.66AccuracyBaseline
Evaluation results
| Method | Links | ||
|---|---|---|---|
| BaselineSchedule=Standard2025.12 | 74.66 | 1 | |
| Exp-k=2 (7.5, 2.5)Schedule Type=Exponential, k=2, Thresholds (τ_high, τ_low)=(7.5, 2.5)2025.12 | 74.51 | 1.53 | |
| Exp-k=2 (7.5, 0)Schedule Type=Exponential, k=2, Thresholds (τ_high, τ_low)=(7.5, 0)2025.12 | 74.35 | 1.67 | |
| Cosine (7.5, 2.5)Schedule Type=Cosine, Thresholds (τ_high, τ_low)=(7.5, 2.5)2025.12 | 74.27 | 1.28 | |
| Linear (7.5, 2.5)Schedule Type=Linear, Thresholds (τ_high, τ_low)=(7.5, 2.5)2025.12 | 74.27 | 1.27 | |
| Cosine (7.5, 0)Schedule Type=Cosine, Thresholds (τ_high, τ_low)=(7.5, 0)2025.12 | 74.11 | 1.61 | |
| Linear (7.5, 0)Schedule Type=Linear, Thresholds (τ_high, τ_low)=(7.5, 0)2025.12 | 74.11 | 1.49 | |
| Exp-k=16 (7.5, 0)Schedule Type=Exponential, k=16, Thresholds (τ_high, τ_low)=(7.5, 0)2025.12 | 73.8 | 5 | |
| Exp-k=16 (7.5, 2.5)Schedule Type=Exponential, k=16, Thresholds (τ_high, τ_low)=(7.5, 2.5)2025.12 | 72.61 | 2.1 | |
| ProphetSchedule=Prophet2025.12 | 66.54 | 1.02 |