Mathematical Reasoning on MATH (Held-out Test)
48.92AccuracyIDPR
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| IDPRSlow Rate=8.20%, Avg. Tokens=519.422026.06 | 48.92 | 1.02 | 47.31 | 27.07 | |
| Low Avg LogProb Same-kSlow Rate=8.20%, Avg. Tokens=582.682026.06 | 48.22 | 0.32 | 46.46 | 21.22 | |
| Always-FastSlow Rate=0.00%, Avg. Tokens=5.262026.06 | 47.9 | — | 47.89 | — | |
| Low Logit Gap Same-kSlow Rate=8.20%, Avg. Tokens=714.732026.06 | 47.84 | -0.06 | 45.71 | 11.46 | |
| Random Same-kSlow Rate=8.20%, Avg. Tokens=517.912026.06 | 46.76 | -1.14 | 45.17 | 13.41 |