Scientific Reasoning on SciBench (Accuracy, Friedman Rank, P(EDEN > competitor))
7.4AccuracyEDEN
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| EDENtotal number of branches explored=10182026.05 | 7.4 | 1.25 | 75 | |
| Beam Searchbeam width=3, total number of branches explored=12002026.05 | 6.9 | 2.25 | 78 | |
| Diverse Beam Searchbeam width=3, total number of branches explored=12002026.05 | 5.4 | 4.75 | 96 | |
| Best-of-nn=52026.05 | 5.2 | 4.62 | 96 | |
| Greedy2026.05 | 4.9 | 6.12 | 99 | |
| Top-k Sampling2026.05 | 4.9 | 7 | 100 | |
| Top-p Sampling2026.05 | 4.8 | 6.62 | 99 | |
| Top-H Sampling2026.05 | 4.5 | 8.12 | 100 | |
| Min-p Sampling2026.05 | 4.3 | 8 | 100 | |
| Majority-nn=52026.05 | 4.1 | 6.25 | 99 |