Multi-step Reasoning on Bamboogle auto-eval (test)
76.1Mean AccuracySelf-improvement, 2nd gen
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Self-improvement, 2nd genModel Size=L2023.12 | 76.1 | 1.3 | |
| Self-improvement, 1st genModel Size=L2023.12 | 74 | 3.3 | |
| Pilot, human filteredModel Size=L2023.12 | 71.5 | 2.2 | |
| Pre-trainedModel Size=L2023.12 | 70.3 | 3.5 | |
| Self-improvement, 2nd genModel Size=S2023.12 | 69.7 | 1.3 | |
| Self-improvement, 2nd genModel Size=XS2023.12 | 65.9 | 2.6 | |
| Self-improvement, 1st genModel Size=S2023.12 | 61.9 | 1.9 | |
| Pilot, human filteredModel Size=S2023.12 | 56.6 | 3.8 | |
| Self-improvement, 1st genModel Size=XS2023.12 | 54.4 | 3.6 | |
| Pilot, human filteredModel Size=XS2023.12 | 44.7 | 3.1 |