Mathematical Reasoning on Swahili and Telugu Math (High Difficulty)
25.6Accuracy (Swahili)Loop-Retry
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Loop-RetryTest-time control method=Loop-Retry, Backbone Model=Qwen3-4B2026.05 | 25.6 | 6.4 | 16 | 3.712 | 54.8 | 2.14 | |
| Formula-RetryTest-time control method=Formula-Retry, Backbone Model=Qwen3-4B2026.05 | 8 | 4 | 6 | 3.735 | 50.8 | 2.16 | |
| BaseTest-time control method=None, Backbone Model=Qwen3-4B2026.05 | 4.8 | 3.2 | 4 | 3.607 | 0 | 1 |