Reading Comprehension on Belebele c
37.11Accuracy (Normalized)Q3 negatives
Evaluation Results
| Method | Links | |
|---|---|---|
| Q3 negativesFiltering Strategy=Q3 negatives2026.04 | 37.11 | |
| HQ Romance (No Fra)Filtering Strategy=HQ Romance (No Fra)2026.04 | 36.89 | |
| ML (Q3)Filtering Strategy=ML (Q3)2026.04 | 36.33 | |
| ML (15%)Filtering Strategy=ML (15%)2026.04 | 36.11 | |
| HQ allFiltering Strategy=HQ all2026.04 | 35.44 | |
| MKC-eFiltering Strategy=MKC-e2026.04 | 35.33 | |
| HQFiltering Strategy=HQ2026.04 | 35.11 | |
| MLFiltering Strategy=ML2026.04 | 35.11 | |
| HQ seedFiltering Strategy=HQ seed2026.04 | 35 | |
| ML (Q3)Filtering Strategy=ML (Q3)2026.04 | 34.78 | |
| ML (Q3)Sampling Strategy=Q32026.04 | 34.44 | |
| No FilteringFiltering Strategy=No Filtering2026.04 | 34.44 | |
| MKC-eFiltering Strategy=MKC-e2026.04 | 34.33 | |
| NordicFiltering Strategy=Nordic2026.04 | 34.22 | |
| HQ seed (LLM Training)Filtering Strategy=HQ seed (LLM Training)2026.04 | 33.89 | |
| HQ (10%)Retention=10%2026.04 | 33.78 | |
| ML (15%)Filtering Strategy=ML (15%)2026.04 | 33.56 | |
| HQ seedVariant=Seed2026.04 | 32.89 | |
| ML (10%)Retention=10%2026.04 | 32.78 | |
| ML (20%)Retention=20%2026.04 | 32.56 | |
| HQ seedFiltering Strategy=HQ seed2026.04 | 32.56 | |
| MKC-e2026.04 | 32.44 | |
| HQ2026.04 | 32.33 | |
| No Filtering2026.04 | 32.22 | |
| MLFiltering Strategy=ML2026.04 | 32.22 | |
| 1B Model (Multilingual classifier)Filtering Strategy=Multilingual (ML), Model Scale=1B2026.04 | 32.22 | |
| HQFiltering Strategy=HQ2026.04 | 32 | |
| 1B Model (HQ filtering)Filtering Strategy=Monolingual high-quality (HQ), Model Scale=1B2026.04 | 32 | |
| ML2026.04 | 31.22 | |
| No filteringFiltering Strategy=No filtering2026.04 | 30.11 | |
| 1B Model (No filtering)Filtering Strategy=No filtering, Model Scale=1B2026.04 | 30.11 | |
| HQ (20%)Retention=20%2026.04 | 29.44 |