Supervised Fine-Tuning on MMLU Danish (test)
46.24MMLU ScoreTV + KMM
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TV + KMMk=10, Pre-trained model=Qwen3-1.7b2026.05 | 46.24 | 20 | |
| One Step + KMMk=5, Pre-trained model=Qwen3-1.7b2026.05 | 46.21 | 29 | |
| DataModel-CSk=5, Pre-trained model=Qwen3-1.7b2026.05 | 46.19 | 27.5 | |
| GradEX-REk=15, Pre-trained model=Qwen3-1.7b2026.05 | 46.18 | 25 | |
| One Step + KMMk=15, Pre-trained model=Qwen3-1.7b2026.05 | 46.15 | 29 | |
| TVk=10, Pre-trained model=Qwen3-1.7b2026.05 | 46.15 | 17 | |
| Randomk=20, Pre-trained model=Qwen3-1.7b2026.05 | 46.15 | 12.5 | |
| One Stepk=15, Pre-trained model=Qwen3-1.7b2026.05 | 46.11 | 17.5 | |
| GradEX-FSk=5, Pre-trained model=Qwen3-1.7b2026.05 | 46.11 | 24.5 | |
| One Step + KMMk=10, Pre-trained model=Qwen3-1.7b2026.05 | 46.1 | 29 | |
| TV + KMMk=15, Pre-trained model=Qwen3-1.7b2026.05 | 46.1 | 20 | |
| One Step + KMMk=25, Pre-trained model=Qwen3-1.7b2026.05 | 46.08 | 29 | |
| TVk=15, Pre-trained model=Qwen3-1.7b2026.05 | 46.08 | 17 | |
| One Stepk=20, Pre-trained model=Qwen3-1.7b2026.05 | 46.06 | 17.5 | |
| One Stepk=5, Pre-trained model=Qwen3-1.7b2026.05 | 46.05 | 17.5 | |
| One Stepk=10, Pre-trained model=Qwen3-1.7b2026.05 | 46.05 | 17.5 | |
| TVk=5, Pre-trained model=Qwen3-1.7b2026.05 | 46.05 | 17 | |
| TVk=20, Pre-trained model=Qwen3-1.7b2026.05 | 46.05 | 17 | |
| TV + KMMk=5, Pre-trained model=Qwen3-1.7b2026.05 | 46.05 | 20 | |
| TV + KMMk=20, Pre-trained model=Qwen3-1.7b2026.05 | 46.05 | 20 | |
| One Step + KMMk=20, Pre-trained model=Qwen3-1.7b2026.05 | 46.02 | 29 | |
| DataModelk=5, Pre-trained model=Qwen3-1.7b2026.05 | 45.99 | 7 | |
| One Stepk=25, Pre-trained model=Qwen3-1.7b2026.05 | 45.98 | 17.5 | |
| TVk=25, Pre-trained model=Qwen3-1.7b2026.05 | 45.98 | 17 | |
| TV + KMMk=25, Pre-trained model=Qwen3-1.7b2026.05 | 45.98 | 20 |