Downstream Evaluation on SEA-HELM
54.1IndonesianQwen3-4B
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Qwen3-4BRole=Teacher2026.06 | 54.1 | 52.7 | 50.7 | 43.1 | 45.9 | 53 | 20 | 45.6 | |
| Qwen2.5-3B-InstructRole=Teacher2026.06 | 42 | 38.7 | 32.3 | 9.8 | 24.6 | 40 | 6.5 | 27.7 | |
| Llama3.2-3B-InstructRole=Teacher2026.06 | 32.4 | 30.3 | 38.2 | 16.7 | 28.3 | 42.8 | 5 | 27.7 | |
| GKDStudent Backbone=Qwen2.5-1.5B, Teacher Model=Qwen2.5-3B-Instruct2026.06 | 28.6 | 28.8 | 19.9 | 7.3 | 15.4 | 29.1 | 4.7 | 19.1 | |
| DuDiStudent Backbone=Qwen2.5-1.5B, Teacher Model=Qwen2.5-3B-Instruct2026.06 | 27.9 | 30.3 | 19.8 | 8.5 | 19 | 30.7 | 4.8 | 20.1 | |
| DuDiStudent Backbone=Qwen3-0.6B-Base, Teacher Model=Qwen3-4B2026.06 | 24.2 | 30.4 | 23.4 | 13.2 | 17.6 | 28.1 | 8.4 | 20.8 | |
| DFTStudent Backbone=Qwen2.5-1.5B, Teacher Model=Qwen2.5-3B-Instruct2026.06 | 22.1 | 17.8 | 16.6 | 9.6 | 13.6 | 22.8 | 6.2 | 15.6 | |
| SFTStudent Backbone=Qwen2.5-1.5B, Teacher Model=Qwen2.5-3B-Instruct2026.06 | 21.4 | 26 | 23.8 | 10 | 16.6 | 26.1 | 6.2 | 18.6 | |
| SPINStudent Backbone=Qwen2.5-1.5B, Teacher Model=Qwen2.5-3B-Instruct2026.06 | 21 | 25.3 | 20.2 | 10.1 | 19.2 | 26.7 | 5.3 | 18.3 | |
| GKDStudent Backbone=Qwen3-0.6B-Base, Teacher Model=Qwen3-4B2026.06 | 20.6 | 27 | 21.9 | 10.9 | 15.1 | 23.9 | 6.5 | 18 | |
| DFTStudent Backbone=Qwen3-0.6B-Base, Teacher Model=Qwen3-4B2026.06 | 18.4 | 18.1 | 17.3 | 14 | 14 | 20.1 | 7.1 | 15.6 | |
| SFTStudent Backbone=Qwen3-0.6B-Base, Teacher Model=Qwen3-4B2026.06 | 15.6 | 19.5 | 17.3 | 10.1 | 10.7 | 17.9 | 6.4 | 13.9 | |
| SPINStudent Backbone=Qwen3-0.6B-Base, Teacher Model=Qwen3-4B2026.06 | 14.4 | 20.5 | 17.7 | 10.4 | 14 | 16.1 | 6.8 | 14.3 | |
| DuDiStudent Backbone=Llama3.2-1B, Teacher Model=Llama3.2-3B-Instruct2026.06 | 14.4 | 16.5 | 10.4 | 6.6 | 8.1 | 18.5 | 5.2 | 11.4 | |
| GKDStudent Backbone=Llama3.2-1B, Teacher Model=Llama3.2-3B-Instruct2026.06 | 11.8 | 17.7 | 11.4 | 5.6 | 7.3 | 16.3 | 4.2 | 10.6 | |
| SFTStudent Backbone=Llama3.2-1B, Teacher Model=Llama3.2-3B-Instruct2026.06 | 4.7 | 5.3 | 3.6 | 3.9 | 3.3 | 6.5 | 5.7 | 4.7 | |
| SPINStudent Backbone=Llama3.2-1B, Teacher Model=Llama3.2-3B-Instruct2026.06 | 2.9 | 3.6 | 3.8 | 3.1 | 2.9 | 7.4 | 2.4 | 3.7 | |
| DFTStudent Backbone=Llama3.2-1B, Teacher Model=Llama3.2-3B-Instruct2026.06 | 0.4 | 0.4 | 0.9 | 0.4 | 0.3 | 1.1 | 0.4 | 0.6 |