Model Merging on Qwen3-4B-Base Transfer 8 benchmarks
32.65Math AccuracyPico
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| PicoBackbone=Qwen3-4B-Base2026.04 | 32.65 | 58.25 | 65.45 | 65.94 | 55.57 | |
| KnOTSBackbone=Qwen3-4B-Base2026.04 | 26.22 | 55.93 | 61.13 | 56.61 | 49.97 | |
| DAREBackbone=Qwen3-4B-Base2026.04 | 26.07 | 57.18 | 61.67 | 55.39 | 50.08 | |
| DELLABackbone=Qwen3-4B-Base2026.04 | 23.26 | 57.37 | 59.96 | 55.74 | 49.08 | |
| No CalibrationBackbone=Qwen3-4B-Base2026.04 | 19.52 | 57.98 | 61.44 | 56.32 | 48.81 | |
| Core-TABackbone=Qwen3-4B-Base2026.04 | 12.64 | 55.63 | 55.15 | 51.87 | 43.82 |