Skill Learning on Medical and Multi-Task Suite (Hellaswag, Humaneval, IFeval, MMLU, TruthfulQA, Winogrande)
40.2Medical ScoreSDFT
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| SDFT2026.01 | 40.2 | 61.4 | 67.7 | 72.3 | 71.5 | 47.3 | 71.9 | 65.4 | |
| DFT2026.01 | 36.2 | 61.9 | 64.6 | 74.6 | 71.6 | 40.1 | 71.3 | 64 | |
| SFT + re-invoke2026.01 | 35.6 | 61.5 | 63.1 | 67.6 | 70 | 42.3 | 71.4 | 62.6 | |
| SFT2026.01 | 35.5 | 59.5 | 62.1 | 56.6 | 70.5 | 39.8 | 72.9 | 60.2 | |
| BaseBackbone=Qwen2.5-7B2026.01 | 30.1 | 62 | 65.8 | 74.3 | 71.7 | 47.9 | 71.1 | 65.5 |