List your company

Multi-Task Skill Learning Suite (ScienceQA, MMLU, Hellaswag, Humaneval, TruthfulQA, Winogrande, IFeval)

70.2ScienceQASDFT
30.57640.86351.1561.437Jan 27, 2026

Evaluation results

MethodLinks
SDFT2026.0170.260.968.966.870.746.573.164.5
SFT2026.0166.25554.835.364.636.873.753.4
SFT + re-invoke2026.016661.663.452.968.745.27060.2
DFT2026.0154.857.66760.469.438.868.260.2
BaseBackbone=Qwen2.5-7B2026.0132.16265.874.371.747.971.165.5