3PO family methods trained on DapoMath-17k using Olmo3-IVON-SFT-7B and Qwen2.5Math-IVON-SFT-7B