智能AI None 离线推理训练的权重空间几何 arXiv:2606.23740v1 Announce Type: new Abstract: Offline reinforcement-learning losses (RFT, RIFT, DFT, Offline GRPO, DPO and the DPO 2026-06-24 Aleksandr Nikolich, Igor Kiselev, Vladimir Platonov, Karina Romanova