作者位次 #4/7ICLR 2026Poster· Deep Learning / Generative Models and Autoencoders
Quagmires in SFT-RL Post-Training: When High SFT Scores Mislead and What to Use Instead
Feiyang Kang · Michael Kuchnik · Karthik Padthe · Marin Vlastelica · Ruoxi Jia · Carole-Jean Wu · Newsha Ardalani
强化学习与后训练OpenReview