12:09
官方账号arXiv cs.LG@Hanqing Zhu, Wenyan Cong, Zhizhou Sha, Sagnik Mukherjee, Xinyuan Song, David González-Martínez, Xiaoxia Wu, Yuandong Tian, Shiwei Liu, David Z. Pan, Zhangyang "Atlas" Wang
推荐理由:如果厌倦了RLVR微调慢慢跑,试试ISO:只用1/3的训练步数就能达到同等效果,Qwen3-8B上实测有效,代码开源。