Reinforcement Mid-Training

Yijun Tian, Shaoyu Chen, Zhichao Xu +4

The development of state-of-the-art large language models is commonly understood as a two-stage process involving pre-training and post-training. We point out the need for an addit…