1 paper · 1 filter
Xinyi Wang, Shawn Tan, Shenbo Xu +4
Reasoning is a core capability of language models (LMs), yet it remains unclear how much model capacity is necessary to support reasoning during pretraining. In this work, we study…