1 paper · 1 filter
Chengying Tu, Hengyu Zhao, Shuaikang Xue +13
Code generation has become a core capability of large language models (LLMs), and code data is central to the pre-training process that develops it. As obtaining public code of suf…