Small language models are often post-trained as students on reasoning traces from stronger teacher models to efficiently learn new skills. However, token-lev…
机构:EPFL
来源:arXiv 2610.06671 | AI4Papers 论文推荐平台