Mid-training is increasingly recognized as a critical stage for shaping the capabilities of large language models. Recent work has shown that targeted mid-tr…
机构:华盛顿大学
来源:arXiv 2608.20314 | AI4Papers 论文推荐平台