Solar Open技术报告
计算与语言
2026-01-13 v1
摘要
我们介绍Solar Open，这是一个1020亿参数的双语Mixture-of-Experts语言模型，专为支持资源匮乏语言而设计。Solar Open系统性地阐述了构建竞争力LLM的方法，聚焦三个相互关联的挑战：首先，针对资源匮乏语言的数据稀缺问题，我们合成4.5万亿个高质量、领域特定且以RL为导向的数据；其次，围绕2万万亿个token的数量，我们通过渐进式课程协调数据，优化组成、质量阈值与领域覆盖；第三，为实现可扩展的推理能力，我们应用提出的SnapPO框架实现高效优化。在英语和韩语基准测试中，Solar Open实现了有竞争力的性能，展示了该方法在支持语言AI开发方面的有效性。
引用
@article{arxiv.2601.07022,
title = {Solar Open Technical Report},
author = {Sungrae Park and Sanghoon Kim and Jungho Cho and Gyoungjin Gim and Dawoon Jung and Mikyoung Cha and Eunhae Choo and Taekgyu Hong and Minbyul Jeong and SeHwan Joo and Minsoo Khang and Eunwon Kim and Minjeong Kim and Sujeong Kim and Yunsu Kim and Hyeonju Lee and Seunghyun Lee and Sukyung Lee and Siyoung Park and Gyungin Shin and Inseo Song and Wonho Song and Seonghoon Yang and Seungyoun Yi and Sanghoon Yoon and Jeonghyun Ko and Seyoung Song and Keunwoo Choi and Hwalsuk Lee and Sunghun Kim and Du-Seong Chang and Kyunghyun Cho and Junsuk Choe and Hwaran Lee and Jae-Gil Lee and KyungTae Lim and Alice Oh},
journal= {arXiv preprint arXiv:2601.07022},
year = {2026}
}