Micius-Flash：面向工程的轻量级智能体模型
原标题：Micius-Flash: Lightweight Agentic Intelligence for Engineering
AI 摘要
研究团队发布 Micius-Flash，一个 270 亿参数、支持原生多模态输入和 256K 上下文窗口的工程领域模型。团队构建知识树驱动的数据飞轮，训练半导体知识、硬件编码、多模态理解及智能体专家模型，再用多教师同策略蒸馏（MOPD）整合为统一模型。在 20 多个数据集上，硬件代码基准平均 64.2%（超同规模基线 15 分），半导体智能体基准 62.4%（超 17.2 分）。团队还开发 Micius Harness 连接 EDA、KiCad 等工具，并以开源形式发布模型。
正文节选
Lightweight Agentic Intelligence for Engineering Abstract We introduce Micius-Flash, a 27-billion-parameter model for engineering with native multimodal input, deep reasoning and a 256K-token context window. We design the Micius data pipeline as a knowledge-tree-driven data flywheel comprising three stages: knowledge tree construction and expansion, data synthesis, and quality validation with hard-example augmentation. From this data, we train expert models for semiconductor knowledge and reason