国内首个！移动端视觉生成大模型长沙上线

5月21日上午，中国首个基于国产算力预训练的视觉基座大模型——“橘洲”V1端侧版在长沙上线。该模型由湖南汇视威智能科技有限公司自主研发，依托中科曙光算力，完成了近4000万张图片训练，是全国首个在国产算力上完成整体训练和推理过程并实现移动端部署的视觉基座大模型，具有成本低、质量高、速度快、轻量级、可离线等特点，创新性地解决了视觉大模型推理成本高、数据隐私缺少保障、带宽和时延要求严苛等痛点。该模型的研发团队成员平均年龄不到30岁，部分还是在读博士。

On the morning of May 21, China’s first visual foundation large model pre-trained on domestic computing power—the "Juzhou" V1 edge version—was officially launched in Changsha. Developed independently by Hunan Huishiwei Intelligent Technology Co., Ltd. and relying on computing power from Sugon, the model completed training on nearly 40 million images. It is China’s first visual foundation large model to complete the entire training and inference process on domestic computing power and achieve mobile deployment. Featuringlow cost, high quality, high speed, lightweight design, and offline capability, it innovatively addresses pain points in visual large models such as high inference costs, insufficient data privacy protection, and strict bandwidth and latency requirements. The R&D team has an average age of under 30, with some members still doctoral students.