商汤与NTU推出NEO-unify：原生统一多模态模型新范式
原标题：NEO-unify: Building Native Multimodal Unified Models End to End sensenova • Mar 5 • 174
AI 摘要
商汤科技与南洋理工大学联合推出NEO-unify（预览版），这是一种原生、统一、端到端的多模态模型范式，无需预训练编码器或VAE。该模型通过自建表示空间直接学习近无损输入，结合Mixture-of-Transformer架构和统一的文本与视觉生成学习。初步实验显示，NEO-unify（2B）在图像重建和编辑任务上表现良好，并展现出更好的数据扩展效率。团队计划未来开源模型。
正文节选
Today, SenseTime, in collaboration with NTU, introduces a native, unified, end-to-end paradigm dubbed NEO-unify (preview) — stepping beyond representation arguments, and breaking free from pre-trained priors or scaling-law bottlenecks. No VE! No VAE! NEO-unify is the first step toward truly end-to-end unified models, learning directly from near-lossless inputs via a representation space shaped by the model itself. NEO-unify incorporates: 1) near-lossless visual interface for input and output, 2)