Full training codebase released for open-source multimodal model SenseNova-U1
Original title:🚀 𝗪𝗲'𝘃𝗲 𝗼𝗽𝗲𝗻-𝘀𝗼𝘂𝗿𝗰𝗲𝗱 𝘁𝗵𝗲 𝗳𝘂𝗹𝗹 𝘁𝗿𝗮𝗶𝗻𝗶𝗻𝗴 𝗰𝗼𝗱𝗲𝗯𝗮𝘀𝗲 𝗳𝗼𝗿 𝗦𝗲𝗻...
AI summary
SenseTime has open-sourced the complete training codebase for SenseNova-U1 (8B dense + A3B MoE). It is a unified framework that supports training for a range of multimodal tasks, including text-to-image, image editing, interleaved generation, and text and visual understanding. Its design emphasizes practicality and large-scale training, adopting hybrid parallelism, a streaming resumable data pipeline, environment-variable configuration and a decoupled modular design, and supports scaling from 1×8 GPUs to multi-node clusters. The codebase is open-sourced under Apache-2.0.
Key Takeaways
- SenseTime has open-sourced the complete training codebase for SenseNova-U1 (8B dense + A3B MoE).
- It is a unified framework that supports training for a range of multimodal tasks,including…
- Its design emphasizes practicality and large-scale training,adopting hybrid parallelism,a…
- The codebase is open-sourced under Apache-2.
Quick Facts
- Published
- Updated
- Category
- Products & Apps
- AI score
- 77
商汤把 SenseNova-U1 的训练代码全量开源,支持多模态任务和 MoE,还给了完整的并行策略,做多模态训练的可以直接 fork 过去用,Apache-2.0 很友好。
Source & copyright
This article is based on curated titles and AI summaries from the AIHOT public API, with attribution and links back. Original copyright remains with the original publishers and authors; this site does not mirror full texts.
How to Cite
When citing this article, keep the title, date, and permalink, and credit Orange News as the source.
Orange News: Full training codebase released for open-source multimodal model SenseNova-U1. May 26, 2026. https://new.nat6.net/en/news/cmpmsj91h0rg1sl01u19v2kyh/