OpenBMB releases two UltraData open-source datasets, topping Hugging Face trending
Original title:🏆 Big news! UltraData just hit #1 AND #2 on HuggingFace Trending worldwide! 🎉 Released by OpenBMB ...
AI summary
OpenBMB, together with Tsinghua NLP and Modelbest, released two open-source datasets: Ultra-FineWeb-L3 (pretraining synthetic data) with 600B+ tokens (over 400B English and 200B+ Chinese), the largest open-source Chinese pretraining synthetic dataset to date; and UltraData-SFT-2605 (post-training SFT data) with 15M+ samples, China's first open-source large-scale SFT dataset with both thinking and non-thinking annotations, covering math, code, knowledge and instruction following. Both are built on the UltraData L0-L4 framework and validated in MiniCPM5-1B training. The datasets are freely available on Hugging Face.
Key Takeaways
- OpenBMB,together with Tsinghua NLP and Modelbest,released two open-source datasets:…
- and UltraData-SFT-2605 (post-training SFT data) with 15M+ samples,China's first open-source…
- Both are built on the UltraData L0-L4 framework and validated in MiniCPM5-1B training.
- The datasets are freely available on Hugging Face.
Quick Facts
- Published
- Updated
- Source
- X:面壁智能 OpenBMB (@OpenBMB)
- Category
- Products & Apps
- AI score
- 78
面壁开源了两个王炸数据集,预训练的 600B+ token 中文合成数据史上最大,SFT 那边 1500 万条带思考链的指令更是头一回见,做中文基础模型的可以无脑下载了。
Related entities
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: OpenBMB releases two UltraData open-source datasets, topping Hugging Face trending. June 1, 2026. https://new.nat6.net/en/news/cmpv8b6wa04jvsl0zc4newlns/