Papers & Research

ModelBest, Tsinghua and OpenBMB open-source the largest Chinese pretraining synthetic dataset and a ten-million-level SFT dataset, disclosing core data for MiniCPM5-1B

Original title:国内首次!面壁智能开源千万级 SFT 与最大中文合成数据,MiniCPM5‑1B 核心数据公开

AI summary

ModelBest, together with Tsinghua University and OpenBMB, released and open-sourced two major datasets: Ultra-FineWeb-L3 (over 600B Tokens, including 200B+ Chinese, currently the largest Chinese pretraining synthetic dataset) and UltraData-SFT-2605 (China's first ten-million-level SFT dataset containing both deep-thinking and non-thinking annotations). Both are built on the UltraData data tiering governance system and fully validated in the MiniCPM5-1B training pipeline, covering the full chain from pretraining annealing to post-training SFT. They are now available for free on the UltraData website and HuggingFace.

Key Takeaways

  • ModelBest,together with Tsinghua University and OpenBMB,released and open-sourced two major…
  • Both are built on the UltraData data tiering governance system and fully validated in the…
  • They are now available for free on the UltraData website and HuggingFace.

Quick Facts

Published
Updated
Source
公众号:面壁智能(MiniCPM)
Category
Papers & Research
AI score
61
Read the original Read the full article at the original source
Why it matters

填补了中文大规模合成数据空白,三年前还在用英文数据做中文模型的日子可以翻篇了,做端侧模型的可以直接拿这份数据跑一版 MiniCPM5-1B 级别的效果。

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.

Original article · AIHOT entry

How to Cite

When citing this article, keep the title, date, and permalink, and credit Orange News as the source.

Orange News: ModelBest, Tsinghua and OpenBMB open-source the largest Chinese pretraining synthetic dataset and a ten-million-level SFT dataset, disclosing core data for MiniCPM5-1B. May 29, 2026. https://new.nat6.net/en/news/cmq2o64uw01fbsl6ns33miceg/

View all →