🏆 Swept the top two spots! UltraData tops Hugging Face Trending
🚀 Great news!
The two UltraData datasets open-sourced by ModelBest × Tsinghua University × OpenBMB
have taken the top two spots on Hugging Face Trending worldwide! 🎉
🔥 Ultra-FineWeb-L3 (synthetic web pre-training data)
· Over 600B tokens in total (400B+ English, 200B+ Chinese)
· Currently the largest open-source Chinese pre-training synthetic dataset
· Focused on improving the "learnability" of each unit of data
🔥 UltraData-SFT-2605 (post-training SFT data)
· China's first open-source SFT dataset at the ten-million scale, with both "deep thinking" and "non-thinking" annotations
· Covers math, code, knowledge, instruction following, and other domains
· The entire data processing pipeline is traceable
🧱 Both are built on the UltraData L0–L4 five-level tiered data governance system and have been validated across the full MiniCPM5-1B training pipeline.
👉 Now open for free — developers worldwide are welcome to download and try them!