OpenBMB Releases Two UltraData Open Datasets, Tops HuggingFace Trending
OpenBMB发布UltraData两大开源数据集,登顶HuggingFace趋势榜
OpenBMB, Tsinghua NLP, and Modelbest released two UltraData open datasets: Ultra-FineWeb-L3 contains 600B+ tokens, including 400B+ English and 200B+ Chinese tokens, while UltraData-SFT-2605 contains 15M+ SFT samples with thinking and non-thinking labels.
Why it matters: OpenBMB with Tsinghua NLP and Modelbest released Ultra-FineWeb-L3 and a second dataset, including over 200B tokens of synthetic Chinese pretraining data ready for pretraining and instruction tuning.